Main
Main
Main
Ulisse Mini
Jesse Li
2
Preface
Huge thanks to the math discord for answering my questions! I don’t know how I’d manage
without them ♡
If you don’t find your exercise here check linearalgebras.com or (god forbid) chegg.
i
ii
Contents
3 Basic Topology of R 69
3.2 Open and Closed Sets . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 69
3.3 Compact Sets . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 76
3.4 Perfect Sets and Connected Sets . . . . . . . . . . . . . . . . . . . . . . . . . 82
3.5 Baire’s Theorem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 87
iii
iv CONTENTS
Solution
(a) Suppose for contradiction that p/q is a fraction in lowest terms, and that (p/q)2 = 3.
Then p2 = 3q 2 implying p is a multiple of 3 since 3 is not a perfect square. Therefore
we can write p as 3r for some r, substituting we get (3r)2 = 3q 2 and 3r2 = q 2 implying
q is √
also a multiple of 3 contradicting the assumption that p/q is in lowest terms.
For 6 the same argument applies, since 6 is not a perfect square.
(b) 4 is a perfect square, meaning p2 = 4q 2 does not imply that p is a multiple of four as
p could be 2.
Exercise 1.2.2
Show that there is no rational number satisfying 2r = 3
Solution
If r = 0 clearly 2r = 1 ̸= 3, if r ̸= 0 set p/q = r to get 2p = 3q which is impossible since 2
and 3 share no factors.
Exercise 1.2.3
Decide which of the following represent true statements about the nature of sets. For any
that are false, provide a specific example where the statement in question does not hold.
(a) If A1 ⊇ A2 ⊇ A3T⊇ A4 · · · are all sets containing an infinite number of elements, then
the intersection ∞
n=1 An is infinite as well.
(b) If A1 ⊇ A2 T⊇ A3 ⊇ A4 · · · are all finite, nonempty sets of real numbers, then the
intersection ∞
n=1 An is finite and nonempty.
1
2 CHAPTER 1. THE REAL NUMBERS
(c) A ∩ (B ∪ C) = (A ∩ B) ∪ C.
(d) A ∩ (B ∩ C) = (A ∩ B) ∩ C.
(e) A ∩ (B ∪ C) = (A ∩ B) ∪ (A ∩ C).
Solution
(a) False, consider A1 = {1, 2, . . . }, A2 = {2, 3, . . . }, ... has ∞
T
n=1 An = ∅.
Exercise 1.2.4
Produce an infinite collection of sets A1 , A2 , A3 , . . . with the
S property that every Ai has an
infinite number of elements, Ai ∩ Aj = ∅ for all i ̸= j, and ∞ i=1 Ai = N
Solution
This question is asking us to partition N into an infinite collection of sets. This is equivalent
to asking us to unroll N into a square, which we can do along the diagonal
1 3 6 10 15 ···
2 5 9 14 ···
4 8 13 ···
7 12 · · ·
11 . . .
..
.
(b) Prove the reverse inclusion (A ∩ B)c ⊇ Ac ∪ B c , and conclude that (A ∩ B)c = Ac ∪ B c
Solution
(a) If x ∈ (A ∩ B)c then x ∈
/ A ∩ B so x ∈ / B implying x ∈ Ac or x ∈ B c which is
/ A or x ∈
c c
the same as x ∈ A ∪ B .
Exercise 1.2.6
(a) Verify the triangle inequality in the special case where a and b have the same sign.
(b) Find an efficient proof for all the cases at once by first demonstrating (a + b)2 ≤
(|a| + |b|)2
(d) Prove ∥a| − |b|| ≤ |a − b|. (The unremarkable identity a = a − b + b may be useful.)
Solution
(a) We have equality |a + b| = |a| + |b| meaning |a + b| ≤ |a| + |b| also holds.
(b) (a + b)2 ≤ (|a| + |b|)2 reduces to 2ab ≤ 2|a||b| which is true as the left side can be
negative but the right side can’t. and since squaring preserves inequality this implies
|a + b| ≤ |a| + |b|.
(c) I would like to do this using the triangle inequality, I notice that (a−c)+(c−d)+(d−b) =
a − b. Meaning I can use the triangle inequality for multiple terms
|a − b| = |(a − c) + (c − d) + (d − b)| ≤ |a − c| + |c − d| + |d − b|
The general triangle inequality is proved by repeated application of the two variable
inequality
|(a + b) + c| ≤ |a + b| + |c| ≤ |a| + |b| + |c|
(d) Since ||a| − |b|| = ||b| − |a|| we can assume |a| > |b| without loss of generality. Then
Exercise 1.2.7
Given a function f and a subset A of its domain, let f (A) represent the range of f over the
set A; that is, f (A) = {f (x) : x ∈ A}.
(a) Let f (x) = x2 . If A = [0, 2] (the closed interval {x ∈ R : 0 ≤ x ≤ 2}) and B = [1, 4],
find f (A) and f (B). Does f (A ∩ B) = f (A) ∩ f (B) in this case? Does f (A ∪ B) =
f (A) ∪ f (B)?
(c) Show that, for an arbitrary function g : R → R, it is always true that g(A ∩ B) ⊆
g(A) ∩ g(B) for all sets A, B ⊆ R
(d) Form and prove a conjecture about the relationship between g(A ∪ B) and g(A) ∪ g(B)
for an arbitrary function g
4 CHAPTER 1. THE REAL NUMBERS
Solution
(a) f (A) = [0, 4], f (B) = [1, 16], f (A ∩ B) = [1, 4] = f (A) ∩ f (B) and f (A ∪ B) = [0, 16] =
f (A) ∪ f (B)
(c) Suppose y ∈ g(A ∩ B), then ∃x ∈ A ∩ B such that g(x) = y. But if x ∈ A ∩ B then
x ∈ A and x ∈ B, meaning y ∈ g(A) and y ∈ g(B) implying y ∈ g(A) ∩ g(B) and thus
g(A ∩ B) ⊆ g(A) ∩ g(B).
Notice why it is possible to have x ∈ g(A) ∩ g(B) but x ∈
/ g(A ∩ B), this happens when
something in A \ B and something in B \ A map to the same thing. If g is 1-1 this
does not happen.
(d) I conjecture that g(A ∪ B) = g(A) ∪ g(B). To prove this we show inclusion both ways,
First suppose y ∈ g(A∪B). then either y ∈ g(A) or y ∈ g(B), implying y ∈ g(A)∪g(B).
Now suppose y ∈ g(A) ∪ g(B) meaning either y ∈ g(A) or y ∈ g(B) which is the same
as y ∈ g(A ∪ B) as above.
Exercise 1.2.8
Here are two important definitions related to a function f : A → B. The function f is one-
to-one (1 − 1) if a1 ̸= a2 in A implies that f (a1 ) ̸= f (a2 ) in B. The function f is onto if,
given any b ∈ B, it is possible to find an element a ∈ A for which f (a) = b Give an example
of each or state that the request is impossible:
Solution
(a) Let f (n) = n + 1 does not have a solution to f (a) = 1
(c) Let f (n) = n/2 for even n, and f (n) = −(n + 1)/2 for odd n.
Exercise 1.2.9
Given a function f : D → R and a subset B ⊆ R, let f −1 (B) be the set of all points from
the domain D that get mapped into B; that is, f −1 (B) = {x ∈ D : f (x) ∈ B}. This set is
called the preimage of B.
(a) Let f (x) = x2 . If A is the closed interval [0, 4] and B is the closed interval [−1, 1],
find f −1 (A) and f −1 (B). Does f −1 (A ∩ B) = f −1 (A) ∩ f −1 (B) in this case? Does
f −1 (A ∪ B) = f −1 (A) ∪ f −1 (B)?
(b) The good behavior of preimages demonstrated in (a) is completely general. Show that
for an arbitrary function g : R → R, it is always true that g −1 (A∩B) = g −1 (A)∩g −1 (B)
and g −1 (A ∪ B) = g −1 (A) ∪ g −1 (B) for all sets A, B ⊆ R
1.2. SOME PRELIMINARIES 5
Solution
(a) f −1 (A) = [−2, 2], f −1 (B) = [−1, 1], f −1 (A ∩ B) = [−1, 1] = f −1 (A) ∩ f −1 (B), f −1 (A ∪
B) = [−2, 2] = f −1 (A) ∪ f −1 (B)
(b) First let x ∈ g −1 (A ∩ B) meaning g(x) ∈ A ∩ B implying g(x) ∈ A and g(x) ∈ B which
is the same as x ∈ g −1 (A) and x ∈ g −1 (B) meaning x ∈ g −1 (A) ∩ g −1 (B).
Second let x ∈ g −1 (A) ∩ g −1 (B), this is the same as x ∈ g −1 (A) and x ∈ g −1 (B) which
is the same as g(x) ∈ A and g(x) ∈ B implying g(x) ∈ A ∩ B and thus x ∈ g −1 (A ∩ B).
Thus g −1 (A ∩ B) = g −1 (A) ∩ g −1 (B).
Seeing g −1 (A ∪ B) = g −1 (A) ∪ g −1 (B) is obvious. see 1.2.7 (d).
Exercise 1.2.10
Decide which of the following are true statements. Provide a short justification for those
that are valid and a counterexample for those that are not:
(a) Two real numbers satisfy a < b if and only if a < b + ϵ for every ϵ > 0.
(b) Two real numbers satisfy a < b if a < b + ϵ for every ϵ > 0.
(c) Two real numbers satisfy a ≤ b if and only if a < b + ϵ for every ϵ > 0.
Solution
(a) False, if a = b then a < b + ϵ for all ϵ > 0 but a ̸< b
(c) True. First suppose a < b + ϵ for all ϵ > 0, We want to show this implies a ≤ b. We
either have a ≤ b or a > b, but a > b is impossible since the gap implies there exists
an ϵ small enough such that a > b + ϵ.
Second suppose a ≤ b, obviously a < b + ϵ for all ϵ > 0.
Exercise 1.2.11
Form the logical negation of each claim. One trivial way to do this is to simply add “It is not
the case that...” in front of each assertion. To make this interesting, fashion the negation
into a positive statement that avoids using the word “not” altogether. In each case, make
an intuitive guess as to whether the claim or its negation is the true statement.
(a) For all real numbers satisfying a < b, there exists an n ∈ N such that a + 1/n < b
(b) There exists a real number x > 0 such that x < 1/n for all n ∈ N.
(c) Between every two distinct real numbers there is a rational number.
Solution
(a) There exist real numbers satisfying a < b where a + 1/n ≥ b for all n ∈ N (false).
(b) For every real number x > 0 there exists an n ∈ N such that x ≥ 1/n (true).
(c) There exist two real numbers a < b such that if r < b then r < a for all r ∈ Q (false).
6 CHAPTER 1. THE REAL NUMBERS
Exercise 1.2.12
Let y1 = 6, and for each n ∈ N define yn+1 = (2yn − 6) /3
(a) Use induction to prove that the sequence satisfies yn > −6 for all n ∈ N.
(b) Use another induction argument to show the sequence (y1 , y2 , y3 , . . .) is decreasing.
Solution
(a) Suppose for induction that yn > −6, our base case clearly satisfies y1 > −6. then
(b) Suppose yn+1 < yn , the base case 2 < 6 works. Now
Exercise 1.2.13
For this exercise, assume Exercise 1.2.5 has been successfully completed.
∞
!c ∞
[ \
Ai = Aci
i=1 i=1
but induction does not apply here. Induction is used to prove that a particular state-
ment holds for every value of n ∈ N, but this does not imply the validity of the infinite
case. To
Tn illustrate this point, find an exampleTof a collection of sets B1 , B2 , B3 , . . .
∞
where i=1 Bi ̸= ∅ is true for every n ∈ N, but i=1 Bi ̸= ∅ fails.
(c) Nevertheless, the infinite version of De Morgan’s Law stated in (b) is a valid statement.
Provide a proof that does not use induction.
1.2. SOME PRELIMINARIES 7
Solution
(a) 1.2.5 Is our base case, Assume (A1 ∪ · · · ∪ An )c = Ac1 ∩ · · · ∩ Acn . We want to show the
n + 1 case. Using associativity we have
(b) Now, state and prove a version of Lemma 1.3.8 for greatest lower bounds.
Solution
(a) We have i = inf A if and only if
(b) Suppose i is a lower bound for A, it is the greatest lower bound if and only if for all
ϵ > 0, there exists an a ∈ A such that i + ϵ > a.
First suppose i = inf A, then for all ϵ > 0, i + ϵ cannot be a lower bound on A because
(ii) implies all lower bounds b obey b ≤ i, therefore there must be some a ∈ A such
that i + ϵ > a.
Second suppose for all ϵ > 0 there exists an a ∈ A such that i + ϵ > a. In other words
i + ϵ is not a lower bound for all ϵ, which is the same as saying every lower bound b
must have b ≤ i implying (ii).
Exercise 1.3.2
Give an example of each of the following, or state that the request is impossible.
(b) A finite set that contains its infimum but not its supremum.
(c) A bounded subset of Q that contains its supremum but not its infimum.
Solution
(a) Let B = {0} we have inf B = 0 and sup B = 0 thus inf B ≥ sup B.
(b) Impossible, finite sets must contain their infimum and supremum.
Exercise 1.3.3
(a) Let A be nonempty and bounded below, and define B = {b ∈ R : b is a lower bound
for A}. Show that sup B = inf A.
(b) Use (a) to explain why there is no need to assert that greatest lower bounds exist as
part of the Axiom of Completeness.
Solution
(a) By definition sup B is the greatest lower bound for A, meaning it equals inf A.
1.3. THE AXIOM OF COMPLETENESS 9
(b) (a) Proves the greatest lower bound exists using the least upper bound.
Exercise 1.3.4
Let A1 , A2 , A3 , . . . be a collection of nonempty sets, each of which is bounded above.
(a) Find a formula for sup (A1 ∪ A2 ). Extend this to sup ( nk=1 Ak ).
S
Solution S
(a) sup ( nk=1 Ak ) = sup {sup Ak | k = 1, . . . , n}
Exercise 1.3.5
As in Example 1.3.7, let A ⊆ R be nonempty and bounded above, and let c ∈ R. This time
define the set cA = {ca : a ∈ A}.
(b) Postulate a similar type of statement for sup(cA) for the case c < 0.
Solution
(a) Assume c > 0 (the c = 0 case is trivial). Let s = c sup A. Suppose ca > s, then
a > sup A which is impossible, meaning s is an upper bound on cA. Now suppose s′
is an upper bound on cA and s′ < s. Then s′ /c < s/c and s′ /c < sup A meaning s′ /c
cannot bound A, so there exists a ∈ A such that s′ /c < a meaning s′ > ca thus s′
cannot be an upper bound on cA, and so s = c sup A is the least upper bound.
Exercise 1.3.6
Given sets A and B, define A + B = {a + b : a ∈ A and b ∈ B}. Follow these steps to prove
that if A and B are nonempty and bounded above then sup(A + B) = sup A + sup B
(b) Now let u be an arbitrary upper bound for A + B, and temporarily fix a ∈ A. Show
t ≤ u − a.
(d) Construct another proof of this same fact using Lemma 1.3.8.
Solution
(a) We have a ≤ s and b ≤ t, adding the equations gives a + b ≤ s + t.
(c) From (a) we know s + t is an upper bound, so we must only show it is the least upper
bound.
Let u = sup(A + B), from (a) we have t ≤ u − a and s ≤ u − b adding and rearranging
gives a + b ≤ 2u − s − t. since 2u − s − t is an upper bound on A + B it is less then
the least upper bound, so u ≤ 2u − s − t implying s + t ≤ u. and since u is the least
upper bound s + t must equal u.
Stepping back, the key to this proof is that a + b ≤ s, ∀a, b implying sup(A + B) ≤ s
can be used to transition from all a + b to a single value sup(A + B), avoiding the
ϵ-hackery I would otherwise use.
(d) Showing s + t − ϵ is not an upper bound for any ϵ > 0 proves it is the least upper
bound by Lemma 1.3.8. Rearranging gives (s − ϵ/2) + (t − ϵ/2) we know there exists
a > (s − ϵ/2) and b > (t − ϵ/2) therefore a + b > s + t − ϵ meaning s + t cannot be
made smaller, and thus is the least upper bound.
Exercise 1.3.7
Prove that if a is an upper bound for A, and if a is also an element of A, then it must be
that a = sup A.
Solution
a is the least upper bound since any smaller bound a′ < a would not bound a.
Exercise 1.3.8
Compute, without proofs, the suprema and infima (if they exist) of the following sets:
(c) {n/(3n + 1) : n ∈ N}
(d) {m/(m + n) : m, n ∈ N}
Solution
(a) sup = 1, inf = 0
Exercise 1.3.9
(a) If sup A < sup B, show that there exists an element b ∈ B that is an upper bound for
A.
(b) Give an example to show that this is not always the case if we only assume sup A ≤
sup B
1.3. THE AXIOM OF COMPLETENESS 11
Solution
(a) By Lemma 1.3.8 we know there exists a b such that (sup B) − ϵ < b for any ϵ > 0, We
set ϵ to be small enough that sup A < (sup B) − ϵ meaning sup A < b for some b, and
thus b is an upper bound on A.
(b) A = {x | x ≤ 1}, B = {x | x < 1} no b ∈ B is an upper bound since 1 ∈ A and 1 > b.
Exercise 1.3.10 (Cut Property)
The Cut Property of the real numbers is the following:
If A and B are nonempty, disjoint sets with A ∪ B = R and a < b for all a ∈ A and
b ∈ B, then there exists c ∈ R such that x ≤ c whenever x ∈ A and x ≥ c whenever x ∈ B.
(a) Use the Axiom of Completeness to prove the Cut Property.
(b) Show that the implication goes the other way; that is, assume R possesses the Cut
Property and let E be a nonempty set that is bounded above. Prove sup E exists.
(c) The punchline of parts (a) and (b) is that the Cut Property could be used in place
of the Axiom of Completeness as the fundamental axiom that distinguishes the real
numbers from the rational numbers. To drive this point home, give a concrete example
showing that the Cut Property is not a valid statement when R is replaced by Q.
Solution
(a) If c = sup A = inf B then a ≤ c ≤ b is obvious. So we must only prove sup A = inf B.
If sup A < inf B then consider c = sup A+inf
2
B
. c > sup A and therefore c ∈
/ A; similarly
c < inf B and therefore c ∈/ B, implying A ∪ B ̸= R. If sup A > inf B then we can
find a such that a > b by subtracting ϵ > 0 and using the least upper/lower bound
facts, similarly to Lemma 1.3.8. Thus sup A must equal inf B since we have shown
both alternatives are impossible.
(b) Let B = {x | e < x, ∀e ∈ E} and let A = B c . Clearly a < b so the cut property
applies. We have a ≤ c ≤ b and must show the two conditions for c = sup E
(i) Since E ⊆ A, a ≤ c implies e ≤ c thus c is an upper bound.
(ii) c ≤ b implies c is the smallest upper bound.
Note: Using (a) here would be wrong, it assumes the axiom of completeness so we
would be making a circular argument.
2 c
√ = {r ∈ Q | r < 2 or r < 0}, B = A does not satisfy the cut property in Q since
(c) A
2∈
/ Q.
(Alternatively, A = {r ∈ Q | r3 < 2} can be used to avoid√ needing to single out
3
negative numbers in defining A, but requires a proof that 2 is irrational - left to the
reader.)
Exercise 1.3.11
Decide if the following statements about suprema and infima are true or false. Give a short
proof for those that are true. For any that are false, supply an example where the claim in
question does not appear to hold.
12 CHAPTER 1. THE REAL NUMBERS
(a) If A and B are nonempty, bounded, and satisfy A ⊆ B, then sup A ≤ sup B.
(b) If sup A < inf B for sets A and B, then there exists a c ∈ R satisfying a < c < b for
all a ∈ A and b ∈ B.
(c) If there exists a c ∈ R satisfying a < c < b for all a ∈ A and b ∈ B, then sup A < inf B.
Solution
(a) True. We know a ≤ sup A and a ≤ sup B since A ⊆ B. since sup A is the least upper
bound on A we have sup A ≤ sup B.
(b) True. Let c = (sup A + inf B)/2, c > sup A implies a < c and c < inf B implies c < b
giving a < c < b as desired.
(c) False. consider A = {x | x < 1}, B = {x | x > 1}, a < 1 < b but sup A ̸< inf B since
1 ̸< 1.
1.4. CONSEQUENCES OF COMPLETENESS 13
(c) Part (a) can be summarized by saying that Q is closed under addition and multipli-
cation. Is I closed under addition and multiplication? Given two irrational numbers s
and t, what can we say about s + t and st?
Solution
(a) Trivial.
Exercise 1.4.2
Let A ⊆ R be nonempty and bounded above, and let s ∈ R have the property that for
all n ∈ N, s + n1 is an upper bound for A and s − n1 is not an upper bound for A. Show
s = sup A.
Solution
This is basically a rephrasing of Lemma 1.3.8 using the archimedean property. The most
straightforward approach is to argue by contradiction:
(i) If s < sup A then there exists an n such that s + 1/n < sup A contradicting sup A
being the least upper bound.
(ii) If s > sup A then there exists an n such that s − 1/n > sup A where s − 1/n is not an
upper bound, contradicting sup A being an upper bound.
Exercise 1.4.3
Prove that ∞
T
n=1 (0, 1/n) = ∅. Notice that this demonstrates that the intervals in the Nested
Interval Property must be closed for the conclusion of the theorem to hold.
Solution
Suppose x ∈ ∞
T
n=1 (0, 1/n), then we have 0 < x < 1/n for all n, which is impossible by the
archimedean property, In other words we can always set n large enough that x lies outside
the interval.
Exercise 1.4.4
Let a < b be real numbers and consider the set T = Q ∩ [a, b]. Show sup T = b
14 CHAPTER 1. THE REAL NUMBERS
Solution
We must show the two conditions for a least upper bound
(i) Clearly t ≤ b for all t ∈ T
(ii) Let a < b′ < b. b′ Cannot be an upper bound for T since the density theorem tells us
we can find r ∈ Q ∩ [a, b] such that b′ < r < b.
Exercise 1.4.5
Using Exercise 1.4.1, supply a proof that I is dense in R by considering the real numbers
√ √
a − 2 and b − 2. In other words show for every two real numbers a < b there exists an
irrational number t with a < t < b.
Solution √ √
The density theorem lets us find a rational number r with a − 2 < r < b − 2, adding
√ √ √
2 to both
√ sides gives a < r + 2 < b. From 1.4.1 we know r + 2 is irrational, so setting
t = r + 2 gives a < t < b as desired.
Exercise 1.4.6
Recall that a set B is dense in R if an element of B can be found between any two real
numbers a < b. Which of the following sets are dense in R ? Take p ∈ Z and q ∈ N in every
case.
(a) The set of all rational numbers p/q with q ≤ 10.
(b) The set of all rational numbers p/q with q a power of 2 .
(c) The set of all rational numbers p/q with 10|p| ≥ q.
Solution
(a) Not dense since we cannot make |p|/q smaller then 1/10.
(b) Dense.
(c) Not dense since we cannot make |p|/q smaller then 1/10.
Exercise 1.4.7
Finish the proof of Theorem 1.4.5 by showing that the assumption α2 > 2 leads to a contra-
diction of the fact that α = sup T
Solution
Recall T = {t ∈ R | t2 < 2} and α = sup T . suppose α2 > 2, we will show there exists
an n ∈ N such that (α − 1/n)2 > 2 contradicting the assumption that α is the least upper
bound.
We expand (α − 1/n)2 to find n such that (α2 − 1/n) > 2
2α 1 1 − 2α
2 < (α − 1/n)2 = α2 − + 2 < α2 +
n n n
Then
1 − 2α
2 < α2 + =⇒ n(2 − α2 ) < 1 − 2α
n
1.4. CONSEQUENCES OF COMPLETENESS 15
Exercise 1.4.8
Give an example of each or state that the request is impossible. When a request is impossible,
provide a compelling argument for why this is the case.
(a) Two sets A and B with A ∩ B = ∅, sup A = sup B, sup A ∈ / A and sup B ∈/ B.
(d) A sequence of T
closed bounded (not necessarilyTnested) intervals I1 , I2 , I3 , . . . with the
∞
property that N n=1 In ̸= ∅ for all N ∈ N, but n=1 In = ∅.
Solution
(a) A = Q ∩ (0, 1), B = I ∩ (0, 1). A ∩ B = ∅, sup A = sup B = 1 and 1 ∈
/ A, 1 ∈
/ B.
(i) Since N
T
n=1 In ̸= ∅ we have Jn ̸= ∅.
(ii) Jn being the intersection of closed intervals makes it a closed interval.
(iii) Jn+1 ⊆ Jn since In+1 ∩ Jn ⊆ Jn
(iv) ∞
T T∞ Tn T∞
n=1 Jn = n=1 ( k=1 Ik ) = n=1 In
T∞
By (i),
T (ii) and (iii) the Nested Interval Property tells us n=1 Jn ̸= ∅. Therefore by
(iv) ∞n=1 In ̸= ∅.
16 CHAPTER 1. THE REAL NUMBERS
1.5 Cardinality
Exercise 1.5.1
Finish the following proof for Theorem 1.5.7. Assume B is a countable set. Thus, there
exists f : N → B, which is 1 − 1 and onto. Let A ⊆ B be an infinite subset of B. We must
show that A is countable.
Let n1 = min{n ∈ N : f (n) ∈ A}. As a start to a definition of g : N → A set
g(1) = f (n1 ). Show how to inductively continue this process to produce a 1-1 function g
from N onto A.
Solution
Let nk = min{n ∈ N | f (n) ∈ A, n ∈
/ {n1 , n2 , . . . , nk−1 }} and g(k) = f (nk ). since g : N → A
is 1-1 and onto, A is countable.
Exercise 1.5.2
Review the proof of Theorem 1.5.6, part (ii) showing that R is uncountable, and then find
the flaw in the following erroneous proof that Q is uncountable:
Assume, for contradiction, that Q is countable. Thus we can write Q = {r1 , r2 , r3 , . . .}
and, as before,
T∞ construct a nested sequence Tof closed intervals with rn ∈
/ In . Our construction
∞
implies n=1 In = ∅ while NIP implies n=1 In ̸= ∅. This contradiction implies Q must
therefore be uncountable.
Solution √
The nested interval property is not true for Q. Consider In being rational bounds for 2
√
with n decimal places, then ∞
T
n=1 In = ∅ since 2∈
/ Q.
Exercise 1.5.3
(a) Prove if A1 , . . . , Am are countable sets then A1 ∪ · · · ∪ Am is countable.
(b) S
Explain why induction cannot be used to prove that if each An is countable, then
∞
n=1 An is countable.
1 3 6 10 15 ···
2 5 9 14 · · ·
4 8 13 ···
7 12 ···
11 ···
..
.
Solution
(a) Let B, C be disjoint countable sets. We use the same trick as with the integers and
list them as
B ∪ C = {b1 , c1 , b2 , c2 , . . . }
1.5. CARDINALITY 17
(b) Induction shows something for each n ∈ N, it does not apply in the infinite case.
∞
! ∞ ∞ ∞
[ [ [ [
f (N) = f Cn = fn (Cn ) = Bn = An
n=1 n=1 n=1 n=1
S∞
Let fn : Cn → Bn be bijective since Bn is countable, define f : N → n=1 Bn as
f1 (n) if n ∈ C1
f (n) = f2 (n) if n ∈ C2
..
.
(i) Since each Cn is disjoint and each fn is 1-1, f (n1 ) = f (n2 ) implies n1 = n2
meaning f is 1-1.
(ii) Since any b ∈ ∞
S
n=1 Bn has b ∈ Bn for some n, we know b = fn (x) has a solution
since fn is onto. Letting x = fn−1 (b) we have f (x) = fn (x) = b since fn−1 (b) ∈ Cn
meaning f is onto.
∞
[ ∞
[
Bn = An
n=1 n=1
S∞
We have that n=1 An is countable, completing the proof.
Exercise 1.5.4
(a) Show (a, b) ∼ R for any interval (a, b).
(b) Show that an unbounded interval like (a, ∞) = {x : x > a} has the same cardinality
as R as well.
(c) Using open intervals makes it more convenient to produce the required 1-1, onto func-
tions, but it is not really necessary. Show that [0, 1) ∼ (0, 1) by exhibiting a 1-1 onto
function between the two sets.
Solution
(a) We will start by finding f : (−1, 1) → R and then transform it to (a, b). Example 1.5.4
gives a suitable f
x
f (x) = 2
x −1
18 CHAPTER 1. THE REAL NUMBERS
The book says to use calculus to show f is bijective, first we will examine the derivative
x2 − 1 − 2x2 x2 + 1
f ′ (x) = = −
(x2 − 1)2 (x2 − 1)2
The denominator and numerator are positive, so f ′ (x) < 0 for all x ∈ (0, 1). This
means no two inputs will be mapped to the same output, meaning f is one to one (a
rigorous proof is beyond our current ability)
To show that f is onto, we examine the limits
x
lim− = −∞
x→1 x2
−1
x
lim + 2 = +∞
x→−1 x − 1
(b) We want a bijective h(x) such that h(x) : (a, ∞) → (−1, 1) because then we could
compose them to get a new bijective function f (h(x)) : (a, ∞) → R.
Let
2
h(x) = −1
x−a+1
We have h : (a, ∞) → (1, −1) since h(a) = 1 and limx→∞ h(x) = 1.
Meaning that f (h(x)) : (a, ∞) → R is our bijective map.
(c) With countable sets adding a single element doesn’t change cardinality since we can
just shift by one to get a bijective map. we’ll use a similar technique here to essentially
outrun our problems. Define f : [0, 1) → (0, 1) as
1/2 if x = 0
1/4 if x = 1/2
f (x) = 1/8 if x = 1/4
..
.
x otherwise
Now we prove f is bijective by showing y = f (x) has exactly one solution for all
y ∈ (0, 1).
If y = 1/2n then the only solution is y = f (1/2n−1 ) (or x = 0 in the special case n = 1),
If y ̸= 1/2n then the only solution is y = f (y).
1.5. CARDINALITY 19
Exercise 1.5.5
(a) Why is A ∼ A for every set A?
(c) For three sets A, B, and C, show that A ∼ B and B ∼ C implies A ∼ C. These three
properties are what is meant by saying that ∼ is an equivalence relation.
Solution
(a) The identity function f (x) = x is a bijection
Exercise 1.5.6
(a) Give an example of a countable collection of disjoint open intervals.
(b) Give an example of an uncountable collection of disjoint open intervals, or argue that
no such collection exists.
Solution
(a) I1 = (0, 1), I2 = (1, 2) and in general In = (n − 1, n)
(b) Let A denote this set. Intuitively no such collection should exist since each In has
nonzero length.
The key here is to try and show A ∼ Q instead of directly showing A ∼ N.
For any nonempty interval In the density theorem tells us there exists an r ∈ Q such
that r ∈ In . Assigning each I ∈ A a rational number r ∈ I proves I ⊆ Q and thus I
is countable.
Exercise 1.5.7
Consider the open interval (0, 1), and let S be the set of points in the open unit square; that
is, S = {(x, y) : 0 < x, y < 1}.
(a) Find a 1-1 function that maps (0, 1) into, but not necessarily onto, S. (This is easy.)
(b) Use the fact that every real number has a decimal expansion to produce a 1−1 function
that maps S into (0, 1). Discuss whether the formulated function is onto. (Keep in
mind that any terminating decimal expansion such as .235 represents the same real
number as .234999 . . . .)
Solution
(a) We scale and shift up into the square. f (x) = 21 x + 1
3
20 CHAPTER 1. THE REAL NUMBERS
(b) Let g : S → (0, 1) be a function that interleaves decimals in the representation without
trailing nines, padding with zeros if necessary. g(0.32, 0.45) = 0.3425, g(0.19̄, 0.2) =
g(0.2, 0.2) = 0.22, g(0.1, 0.23) = 0.1203, g(0.1, 0.2̄) = 0.1202, etc.
Every real number can be written with two digit representations, one with trailing
9’s and one without. However g(x, y) = 0.d1 d2 . . . 9̄ is impossible since it would imply
x = 0.d1 . . . 9̄ and y = 0.d2 . . . 9̄ but the definition of g forbids this. therefore g(s) is
unique, and so g is 1-1.
Is g onto? No since g(x, y) = 0.1 has no solutions, since we would want x = 0.1 and
y = 0 but 0 ∈
/ (0, 1).
Exercise 1.5.8
Let B be a set of positive real numbers with the property that adding together any finite
subset of elements from B always gives a sum of 2 or less. Show B must be finite or countable.
Solution
Notice B ∩ (a, 2) is finite for all a > 0, since if it was infinite we could S
make a set with sum
greater then two. And since B is the countable union of finite sets ∞ n=1 B ∩ (1/n, 2), B
must be countable or finite.
Exercise 1.5.9
A real number x ∈ R is called algebraic if there exist integers a0 , a1 , a2 , . . . , an ∈ Z, not all
zero, such that
an xn + an−1 xn−1 + · · · + a1 x + a0 = 0
Said another way, a real number is algebraic if it is the root of a polynomial with integer
coefficients. Real numbers that are not algebraic are called transcendental numbers. Reread
the last paragraph of Section 1.1. The final question posed here is closely related to the
question of whether or not transcendental numbers exist.
√ √ √ √
(a) Show that 2, 3 2, and 3 + 2 are algebraic.
(b) Fix n ∈ N, and let An be the algebraic numbers obtained as roots of polynomials with
integer coefficients that have degree n. Using the fact that every polynomial has a
finite number of roots, show that An is countable.
(c) Now, argue that the set of all algebraic numbers is countable. What may we conclude
about the set of transcendental numbers?
Solution √ √
(a) x2 − 2 = 0, x3 − 2 = 0 are obvious. Now consider 3 + 2. The key is setting
√ √
x = 3+ 2 then using algebra on x to concoct an integer, and thus find the polynomial
with x as a root.
√ √
We have x2 = 5+2 6√ meaning
√ x 2
−5 = 2 6 and thus (x2 −5)2 = 24 so (x2 −5)2 −24 = 0
is a polynomial with 3 + 2 as a root.
(b) Basically An ∼ Zn ∼ Nn ∼ N.
1.5. CARDINALITY 21
Exercise 1.5.10
(a) Let C ⊆ [0, 1] be uncountable. Show that there exists a ∈ (0, 1) such that C ∩ [a, 1] is
uncountable.
(b) Now let A be the set of all a ∈ (0, 1) such that C ∩ [a, 1] is uncountable, and set
α = sup A. Is C ∩ [α, 1] an uncountable set?
(c) Does the statement in (a) remain true if “uncountable” is replaced by “infinite”?
Solution
(a) Suppose a does not exist, then C ∩ [a, 1] is countable for all a ∈ (0, 1) meaning
∞
[
C ∩ [1/n, 1] = C ∩ [0, 1]
n=1
(b) If α = 1 then C ∩ [α, 1] is finite. Now if α < 1 we have C ∩ [α + ϵ, 1] countable for ϵ > 0
(otherwise the set would be in A, and hence α would not be an upper bound). Take
∞
[
C ∩ [α + 1/n, 1] = C ∩ [α, 1]
n=1
(c) No, consider the set C = {1/n : n ∈ N} it has C ∩ [α, 1] finite for every α, but C ∩ [0, 1]
is infinite.
Exercise 1.5.11 (Schröder-Bernstein Theorem)
Assume there exists a 1-1 function f : X → Y and another 1-1 function g : Y → X. Follow
the steps to show that there exists a 1-1, onto function h : X → Y and hence X ∼ Y . The
strategy is to partition X and Y into components
X = A ∪ A′ and Y = B ∪ B′
with A ∩ A′ = ∅ and B ∩ B ′ = ∅, in such a way that f maps A onto B, and g maps B ′ onto
A′ .
(c) Let A = ∞
S S∞
n=1 An and B = n=1 f (An ). Show that f maps A onto B.
Solution
(a) f : A → B and g : B ′ → A′ are bijective, therefore we can define
(
f (x) if x ∈ A
h(x) =
g (x) if x ∈ A′
−1
which is bijective.
Solution
In Exercise 1.5.4 (a) we found a bijection f : (0, 1) → R. Now suppose g : (0, 1) → N is
some map, we must show g is bijective if and only if (g ◦ f ) : R → N is bijective. This is
clearly true as if g is bijective then (g ◦ f ) is bijective (composition of bijective functions),
and if (g ◦ f ) is bijective then (g ◦ f ) ◦ f −1 = g is bijective.
Exercise 1.6.2
Let f : N → R be a way to list every real number (hence show R is countable).
Define a new number x with digits b1 b2 . . . given by
(
2 if ann ̸= 2
bn =
3 if ann = 2
(b) Now, explain why x ̸= f (2), and in general why x ̸= f (n) for any n ∈ N.
(c) Point out the contradiction that arises from these observations and conclude that (0, 1)
is uncountable.
Solution
(a) The first digit is different
(c) Therefore x is not in the list, since the nth digit is different
Exercise 1.6.3
Supply rebuttals to the following complaints about the proof of Theorem 1.6.1.
(a) Every rational number has a decimal expansion, so we could apply this same argument
to show that the set of rational numbers between 0 and 1 is uncountable. However,
because we know that any subset of Q must be countable, the proof of Theorem 1.6.1
must be flawed.
(b) Some numbers have two different decimal representations. Specifically, any decimal
expansion that terminates can also be written with repeating 9’s. For instance, 1/2
can be written as .5 or as .4999 . . . Doesn’t this cause some problems?
Solution
(a) False, since the constructed number has an infinite number of decimals it is irrational.
(b) No, since if we have 9999 . . . and change the nth digit 9992999 = 9993 is still different.
24 CHAPTER 1. THE REAL NUMBERS
Exercise 1.6.4
Let S be the set consisting of all sequences of 0 ’s and 1 ’s. Observe that S is not a particular
sequence, but rather a large set whose elements are sequences; namely,
S = {(a1 , a2 , a3 , . . .) : an = 0 or 1}
Solution
We flip every bit in the diagonal just like with R. Another way would be to show S ∼ R by
writing real numbers in base 2.
Exercise 1.6.5
(a) Let A = {a, b, c}. List the eight elements of P (A). (Do not forget that ∅ is considered
to be a subset of every set.)
Solution
(a) A = {∅, {a}, {b}, {c}, {a, b}, {a, c}, {b, c}, {a, b, c}}.
(b) There are n elements, we can include or exclude each element so there are 2n subsets.
Exercise 1.6.6
(a) Using the particular set A = {a, b, c}, exhibit two different 1 − 1 mappings from A into
P (A).
(c) Explain why, in parts (a) and (b), it is impossible to construct mappings that are onto.
Solution
(a) f (x) = {x}, f (x) = {x, b} for x ̸= b and f (x) = {a, b, c} for x = b.
(c) We can hit at most n elements in the power set out of the 2n total elements.
Proof
Suppose f : A → P (A) is onto. We want to use the self referential nature of P (A) to find a
contradiction. Define
B = {a : a ∈
/ f (a)}
Since f is onto we must have f (a) = B for some a ∈ A. Then either
(i) a ∈ B implies a ∈ f (a) which by the definition of B implies a ∈
/ B, so a ∈ B is
impossible.
1.6. CANTOR’S THEOREM 25
(ii) a ∈
/ B implies a ∈/ f (a) since f (a) = B. but if a ∈
/ f (a) then a ∈ B by the definition
of B, contradicting a ∈/ B.
Therefore f cannot be onto, since we have found a B ∈ P (A) where f (a) = B is impossible.
Stepping back, the pearl of the argument is that if B = f (a) then B = {a : a ∈/ B} is
undecidable/impossible.
Exercise 1.6.7
See the proof of Cantor’s theorem above (the rest is a computation)
Exercise 1.6.8
See the proof of Cantor’s theorem above
Exercise 1.6.9
Using the various tools and techniques developed in the last two sections (including the
exercises from Section 1.5), give a compelling argument showing that P (N) ∼ R.
Solution
Recall from Exercise 1.6.4 that if
S = {(a1 , a2 , a3 , . . .) : an = 0 or 1}
Exercise 1.6.10
As a final exercise, answer each of the following by establishing a 1 − 1 correspondence with
a set of known cardinality.
Solution
(a) The set of functions from {0, 1} to N is the same as N2 which we found was countable
in Exercise 1.5.9.
(b) This is the same as an infinite list of zeros and ones which we showed was uncountable
in Exercise 1.6.4.
(c) Let A be an antichain of P (N), let Al be the sets in A of size l. For finite l, Al is
countable
S∞ since Al ⊆ Nl is countable (shown in 1.5.9). Therefore the countable union
l=0 Al = A is countable. Thus, if P (N) contains an uncountable antichain, “most”
sets in the antichain must be infinite (in that there will be uncountably many sets in
26 CHAPTER 1. THE REAL NUMBERS
the antichain will be infinite, whereas only countably many sets in the antichain will
be finite).
This observation inspires the following construction, using a variant of the set S from
Exercise 1.6.4. Define the set
where d(x, n) is the n’th digit in the decimal expansion of x. To avoid the issue of some
numbers having two equivalent decimal representations, always use the representation
with repeating 9’s. In this manner, each element of A encodes a particular real number,
in a similar way that each element of S encodes a particular real number through its
binary expansion.
Note that each element of A is infinite. Note also that since any two distinct real
numbers will differ in at least one place in their decimal expansions, the corresponding
elements in A will differ there as well, and hence A is an antichain.
Formally, let x1 , x2 be two distinct real numbers, A1 , A2 be the elements of A corre-
sponding to x1 , x2 respectively, and n be the first decimal position where x1 and x2
differ. Then 10n + d(x1 , n) will be in A1 but not A2 , and 10n + d(x2 , n) will be in A2
but not A1 . Thus, neither A1 ⊆ A2 nor A2 ⊆ A1 . Since (0, 1) is uncountable, A is an
uncountable antichain in P (N).
Chapter 2
sin(n2 )
(c) lim √
3n = 0.
Solution
(a) We have
2n + 1 2 5(2n + 1) − 2(5n + 4) −3 3
5n + 4 − 5 = =
= <ϵ
5(5n + 4) 5(5n + 4) 5(5n + 4)
27
28 CHAPTER 2. SEQUENCES AND SERIES
You could also solve for the smallest n, which would give you
3 3 3 4
< ϵ =⇒ 5n + 4 > =⇒ n > −
5(5n + 4) 5ϵ 25ϵ 5
I prefer the first approach, the second is better if you were doing numerical analysis
and wanted a precise convergence rate.
(b) We have
2n2 2 2
= 2n < 2n = 2 < ϵ =⇒ n > 2
n3 + 3 − 0 n3 + 3 n3 n ϵ
(c) We have
sin(n2 ) 1 1
≤ < ϵ =⇒ n >
n1/3 n1/3 ϵ3
Really slow convergence! if ϵ = 10−2 we would require n > 106
Exercise 2.2.3
Describe what we would have to demonstrate in order to disprove each of the following
statements.
(a) At every college in the United States, there is a student who is at least seven feet tall.
(b) For all colleges in the United States, there exists a professor who gives every student
a grade of either A or B.
(c) There exists a college in the United States where every student is at least six feet tall.
Solution
(a) Find a collage in the United States with no students over seven feet tall.
(b) Find a collage in the United States with no professors that only give grades of A or B.
(c) Show that all collages in the United States have at least one student under six feet tall.
Exercise 2.2.4
Give an example of each or state that the request is impossible. For any that are impossible,
give a compelling argument for why that is the case.
(a) A sequence with an infinite number of ones that does not converge to one.
(b) A sequence with an infinite number of ones that converges to a limit not equal to one.
(c) A divergent sequence such that for every n ∈ N it is possible to find n consecutive
ones somewhere in the sequence.
2.2. THE LIMIT OF A SEQUENCE 29
Solution
(a) an = (−1)n
(b) Impossible, if lim an = a ̸= 1 then for any n ≥ N we can find a n with an = 1 meaning
ϵ < |1 − a| is impossible.
(c) an = (1, 2, 1, 1, 3, 1, 1, 1, . . . )
Exercise 2.2.5
Let [[x]] be the greatest integer less than or equal to x. For example, [[π]] = 3 and [[3]] = 3.
For each sequence, find lim an and verify it with the definition of convergence.
(a) an = [[5/n]],
(ii) A sequence (an ) is frequently in a set A ⊆ R if, for every N ∈ N, there exists an n ≥ N
such that an ∈ A.
(d) Suppose an infinite number of terms of a sequence (xn ) are equal to 2 . Is (xn )
necessarily eventually in the interval (1.9, 2.1)? Is it frequently in (1.9, 2.1)?
Solution
(a) Frequently, but not eventually.
(d) (xn ) is frequently in (1.9, 2.1) but not necessarily eventually (consider xn = 2(−1)n ).
Exercise 2.2.8
For some additional practice with nested quantifiers, consider the following invented defini-
tion:
Let’s call a sequence (xn ) zero-heavy if there exists M ∈ N such that for all N ∈ N there
exists n satisfying N ≤ n ≤ N + M where xn = 0
(d) Form the logical negation of the above definition. That is, complete the sentence: A
sequence is not zero-heavy if ....
Solution
(a) Yes. Choose M = 1; since the sequence has a 0 in every two spaces, for all N either
xn = 0 or xn+1 = 0.
(b) Yes. If there were a finite number of zeros, with the last zero at position K, then
choosing N > K would lead to a contradiction.
(c) No, consider (0, 1, 0, 1, 1, 0, 1, 1, 1, 0, 1, 1, 1, 1, 0, . . . ) where the gap between 0’s grows
indefinitely. For any value of M , for large enough N the gap between zeros will be
greater than M . Then we simply choose N so that xN is the first 1 in a streak of at
least M + 1 1’s.
(d) A sequence is not zero-heavy if for all M ∈ N, there exists some N ∈ N such that for
all n ∈ N, N ≤ n ≤ N + M , xn ̸= 0.
2.3. THE ALGEBRAIC AND ORDER LIMIT THEOREMS 31
Exercise 2.3.4
Let (an ) → 0, and use the Algebraic Limit Theorem to compute each of the following limits
(assuming the fractions are always defined):
1+2an
(a) lim 1+3an −4a2
n
(an +2)2 −4
(b) lim an
2
an
+3
(c) lim 1
+5
.
an
Solution
(a) Apply the ALT
1 + 2an lim (1 + 2an )
lim =
1 + 3an − 4a2n lim (1 + 3an − 4a2n )
1 + 2 lim(an )
=
1 + 3 lim an − 4 lim a2n
=1
(c) Multiply the top and bottom by an then apply the ALT
2
!
+3
an 2 + 3an
lim 1 = lim
an
+5 1 + 5an
2 + 3 lim an
=
1 + 5 lim an
=2
Exercise 2.3.5
Let (xn ) and (yn ) be given, and define (zn ) to be the “shuffled” sequence (x1 , y1 , x2 , y2 , x3 , y3 , . . . , xn , yn , . . .).
Prove that (zn ) is convergent if and only if (xn ) and (yn ) are both convergent with lim xn =
lim yn .
Solution
Obviously if lim xn = lim yn = l then zn → l. To show the other way suppose (zn ) → l, then
|zn − l| < ϵ for all n ≥ N meaning |yn − l| < ϵ and |xn − l| < ϵ for n ≥ N aswell. Thus
lim xn = lim yn = l.
2.3. THE ALGEBRAIC AND ORDER LIMIT THEOREMS 33
Exercise 2.3.6 √
Consider the sequence given by bn = n − n2 + 2n. Taking (1/n) → 0 as given, and using
both the Algebraic Limit Theorem and the result in Exercise 2.3.1, show lim bn exists and
find the value of the limit.
Solution
I’m going to find the value of the limit before proving it. We have
√ p
n− n2 + 2n = n − (n + 1)2 − 1
p
For large n, (n + 1)2 − 1 ≈n + 1 so lim bn = −1.
p
Factoring out n we get n 1 − 1 + 2/n . Tempting as it is to apply the ALT here to
say (bn ) → 0 it doesn’t work since n diverges.
How about if I get rid of the radical, then use the ALT to go back to what we had before?
√ √
(n − n2 + 2n)(n + n2 + 2n) = n2 − (n2 + 2n) = −2n
Then we have
√ −2n −2
bn = n − n2 + 2n = √ = p
2
n + n + 2n 1 + 1 + 2/n
Now we can finally use the algebraic limit theorem!
!
−2 −2 −2
lim p = p = √ = −1
1 + 1 + 2/n 1 + 1 + lim (2/n) 1+ 1+0
Stepping back the key to this technique is removing the radicals via a difference of squares,
then dividing both sides by the growth rate n and applying the ALT.
Exercise 2.3.7
Give an example of each of the following, or state that such a request is impossible by
referencing the proper theorem(s):
(a) sequences (xn ) and (yn ), which both diverge, but whose sum (xn + yn ) converges;
(b) sequences (xn ) and (yn ), where (xn ) converges, (yn ) diverges, and (xn + yn ) converges;
(c) a convergent sequence (bn ) with bn ̸= 0 for all n such that (1/bn ) diverges;
(d) an unbounded sequence (an ) and a convergent sequence (bn ) with (an − bn ) bounded;
(e) two sequences (an ) and (bn ), where (an bn ) and (an ) converge but (bn ) does not.
Solution
(a) (xn ) = n and (yn ) = −n diverge but xn + yn = 0 converges
(b) Impossible, the algebraic limit theorem implies lim(xn +yn )−lim(xn ) = lim yn therefore
(yn ) must converge if (xn ) and (xn + yn ) converge.
34 CHAPTER 2. SEQUENCES AND SERIES
(c) bn = 1/n has bn → 0 and 1/bn diverges. If bn → b ̸= 0 then 1/bn → 1/b, but since
b = 0 ALT doesn’t apply.
(d) Impossible, |bn | is convergent and therefore bounded (Theorem 2.3.2) so |bn | ≤ M1 ,
and |an − bn | ≤ M2 is bounded, therefore
must be bounded.
(e) bn = n and an = 0 works. However if (an ) → a, a ̸= 0 and (an bn ) → p then the ALT
would imply (bn ) → p/a.
Exercise 2.3.8
Let (xn ) → x and let p(x) be a polynomial.
(b) Find an example of a function f (x) and a convergent sequence (xn ) → x where the
sequence f (xn ) converges, but not to f (x).
Solution
(a) Applying the algebraic limit theorem multiple times gives (xdn ) → xd meaning
As a cute corollary, any continuous function f has lim f (xn ) = f (x) since polynomials
can approximate continuous functions arbitrarily well by the Weierstrass approxima-
tion theorem.
Exercise 2.3.9
(a) Let (an ) be a bounded (not necessarily convergent) sequence, and assume lim bn = 0.
Show that lim (an bn ) = 0. Why are we not allowed to use the Algebraic Limit Theorem
to prove this?
(b) Can we conclude anything about the convergence of (an bn ) if we assume that (bn )
converges to some nonzero limit b ?
(c) Use (a) to prove Theorem 2.3.3, part (iii), for the case when a = 0.
2.3. THE ALGEBRAIC AND ORDER LIMIT THEOREMS 35
Solution
(a) We can’t use the ALT since an is not necessarily convergent. an being bounded gives
|an | ≤ M for some M giving
|an bn | ≤ M |bn | < ϵ
Which can be accomplished by letting |bn | < ϵ/M since (bn ) → 0.
(b) No
(c) In (a) we showed lim(an bn ) = 0 = ab for b = 0 which proves part (iii) of the ALT.
Exercise 2.3.10
Consider the following list of conjectures. Provide a short proof for those that are true and
a counterexample for any that are false.
Solution
(a) False, consider an = n and bn = −n.
(b) True since if |bn − b| < ϵ then ||bn | − |b|| ≤ |bn − b| < ϵ by Exercise 1.2.6 (d).
(d) True, since 0 ≤ |bn − b| ≤ an we have an ≥ 0. Let ϵ > 0 and pick N such that an < ϵ
for all n ≥ N . Therefor
|bn − b| ≤ an < ϵ
Proving (bn ) → b.
x1 + x2 + · · · + xn
yn =
n
also converges to the same limit.
(b) Give an example to show that it is possible for the sequence (yn ) of averages to converge
even if (xn ) does not.
36 CHAPTER 2. SEQUENCES AND SERIES
Solution
(a) Let D = sup{|xn − x| : n ∈ N} and let 0 < ϵ < D, we have
x1 + · · · + xn |x1 − x| + · · · + |xn − x|
|yn − x| =
− x ≤
≤D
n n
Let N2 be large enough that for all n > N2 (remember 0 < ϵ < D so (D − ϵ/2) > 0.)
N1 (D − ϵ/2)
0< < ϵ/2
n
Therefor
N1 (D − ϵ/2)
|yn − x| ≤ + ϵ/2 < ϵ
n
Letting N = max{N1 , N2 } completes the proof as |yn − x| < ϵ for all n > N .
(Note: I could have used any ϵ′ < ϵ instead of ϵ/2, I just needed some room.)
Exercise 2.3.12
A typical task in analysis is to decipher whether a property possessed by every term in a
convergent sequence is necessarily inherited by the limit. Assume (an ) → a, and determine
the validity of each claim. Try to produce a counterexample for any that are false.
(a) If every an is an upper bound for a set B, then a is also an upper bound for B.
(b) If every an is in the complement of the interval (0, 1), then a is also in the complement
of (0, 1).
Solution
(a) True, let s = sup B, we know s ≤ an so by the order limit theorem s ≤ a meaning a is
also an upper bound on B.
(b) True, since if a ∈ (0, 1) then there would exist an ϵ-neighborhood inside (0, 1) that an
would have to fall in, contradicting the fact that an ∈/ (0, 1).
√
(c) False, consider the sequence of rational approximations to 2
Define limm,n→∞ amn = a to mean that for all ϵ > 0 there exists an N ∈ N such that
if both m, n ≥ N , then |amn − a| < ϵ
(b) Let amn = 1/(m + n). Does limm,n→∞ amn exist in this case? Do the two iterated
limits exist? How do these three values compare? Answer these same questions for
amn = mn/ (m2 + n2 )
(c) Produce an example where limm,n→∞ amn exists but where neither iterated limit can
be computed.
(d) Assume limm,n→∞ amn = a, and assume that for each fixed m ∈ N, limn→∞ (amn ) →
bm . Show limm→∞ bm = a
(e) Prove that if limm,n→∞ amn exists and the iterated limits both exist, then all three
limits must be equal.
Solution
(a)
m m
lim lim = 1, and lim lim =0
n→∞ m→∞ m + n m→∞ n→∞ m + n
(b) For amn = 1/(m + n) all three limits are zero. For amn = mn/(m2 + n2 ) iterated limits
are zero, and limm,n→∞ amn does not exist since for m, n ≥ N setting m = n gives
n2 1
2 2
=
n +n 2
Which cannot be made smaller then ϵ = 1/2.
The reason you would think to set m = n is in trying to maximize mn/(m2 +n2 ) notice
if m > n then mn > n2 so we are adding more to the numerator then the denominator,
hence the ratio is increasing. And if m < n then the ratio is decreasing. therefore the
maximum point is at m = n.
(c) Intuitively, in order for limm,n→∞ amn to exist, neither iterated limit can diverge to
infinity - otherwise, amn can also diverge to infinity, by holding letting the index that
causes divergence to grow while holding the other index fixed. Therefore, we must rely
on each iterated limit diverging due to oscillation.
The key additional “ability” that limm,n→∞ gives over an iterated limit is that both m
and n can be forced to grow big at the same time, whereas with an iterated limit only
one of them is forced to grow big.
Note that since iterated limits can only increase one of m and n, min{m, n} can’t
be increased indefinitely - but with limm,n→∞ , it can. Thus, the idea is to introduce
38 CHAPTER 2. SEQUENCES AND SERIES
oscillation in the sequence, then use min{m, n} to cause the oscillation to die out.
Define
(−1)m+n
amn =
min{m, n}
For a fixed m, once n > m, amn will oscillate between 1/m and −1/m, and thus
limn→∞ amn does not exist. Similar reasoning shows that for a fixed n, limm→∞ amn
does not exist either. But clearly limm,n→∞ amn = 0.
(d) Choose ϵ > 0 and let 0 < ϵ′ < ϵ. We need to find N so that |bm − a| < ϵ for all m > N .
Set N such that |amn − a| < ϵ′ when n, m ≥ N . Then fix m ≥ N , I will show
|bm − a| < ϵ. apply the triangle inequality to get
This inequality is true for all n, we will pick n to make it strict enough to complete
the proof. Set n ≥ max{N, Nm } where Nm (dependent on m) is big enough that
|bm − amn | < ϵ − ϵ′ . We also have |amn − a| < ϵ′ since m ≥ N and n ≥ N . So finally
And we are done. The key is that we can make |bm − amn | as small as we want
independent of m, so we take the limit as n → ∞ to show |bm − a| ≤ |amn − a|.
(e) Let bm = limn→∞ (amn ), cn = limm→∞ (amn ), and a = limm,n→∞ (amn ). In (d) we showed
(bm ) → a; a similar argument shows (cn ) → a. Thus all three limits are equal to a.
2.4. THE MONOTONE CONVERGENCE THEOREM AND A FIRST LOOK AT INFINITE SERIES39
(b) Now that we know lim xn exists, explain why lim xn+1 must also exist and equal the
same value.
(c) Take the limit of each side of the recursive equation in part (a) to explicitly compute
lim xn .
Solution
(a) x2 = 1 makes me conjecture xn is monotonic. For induction suppose xn > xn+1 then
we have
1 1
4 − xn < 4 − xn+1 =⇒ > =⇒ xn+1 > xn+2
4 − xn 4 − xn+1
Thus xn is decreasing, to show xn is bounded notice xn cannot be negative since xn < 3
means xn+1 = 1/(4 − xn ) > 0. therefore by the monotone convergence theorem (xn )
converges.
(b) Clearly skipping a single term does not change what the series converges to.
yn+1 = 3 − yn
and set y = lim yn . Because (yn ) and (yn+1 ) have the same limit, taking the limit across
the recursive equation gives y = 3 − y. Solving for y, we conclude lim yn = 3/2 What
is wrong with this argument?
(b) This time set y1 = 1 and yn+1 = 3 − y1n . Can the strategy in (a) be applied to compute
the limit of this sequence?
Solution
(a) The sequence yn = (1, 2, 1, 2, . . . ) does not converge.
40 CHAPTER 2. SEQUENCES AND SERIES
(b) Yes, yn converges by the monotone convergence theorem since 0 < yn < 3 and yn is
increasing.
Exercise 2.4.3
(a) Show that
r
√ √ √
q q
2, 2 + 2, 2+ 2+ 2, . . .
converges and find the limit.
Solution √ √
(a) Let x1 = 2 and xn+1 = 2 + xn clearly x2 > x1 . assuming xn+1 > xn gives
p √
2 + xn+1 > 2 + xn ⇐⇒ 2 + xn+1 > 2 + xn ⇐⇒ xn+2 > xn+1
Now the monotone convergence theorem tells us (xn ) converges. To find the limit use
lim xn = lim xn+1 = x to get
x = (2x)1/2 =⇒ x2 = 2x =⇒ x = ±2
Since xn ≥ 0 we have x = 2.
Exercise 2.4.4
(a) In Section 1.4 we used the Axiom of Completeness (AoC) to prove the Archimedean
Property of R (Theorem 1.4.2). Show that the Monotone Convergence Theorem can
also be used to prove the Archimedean Property without making any use of AoC.
2.4. THE MONOTONE CONVERGENCE THEOREM AND A FIRST LOOK AT INFINITE SERIES41
(b) Use the Monotone Convergence Theorem to supply a proof for the Nested Interval
Property (Theorem 1.4.1) that doesn’t make use of AoC.
These two results suggest that we could have used the Monotone Convergence Theorem
in place of AoC as our starting axiom for building a proper theory of the real numbers.
Solution
(a) MCT tells us (1/n) converges, obviously it must converge to zero therefore we have
|1/n − 0| = 1/n < ϵ for any ϵ, which is the Archimedean Property.
(b) We have In = [an , bn ] with an ≤ bn since In ̸= ∅. Since In+1 ⊆ In we must have
bn+1 ≤ bn and an+1 ≥ an the MCT tells us that (an ) → a and (bn ) → b. by the
Order
T Limit Theorem T we have a ≤ b since an ≤ bn , therefore a ∈ In for all n meaning
a∈ ∞ I
n=1 n and thus ∞
n=1 In ̸= ∅.
(b) Let
1 c
xn+1 = xn +
2 xn
I won’t go through the convergence analysis again, but the only fixed point is
1 c 1 1
x= x+ =⇒ x2 = c =⇒ x2 = c
2 x 2 2
So if xn converges, it must converge to x2 = c.
Solution
(a) We have
√
xy ≤ (x + y)/2 ⇐⇒ 4xy ≤ x2 + 2xy + y 2 ⇐⇒ 0 ≤ (x − y)2
(b) The only fixed point is xn = yn so we only need to show both sequences converge.
The inequality x1 ≤ y1 is always true since
√ xn + y n
x n yn ≤ =⇒ xn+1 ≤ yn+1
2
√
Also xn ≤ yn implies (xn + yn )/2 = yn+1 ≤ yn , similarly xn yn = xn+1 ≥ xn meaning
both sequences converge by the monotone convergence theorem.
where yn is the sequence from part (a) of this exercise. Provide a reasonable definition
for lim inf an and briefly explain why it always exists for any bounded sequence.
(c) Prove that lim inf an ≤ lim sup an for every bounded sequence, and give an example of
a sequence for which the inequality is strict.
2.4. THE MONOTONE CONVERGENCE THEOREM AND A FIRST LOOK AT INFINITE SERIES43
(d) Show that lim inf an = lim sup an if and only if lim an exists. In this case, all three
share the same value.
Solution
(a) (yn ) is decreasing and converges by the monotone convergence theorem.
(b) Define lim inf an = lim zn for zn = inf{an : k ≥ n}. zn converges since it is increasing
and bounded.
(c) Obviously inf{ak : k ≥ n} ≤ sup{an : k ≥ n} so by the Order Limit Theorem
lim inf an ≤ lim sup an .
(d) If lim inf an = lim sup an then the squeeze theorem (Exercise 2.3.3) implies an converges
to the same value, since inf{ak≥n } ≤ an ≤ sup{ak≥n }.
Exercise 2.4.8
For each series, find an explicit formula for the sequence of partial sums and determine if
the series converges.
P∞ 1
(a) n=1 2n
P∞ 1
(b) n=1 n(n+1)
P∞ n+1
(c) n=1 log n
(In (c), log(x) refers to the natural logarithm function from calculus.)
Solution
(a) This is a geometric series, we can use the usual trick to derive sn . Let r = 1/2 for
convenience
sn = 1 + r + r 2 + · · · + r n
rsn = r + r2 + · · · + rn+1
rn+1 − 1
rsn − sn = rn+1 − 1 =⇒ sn =
r−1
This is the formula when n starts at zero, but the sum in question starts at one so we
subtract the first term to correct this
∞ ∞
X 1 X 1 (1/2)n+1 − 1 −1
= −1 + = −1 + lim = −1 + =1
n=1
2n n=0
2n n→∞ 1/2 − 1 −1/2
(a) Find an explicit formula for the sequence of partial products in the case where an = 1/n
and decide whether the sequence converges. Write out the first few terms in the
sequence of partial products in the case where an = 1/n2 and make a conjecture about
the convergence of this sequence.
(b) Show, in general, that the sequence of partial products converges if and only if ∞
P
n=1 an
x
converges. (The inequality 1 + x ≤ 3 for positive x will be useful in one direction.)
Solution
(a) This is a telescoping product, most of the terms cancel
m m
Y Y n+1 2 3 4 m+1
pm = (1 + 1/n) = = · · ··· =m+1
n=1 n=1
n 1 2 2 m
(a) A sequence that has a subsequence that is bounded but contains no subsequence that
converges.
(b) A sequence that does not contain 0 or 1 as a term but contains subsequences converging
to each of these values.
(c) A sequence that contains subsequences converging to every point in the infinite set
{1, 1/2, 1/3, 1/4, 1/5, . . .}.
(d) A sequence that contains subsequences converging to every point in the infinite set
{1, 1/2, 1/3, 1/4, 1/5, . . .}, and no subsequences converging to points outside of this
set.
Solution
(a) Impossible, the Bolzano–Weierstrass theorem tells us a convergent subsequence of that
subsequence exists, and that sub-sub sequence is also a subsequence of the original
sequence.
(b) (1 + 1/n) → 1 and (1/n) → 0 so (1/2, 1 + 1/2, 1/3, 1 + 1/3, . . . ) has subsequences
converging to 0 and 1.
(c) Copy the finitely many previous terms before proceeding to a new term
The sequence contains infinitely many terms in {1, 1/2, 1/3, . . . } hence subsequences
exist converging to each of these values.
(d) Impossible, the sequence must converge to zero which is not in the set.
Proof: Let ϵ > 0 be arbitrary, pick N large enough that 1/n < ϵ/2 for n > N . We
can find a subsequence (bm ) → 1/n meaning |bm − 1/n| < ϵ/2 for some m. using the
triangle inequality we get
therefore we have found a number bm in the sequence am with |bm | < ϵ. This process can
be repeated for any ϵ therefore a sequence which converges to zero can be constructed.
Exercise 2.5.2
Decide whether the following propositions are true or false, providing a short justification
for each conclusion.
(a) If every proper subsequence of (xn ) converges, then (xn ) converges as well.
2.5. SUBSEQUENCES AND THE BOLZANO–WEIERSTRASS THEOREM 47
Solution
Let A be a bounded set, we’re basically going to binary search for sup A and then use NIP
to prove the limit exists.
Let M be an upper bound on A, and pick any L ∈ A as our starting lower bound for
sup A and define I1 = [L, M ]. Doing binary search gives In+1 ⊆ In with length proportional
to (1/2)n . Applying the Nested Interval Property gives
∞
\
In ̸= ∅
n=1
Exercise 2.5.7
Extend the result proved in Example 2.5.3 to the case |b| < 1; that is, show lim (bn ) = 0 if
and only if −1 < b < 1.
Solution
If |b| ≥ 1 then lim(bn ) ̸= 0 (diverges for b ̸= 1).
Now for the other direction, if |b| < 1 we immediately get |bn | < 1 thus bn is bounded.
Since it is decreasing the monotone convergence theorem implies it converges. To find the
limit equating terms bn+1 = bn gives b = 0 or b = 1, since b is strictly decreasing we have
b = 0.
Exercise 2.5.8
Another way to prove the Bolzano-Weierstrass Theorem is to show that every sequence
contains a monotone subsequence. A useful device in this endeavor is the notion of a peak
term. Given a sequence (xn ), a particular term xm is a peak term if no later term in the
sequence exceeds it; i.e., if xm ≥ xn for all n ≥ m.
(a) Find examples of sequences with zero, one, and two peak terms. Find an example of
a sequence with infinitely many peak terms that is not monotone.
(b) Show that every sequence contains a monotone subsequence and explain how this
furnishes a new proof of the Bolzano-Weierstrass Theorem.
Solution
(a) (1, 2, . . . ) has zero peak terms, (1, 0, 1/2, 2/3, 3/4, . . . ) has a single peak term, (2, 1, 1/2, 2/3, . . . )
has two peak terms (a similar argument works for k peak terms) and (1, 1/2, 1/3, . . . )
has infinitely many peak terms. The sequence (1, −1/2, 1/3, −1/4, . . . ) has infinitely
many peak terms, but is not monotone.
(b) Note that the (possibly finite) sequence of peak terms is monotonic decreasing. There
are two possibilities - either there are infinitely many peak terms, or only finitely many.
If there are infinitely many peak terms, simply take the subsequence of peak terms; if
the parent sequence is bounded we have found a subsequence which converges (by the
Monotone Convergence Theorem), hence proving BW in this case.
If there are finitely many peak terms, let the last peak term be at position k. Consider
the terms after the last peak term. Since after this point there are no more peak terms,
then for every term xn there must be at least one term xm ≥ xn where m > n > k.
Therefore we can define the monotone subsequence x′ as x′1 = xk+1 , x′n as the first term
after x′n−1 such that x′n > x′n−1 . By MCT this subsequence converges, hence proving
BW in this case as well.
Exercise 2.5.9
Let (an ) be a bounded sequence, and define the set
Show that there exists a subsequence (ank ) converging to s = sup S. (This is a direct proof
of the Bolzano-Weierstrass Theorem using the Axiom of Completeness.)
50 CHAPTER 2. SEQUENCES AND SERIES
Solution
For every ϵ > 0 there exists an x ∈ S with x > s − ϵ implying |s − x| < ϵ. therefore we can
get arbitrarily close to s = sup S so there is a subsequence converging to this value.
To make this more rigorous, pick xn ∈ S such that |xn − s| < 1/n then pick N > 1/ϵ to
get |xn − s| < ϵ for all n > N .
2.6. THE CAUCHY CRITERION 51
Exercise 2.6.4
Let (an ) and (bn ) be Cauchy sequences. Decide whether each of the following sequences is a
Cauchy sequence, justifying each conclusion.
(a) cn = |an − bn |
(b) cn = (−1)n an
(c) cn = [[an ]], where [[x]] refers to the greatest integer less than or equal to x.
Solution
(a) Yes. Note that by the Triangle Inequality,
|an −am |+|bm −bn |+|am −bm | ≥ |an −bn | ⇒ |an −am |+|bm −bn | ≥ |an −bn |−|bm −am |
and
|am −an |+|bn −bm |+|an −bm | ≥ |am −bm | ⇒ |an −am |+|bm −bn | ≥ |am −bm |−|bn −an |
therefore
|cn − cm | = |an − bn | − |am − bm | ≤ |an − am | + |bm − bn | < ϵ/2 + ϵ/2 = ϵ
(c) No, if an = 1−(−1)n /n then [[an ]] fluctuates between 0 and 1 and so cannot be Cauchy.
Exercise 2.6.5
Consider the following (invented) definition: A sequence (sn ) is pseudo-Cauchy if, for all
ϵ > 0, there exists an N such that if n ≥ N , then |sn+1 − sn | < ϵ
Decide which one of the following two propositions is actually true. Supply a proof for
the valid statement and a counterexample for the other.
(ii) If (xn ) and (yn ) are pseudo-Cauchy, then (xn + yn ) is pseudo-Cauchy as well.
Solution
(i) False, consider sn = log n. clearly |sn+1 − sn | can be made arbitrarily small but sn is
unbounded.
(ii) True, as |(xn+1 + yn+1 ) − (xn + yn )| ≤ |xn+1 − xn | + |yn+1 − yn | < ϵ/2 + ϵ/2 = ϵ.
Exercise 2.6.6
Let’s call a sequence (an ) quasi-increasing if for all ϵ > 0 there exists an N such that whenever
n > m ≥ N it follows that an > am − ϵ
(a) Give an example of a sequence that is quasi-increasing but not monotone or eventually
monotone.
2.6. THE CAUCHY CRITERION 53
(b) Give an example of a quasi-increasing sequence that is divergent and not monotone or
eventually monotone.
(c) Is there an analogue of the Monotone Convergence Theorem for quasiincreasing se-
quences? Give an example of a bounded, quasi-increasing sequence that doesn’t con-
verge, or prove that no such sequence exists.
Solution
Think of “quasi-increasing” as “eventually the n’th term will be almost smaller then all
terms after it”
(a) an = (−1)n /n is quasi-increasing since (−1)m /m − (−1)n /n ≤ 1/m + 1/n < 2/N < ϵ
after picking some N > 2/ϵ.
(b) an = (2, 2 − 1/2, 3, 3 − 1/3, . . . ) is quasi-increasing. Let ϵ > 0 and set N > 1/ϵ, for
n > m ≥ N consider two cases We have an > am as long as an ̸= m − 1/m. If
an = m − 1/m then an > am − ϵ since m − 1/m > m − ϵ as 1/m < ϵ.
(c) Suppose (an ) is quasi-increasing and bounded and let ϵ > 0.
Let N1 be large enough that n > m ≥ N1 implies an > am − ϵ.
Since (an ) is bounded we can set s = sup an applying Lemma 1.3.8 tells us there exists
an N > N1 such that aN > s − ϵ.
Now for all n > N we have an > aN − ϵ, and since aN > s − ϵ we have an > s − 2ϵ.
This completes the proof as s ≥ an > s − 2ϵ implies |an − s| < 2ϵ for all n ≥ N , thus
s = lim an .
Exercise 2.6.7
Exercises 2.4.4 and 2.5.4 establish the equivalence of the Axiom of Completeness and the
Monotone Convergence Theorem. They also show the Nested Interval Property is equivalent
to these other two in the presence of the Archimedean Property.
(a) Assume the Bolzano-Weierstrass Theorem is true and use it to construct a proof of
the Monotone Convergence Theorem without making any appeal to the Archimedean
Property. This shows that BW, AoC, and MCT are all equivalent.
(b) Use the Cauchy Criterion to prove the Bolzano-Weierstrass Theorem, and find the
point in the argument where the Archimedean Property is implicitly required. This
establishes the final link in the equivalence of the five characterizations of completeness
discussed at the end of Section 2.6.
(c) How do we know it is impossible to prove the Axiom of Completeness starting from
the Archimedean Property?
Solution
(a) Suppose (xn ) is increasing and bounded, BW tells us there exists a convergent subse-
quence (xnk ) → x, We will show (xn ) → x. First note xk ≤ xnk implies xn ≤ x by the
Order Limit Theorem.
Pick K such that for k ≥ K we have |xnk − x| < ϵ. Since (xn ) is increasing and xn ≤ x
every n ≥ nK satisfies |xn − x| < ϵ as well. Thus (xn ) converges, completing the proof.
54 CHAPTER 2. SEQUENCES AND SERIES
(b) We’re basically going to use the Cauchy criterion as a replacement for NIP in the proof
of BW. Recall we had In+1 ⊆ In with ank ∈ Ik , we will show ank is Cauchy.
The length of Ik is M (1/2)k−1 by construction, so clearly |ank − anj | < M (1/2)N −1 for
k, j ≥ N , implying (ank ) converges by the Cauchy criterion.
We needed the Archimedean Property to conclude M (1/2)N −1 ∈ Q can be made
smaller then any ϵ ∈ R+ .
(c) The Archimedean Property is true for Q meaning it cannot prove AoC which is only
true for R. (If we did, then we would have proved AoC for Q which is obviously false.)
2.7. PROPERTIES OF INFINITE SERIES 55
(a) Prove the Alternating Series Test by showing that (sn ) is a Cauchy sequence.
(b) Supply another proof for this result using the Nested Interval Property (Theorem
1.4.1).
(c) Consider the subsequences (s2n ) and (s2n+1 ), and show how the Monotone Convergence
Theorem leads to a third proof for the Alternating Series Test.
Solution
(a) Let N ∈ N be even and let n ≥ N . because the series is alternating we have
sN ≤ sn ≤ sN +1
Which shows (sn ) is Cauchy, and hence converges by the Cauchy Criterion.
(b) Let I1 be the interval [a1 − a2 , a1 ] and in general In = [an − an+1 , an ], since (an ) is
decreasing we have In+1 ⊆ In . Applying the nested interval property gives
∞
\
In ̸= ∅
n=1
Let x ∈ ∞
T
n=1 In , since an ∈ In and x ∈ In the distance |an − x| must be less then the
length |In |. and since the length goes to zero |an − x| can be made less then any ϵ.
(c) If we can show lim s2n = lim s2n+1 = s that will imply lim sn = s since each sn is either
in (s2n+1 ) or in (s2n ) as n is must be even or odd.
We have s2n+1 ≤ a1 since
(c) 1 − 43 + 46 − 85 + 6
10
− 7
12
+ ···
(d) 1 + 12 − 13 + 14 + 15 − 61 + 17 + 18 − 19 + · · ·
1
(e) 1 − 22
+ 13 − 1
42
+ 51 − 1
62
+ 17 − 1
82
+ ···
Solution
(a) Converges by a comparison test with ∞ 1
P
n=1 2n
.
P∞ 1
(b) Converges by a comparison test with n=1 n2
.
(c) Diverges since (n + 1)/2n = 1/2 + 1/2n never gets smaller then 1/2.
(e) Intuitively this should diverge since it is a mixture of 1/n (divergent) and 1/n2 (con-
vergent). To make this rigorous examine the subsequence (s2n )
A comparison test with the harmonic series (after some manipulation) shows that (tn )
diverges, and p-series tells us (vn ) converges. Therefore their difference s2n = tn − vn
must diverge, which implies (sn ) diverges as desired.
Exercise 2.7.3
(a) Provide the details for the proof of the Comparison Test (Theorem 2.7.4) using the
Cauchy Criterion for Series.
(b) Give another proof for the Comparison Test, this time using the Monotone Convergence
Theorem.
Solution
Suppose an , bn ≥ 0, an ≤ bn and define sn = a1 + · · · + an , tn = b1 + · · · + bn .
2.7. PROPERTIES OF INFINITE SERIES 57
(b) Since (tn ) → t. This implies that sn is bounded, and since sn ≤ tn implies sn ≤ t by
the order limit theorem, we can use the monotone convergence theorem to conclude
(sn ) converges.
Exercise 2.7.4
Give an example of each or explain why the request is impossible referencing the proper
theorem(s).
P P P
(a) Two series xn and yn that both diverge but where xn yn converges.
P P
(b) A convergent series xn and a bounded sequence (yn ) such that xn yn diverges.
P P P
(c) Two sequences (xn ) and (yn ) where xn and (xn + yn ) both converge but yn
diverges.
Solution
1/n2
P P
(a) xn = 1/n and yn = 1/n have their respective series diverge, but xn y n =
converges since it is a p-series with p > 1.
Exercise 2.7.5 P
Prove the series ∞ p
n=1 1/n converges if and only if p > 1. (Corollary 2.4.7)
Solution
Eventually we have 1/np < 1/pn P
for p > 1 (polynomial vs exponential) meaning we can use
the comparison test to conclude ∞ p
n=1 1/n converges if p > 1.
Now suppose p ≤ 1, since 1/np ≤ 1/n a comparsion test with the harmonic series implies
p ≤ 1 diverges.
Exercise 2.7.6
Let’s say that a series subverges if the sequence of partial sums contains a subsequence that
converges. Consider this (invented) definition for a moment, and then decide which of the
following statements are valid propositions about subvergent series:
58 CHAPTER 2. SEQUENCES AND SERIES
P
(a) If (an ) is bounded, then an subverges.
Solution
(a) False, consider an = 1 then sn = n does not have a convergent subsequence.
(c) True, since sn = nk=1 |ak | converges it is bounded |sn | ≤ M , and since tn = nk=1 ak
P P
is smaller tn ≤ sn it is bounded tn ≤ M which by BW implies there exists a convergent
subsequence (tnk ).
(d) False, an = (1, −1, 2, −2, . . . ) has no convergent subsequence but the sum sn = nk=1 ak
P
has the subsequence (s2n ) → 0.
Exercise 2.7.7 P
(a) Show that if an > 0 and lim (nan ) = l with l ̸= 0, then the series an diverges.
Solution
Note: This is kind of like a wierd way to do a comparison with 1/n and 1/n2 .
(a) If lim(nan ) = l ̸= 0 then nan ∈ (l − ϵ, l + ϵ), setting P
ϵ = l/2 gives nan ∈ (l/2, 3l/2)
implying an > (l/2)(1/n). But if an > (l/2)(1/n) then an diverges as it is a multiple
of the harmonic series. (note that an > 0 ensures l ≥ 0.)
Exercise 2.7.8
Consider each of the following propositions. Provide short proofs for those that are true and
counterexamples for any that are not.
P P 2
(a) If an converges absolutely, then an also converges absolutely.
P P
(b) If an converges and (bn ) converges, then an bn converges.
P P 2
(c) If an converges conditionally, then n an diverges.
Solution
(a) True sinceP(an ) → 0 so eventually a2n ≤ |an | meaning
P 2
an converges by a comparsion
test with |an |.
√ √
(−1)n / nPand bn = (−1)n / n.
P
(b) False, let an = P an converges by the alternating
series test, but an bn = 1/n diverges.
2.7. PROPERTIES OF INFINITE SERIES 59
(c) True, suppose (n2 an ) converges, since (n2 an ) → 0 we have |n2 an | < 1 for n > N ,
implying |an | < 1/n2 . But if |an | < 1/n2 then a comparsion test with 1/n2 implies
an converges
P absolutely, contradicting the assumption that an converges conditionally.
Therefore n2 an must diverge.
(a) Let r′ satisfy r < r′ < 1. Explain why there exists an N such that n ≥ N implies
|an+1 | ≤ |an | r′ .
P P
(c) Now, show that |an | converges, and conclude that an converges.
Solution
(a) We are given
an+1
an − r <ϵ
Since 1 > r′ > r we can set ϵ = r′ − r meaning the neighborhood
an+1
∈ (r − ϵ, r + ϵ) = (2r − r′ , r′ )
an
Is all less then r′ meaning
an+1 ′ ′
an ≤ r =⇒ |an+1 | ≤ r |an |
(b) Let N be large enough that for n > N we have |an+1 | ≤ |an |r′ . Applying this multiple
times gives |an | ≤ (r′ )n−N |aN | which gives
Which converges asPn → ∞ since |r′ | < 1 and |aN | is constant. Implying nk=N |ak |
P
converges and thus nk=1 |ak | also converges since we only omitted finitely many terms.
Show that the left side of this identity at least converges to something. (A complete
proof of this result is taken up in Section 8.3.)
Solution
(a) Rewriting the terms asPan = (1 + 1/n) and using the result from 2.4.10 implies the
product diverges since 1/n diverges.
(b) Converges by the monotone convergence theorem since the partial products are de-
creasing and greater then zero. To show that the product converges to zero, the key
insight is to rewrite each term an = (2n − 1)/(2n) = 1/(2n/(2n − 1)) = 1/bn , where
bn = 2n/(2n − 1). Then the partial products
n
Y 1
pn = an = Q n
i=1 i=1 bn
But
n n
Y Y 1
bn = 1+
i=1 i=1
2n − 1
P
and since
Q 1/(2n − 1) diverges by comparison against a multiple of the harmonic
Qn exists some N so that n ≥ N implies pn < ϵ,
series, bn diverges. Thus, to show there
simply take N large enough so that i=1 bn > 1/ϵ.
Exercise 2.7.11 P P P
Find examples of two series an and bn both of which diverge but for which min {an , bn }
converges. To make it more challenging, produce examples where (an ) and (bn ) are strictly
positive and decreasing.
Solution
Let mn = min{an , bn }. Clearly (mn ) must take an infiniteP amount Pof (an ) and (bn ) terms,
as otherwise removing the finite terms would imply one of an or bn converged.
The key insight is that as long as an , bn > 0, we can simply repeat terms in one sequence
(while letting mn be governed by the other sequence) for as long as we want - say, until we
have enough terms to e.g. sum to 1. Then we can switch the roles of the sequences. To start
with the construction, take some converging series with all terms positive - say, mn = 1/2n .
Then, define the first few terms of mn , an , and bn as:
11
n 1 2 3 [4, 4 + 8 = 12) [12, 12 + 211 ) [12 + 211 , 12 + 211 + 212+2 )
mn 1/2 1/4 1/8 1/2n 1/2n 1/2n
min{an , bn } an bn bn an bn an
n 11
an 1/2 1/2 1/2 1/2 1/2 1/2n
11
bn 1 1/4 1/8 1/8 1/2n 1/212+2
Specifically, min{an , bn } will alternate between following an and bn . Every “block” of the
sequence (examples: each of the last three columns) that isn’t being followed by min{an , bn }
sums to 1. SincePeach block P is finite, min{an , bn } will alternate between an and bn infinitely,
and thus both an and bn will diverge.
For the sake of completeness, an and bn are defined more formally below. Let k1 = 1,
kn = kn−1 + 2kn−1 −1 . Then
( (
1/2n if k2p−1 ≤ n < k2p 1/2n if k2p ≤ n < k2p+1
an = and b n =
1/2k2p −1 if k2p ≤ n < k2p+1 1/2k2p−1 −1 if k2p−1 ≤ n < k2p
Solution
Since xj = sj − sj−1 we can rewrite the sum as
n
X n
X n
X
xj y j = yj (sj − sj−1 ) = sn yn+1 − sm ym−1 + sj (yj − yj+1 )
j=m j=m j=m
In the last part we combine each sj yj term with the −sj yj+1 term which is next in the sum,
then we add some correction terms for the start P and ending points. P
Note the symmetry here, we can turn a sum yj (sj − sj−1 ) into a sum sj (yj − yj+1 )
(at the cost of some correction terms). This is a useful pattern to keep in mind.
62 CHAPTER 2. SEQUENCES AND SERIES
where sn = x1 + x2 + · · · + xn .
Solution
(a) Exercise 2.7.12 combined with s0 = 0 gives
n
X n
X
xk yk = sn yn+1 + sk (yk − yk+1 )
k=1 k=1
as desired.
(b) sn yn+1 clearly converges since yn+1 is “eventually constant”, so we must only show the
right hand side converges.
We will show absolute convergence, note yk − yk+1 ≥ 0 and so
n
X
|sk |(yk − yk+1 ) ≥ 0
k=1
Implying ∞
P
k=1 |sk |(yk − yk+1 ) converges since it is bounded and increasing. And since
the series converges absolutely so does the original ∞
P
k=1 k k − yk+1 ).
s (y
P
Summary: Bound |sk | ≤ M and use the fact that (yk − yk+1 ) is telescoping.
2.7. PROPERTIES OF INFINITE SERIES 63
(a) Point out how the hypothesis of Dirichlet’s Test differs from that of Abel’s Test in
Exercise 2.7.13, but show that essentially the same strategy can be used to provide a
proof.
(b) Show how the Alternating Series Test (Theorem 2.7.7) can be derived as a special case
of Dirichlet’s Test.
Solution P
(a) Abel’s test gets it’s convergence from xn converging, while Dirichlet’s test gets its
convergence from (yn ) → 0. Expanding on that, the proof that the nk=1 sk (yk − yk−1 )
P
term converges is the same in Abel and Dirichlet, but the proof that sn yn+1 differs
depending on if we get our convergence from (yn ) → 0 and (sn ) bounded, or (sn ) → s
and (yn ) bounded.
The proof is almost identical to Abel’s test, bound |sn | < M and use the triangle
inequality on the right hand side to get (note yk − yk+1 > 0 because decreasing)
n
X n
X
|sk |(yk − yk+1 ) ≤ M (yk − yk+1 ) = M (y1 − yn+1 ) ≤ M y1
k=1 k=1
P
Thus sk (yk − yk+1 ) is bounded and increasing, so it converges by MCT. To see the
sn yn+1 term converges, simply note that (yn ) → 0 and |sn | < M .
Thus the original series converges, and furthermore
∞
X ∞
X
xk yk = sk (yk − yk+1 )
k=1 k=1
(−1)n is bounded, so
P
(b) Let an ≥ 0 with a1 ≥ aP 2 ≥ · · · ≥ 0 and lim an = 0. The series
Dirichlet’s test implies (−1)n an converges.
64 CHAPTER 2. SEQUENCES AND SERIES
Solution
By inspection snn = −2 + 1/2n−1 , so limn→∞ snn = −2. This is the same as the result when
fixing j and summing down each column, since each column series has finitely many non-zero
elements.
Exercise 2.8.2
Show that if the iterated series
∞ X
X ∞
|aij |
i=1 j=1
P∞
P∞each fixed i ∈ N the series j=1 |aij | converges to some real
converges (meaning that for
number bi , and the series i=1 bi converges as well), then the iterated series
∞ X
X ∞
aij
i=1 j=1
converges.
Solution
Since ∞
P P∞ P∞
P∞ j=1 aij converges as well; moreover |ci | ≤ |bi |, so i=1 ci
j=1 |aij | converges, ci =
converges by comparison with i=1 |bi |.
Exercise 2.8.3
Define
m X
X n
tmn = |aij |
i=1 j=1
(b) Now, use the fact that (tnn ) is a Cauchy sequence to argue that (snn ) converges.
Solution
(a) Note that tnn is monotone increasing; moreover
∞ X
X ∞ ∞ X
X n n X
X n
|aij | ≥ |aij | ≥ |aij | = tnn
i=1 j=1 i=1 j=1 i=1 j=1
and therefore tnn is bounded; by the Monotone Convergence Theorem tnn converges.
2.8. DOUBLE SUMMATIONS AND PRODUCTS OF INFINITE SERIES 65
(b) Since (tnn ) is a Cauchy sequence, for any ϵ > 0 there exists N such that if p > q > N ,
p p q q
X X X X
ϵ > |tpp − tqq | = |aij | − |aij |
i=1 j=1 i=1 j=1
p p q p q q q q
X X X X X X X X
= |aij | + |aij | + |aij | − |aij |
i=q+1 j=1 i=1 j=q+1 i=1 j=1 i=1 j=1
p p q p
X X X X
= |aij | + |aij |
i=q+1 j=1 i=1 j=q+1
p p q p
X X X X
≥ aij + aij
i=q+1 j=1 i=1 j=q+1
p p q p q q q q
X X X X X X X X
= aij + aij + aij − aij
i=q+1 j=1 i=1 j=q+1 i=1 j=1 i=1 j=1
p p q q
X X X X
= aij − aij
i=1 j=1 i=1 j=1
= |spp − sqq |
Exercise 2.8.5
(a) Show that for all m ≥ N
|(r1 + r2 + · · · + rm ) − S| ≤ ϵ
P∞ P∞
Conclude that the iterated sum i=1 aij converges to S.
j=1
(b) Finish the proof by showing that the other iterated sum, ∞
P P∞
j=1 i=1 aij , converges to
S as well. Notice that the same
P argument can be used once it is established that, for
each fixed column j, the sum ∞ i=1 aij converges to some real number cj .
Solution
(a) For any given m, there must be some N3 such that for n > N3 , k ∈ N ≤ m,
n
X ϵ
rk − akj <
2m
j=1
and thus ∞
P P∞
i=1 j=1 aij converges to S.
P∞
by comparison with ∞
P P∞
(b) i=1 |aij | converges for any fixed jP i=1 k=1 |aik | which con-
∞
verges by the hypothesis, and thus i=1 aij converges to some real number cj . Then
a similar argument to (a) can be used to show that there must be some N such that
when n ≥ N ,
X n
c − S ≤ϵ
j
j=1
and thus ∞
P P∞
j=1 i=1 aij converges to S.
Exercise 2.8.6
(a) P
Assuming the hypothesis - and hence the conclusion - of Theorem 2.8.1, show that
∞
k=2 dk converges absolutely.
2.8. DOUBLE SUMMATIONS AND PRODUCTS OF INFINITE SERIES 67
P∞
(b) Imitate the strategy in the proof of Theorem 2.8.1 to show that k=2 dk converges to
S = limn→∞ snn .
Solution Pn Pn P∞
(a) Note
Pnthat
Pn i=1 j=1 Pa∞ij contains all of the terms of k=2 dk , and thus by comparison
to i=1 j=1 |aij |, k=2 |dk | must converge.
(b) WhatP we need to show is that for allP ϵ > 0 there exists N such that for all n > N ,
|S − nk=2 dk | < ϵ. Note first that n
k=2 dk contains
Pn all the elements of spp when
p ≤ n/2, and that sqq contains all the elements of k=2 dk as long as q ≥ n − 1.
Since (snn ) → S, for arbitrary ϵ1 > 0 we can choose nP
1 large enough such that |sn1 n1 −
S| < ϵ1 . If we choose N = 2n then whenever n > N , nk=2 dk will contain
Pn all terms in
sn1 sn1 , and if we choose m = n − 1 then smm will contain all terms in k=2 dk . Thus
n
X
|dk | − tn1 n1 ≤ tmm − tn1 n1
k=2
where tnn was defined near the start of the proof of Theorem 2.8.1. Moreover since tnn
converges (as proved in Exercise 2.8.3a) and is thus a Cauchy sequence, for arbitrary
ϵ2 > 0, we can also choose n1 large enough to ensure for any m1 > n1 , tm1 m1 −tn1 n1 < ϵ2 .
Putting it all together, choosing ϵ1 = ϵ2 = ϵ/2, n1 large enough to satisfy the two
conditions discussed above, and N = 2n1 :
Xn X n
S − dk ≤ |S − sn1 n1 | + dk − sn1 n1
k=2 k=2
Xn
< ϵ1 + |dk | − tn1 n1
k=2
≤ ϵ1 + tmm − tn1 n1 < ϵ1 + ϵ2
=ϵ
Exercise 2.8.7
Assume that ∞
P P∞
i=1 ai converges absolutely to A, and j=1 bj converges absolutely to B.
(b) Let snn = ni=1 nj=1 ai bj , and prove that limn→∞ snn = AB. Conclude that
P P
∞ X
X ∞ ∞ X
X ∞ ∞
X
ai b j = ai b j = dk = AB,
i=1 j=1 j=1 i=1 k=2
Solution P
(a) Let ∞ ′
P∞ ′
i=1 |ai | converge to A and j=1 |bj | converge to B . By the Algebraic Limit
Theorem for Series,
∞ X ∞ ∞ ∞
! ∞
! ∞ !
X X X X X
|ai bj | = |ai | |bj | = |ai | |bj | = A′ B ′
i=1 j=1 i=1 j=1 i=1 j=1
Basic Topology of R
(b) Give anTexample of a countable collection of open sets {O1 , O2 , O3 , . . .} whose inter-
section ∞n=1 On is closed, not empty and not all of R.
Solution
(a) Taking min{ϵ1 , . . . , ϵN } is only possible for finite sets.
Exercise 3.2.2
Let
n 2
A = (−1) + : n = 1, 2, 3, . . . and B = {x ∈ Q : 0 < x < 1}
n
Answer the following questions for each set:
Solution
(a) The set of B’s limit points is [0, 1]. The set of A’s limit points is {1, −1}.
(b) B is not open since every (a, b) ̸⊆ B and B is not closed since we can construct limits
to irrational values outside B. A is closed since {1, −1} ⊆ A, but not open as it does
not contain any irrationals meaning (a, b) ̸⊆ A for all a, b ∈ R.
69
70 CHAPTER 3. BASIC TOPOLOGY OF R
(c) Every point of A except the limit points {1, −1} is isolated, as if it were not isolated it
would be a limit point. B has no isolated points since B \ [0, 1] = ∅, or in other words
since B is dense in [0, 1] every b ∈ B ⊆ [0, 1] can be reached via a limit.
Exercise 3.2.3
Decide whether the following sets are open, closed, or neither. If a set is not open, find a
point in the set for which there is no ϵ-neighborhood contained in the set. If a set is not
closed, find a limit point that is not contained in the set.
(a) Q.
(b) N.
(c) {x ∈ R : x ̸= 0}.
Solution
(a) Neither, not open as (a, b) ⊆ Q is impossible since Q contains no irrationals but (a, b)
√
does. and not closed since every irrational can be reached as a limit of rationals ( 2
is a simple example).
(b) Clearly not open, but ironically closed since it has no limit points.
(d) Neither, not closed, as the limit nk 1/n2 = π 2 /6 is irrational but every term is rational.
P
and not open as it does not contain any irrationals.
(e) Closed as it has no limit points, every sequence diverges. Not open because it contains
no irrationals.
Exercise 3.2.4
Let A be nonempty and bounded above so that s = sup A exists.
Solution
(a) Since every s − ϵ has an a ∈ A with a > s − ϵ we can find a ∈ Vϵ (s) for any ϵ > 0,
meaning s is a limit point of A and thus contained in A.
(b) No, as (s, s + ϵ) contains no a’s we have Vϵ (s) ̸⊆ A for every ϵ > 0.
3.2. OPEN AND CLOSED SETS 71
Exercise 3.2.5
Prove that a set F ⊆ R is closed if and only if every Cauchy sequence contained in F has
a limit that is also an element of F .
Solution
Let F ⊆ R be closed and suppose (xn ) is a Cauchy sequence in F , since Cauchy sequences
converge (xn ) → x and finally since x ∈ F since F contains its limit points.
Now suppose every Cauchy sequence (xn ) in F converges to a limit in F and let l be a
limit point of F , as l is a limit point of F there exists a sequence (yn ) in F with lim(yn ) = l.
since (yn ) converges it must be Cauchy, and since every Cauchy sequence converges to a
limit inside F we have l ∈ F .
Exercise 3.2.6
Decide whether the following statements are true or false. Provide counterexamples for those
that are false, and supply proofs for those that are true.
(a) An open set that contains every rational number must necessarily be all of R.
(b) The Nested Interval Property remains true if the term “closed interval” is replaced by
“closed set.”
(c) Every nonempty open set contains a rational number.
(d) Every bounded infinite closed set contains a rational number.
(e) The Cantor set is closed.
Solution √ √ √
(a) False, A = (−∞, 2) ∪ ( 2, ∞) contains every rational number but not 2.
(b) False, Cn = [n, ∞) is closed, has Cn+1 ⊆ Cn and Cn ̸= ∅ but ∞
T
n=1 Cn = ∅.
(c) True, let x ∈ A since A is open we have (a, b) ⊆ A with x ∈ (a, b) the density theorem
implies there exists an r ∈ Q with r ∈ (a, b) and thus r ∈ A.
√ √
(d) False, A = {1/n + 2 : n ∈ N} ∪ { 2} is closed and contains no rational numbers.
(e) True, as it is the intersection of countably many closed intervals.
Exercise 3.2.7
Given A ⊆ R, let L be the set of all limit points of A.
(a) Show that the set L is closed.
(b) Argue that if x is a limit point of A∪L, then x is a limit point of A. Use this observation
to furnish a proof for Theorem 3.2.12.
Solution
(a) Every xn ∈ L is xn = limm→∞ amn for amn ∈ A. Meaning if lim xn = x then for n > N
and m > M we have
|amn − x| ≤ |amn − xn | + |xn − x| < ϵ/2 + ϵ/2 = ϵ
and thus x is a limit point of A, so x ∈ L.
72 CHAPTER 3. BASIC TOPOLOGY OF R
(b) Let xn ∈ A ∪ L and x = lim xn . Since xn is infinite there must be at least one
subsequence (xnk ) → x which is either all in A or all in L. If every xnk ∈ L then we
know x ∈ L from (a), and if every xnk ∈ A then x ∈ L aswell.
Exercise 3.2.8
Assume A is an open set and B is a closed set. Determine if the following sets are definitely
open, definitely closed, both, or neither.
(a) A ∪ B
(b) A\B = {x ∈ A : x ∈
/ B}
(d) (A ∩ B) ∪ (Ac ∩ B)
c
(e) A ∩ Ac
Solution
For all of these keep in mind the only open and closed sets are R and ∅, and if A is open Ac
is closed and vise versa.
(b) Open since B being closed implies B c is open and thus A ∩ B c is open as it is an
intersection of open sets.
(c) De Morgan’s laws give (Ac ∪ B)c = A ∩ B c which is the same as (b)
(b) Now, provide the details for the proof of Theorem 3.2.14.
3.2. OPEN AND CLOSED SETS 73
Solution c
/ Eλ for all λ ∈ Λ implying x ∈ Eλc
S S
(a) If x ∈ λ∈Λ Eλ then x ∈
/ λ∈Λ
T Eλ meaning x∈
for all λ ∈ Λ and so finally x ∈ λ∈Λ Eλc . This shows
!c
[ \
Eλ ⊆ Eλc
λ∈Λ λ∈Λ
Eλc then x ∈
T
S the reverse inclusion suppose x ∈
To show λ∈Λ / Eλ for all λ meaning
x∈/ λ∈Λ Eλ and so the reverse inclusion
!c
\ [
Eλc ⊆ Eλ
λ∈Λ λ∈Λ
Each Fλc
is open by Theorem 3.2.13, thus Theorem 3.2.3 (ii) implies F c is open, and
c c
so (F ) = F is closed.
Exercise 3.2.10
Only one of the following three descriptions can be realized. Provide an example that
illustrates the viable description, and explain why the other two cannot exist.
Solution
(i) Cannot exist because taking any sequence (xn ) BW tells us there exists a convergent
subsequence.
(iii) Impossible, let A ⊆ R and let x be an isolated point of A. From the definition
there exists a δ > 0 with Vδ (x) ∩ A = {x}. in Exercise 1.5.6 we proved there cannot
exist an uncountable collection of disjoint open intervals, meaning we cannot have an
uncountable set of isolated points as we can map them to open sets in a 1-1 fashion.
Exercise 3.2.11
(a) Prove that A ∪ B = A ∪ B.
74 CHAPTER 3. BASIC TOPOLOGY OF R
(b) Does this result about closures extend to infinite unions of sets?
Solution
(a) Recall that the set of limit points of a set is closed (Exercise 3.2.7). Let L be the set of
limit points of A ∪ B and let La , Lb be the set of limit points for A and B respectively.
Let x ∈ L, thus there exists a sequence xn ∈ A ∪ B with x = lim xn , since (xn ) is
infinite there exists a subsequence (xnk ) where every term is in A or B. Thus the
limit lim(xnk ) = x must be a limit point of A or B meaning x ∈ La ∪ Lb . This shows
A ∪ B ⊆ A ∪ B.
Now let x ∈ La (Lb is the same). there exists a sequence xn ∈ A with x = lim xn , now
since xn ∈ A ∪ B as well, x ∈ L. Thus we have shown A ∪ B ⊆ A ∪ B completing the
proof.
∞
[ ∞
[
An = {1/n : n ∈ N} ∪ {0}, but An = {1/n : n ∈ N}
n=1 n=1
Exercise 3.2.12
Let A be an uncountable set and let B be the set of real numbers that divides A into two
uncountable sets; that is, s ∈ B if both {x : x ∈ A and x < s} and {x : x ∈ A and x > s}
are uncountable. Show B is nonempty and open.
Solution
Our primary tool will be that countably infinite unions preserve countability (see Exercise
1.5.3).
Consider B1 = {x ∈ R : (−∞, x) ∩ A is uncountable}. B1 must be nonempty; otherwise,
A = ∞
S
n=1 (−∞, n) ∩ A is a union of countable or finite sets, which by 1.5.3 means that
A is countable (which it isn’t). Note that if x ∈ B1 and y > x, then y ∈ B1 . Moreover,
∃ϵ > 0 so that x − ϵ ∈ B1 ; we can prove this by contradiction. If there is no such ϵ, then
(−∞, x − 1/n) ∩ A must be countable for all n ∈ N; by 1.5.3,
∞
[
(−∞, x − 1/n) ∩ A = (−∞, x) ∩ A
n=1
is also countable, a contradiction. Therefore, B1 is open. Now note that B1 must be of the
form (−∞, b1 ), where b1 = inf B1 (or −∞ if inf B1 is undefined). Similarly, B2 = {x ∈ R :
(x, ∞) ∩ A is uncountable} is of the form (b2 , ∞).
Note that B1 ∪ B2 = R; therefore b1 > b2 and so B = B1 ∩ B2 ̸= ∅. Moreover since B1
and B2 are both open, so is B.
Exercise 3.2.13
Prove that the only sets that are both open and closed are R and the empty set ∅.
3.2. OPEN AND CLOSED SETS 75
Solution
Let A ̸= ∅ be open and closed, and suppose for contradiction that A ̸= R and r ∈ / A. Note
that every closed set must contain its supremum and infimum, but Exercise 3.2.4b shows
that every open set cannot contain its supremum or its infimum; thus A must be unbounded.
A ∩ (−∞, r) is open and closed since A ∩ (−∞, r) is an intersection of open sets, and
A ∩ (−∞, r) = A ∩ (−∞, r] (since r ∈ / A) is an intersection of closed sets. Moreover, since
A is unbounded below, A ∩ (−∞, r) ̸= ∅.
Attempting to take s = sup A ∩ (−∞, r) gives a contradiction, since s ∈ A ∩ (−∞, r)
(because closed and bounded above) we can find ϵ > 0 with Vϵ (s) ⊆ A ∩ (−∞, r) (because
open) which contradictions s being an upper bound of A ∩ (−∞, r).
Therefore if A ̸= ∅ we must have A = R. The converse is simple, suppose A ̸= R is
open and closed, this happens iff Ac is open and closed, but since Ac ̸= ∅ we have Ac = R
implying A = ∅.
Exercise 3.2.14
A dual notion to the closure of a set is the interior of a set. The interior of E is denoted E ◦
and is defined as
E ◦ = {x ∈ E : there exists Vϵ (x) ⊆ E}
Results about closures and interiors possess a useful symmetry.
(a) Show that E is closed if and only if E = E. Show that E is open if and only if E ◦ = E.
c
(b) Show that E = (E c )◦ , and similarly that (E ◦ )c = E c .
Solution
(a) (i) If E = E then E contains its limit points and so is closed. If E is closed then E
contains its limit points so E = E.
(ii) If E ◦ = E then every x ∈ E has Vϵ (x) ⊆ E therefore E is open. If E is open then
every x ∈ E has Vϵ (x) ⊆ E therefore E = E ◦ .
c
(b) x ∈ E iff x ∈ / E and x is not a limit point of E, x ∈ (E c )◦ iff x ∈/ E and there
c
exists Vϵ (x) ⊆ E . Notice “x is not a limit point of E” is equivalent to “there exists
Vϵ (x) ⊆ E c ” therefore the sets are the same.
To show (E ◦ )c = E c let D = E c yielding ((Dc )◦ )c = D taking the complement of both
c
sides yields (Dc )◦ = D which we showed earlier.
Exercise 3.2.15
A set A is called an Fσ set if it can be written as the countable union of closed sets. A set
B is called a Gδ set if it can be written as the countable intersection of open sets.
(b) Show that the half-open interval (a, b] is both a Gδ and an Fσ set.
(c) Show that Q is an Fσ set, and the set of irrationals I forms a Gδ set. (We will see in
Section 3.5 that Q is not a Gδ set, nor is I an Fσ set.)
76 CHAPTER 3. BASIC TOPOLOGY OF R
Solution
(a) [a, b] = ∞
T
n=1 (a − 1/n, b + 1/n)
(b) (a, b] = ∞
T S∞
n=1 (a, b + 1/n) = n=1 [a + 1/n, b]
Applying De Morgan’s laws combined with the complement of a closed set being open
we get
\∞
c
Q = [rn , rn ]c
n=1
Exercise 3.3.2
Decide which of the following sets are compact. For those that are not compact, show how
Definition 3.3.1 breaks down. In other words, give an example of a sequence contained in
the given set that does not possess a subsequence converging to a limit in the set.
(a) N.
Solution
(a) Not compact, the sequence xn = n in N has no convergent subsequence in N.
√
(b) Not compact, as we can construct a sequence (xn ) → 1/ 2 ∈
/ Q ∩ [0, 1] implying K is
not closed, and thus cannot be compact.
3.3. COMPACT SETS 77
(c) Compact, sinceTthe cantor set is bounded and closed since it is the infinite intersection
∞
of closed sets n=1 Cn where C1 = [0, 1/3] ∪ [2/3, 1] etc where you keep removing the
middle thirds of each interval.
(d) Not compact as every sequence (xn ) contained in the set converges to π 2 /6 which is
not in the set, meaning the set isn’t closed and thus cannot be compact.
(e) Compact since it is bounded and closed, with every sequence in the set converging to
one.
Exercise 3.3.3
Prove the converse of Theorem 3.3.4 by showing that if a set K ⊆ R is closed and bounded,
then it is compact.
Solution
Let K be closed and bounded and let (xn ) be a sequence contained in K. BW tells us a
convergent subsequence (xnk ) → x exists since K is bounded, and since K is closed x ∈ K.
Thus every sequence in K contains a subsequence convering to a limit in K, which is the
definition of K being compact.
Exercise 3.3.4
Assume K is compact and F is closed. Decide if the following sets are definitely compact,
definitely closed, both, or neither.
(a) K ∩ F
(b) F c ∪ K c
(c) K\F = {x ∈ K : x ∈
/ F}
(d) K ∩ F c
Solution
(a) Compact since K ∩ F is closed (finite intersection of closed sets) and bounded (since
K is bounded)
(b) Closed but not Compact since K being bounded implies K c is unbounded, meaning
F c ∪ K c is unbounded.
Exercise 3.3.5
Decide whether the following propositions are true or false. If the claim is valid, supply a
short proof, and if the claim is false, provide a counterexample.
(c) Let A be arbitrary, and let K be compact. Then, the intersection A ∩ K is compact.
Solution
(a) True, as it will be bounded and closed (since arbitrary intersections of closed sets are
closed).
(b) False, ∞
S
n=1 [0, n] is unbounded and thus not compact.
(c) False, let K = [0, 1] and A = (0, 1). The intersection K ∩ A = (0, 1) is not compact.
(d) False as ∞
T
n=1 [n, ∞) = ∅ (It is true for compact sets though)
Exercise 3.3.6
This exercise is meant to illustrate the point made in the opening paragraph to Section 3.3.
Verify that the following three statements are true if every blank is filled in with the word
“finite.” Which are true if every blank is filled in with the word “compact”? Which are true
if every blank is filled in with the word “closed”?
(c) If {An : n ∈ N} is a collection of setsTwith the property that every finite subcol-
lection has a nonempty intersection, then ∞ n=1 An is nonempty as well.
Solution
(a) Finite (by taking the maximum value), compact (by taking the supremum, which exists
because of boundedness and is in the set because of closed-ness), but not closed (R is
closed)
(b) Finite (by just enumerating through all possibilities of a + b). Compact - boundedness
is obviously preserved. For closed-ness, note that if c ∈ A + B then we can find a
convergent sequence (an + bn ) → c in A + B. Now since (an ) is bounded there must be
a convergent subsequence which converges to some a ∈ A, and taking the corresponding
elements in (bn ) we have a new bounded sequence from which we can get a convergent
subsequence which converges to some b ∈ B. Taking the corresponding elements in
(an + bn ) shows that c = a + b. This argument fails if A and B are only closed, as
boundedness no longer applies. E.g. for A = {n : n ∈ N}, B = {−n + 1/n : n ∈ N}
then the sequence (1/n) is in A + B which converges to 0, which is not in A + B.
(c) Finite - the only way every finite subcollection has a nonempty intersection
Tn is if there
is at least one element all sets include. Compact, since letting Kn = k=1 Ak gives
3.3. COMPACT SETS 79
Exercise 3.3.7
As some more evidence of the surprising nature of the Cantor set, follow these steps to show
that the sum C + C = {x + y : x, y ∈ C} is equal to the closed interval [0, 2]. (Keep in mind
that C has zero length and contains no intervals.)
Because C ⊆ [0, 1], C + C ⊆ [0, 2], so we only need to prove the reverse inclusion [0, 2] ⊆
{x + y : x, y ∈ C}. Thus, given s ∈ [0, 2], we must find two elements x, y ∈ C satisfying
x+y =s
(a) Show that there exist x1 , y1 ∈ C1 for which x1 + y1 = s. Show in general that, for an
arbitrary n ∈ N, we can always find xn , yn ∈ Cn for which xn + yn = s.
(b) Keeping in mind that the sequences (xn ) and (yn ) do not necessarily converge, show
how they can nevertheless be used to produce the desired x and y in C satisfying
x + y = s.
Solution
(a) Recall the definition for sets A and B, A + B = {x + y : x ∈ A, y ∈ B}. Note that
[a, b] + [c, d] = [a + c, b + d]. Recall C1 = [0, 1/3] ∪ [2/3, 1]; we have
C1 + C1 = ([0, 1/3] + [0, 1/3]) ∪ ([2/3, 1] + [2/3, 1]) ∪ ([0, 1/3] + [2/3, 1])
= [0, 2/3] ∪ [4/3, 2] ∪ [2/3, 4/3]
= [0, 2]
We prove the general case through induction; the base case has been demonstrated
above. Define A + r = {x + r : x ∈ A}, r · A = {rx : x ∈ A} for r ∈ R. Note that
[a, b] + c = [a + c, b + c] and r · (A + B) = r · A + r · B.
The inductive hypothesis is that Cn + Cn = [0, 2]. We can complete the inductive step
by noticing that if we scale up Cn+1 a factor of 3, we simply get two copies of Cn , with
one being offset 2 away; this ultimately makes it easy to express Cn+1 in terms of Cn .
′
Let Cn+1 = Cn ∪ (Cn + 2) = 3 · Cn+1 be this upscaled Cn+1 . Then
′ ′
Cn+1 + Cn+1 = (Cn + Cn ) ∪ (Cn + (Cn + 2)) ∪ ((Cn + 2) + (Cn + 2))
= [0, 2] ∪ ([0, 2] + 2) ∪ ([0, 2] + 4)
= [0, 6]
1 ′ ′ 1
· (Cn+1 + Cn+1 ) = Cn+1 + Cn+1 = · [0, 6] = [0, 2]
3 3
and the inductive step is complete.
80 CHAPTER 3. BASIC TOPOLOGY OF R
(b) Since C is compact, there exists a subsequence (xnk ) → x with x ∈ C. Now since
xnk + ynk = s for all k, we have lim ynk = lim s − xnk = s − x. Now since each ynk ∈ C
the limit y = s − x ∈ C as well, thus we have found x, y ∈ C with x + y = s.
Exercise 3.3.8
Let K and L be nonempty compact sets, and define
d = inf{|x − y| : x ∈ K and y ∈ L}
This turns out to be a reasonable definition for the distance between K and L.
(a) If K and L are disjoint, show d > 0 and that d = |x0 − y0 | for some x0 ∈ K and y0 ∈ L.
(b) Show that it’s possible to have d = 0 if we assume only that the disjoint sets K and L
are closed.
Solution
(a) The set |K − L| = {|x − y| : x ∈ K, y ∈ L} is compact since K − L is compact by 3.3.6
(b) and | · | preserves compactness. Thus d = inf |K − L| has d = |x0 − y0 | for some
x0 ∈ K and y0 ∈ L.
Exercise 3.3.9
Follow these steps to prove that being compact implies every open cover has a finite subcover.
Assume K is compact, and let {Oλ : λ ∈ Λ} be an open cover for K. For contradiction,
let’s assume that no finite subcover exists. Let I0 be a closed interval containing K.
(a) Show that there exists a nested sequence of closed intervals I0 ⊇ I1 ⊇ I2 ⊇ · · · with
the property that, for each n, In ∩ K cannot be finitely covered and lim |In | = 0.
(c) Because x ∈ K, there must exist an open set Oλ0 from the original collection that
contains x as an element. Explain how this leads to the desired contradiction.
Solution
(a) Bisect I0 into two intervals, and let I1 be the interval where I1 ∩ K cannot be finitely
covered. Repating in this fashion we have lim |In | = lim |I0 |(1/2)n = 0.
(c) Since x ∈ Oλ0 and |In | → 0 with x ∈ In for all n, there exists an N where n > N
implies In ⊆ Oλ0 contradicting the assumption that In ∩ K cannot be finitely covered
since {Oλ0 } is a finite subcover for In ∩ K.
3.3. COMPACT SETS 81
Exercise 3.3.10
Here is an alternate proof to the one given in Exercise 3.3.9 for the final implication in the
Heine-Borel Theorem.
Consider the special case where K is a closed interval. Let {Oλ : λ ∈ Λ} be an open cover
for [a, b] and define S to be the set of all x ∈ [a, b] such that [a, x] has a finite subcover from
{Oλ : λ ∈ Λ}.
(a) Argue that S is nonempty and bounded, and thus s = sup S exists.
(b) Now show s = b, which implies [a, b] has a finite subcover.
(c) Finally, prove the theorem for an arbitrary closed and bounded set K.
Solution
(a) S is nonempty since x = a has the finite subcover {Oλ0 } for a ∈ Oλ0 . S is bounded
since x ≤ b for all x ∈ S.
(b) Suppose for contradiction that s < b, letting s ∈ Oλ0 implies [a, s] is finitely coverable
since we can take the finite cover of an x ∈ Oλ0 with x < s. This is causes a contradic-
tion however since there exist points y > s with y ∈ Oλ0 meaning [a, y] is also finitely
coverable. therefore the only option is s = b, since any s < b doesn’t work.
(c) (a) still works, for (b) we must also consider the case where y does not exist / there is
a gap. Let y = inf [s, b] ∩ K and suppose y ̸= s. since y ∈ [s, b] ∩ K we know
therefore if {Oλ1 , . . . , Oλn } covered [a, s] then letting y ∈ Oλn+1 would give the finite
cover {Oλ1 , . . . , Oλn+1 } contradicting the assumption that s < b, therefore s = b is the
only option, and so K can be finitely covered.
Exercise 3.3.11
Consider each of the sets listed in Exercise 3.3.2. For each one that is not compact, find an
open cover for which there is no finite subcover.
Solution
(a) N and {V1 (n) : n ∈ N} has no finite subcover since each V1 (n) covers exactly one
n ∈ N, meaning there are no subcovers at all!
√
(b) Q ∩ [0, 1]: Choose some y ∈ Qc ∩ [0, 1], for example y = 2/2. Consider the open
cover {(−1, y)} ∪ {(y + 1/n, 2) : n ∈ N}. Since Q is dense in R, for any finite subcover
there must be some rational number q ∈ (y, y + 1/n) where n is finite.
(c) The Cantor is compact
n o
(d) K = {1 + 1/22 + 1/32 + · · · + 1/n2 : n ∈ N} and V 1 |x−L| (x) : x ∈ K for L = π 2 /6
n o2
since any finite cover V 1 |x1 −L| (x1 ), . . . , V 1 |xn −L| (xn ) , lettingϵ = min{ 21 |xi − L||} will
2 2
make Vϵ (L) not in the finite cover, meaning there exists an x ∈ Vϵ (L) with x ∈ K
(since K gets arbitrarily close to L) but x not in the finite cover.
82 CHAPTER 3. BASIC TOPOLOGY OF R
Exercise 3.3.12
Using the concept of open covers (and explicitly avoiding the Bolzano-Weierstrass Theorem),
prove that every bounded infinite set has a limit point.
Solution
Let A be an infinite set bounded by M (i.e. |x| < M for all x ∈ A), suppose for contradiction
that A has no limit points, meaning there exists an ϵ > 0 such that Vϵ (x) ∩ A = {x} for all
x ∈ A.
This immediately implies there are only a finite number of sets in our cover, otherwise
the union would be unbounded. Contradiction.
TODO Finish tikz picture and proof (its visually obivous)
Exercise 3.3.13
Let’s call a set clompact if it has the property that every closed cover (i.e., a cover consisting
of closed sets) admits a finite subcover. Describe all of the clompact subsets of R.
Solution
K is clompact if and only if K is finite, since the closed cover {[x, x] : x ∈ K} = K having
a finite subcover implies {[x, x] : x ∈ K} is finite (since it is the only subcover that works)
therefore K is finite. If K is finite then it obviously permits a finite subcover.
Solution
Recall a perfect set is a closed set with no isolated points. Thus the intersection of a closed
set P and a closed bounded set K gives a closed bounded (and thus compact) set P ∩ K.
Now take P = R, we get P ∩ K = K which is not nessesarily perfect.
Exercise 3.4.2
Does there exist a perfect set consisting of only rational numbers?
Solution
No, since any nonempty set P ⊆ Q is countable but, nonempty perfect sets are uncountable
by Theorem 3.4.3
Exercise 3.4.3
Review the portion of the proof given in Example 3.4.2 and follow these steps to complete
the argument.
3.4. PERFECT SETS AND CONNECTED SETS 83
(b) Finish the proof by showing that for each n ∈ N, there exists xn ∈ C ∩ Cn , different
from x, satisfying |x − xn | ≤ 1/3n .
Solution
(a) Noting that C1 is the union of disjoint intervals of length 1/3, and that C2 divides each
interval in C1 into two, consider the intervals [a, b] ⊆ C1 and [c, d] ⊆ C2 that x is in.
Then choose x1 to be any other point c ∈ C ∩ ([a, b]\[c, d]) - i.e. it shares an interval
with x in C1 but is in a different interval in C2 ; therefore it must be within 1/3 of x
but is different from x.
(b) Identical argument to part (a), replacing C1 with Cn , C2 with Cn+1 , 1/3 with 1/3n ,
and x1 with xn .
Exercise 3.4.4
Repeat the Cantor construction from Section 3.1 starting with the interval [0, 1]. This time,
however, remove the open middle fourth from each component.
(b) Using the algorithms from Section 3.1, compute the length and dimension of this
Cantor-like set.
Solution
(a) The proofs that the Cantor set is compact and perfect can be copied and applied nearly
word for word here. The Cantor-like set is obviously bounded, and it is closed because it
is the intersection of countably many closed sets (see Exercise 3.2.6e); therefore it must
be compact. Using the same strategy as Exercise 3.4.3, for any x in the Cantor-like
set, we can find a sequence (xn ) where xn ̸= x but |xn − x| ≤ (3/8)n .
Exercise 3.4.5
Let A and B be nonempty subsets of R. Show that if there exist disjoint open sets U and
V with A ⊆ U and B ⊆ V , then A and B are separated.
Solution
Disjoint open sets are separated, therefore so are their subsets.
84 CHAPTER 3. BASIC TOPOLOGY OF R
Exercise 3.4.6
Prove that A set E ⊆ R is connected if and only if, for all nonempty disjoint sets A and
B satisfying E = A ∪ B, there always exists a convergent sequence (xn ) → x with (xn )
contained in one of A or B, and x an element of the other. (Theorem 3.4.6)
Solution
Both are obvious if you think about the definitions, here’s some formal(ish) garbage though
Suppose A ∪ B is nonempty and let x be an element in both, x ∈ B implies x ∈ / A
therefore x ∈ L (the set of limit points of A) meaning there must exist a sequence (xn ) → x
contained in A.
Now suppose there exists an (xn ) → x in A with limit in B, then clearly A ∩ B ⊆ {x} is
nonempty.
Exercise 3.4.7
A set E is totally disconnected if, given any two distinct points x, y ∈ E, there exist separated
sets A and B with x ∈ A, y ∈ B, and E = A ∪ B.
Solution
(a) Let x, y ∈ Q, and let z ∈ (x, y) with z ∈ I. The sets A = (−∞, z) ∩ Q and B =
(z, ∞) ∩ Q are separated and have A ∪ B = Q.
(b) Now let x, y ∈ I, and let z ∈ (x, y) with z ∈ Q. The sets A = (−∞, z) ∩ I and
B = (z, ∞) ∩ I are separated and have A ∪ B = I.
Exercise 3.4.8
Follow these steps to show that
T∞ the Cantor set is totally disconnected in the sense described
in Exercise 3.4.7. Let C = n=0 Cn , as defined in Section 3.1.
(a) Given x, y ∈ C, with x < y, set ϵ = y − x. For each n = 0, 1, 2, . . ., the set Cn consists
of a finite number of closed intervals. Explain why there must exist an N large enough
so that it is impossible for x and y both to belong to the same closed interval of CN .
Solution
(a) Since the length of every interval goes to zero, we set N large enough that the length
of every interval is less then ϵ, meaning x and y cannot be in the same interval.
(b) Obvious
Exercise 3.4.9
Let {r1 , r2 , r3 , . . .} be an
S enumeration of the rational numbers, and for each n ∈ N set
ϵn = 1/2n . Define O = ∞ V
n=1 ϵn (rn ), and let F = O c
.
(a) Argue that F is a closed, nonempty set consisting only of irrational numbers.
3.4. PERFECT SETS AND CONNECTED SETS 85
(b) Does F contain any nonempty open intervals? Is F totally disconnected? (See Exercise
3.4.7 for the definition.)
(c) Is it possible to know whether F is perfect? If not, can we modify this construction to
produce a nonempty perfect set of irrational numbers?
Solution
(a) O is a union of an arbitrary collection of open sets and therefore is open; therefore
Oc = F is closed. O contains all rational numbers, therefore F must consist only of
irrational numbers. Intuitively, F must be nonempty, because the sum of the lengths
of the intervals in O is 2 and can’t cover the infinite real line - but the properties
of interval lengths haven’t been rigorously defined, and maybe something weird can
happen (e.g. Banach-Tarski paradox), so it’s best to prove it.
Let On = ni=1 Vϵi (ri ), Fn = Onc , and Gn = Fn ∩ [1, 3). Clearly Gn ⊇ Gn+1 and Gn is
S
compact, so if we can showTGn is always T∞ nonempty, we can use the Nested Compact
∞
Set Property to show F = n=1 Fn ⊇ n=1 Gn is nonempty.
Consider the set M = {m/2n−1 : m ∈ N} ∪ [1, 3), a set of evenly spaced numbers with
1/2n−1 between each number. Some arithmetic shows that there are 2n elements in
M . Since the length of Oi is 1/2i−1 , there can only be at most 2n−i elements of M
in Oi , and therefore 2n − 1 elements of M in O. Since Gn ⊇ M \O, Gn is nonempty,
completing the proof.
(b) Since F contains no rational numbers, and any nonempty open interval will contain
rational numbers (since Q is dense in R), F cannot contain any open interval. The
proof that F is totally disconnected is the same as that for I in Exercise 3.4.7b.
(c) It is possible
√ for F to be not perfect. Our approach will be to create an isolated point
in F , say, 2. To do so, we need to spilt the√sequence of sets in O into three parts; the
√ limits approach 2 from below, the second will have their
first will have their upper
lower limits approach 2 from above, and the third will be used to enumerate through
the rest of the rational numbers.
First, we prove the following lemma - for any L ∈ R and a positive sequence (ζn ) → 0, it
that [L, L + 2ζ1 )\ ∞
S
is possible to construct a sequence (xn ) ∈ Q such √ i=1 ζi (xi ) = {L}.
V
(We’ll use this lemma for the sequences isolating 2.) To do so, let Vζn (xn ) = (αn , βn ).
Choose L < αn < L + 2ζn+1 .
We show by induction that [L, L+2ζ1 )\ ni=1 Vζi (xi ) ⊆ [L, S
S
L+2ζn+1 ). The base case n =
1 is trivial. For the inductive case: assume [L, L + 2ζ1 )\ ni=1 Vζi (xi ) ⊆ [L, L + 2ζn+1 ).
Then
n+1
[
[L, L + ζ1 )\ Vζi (xi ) ⊆ [L, L + 2ζn+1 )\Vζn+1 (xn+1 ) = [L, L + 2ζn+1 )\(αn+1 , βn+1 )
i=1
Since αn+1 > L, βn+1 > L + 2ζn+1 . Recall also that αn+1 < L + 2ζn+2 , and therefore
[L, L + 2ζn+1 )\(αn+1 , βn+1 ) ⊆ [L, L + 2ζn+2 ), completing the inductive step.
86 CHAPTER 3. BASIC TOPOLOGY OF R
Now, for any l ∈ [L, L + 2ζ1 ) > L, since (ζn ) → 0 there must be some ζj so that
L + 2ζj ≤ l and j > 1, and therefore
l∈
/ [L, L + 2ζj )
j−1
[
⊇ [L, L + 2ζ1 )\ Vζi (xi ) (by induction above)
i=1
[∞
⊇ [L, L + 2ζ1 )\ Vζi (xi )
i=1
∞
[
=⇒ l ∈
/ [L, L + 2ζ1 )\ Vζi (xi )
i=1
S∞
Also, since αn > L ∀n, L ∈ [L, L + 2ζ1 )\ i=1 Vζi (xi ). This completes the proof of the
lemma.
Returning to the original
√ problem of making F not perfect, we will construct a sequence
(rn ) which isolates 2. Start with any enumeration of the rational numbers√ √ qn . Our
lemma √ above means we can assign r3n+1 and ϵ3n+1 for n ≥ 0 to ensuring ( 2, 2+2ϵ1 ) ∈
O but 2 itself not in O. Similarly, a√slight modification
√ to the lemma
√ lets us assign
r3n+2 and ϵ3n+2 for n ≥ 0 to ensuring ( 2−2ϵ2 , 2) ∈ O while leaving 2 out. Finally,
we assign r3n+3 , n ≥ 0 to enumerating through the elements of qn , skipping over any
elements
√ that will be present in r3n+1 and r3n+2 , and deferring any elements that would
cause 2 ∈ V (q) until ϵ becomes small enough that this is no longer the case.
√ √
In this manner, 2 has been surrounded, and F = Oc will have 2 as an isolated
point, and thus F is not perfect.
We can also construct (rn ) so that
Sn F is perfect. (Note: To simplify the notation a bit
let Vi = Vϵi (ri ).) Define Rn = i=1 Vi , with R0 = ∅. We will rely on the lemma that
F is perfect if for all i ∈ N, either Vi ∩ Ri−1 = ∅ or Vi ⊆ Ri−1 . Informally, if Vi isn’t
redundant (in that it covers new numbers), then it is disjoint from all previous Vi . To
prove this lemma, consider any element x ∈ F and let ϵ > 0 be arbitrary. Consider the
interval (x, x + ϵ), and ignore any Vp if Vp ⊆ Rp−1 . Since none of Vi overlap partially,
this interval cannot be covered completely by some union of several Vi since any union
would have gaps. Moreover, (x, x + ϵ) cannot be covered entirely by a single Vϵj , since
then Vj would be centered on rj = x + ϵj which is irrational. Thus, there must be some
other element y ̸= x, y ∈ F so that y ∈ (x, x + ϵ) and therefore x is not an isolated
point and F must be perfect.
To constructSan (rn ) which satisfies this condition, we start with an arbitrary (qn ).
Define Rn = ni=1 ri . For each element of (qn ), we add qi to (rn ) only if either Vϵn (qi ) ⊆
Rn or Vϵn (qi ) ∩ Rn = ∅. Otherwise, we procrastinate on adding qi by appending any
rational number s > max(Rn )+2ϵn . Clearly, for any qi there will eventually be ϵn small
enough that qi can be added to (rn ) without violating our restrictions, and we don’t
need to worry about s’s being added since they’re far enough away from everything
that they can’t affect the restrictions.
3.5. BAIRE’S THEOREM 87
Solution
If A is a Gδ set, then Ac is a Fσ set by De Morgan’s laws. Likewise if A is an Fσ set then Ac
must be a Gδ set.
Exercise 3.5.2
Replace each with the word finite or countable, depending on which is more appropriate.
Solution
(a) Countable, since two countable union can be written as a single countable union over
the diagonal (see Exercise 1.2.4). Another way of seeing this is that we can form a
bijection between N and N2 , therefore a double infinite union can be written as a
single infinite union.
(b) Finite
(c) Finite
(d) Countable, by the same logic as in (a) we can write two countable intersections as a
single countable intersection.
Exercise 3.5.3
(a) Show that a closed interval [a, b] is a Gδ set.
(b) Show that the half-open interval (a, b] is both a Gδ and an Fσ set.
(c) Show that Q is an Fσ set, and the set of irrationals I forms a Gδ set.
Solution
This exercise has already appeared as Exercise 3.2.15.
Exercise 3.5.4
Let {G1 , G2 ,TG3 , . . .} be a countable collection of dense, open sets, we will prove that the
intersection ∞ n=1 Gn is not empty.
Starting with n = 1, inductively construct a nested sequence of closed intervals I1 ⊇ I2 ⊇
I3 ⊇ · · · satisfying In ⊆ Gn . Give special attention to the issue of the endpoints of each In .
Show how this leads to a proof of the theorem.
88 CHAPTER 3. BASIC TOPOLOGY OF R
Solution
Because G1 is open there exists an open interval (a1 , b1 ) ⊆ G1 , letting [c1 , d1 ] be a closed
interval contained in (a1 , b1 ) gives I1 ⊆ G1 as desired.
Now suppose In ⊆ Gn . because Gn+1 is dense and (cn , dn ) ∩ Gn+1 is open there exists
an interval (an+1 , bn+1 ) ⊆ Gn ∩ (cn , dn ). Letting [cn+1 , dn+1 ] ⊆ (an+1 , bn+1 ) gives us our new
closed interval.
This gives us our collection of sets with In+1 ⊆ In , In ⊆ Gn and In ̸= ∅ allowing us to
apply the Nested Interval Property to conclude
∞
\
In ̸= ∅
n=1
T∞
and thus n=1 Gn ̸= ∅ since each In ⊆ Gn .
Exercise 3.5.5
Show that it is impossible to write
∞
[
R= Fn
n=1
Solution
This is just the complement of Exercise 3.5.4, If we had R = ∞
S
n=1 Fn then we would also
have ∞
\
∅= Gn
n=1
Exercise 3.5.6
Show how the previous exercise implies that the set I of irrationals cannot be an Fσ set, and
Q cannot be a Gδ set.
Solution
Recall from 3.5.3 that Q is an Fσ set, suppose for contradiction that I were also an Fσ set.
Then we could write ∞ ∞
[ [
Q= Fn and I = Fn′
n=1 n=1
which gives our desired contradiction, hence I is not an Fσ set and Q is not a Gδ set (take
complements).
Exercise 3.5.7
Using Exercise 3.5.6 and versions of the statements in Exercise 3.5.2, construct a set that is
neither in Fσ nor in Gδ .
Solution
For a set A ⊆ R define −A = {−x : x ∈ A}. Note that if A is closed (open), then so is −A,
and if A is a Fσ set (Gδ ), so is −A.
Define I + = I ∩ [0, ∞), I − = I ∩ (−∞, 0] , and Q+ and Q− be defined similarly for Q.
Consider the set A = I + ∪ Q− . If A is a Fσ set, then by Exercise 3.5.2 so is A ∩ [0, ∞) = I + ,
but so is I + ∪ −I + = I + ∪ I − = I, which is a contradiction. Similarly, A being a Gδ set
implies A ∩ (−∞, 0] = Q− and Q− ∪ −Q− = Q are both Gδ sets, a contradiction.
Exercise 3.5.8
Show that a set E is nowhere-dense in R if and only if the complement of E is dense in R.
Solution
First suppose E is nowhere-dense, then E contains no nonempty open intervals meaning for
every a, b ∈ R we have (a, b) ̸⊆ E meaning we can find a c ∈ (a, b) with c ∈ / E. But this is
c c c
just saying c ∈ E which implies E is dense since for every a, b ∈ R we can find a c ∈ E
with a < c < b.
c c
Now suppose E is dense in R, then then every interval (a, b) contains a point c ∈ E ,
implying that (a, b) ̸⊆ E since c ∈
/ E and c ∈ (a, b). therefore E contains no nonempty open
intervals and so E is nowhere-dense by defintion 3.5.3.
Exercise 3.5.9
Decide whether the following sets are dense in R, nowhere-dense in R, or somewhere in
between.
Solution
(a) between, since A is dense in [0, 5] but not in all of R.
(d) nowhere-dense since the Cantor set is closed, so C = C, and C contains no intervals
as desired.
Chapter 4
(b) Now, write another proof of Corollary 4.2.4 part (ii) directly from Definition 4.2.1
without using the sequential criterion in Theorem 4.2.3.
(c) Repeat (a) and (b) for Corollary 4.2.4 part (iii).
Solution
(a) By the Sequential Criterion for Functional Limits, since limx→c f (x) = L, limx→c g(x) =
M , all sequences (xn ) → c (where every xn ̸= c) have f (xn ) → L and g(xn ) → M ,
which implies that f (xn ) + g(xn ) → L + M by the Algebraic Limit Theorem, which
implies that limx→c [f (x) + g(x)] = L + M .
(b) Let ϵ > 0, set δ1 such that 0 < |x − c| < δ1 implies |f (x) − f (c)| < ϵ/2 and set δ2 such
that 0 < |x − c| < δ2 implies |g(x) − g(c)| < ϵ/2. Now let δ = min{δ1 , δ2 } and use the
triangle inequality to get
|f (x) + g(x) − (f (c) + g(c))| ≤ |f (x) − f (c)| + |g(x) − g(c)| < ϵ/2 + ϵ/2 = ϵ
(c) (a) is the same, if f (xn ) → L and g(xn ) → M then (f (xn )g(xn )) → LM by the
sequential criterion for functional limits.
For (b) we add and subtract f (c)g(x) then factor and use the triangle inequality (this
is a common trick)
91
92 CHAPTER 4. FUNCTIONAL LIMITS AND CONTINUITY
Now we want a few things, (1) to bound |g(x)| (2) to make |f (x) − f (c)| small and
(3) to make |g(x) − g(c)| small. Whenever you want multiple things start thinking
min/max!
In this case, set δ1 so |g(x) − g(c)| < 1 giving the bound |g(x)| < M + 1. Set δ2 so
|g(x) − g(c)| < Mϵ/2
+1
and set δ3 so |f (x) − f (c)| < fϵ/2
(c)
. Finally set δ = min{δ1 , δ2 , δ3 }
to get
|f (x)g(x) − f (c)g(c)| < ϵ/2 + ϵ/2 = ϵ
as desired.
Exercise 4.2.2
For each stated limit, find the largest possible δ-neighborhood that is a proper response to
the given ϵ challenge.
(c) limx→π [[x]] = 3, where ϵ = 1. (The function [[x]] returns the greatest integer less than
or equal to x.)
Solution
(a) |(5x − 6) − 9| = |5x − 15| = 5|x − 3| < 5δ implies δ = 1/5 for ϵ = 1.
√ √
(b) Consider edge cases: We have | 9 − 2| = 1 (x is 5 above)√ and | 1 − 2| = 1 (x is 3
below) leading us to set δ = 3. This δ must work since x is monotone.
(d) Since [[x]] is an integer ϵ = .01 is the same as saying [[x]] = 3. This happens precisely
when x ∈ (3, 4) hence we need δ = min{|π − 3|, |π − 4|} = π − 3.
Exercise 4.2.3
Review the definition of Thomae’s function t(x) from Section 4.1.
1
if x = 0
t(x) = 1/n if x = m/n ∈ Q\{0} is in lowest terms with n > 0
0 if x ∈
/ Q.
(a) Construct three different sequences (xn ) , (yn ), and (zn ), each of which converges to 1
without using the number 1 as a term in the sequence.
(b) Now, compute lim t (xn ) , lim t (yn ), and lim t (zn ).
4.2. FUNCTIONAL LIMITS 93
(c) Make an educated conjecture for limx→1 t(x), and use Definition 4.2.1 B to verify the
claim. (Given ϵ > 0, consider the set of points {x ∈ R : t(x) ≥ ϵ} Argue that all the
points in this set are isolated.)
Solution
(a) xn = (1 + n)/n, yn = 1 − 1/n2 and zn = 1 + 1/2n .
(b) lim t(xn ) = 0 since the size of the denominator becomes arbitrarily large. Same for the
others
(c) I claim limx→1 t(x) = 0. Let ϵ > 0 be arbitrary; we must show there exists a δ where
every |x − 1| < δ has t(x) < ϵ. For x ∈
/ Q we have t(x) = 0 < ϵ, and we can easily set
δ small enough that t(0) = 1 is excluded. That leaves us with the case x ∈ Q in which
case we can write x − 1 = m/n in lowest terms.
To get t(x) = 1/n < ϵ we observe that |m/n| < δ implies t(x) = 1/n ≤ |m/n| < δ so
setting δ = ϵ gives t(x) < ϵ. To complete the proof set δ = min{ϵ, 1}.
Exercise 4.2.4
Consider the reasonable but erroneous claim that
(a) Find the largest δ that represents a proper response to the challenge of ϵ = 1/2
(c) Find the largest ϵ challenge for which there is no suitable δ response possible.
Solution
(a) δ = 13
(b) δ = 1
(c) No matter how small δ is, [[10 − δ/2]] = 9 can be obtained, meaning
Exercise 4.2.5
Use Definition 4.2.1 to supply a proper proof for the following limit statements.
(b) limx→0 x3 = 0
Solution
(Note that I use the largest δ choice that’s easy to use)
(a) Since |3x − 6| = 3|x − 2| setting δ = ϵ/3 gives |3x − 6| < ϵ as desired.
(b) Since |x3 | = |x|3 setting δ = ϵ1/3 gives |x|3 < ϵ as desired.
(c) Since |x2 + x − 6| = |x − 2||x + 3| < δ(5 + δ) setting δ = min{1, ϵ/6} gives δ(5 + δ) <
δ(6) < ϵ as desired. Another approach is to write |x2 + x − 6| in the “(x − 2)n basis”
(c) If limx→a f (x) = L, then limx→a 3[f (x) − 2]2 = 3(L − 2)2
(d) If limx→a f (x) = 0, then limx→a f (x)g(x) = 0 for any function g (with domain equal to
the domain of f .)
Solution
(a) Obviously, since if δ ′ < δ then |x − a| < δ ′ implies |x − a| < δ.
(b) False, consider f (0) = 1 and f (x) = 0 otherwise, the definition of a functional limit
requires |x − a| < δ to imply |f (x) − L| < ϵ for all x not equal to a (This is the
0 < |x − a| part)
(c) True by the algebraic limit theorem for functional limits. (or composition of continuous
functions, but that’s unnecessary here)
(d) False, consider how f (x) = x has limx→0 f (x) = 0 but g(x) = 1/x has limn→0 f (x)g(x) =
1. (Fundementally this is because 1/x is not continuous at 0)
Exercise 4.2.7
Let g : A → R and assume that f is a bounded function on A in the sense that there
exists M > 0 satisfying |f (x)| ≤ M for all x ∈ A. Show that if limx→c g(x) = 0, then
limx→c g(x)f (x) = 0 as well.
4.2. FUNCTIONAL LIMITS 95
Solution
We have |g(x)f (x)| ≤ M |g(x)|, set δ small enough that |g(x)| < ϵ/M to get
ϵ
|g(x)f (x)| ≤ M |g(x)| < M =ϵ
M
for all |x − a| < δ.
Exercise 4.2.8
Compute each limit or state that it does not exist. Use the tools developed in this section
to justify each conclusion.
|x−2|
(a) limx→2 x−2
|x−2|
(b) limx→7/4 x−2
Solution
(a) Does not exist, the sequence xn = 2 + 1/n makes |x − 2|/(x − 2) converge to 1, but
xn = 2 − 1/n makes |x − 2|/x − 2 converge to −1. (x → |x| is not differentiable at zero
for the same reason)
(b) −1. For δ < 1/4 then x < 2 and we just have −1.
(c) Does not exist, let (xn ) = 1/2n and (yn ) = 1/(2n + 1), then clearly lim xn = lim yn = 0
but lim f (xn ) = 1 ̸= lim f (yn ) = −1
√ √
(d) (−1)[[1/x]] is bounded and limx→0 3 x = 0, so by Exercise 4.2.7 limx→0 3 x(−1)[[1/x]] = 0.
√ √
We can also show this directly, since | 3 x(−1)[[1/x]] | = | 3 x| < ϵ when δ = ϵ3 .
(a) Show limx→0 1/x2 = ∞ in the sense described in the previous definition.
(b) Now, construct a definition for the statement limx→∞ f (x) = L. Show limx→∞ 1/x = 0.
Solution √
(a) For a given M > 0, if 0 < |x − 0| = |x| < 1/ M = δ then 1/|x|2 = 1/x2 < M as
desired.
96 CHAPTER 4. FUNCTIONAL LIMITS AND CONTINUITY
(b) limx→∞ f (x) = L means that for all ϵ > 0 we can find a N such that when x > N
it follows that |f (x) − L| < ϵ. For a given ϵ > 0, choosing N = 1/ϵ leaves us with
x > N =⇒ 1/N = ϵ > 1/x hence limx→∞ 1/x = 0.
Exercise 4.2.10
Introductory calculus courses typically refer to the right-hand limit of a function as the limit
obtained by “letting x approach a from the right-hand side.”
(a) Give a proper definition in the style of Definition 4.2.1 for the right-hand and left-hand
limit statements:
lim+ f (x) = L and lim− f (x) = M
x→a x→a
(b) Prove that limx→a f (x) = L if and only if both the right and left-hand limits equal L.
Solution
(a) Let f : A → R, and let c be a limit point of the domain A. We say that limx→c+ f (x) =
L provided that, for all ϵ > 0, there exists a δ > 0 such that whenever 0 < x − c < δ
(and x ∈ A) it follows that |f (x) − L| < ϵ. We say that limx→c− f (x) = L provided
that, for all ϵ > 0, there exists a δ > 0 such that whenever 0 < c − x < δ (and x ∈ A)
it follows that |f (x) − L| < ϵ.
(b) ( =⇒ ) If limx→a f (x) = L then for any ϵ > 0, there exists a δ > 0 so that 0 < |x−c| < δ
implies |f (x) − L| < ϵ. Since both 0 < x − c < δ and 0 < c − x < δ will satisfy the
requirement that 0 < |x − c| < δ, then limx→a+ f (x) = limx→a− f (x) = L.
( ⇐= ) For a given ϵ > 0, there exists δ1 , δ2 > 0 so that either 0 < x − c < δ1 or
0 > x − c > −δ2 implies |f (x) − L| < ϵ. If 0 < |x − c| < δ = min{δ1 , δ2 } then at least
one of the preconditions is always true, so limx→a f (x) = L.
(b) Prove that g is continuous at a point c ̸= 0. (The identity a3 −b3 = (a−b) (a2 + ab + b2 )
will be helpful.)
Solution
(a) Let ϵ > 0 be arbitrary and set δ = ϵ3 . If |x − 0| < δ = ϵ3 then taking the cube root of
both sides gives |x|1/3 < 1/ϵ and since (−x)1/3 = −(x1/3 ) we have |x|1/3 = |x1/3 | < ϵ.
(b) We must make |x1/3 − c1/3 | < ϵ by making |x − c| small. The identity given allows us
to write
|x1/3 − c1/3 | = |x − c| · |x2/3 + x1/3 c1/3 + c2/3 |
If
√ we choose
√ δ < c then 0 < |x| < 2|c|. Keeping in mind that if a > b > 0 then
3
3
a > b, we can now bound
|x1/3 − c1/3 | ≤ |x − c| · K
ϵ
Setting δ = K
gives |x1/3 − c1/3 | ≤ ϵ completing the proof.
Exercise 4.3.2
To gain a deeper understanding of the relationship between ϵ and δ in the definition of
continuity, let’s explore some modest variations of Definition 4.3.1. In all of these, let f be
a function defined on all of R.
(a) Let’s say f is onetinuous at c if for all ϵ > 0 we can choose δ = 1 and it follows that
|f (x)−f (c)| < ϵ whenever |x−c| < δ. Find an example of a function that is onetinuous
on all of R.
(b) Let’s say f is equaltinuous at c if for all ϵ > 0 we can choose δ = ϵ and it follows
that |f (x) − f (c)| < ϵ whenever |x − c| < δ. Find an example of a function that is
equaltinuous on R that is nowhere onetinuous, or explain why there is no such function.
(c) Let’s say f is lesstinuous at c if for all ϵ > 0 we can choose 0 < δ < ϵ and it follows that
|f (x)−f (c)| < ϵ whenever |x−c| < δ. Find an example of a function that is lesstinuous
on R that is nowhere equaltinuous, or explain why there is no such function.
(d) Every lesstinuous function is continuous, since the definition of lesstinuous is just con-
tinuous plus the requirement that 0 < δ < ϵ.
And every continuous function is lesstinuous since if δ > 0 works we can set δ ′ < δ and
δ ′ < ϵ so that |x − c| < δ ′ < δ still implies |f (x) − f (c)| < ϵ
Exercise 4.3.3
(a) Supply a proof for Theorem 4.3.9 (Composition of continuous functions) using the ϵ−δ
characterization of continuity.
(b) Give another proof of this theorem using the sequential characterization of continuity
(from Theorem 4.3.2 (iii)).
Solution
(a) Let f is continuous at c and g be continuous at f (c). We will show g ◦ f is continuous
at c. Let ϵ > 0 be arbitrary, we want |g(f (x)) − g(f (c))| < ϵ for |x − c| < δ. Pick
α > 0 so that |y − f (c)| < α implies |g(y) − g(f (c))| < ϵ (possible since g is continuous
at f (c)) and pick δ > 0 so that |x − c| < δ implies |f (x) − f (c)| < α. Putting all of
this together we have
(b) Let (xn ) → c, we know f (xn ) is a sequence converging to f (c) since f is continuous
at c, and since g is continuous at f (c) any sequence (yn ) → f (c) has g(yn ) → g(f (c)).
Letting yn = f (xn ) gives g(f (xn )) → g(f (c)) as desired.
Exercise 4.3.4
Assume f and g are defined on all of R and that limx→p f (x) = q and limx→q g(x) = r.
(b) Show that the result in (a) does follow if we assume f and g are continuous.
(c) Does the result in (a) hold if we only assume f is continuous? How about if we only
assume that g is continuous?
Solution
(a) Let f (x) = q be constant and define g(x) as
(
(r/q)x if x ̸= q
g(x) =
0 if x = q
4.3. CONTINUOUS FUNCTIONS 99
(c) Let f (x) = 0, then f (x)g(x) = 0 is continuous at zero for any g(x).
(d) Let (
2 if x ≥ 0
f (x) =
1/2 if x < 0
Then f (x) + 1/f (x) = 2.5 is continuous at zero.
1/3
(e) Impossible, if [f (x)]3 was continuous at zero then ([f (x)]3 ) = f (x) would also be
continuous at zero since the composition of continuous functions is continuous
Exercise 4.3.7
(a) Referring to the proper theorems, give a formal argument that Dirichlet’s function
from Section 4.1 is nowhere-continuous on R.
(b) Review the definition of Thomae’s function in Section 4.1 and demonstrate that it fails
to be continuous at every rational point.
(c) Use the characterization of continuity in Theorem 4.3.2 (iii) to show that Thomae’s
function is continuous at every irrational point in R. (Given ϵ > 0, consider the set of
points {x ∈ R : t(x) ≥ ϵ}.)
Solution
Recall Dirichlet’s function is (
1 if x ∈ Q
g(x) =
0 if x ∈ I
And Thomae’s function is
1
if x = 0
t(x) = 1/n if x = m/n in lowest terms with m, n ̸= 0
0 if x ∈ I
(a) Let a ∈ Q and set ϵ = 1. For any δ > 0 there will exist points x ∈ (a − δ, a + δ) ∩ I by
the density of I in R with |f (x) − f (a)| = |0 − 1| = 1 not less then ϵ, therefore there
does not exist a δ to match ϵ = 1 and so f is discontinuous at a. Since a was arbitrary
(the a ∈ I case is identical) g must be discontinuous at all of R.
(b) By the same argument as in (a) for any m/n ∈ Q no matter how small δ is, we can
find an irrational number within δ of m/n meaning ϵ cannot be made smaller then
|f (m/n) − f (x)| = 1/n.
(c) Let a ∈ I, we want to show t(x) < ϵ for |x − a| < δ. I claim the set {x ∈ V1 (a) :
t(x) ≥ ϵ} is finite, this can be seen since the requirement that t(x) ≥ ϵ is the same
as x = m/n and 1/n ≥ ϵ. It is easy to see there are finitely many points like this
(consider how there are finitely many n and finitely many m given n) thus we can say
{x ∈ V1 (a) : t(x) ≥ ϵ} = {x1 , . . . , xn } and set δ = min{|xi − a| : i ∈ {1, . . . , n}} to
ensure every x ∈ Vδ (a) has t(x) < ϵ.
4.3. CONTINUOUS FUNCTIONS 101
Exercise 4.3.8
Decide if the following claims are true or false, providing either a short proof or counterex-
ample to justify each conclusion. Assume throughout that g is defined and continuous on
all of R.
(c) If g (x0 ) > 0 for a single point x0 ∈ R, then g(x) is in fact strictly positive for
uncountably many points.
Solution
(a) True, using the sequential definition for functional limits letting (xn ) → 1 we have
g(xn ) ≥ 0 and g(xn ) → g(1) so by the Order Limit Theorem g(1) ≥ 0
(b) True, since if there was some x ∈ R with g(x) ̸= 0 then g would not be continuous at
x because we could never make ϵ smaller then |g(x) − g(r)| = |g(x)| as we can always
find rational numbers satisfying g(r) = 0 inside any δ-neighborhood.
(c) True, let ϵ < g(x0 ) and pick δ so that every x ∈ Vδ (x0 ) satisfies g(x) ∈ (g(x0 ) −
ϵ, g(x0 ) + ϵ) and thus g(x) > 0 since g(x0 ) − ϵ > 0.
Exercise 4.3.9
Assume h : R → R is continuous on R and let K = {x : h(x) = 0}. Show that K is a closed
set.
Solution
Let (xn ) be a convergent sequence in K and set x = lim xn . Since h is continuous the limit
lim h(xn ) = h(x) and since h(xn ) = 0 for all n lim h(xn ) = 0. therefore h(x) = 0 and x ∈ K.
Exercise 4.3.10
Observe that if a and b are real numbers, then
1
max{a, b} = [(a + b) + |a − b|]
2
(a) Show that if f1 , f2 , . . . , fn are continuous functions, then
is a continuous function.
(b) Let’s explore whether the result in (a) extends to the infinite case. For each n ∈ N,
define fn on R by (
1 if |x| ≥ 1/n
fn (x) =
n|x| if |x| < 1/n
Now explicitly compute h(x) = sup {f1 (x), f2 (x), f3 (x), . . .}
102 CHAPTER 4. FUNCTIONAL LIMITS AND CONTINUITY
Solution
(a) We will prove this by induction. The base case is
1
max{f1 (x), f2 (x)} = [(a + b) + |a − b|]
2
Which is obviously continuous. Now assume max{f1 , . . . , fn−1 } is continuous, letting
m(x) = max{f1 , . . . , fn−1 } we can write
Now since fn (x) and m(x) are continuous functions max{fn , m} is continuous by the
base case!
Which is not continuous at x = 0, therefore (a) does not hold in the infinite case.
In general, if yn+1 = f (yn ), show that the resulting sequence (yn ) is a Cauchy sequence.
Hence we may let y = lim yn .
(c) Prove that y is a fixed point of f (i.e., f (y) = y ) and that it is unique in this regard.
(d) Finally, prove that if x is any arbitrary point in R, then the sequence (x, f (x), f (f (x)), . . .)
converges to y defined in (b).
Solution
(a) Let δ = ϵ/c to get |f (x) − f (y)| ≤ c|x − y| < ϵ whenever |x − y| < δ. (This is the
general proof, we could make it shorter by letting δ = ϵ since 0 < c < 1)
4.3. CONTINUOUS FUNCTIONS 103
(b) We want |yn − ym | < ϵ. Since 0 < c < 1 we have |yn+1 − ym+1 | ≤ c|yn − ym |, implying
|yn − ym | ≤ cm |yn−m+1 − y1 |. Thus if we can bound |yk − y1 | ≤ M for some constant
M (which may depend on y1 ) we will be done, by choosing m large enough so that
cm ≤ ϵ/M .
In general, |ya+1 − ya | ≤ c|ya − ya−1 |, so
(c) Since f is continuous, f (limn→∞ yn ) = limn→∞ f (yn ) which is just the same as y shifted
one element forward, so clearly lim f (yn ) = y, showing that y is a fixed point.
Now, consider two similar sequences (an ) → a and (bn ) → b where a1 , b1 ∈ R, an+1 =
f (an ), and bn+1 = f (bn ). By the Algebraic Limit Theorem b = a + limn→∞ bn − an .
Now note
lim bn − an ≤ lim |bn − an | ≤ lim cn−1 |b1 − a1 | = 0
n→∞ n→∞ n→∞
Therefore a = b; this implies that regardless of our starting choice of y1 we will end
up at the same fixed point y. In particular, for a given fixed point z, if we start at
z1 = z then clearly limn→∞ zn = z but also limn→∞ zn = y and therefore z = y and y
is a unique fixed point.
Exercise 4.3.12
Let F ⊆ R be a nonempty closed set and define g(x) = inf{|x − a| : a ∈ F }. Show that g is
continuous on all of R and g(x) ̸= 0 for all x ∈
/ F.
Solution
Let x ∈ R and let a ∈ F be the element of F closest to x (must exist since F is closed), we
have 0 ≤ g(y) ≤ |y − a| and g(x) = |x − a|. Similarly, let b ∈ F be the element of F closest
to y. For the rest of the argument to make sense, we need to pick a as our comparison point
if |x − a| ≤ |y − b| or b otherwise. Suppose, without loss of generality, that we pick a. Thus,
not only g(y) − g(x) ≤ |y − a| − |x − a| but also |x − a| ≤ |y − b| =⇒ 0 ≤ g(y) − g(x). This
allows us to write
|g(y) − g(x)| ≤ ||y − a| − |x − a||
Applying the bound from Exercise 1.2.6 (d) we get
Setting δ = ϵ gives |g(x) − g(y)| < ϵ as desired. If we had to pick b, the argument is the
same just replacing y with x and vice-versa.
To see g(x) ̸= 0 for x ∈/ F notice that F c is open so there exists an α > 0 so that
Vα (x) ∩ F = ∅ meaning g(x) ≥ α and so g(x) ̸= 0.
Exercise 4.3.13
Let f be a function defined on all of R that satisfies the additive condition f (x + y) =
f (x) + f (y) for all x, y ∈ R.
(a) Show that f (0) = 0 and that f (−x) = −f (x) for all x ∈ R.
(b) Let k = f (1). Show that f (n) = kn for all n ∈ N, and then prove that f (z) = kz for
all z ∈ Z. Now, prove that f (r) = kr for any rational number r.
(c) Show that if f is continuous at x = 0, then f is continuous at every point in R and
conclude that f (x) = kx for all x ∈ R. Thus, any additive function that is continuous
at x = 0 must necessarily be a linear function through the origin.
Solution
(a) f (0 + 0) = f (0) + f (0) implies f (0) = 0 and thus f (x + (−x)) = 0 = f (x) + f (−x)
meaning f (−x) = −f (x).
(b) f (n) = f (1) + · · · + f (1) = k + · · · + k = kn. Now since f (−n) = −f (n) = −kn we
have f (z) = kz for all z ∈ Z. Finally let r = p/q for p ∈ Z and q ∈ N. notice that
f (qr) = k(qr) and since f (qr) = f (r + · · · + r) = f (r) + · · · + f (r) = qf (r) we have
f (r) = k(qr)/q = kr.
(c) Assume f is continuous at 0 and let x ∈ R be arbitrary. Let xn be a sequence
approaching x. since (x − xn ) → 0 we have f (x − xn ) → 0 because f is continuous
at zero. Now since f is additive f (x − xn ) = f (x) − f (xn ) → 0 implies f (xn ) → f (x)
meaning f is continuous at x ∈ R by the sequential characterization of continuity.
Now to see that f (x) = kx for x ∈ I simply take a limit of rationals approaching x.
Exercise 4.3.14
(a) Let F be a closed set. Construct a function f : R → R such that the set of points
where f fails to be continuous is precisely F . (The concept of the interior of a set,
discussed in Exercise 3.2.14, may be useful.)
(b) Now consider an open set O. Construct a function g : R → R whose set of discontin-
uous points is precisely O. (For this problem, the function in Exercise 4.3.12 may be
useful.)
Solution
(a) Using the notation from Exercise 3.2.14, note that F o , F c , and F c \F c are disjoint but
their union is R; moreover F o ∪ F c \F c = F . Let d(x) denote Dirichlet’s function (1
on rationals, 0 on irrationals), and consider
o
d(x) x ∈ F
f (x) = 2 x ∈ F c \F c
3 x ∈ Fc
4.4. CONTINUOUS FUNCTIONS ON COMPACT SETS 105
If x ∈ F o (which is open) then we can find Vϵ (x) ⊆ F o where there will be both
irrational and rational numbers, indicating that f is discontinuous over F o .
If x ∈ F c \F c , x must be a limit point of F c , and therefore all Vϵ (x) will intersect F c at
some point, and thus f (y) = 3 for some y ∈ Vϵ (x), preventing f from being continuous
in F c \F c .
If x ∈ F c (which is open) then we can find Vϵ (x) ⊆ F c which is a constant 3, and
therefore f is continuous over F c . Thus, f is discontinous only over F .
(b) Define
f (x) = d(x) (inf{|x − a| : a ∈ F c })
Exercise 4.4.2
(a) Is f (x) = 1/x uniformly continuous on (0, 1) ?
√
(b) Is g(x) = x2 + 1 uniformly continuous on (0, 1) ?
Solution
(a) No, intuitively because slope becomes unbounded as we approach zero. Rigorously
consider xn = 2/n and yn = 1/n we have |xn − yn | → 0 but |1/xn − 1/yn | = |n/2 − n| =
n/2 is unbounded meaning f cannot be uniformly continuous by Theorem 4.4.5.
(b) Yes, since it’s continuous on [0, 1] Theorem 4.4.7 implies it is uniformly continuous on
[0, 1] and hence on any subset as well.
(c) Yes, since h is continuous over [0, 1] implying it is also uniformly continuous over [0, 1]
by Theorem 4.4.7
Exercise 4.4.3
Show that f (x) = 1/x2 is uniformly continuous on the set [1, ∞) but not on the set (0, 1].
Solution
By Lipschitz over [1, ∞)
Exercise 4.4.4
Decide whether each of the following statements is true or false, justifying each conclusion.
(a) If f is continuous on [a, b] with f (x) > 0 for all a ≤ x ≤ b, then 1/f is bounded on
[a, b] (meaning 1/f has bounded range).
Solution
(a) True, the Algebraic Limit Theorem implies 1/f is continuous (well defined since f > 0)
and the Extreme Value Theorem implies 1/f attains a maximum and minimum and
so is bounded.
4.4. CONTINUOUS FUNCTIONS ON COMPACT SETS 107
(b) Let ϵ = 1 and choose δ > 0 so that |x − y| < δ implies |f (x) − f (y)| < ϵ. Define the
set X = {x0 , . . . , xn } consisting of evenly spaced values xi , ranging from x0 = inf A to
xn = sup A, with the spacing between each value less than δ/2 (i.e. ∀k, xk − xk−1 <
δ/2). Now define the set P = {p0 , . . . , pm } where for each xi ∈ X, we add one element
pi ∈ A ∩ Vδ/2 (x), if A ∩ Vδ/2 (x) ̸= ∅ (and do not add anything if A ∩ Vδ/2 (x) = ∅).
Now, every element a ∈ A is at most δ from an element p ∈ P (i.e. |a − x| < δ). To
see this, for any a ∈ A, there must be some xi ∈ X so that |a − xi | < δ/2, and since
A ∩ Vδ/2 (x) ̸= ∅ (it at least contains a), there must also be an element pi ∈ P so that
|xi − p| < δ/2. By the Triangle Inequality, |a − p| < δ for some p ∈ P .
Noting that P ⊆ A is finite, we can consider M = max(f (P )). Let a ∈ A be arbitrary,
and identify the nearest p ∈ P . We have |a − p| < δ so |f (y) − f (p)| < ϵ and since
f (p) ≤ M , f (a) < ϵ + M , completing the proof.
Alternative proof approach: Ā is closed, bounded, and thus compact. Extend the
definition of f to cover limit points of A via a limit on f ; these limits exist since f
is uniformly continuous (TODO write explicit proof). The extended f is continuous,
and thus preserves the compactness of Ā; therefore f (A) is bounded.
(c) Any function with finite range preserves compact sets, since all finite sets are compact.
Meaning Dirichlet’s function
(
1 if x ∈ Q
g(x) =
0 if x ∈ Q
(b) A uniformly continuous function f : (0, 1) → R and a Cauchy sequence (xn ) such that
f (xn ) is not a Cauchy sequence;
108 CHAPTER 4. FUNCTIONAL LIMITS AND CONTINUITY
(c) A continuous function f : [0, ∞) → R and a Cauchy sequence (xn ) such that f (xn ) is
not a Cauchy sequence;
Solution
(a) f (x) = 1/x and xn = 1/n has f (xn ) diverging, hence f (xn ) is not Cauchy.
(b) Impossible since for all ϵ > 0 we can find an N so that all n ≥ N has |xn − xm | < δ
(since xn is Cauchy) implying |f (xn ) − f (xm )| < ϵ and thus f (xn ) is Cauchy. (Uniform
continuity is needed for the ∀n ≥ N part)
(c) Impossible since [0, ∞) is closed (xn ) → x ∈ [0, ∞) implying f (xn ) → f (x) since f is
continuous, thus f (xn ) is a Cauchy sequence.
Exercise 4.4.7 √
Prove that f (x) = x is uniformly continuous on [0, ∞).
Solution
We will show f is uniformly continuous on [0, 1] and [1, ∞) then combine them similar to
Exercise 4.4.5.
(i) Since f is continuous over [0, 1] Theorem 4.4.7 implies f is uniformly continuous on
[0, 1].
√
(ii) f is Lipschitz on [1, ∞) since x is sublinear over [1, ∞)
√ √ √ √
x − y ( x − y) 1
= √ √ = √
x − y ( x − y)( x + y) x + y ≤ 1
√ √ √
Let δ = min{δ1 , δ2 } where δ1 is for [0, 1] and δ2 is for [1, ∞). If x, y are both in one of
[0, 1] or [1, ∞) we have |f (x) − f (y)| < ϵ/2 and are done. If x ∈ [0, 1] and y ∈ [1, ∞) then
Exercise 4.4.8
Give an example of each of the following, or provide a short argument for why the request
is impossible.
Solution
(a) Impossible as continuous functions preserve compact sets and (0, 1) is not compact.
4.4. CONTINUOUS FUNCTIONS ON COMPACT SETS 109
(b) Define
3x − 1
if x ∈ [1/3, 2/3]
f (x) = 1 if x ∈ (2/3, 1)
0 if x ∈ (0, 1/3)
(c) Consider g(x) = sin(1/x)(1 − x) over x ∈ (0, 1]; clearly g(x) is continuous over this
interval. The 1 − x term bounds g(x) to (−1, 1), while the sin(1/x) ensures that g(x)
will approach this bound arbitrarily close as x → 0. Thus, the range of g(x) is (−1, 1).
We now just need to shape this to (0, 1) by defining f (x) = (g(x) + 1)/2.
(b) Is the converse statement true? Are all uniformly continuous functions necessarily
Lipschitz?
Solution
(a) Choose ϵ > 0 and set δ = ϵ/M to get |f (x) − f (y)| ≤ M |x − y| < M δ < ϵ.
√
(b) No, consider f (x) = x over [0, 1] which is uniformly continuous by Theorem 4.4.7
however at x = 0 we have
√
f (x) − f (y) − y 1
x − y = −y = √y
Exercise 4.4.10
Assume that f and g are uniformly continuous functions defined on a common domain A.
Which of the following combinations are necessarily uniformly continuous on A:
f (x)
f (x) + g(x), f (x)g(x), , f (g(x))?
g(x)
(Assume that the quotient and the composition are properly defined and thus at least con-
tinuous.)
110 CHAPTER 4. FUNCTIONAL LIMITS AND CONTINUITY
Solution
(i) f (x) + g(x) is clearly uniformly continuous
(ii) f (x) = g(x) = x are individually uniformly continuous over R but f (x)g(x) = x2 is
not.
Note this counterexample only works when A is unbounded. If A is bounded you can
prove f (x)g(x) must be uniformly continuous the same way you prove the product
rule.
(iii) False, consider f (x) = 1 and g(x) = x over (0, 1). Both are uniformly continuous on
(0, 1) but f /g = 1/x is not.
(iv) Want |f (g(x)) − f (g(y))| < ϵ. Since f is uniformly continuous we can find an α > 0
so that
|g(x) − g(y)| < α =⇒ |f (g(x)) − f (g(y))| < ϵ
Now since g is uniformly continuous we can find a δ > 0 so that
Show that g is continuous if and only if g −1 (O) is open whenever O ⊆ R is an open set.
Solution
A fact we’ll use is that g(A) ⊆ B if and only if A ⊆ g −1 (B). Which is true since
Fix x ∈ R, we are given that g −1 (Vϵ (x)) is open, meaning there exists a neighborhood
Vδ (x) with Vδ (x) ⊆ g −1 (Vϵ (x)) implying g(Vδ (x)) ⊆ Vϵ (x) and thus g is continuous.
Now suppose g is continuous and let O ⊆ R be an open set. If x ∈ g −1 (O) then g(x) ∈ O
and (since O is open) there exists a neighborhood Vϵ (g(x)) ⊆ O, now there exists Vδ (x)
where g(Vδ (x)) ⊆ Vϵ (g(x)) ⊆ O so we have Vδ (x) ⊆ g −1 (O) and are done.
Exercise 4.4.12
Review Exercise 4.4.11, and then determine which of the following statements is true about
a continuous function defined on R:
Solution
(a) False, f (x) = 0 has f −1 ({0}) = R
(b) False, f (x) = 0 has {0} compact but f −1 ({0}) = R is not compact
(b) Let g be a continuous function on the open interval (a, b). Prove that g is uniformly
continuous on (a, b) if and only if it is possible to define values g(a) and g(b) at the
endpoints so that the extended function g is continuous on [a, b]. (In the forward
direction, first produce candidates for g(a) and g(b), and then show the extended g is
continuous.)
Solution
(a) Let ϵ > 0 and set N large enough that n, m ≥ N has |xn − xm | < δ implying |f (xn ) −
f (xm )| < ϵ by the uniformly continuity of f .
(b) Define g(a) = limx→a g(x) and g(b) = limx→b g(x) if both limits exist, then g is contin-
uous on [a, b] meaning it is uniformly continuous on [a, b] by Theorem 4.4.7, and thus
is uniformly continuous the subset (a, b).
If f were uniformly continuous (a, b) Cauchy sequences are preserved meaning the
sequential definition for functional limits (Theorem 4.2.3) implies the limits limx→a g(x)
and limx→b g(x) exist.
Exercise 4.4.14
Construct an alternate proof of Theorem 4.4.7 (Continuous on K implies Uniformly Con-
tinuous on K) using the open cover characterization of compactness from the Heine-Borel
Theorem (Theorem 3.3.8 (iii)).
Solution
Let f be continuous on K, and choose ϵ > 0. We can create the open cover {Vδx /2 (x) : x ∈ K}
where δx is chosen so every y ∈ Vδ (x) has |f (x) − f (y)| < ϵ/2. Now Snsince K is compact there
exists a finite subcover O = {Vδ1 /2 (x1 ), . . . , Vδn /2 (xn )} with K ⊆ k=1 Vδk /2 (xk ).
112 CHAPTER 4. FUNCTIONAL LIMITS AND CONTINUITY
Also, |x − xi | < δi /2 < δi . Since f is continuous at xi , this implies |f (y) − f (xi )| < ϵ/2
and |f (x) − f (xi )| < ϵ/2, so by the Triangle Inequality |f (x) − f (y)| < ϵ.
Solution
Since [a, b] is connected, so too must be f ([a, b]), which must be an interval containing both
f (a) and f (b); therefore [f (a), f (b)] (or [f (b), f (a)] if f (b) < f (a)) is a subset of f ([a, b]),
completing the proof.
Exercise 4.5.2
Provide an example of each of the following, or explain why the request is impossible
(a) A continuous function defined on an open interval with range equal to a closed interval.
(b) A continuous function defined on a closed interval with range equal to an open interval.
(c) A continuous function defined on an open interval with range equal to an unbounded
closed set different from R.
Solution
(a) Possible, see Exercise 4.4.8 (b)
Exercise 4.5.3
A function f is increasing on A if f (x) ≤ f (y) for all x < y in A. Show that if f is increasing
on [a, b] and satisfies the intermediate value property (Definition 4.5.3), then f is continuous
on [a, b].
4.5. THE INTERMEDIATE VALUE THEOREM 113
Solution
Let x ∈ [a, b] and choose ϵ > 0. Let L ∈ (f (a), f (x)) ∩ (f (x) − ϵ, f (x)) IVP lets us find
c ∈ (a, x) with f (c) = L, thus |f (x) − f (c)| < ϵ. Likewise we can find d ∈ (x, b) with
|f (d) − f (x)| < ϵ. Because f is increasing f (x) − f (c) < ϵ implies f (x) − f (c′ ) < ϵ for
c′ ∈ (c, x) (and likewise for d) meaning every y ∈ (c, d) has |f (x) − f (y)| < ϵ. To get a
δ-neighborhood simply set δ = min{x − c, d − x}.
Exercise 4.5.4
Let g be continuous on an interval A and let F be the set of points where g fails to be
one-to-one; that is,
F = {x ∈ A : f (x) = f (y) for some y ̸= x and y ∈ A}.
Show F is either empty or uncountable.
Solution
Suppose F is nonempty, let x, y ∈ A with x ̸= y and f (x) = f (y). Pick z ∈ (x, y) such that
f (z) ̸= f (x) (if z does not exist f is constant over (x, y) and we are finished early). By the
Intermediate Value Theorem every L ∈ (f (x), f (z)) has an x′ ∈ (x, z) with f (x′ ) = L. And
since L ∈ (f (z), f (y)) as well we can find y ′ ∈ (z, y) with f (y ′ ) = L, thus f (y ′ ) = f (x′ ) and
so f is not 1-1 at every L ∈ (f (x), f (z)) which is uncountable.
Exercise 4.5.5
(a) Finish the proof of the Intermediate Value Theorem using the Axiom of Completeness
started previously.
(b) Finish the proof of the Intermediate Value Theorem using the Nested Interval Property
started previously.
Solution
Let f be continuous, and let L ∈ (f (a), f (b)), we must find c ∈ (a, b) with f (c) = L. (If
f (a) > f (b) then instead consider f ′ (x) = −f (x) and L′ = −L)
(a) Let c = sup{x : f (x) ≤ L}. f (c) < L is not possible since we could find δ small enough
that f (c + δ) < L contradicting c being an upper bound. And f (c) > L is impossible
since we could find δ small enough that f (c − δ) > L contradicting c being the least
upper bound (since we found a smaller upper bound). Thus we must have f (c) = L
completing the proof.
A detail we glossed over is c ∈ (a, b), which can be seen since f (b) > L and f (a) < L
has f (a + δ) < L meaning a cannot be the least upper bound.
(b) Let I0 = [a, b] and bisect into two intervals, let I1 be the interval where L is still
T the endpoints. Continue like this to get a sequence In ⊆ In−1 , In ̸= ∅,
between f at
and let c ∈ ∞n=1 In .
Suppose for contradiction that f (c) < L; then there must be some ϵ > 0 so that
Vϵ (f (c)) < L. Since f is continuous, there must also be some δ so that f (Vδ (c)) < L -
but this contradicts our construction in that one endpoint of In is mapped to a number
greater than L, no matter how large n gets and how small In is as a result. A similar
argument shows f (c) cannot be larger than L, and therefore f (c) = L.
114 CHAPTER 4. FUNCTIONAL LIMITS AND CONTINUITY
Exercise 4.5.6
Let f : [0, 1] → R be continuous with f (0) = f (1)
(a) Show that there must exist x, y ∈ [0, 1] satisfying |x − y| = 1/2 and f (x) = f (y).
(b) Show that for each n ∈ N there exist xn , yn ∈ [0, 1] with |xn − yn | = 1/n and f (xn ) =
f (yn ).
(c) If h ∈ (0, 1/2) is not of the form 1/n, there does not necessarily exist |x − y| = h
satisfying f (x) = f (y). Provide an example that illustrates this using h = 2/5.
Solution
(a) Let g(x) = f (x) − f (x + 1/2) and note that g(x) is continuous over [0, 1/2]. Note also
that g(0) = f (0) − f (1/2) = −g(1/2) = f (1/2) − f (1), and therefore we can apply
IVT to g(x) over [0, 1/2] to conclude that there must be a root of g(x), and therefore
f (x) = f (x + 1/2), for some x ∈ [0, 1/2].
(b) Let g(x) = f (x) − f (x + 1/n) and note that g(x) is continuous over [0, n−1
n
]. Note also
Pn−1 Pn−1
that g(0) = − i=1 g(i/n), and since g(0) and i=1 g(i/n) have opposite sign there
must be some natural number 1 ≤ i ≤ n − 1 where g(i/n) is opposite in sign from
g(0), at which point we can apply IVT in a similar fashion to part (a) and find a root
of g(x), completing the proof.
(c) Consider
−10(x − 0) + 0
0 < x ≤ 1/5
15(x − 1/5) − 2 1/5 < x ≤ 2/5
f (x) = −10(x − 2/5) + 1 2/5 < x ≤ 3/5
15(x − 3/5) − 1 3/5 < x ≤ 4/5
−10(x − 4/5) + 2 4/5 < x ≤ 1
You could go through the grunt work of verifying that this meets the requirements,
but it’s easier to just plot the function (see graphic) - the function takes a corner every
1/5 along x.
Exercise 4.5.7
Let f be a continuous function on the closed interval [0, 1] with range also contained in
[0, 1]. Prove that f must have a fixed point; that is, show f (x) = x for at least one value of
x ∈ [0, 1].
4.6. SETS OF DISCONTINUITY 115
Solution
Since 0 and 1 are both in the range of f , choose a and b such that f (a) = 0 and f (b) = 1.
Define g(x) = f (x) − x; clearly g is continuous, g(a) = 0 − a = −a ≤ 0, and g(b) = 1 − b ≥ 0.
By IVT there must be some c ∈ [a, b] so that g(c) = 0 and hence f (c) = c.
Solution
Define a function f to be strictly increasing (decreasing) on A if f (x) < f (y) (f (x) > f (y))
for all x < y in A, and define strictly monotone to mean either strictly increasing or strictly
decreasing. It’s easy to show by contradiction using IVT that a one-to-one continuous
function f must be strictly monotone, and similarly so is f −1 . Moreover, if f is strictly
increasing then so is f −1 , and if f is strictly decreasing then so is f −1 .
We now show that f −1 satisfies the intermediate value property. Assume for now that
f is strictly increasing - the proof for the case of strictly decreasing is similar. Consider
f (x1 ) = y1 < f (x2 ) = y2 and any L between f −1 (y1 ) and f −1 (y2 ), i.e. x1 < L < x2 ; we
need to show there is some y3 ∈ [y1 , y2 ] such that f −1 (y3 ) = L; clearly y3 = f (L) works. By
Exercise 4.5.3, this shows f −1 (x) is continuous.
(a) Df = Zc .
Solution
(a) Modify the function continuous only at zero to be continuous around integers.
(
x − round(x) if x ∈ Q
f (x) =
0 if x ∈ I
(b) Modify the function continuous only at zero to be continuous below 0 and above 1.
Let (
x if x ∈ Q
d(x) =
0 if x ∈ I
116 CHAPTER 4. FUNCTIONAL LIMITS AND CONTINUITY
then
0
if x ≤ 0
g(x) = d(x) if 0 < x ≤ 1
2 if 1 < x
Exercise 4.6.2
Given a countable set A = {a1 , a2 , a3 , . . .}, define f (an ) = 1/n and f (x) = 0 for all x ∈
/ A.
Find Df .
Solution
To find Df consider x ∈ Df for the two cases x ∈ A and x ∈ / A. If x ∈ A then f is not
continuous, since f (x) > 0 but for any δ > 0 we can find y ∈ Vδ (x) with y ∈
/ A (because A is
c
countable, A must be dense) hence there is an unavoidable error of |f (x)−f (y)| = f (x) > 0.
Now consider x ∈ / A, using the sequential criterion for continuity notice every sequence
(xn ) → x has f (xn ) → 0 (since if xn ∈ A converge to 0 as n → ∞, and xn ∈ / A are always
0), now since f (x) = 0 this shows f is continuous at x ∈/ A.
Together we’ve shown Df = A. Setting A = Q and using a particular ordering recovers
Thomae’s function. Hence we can view this as a generalization of Thomae’s function for
arbitrary countable sets.
Exercise 4.6.3
State a similar definition for the left-hand limit
lim f (x) = L
x→c−
Solution
For all ϵ > 0 there exists a δ > 0 such that 0 < c − x < δ implies |f (x) − L| < ϵ.
Exercise 4.6.4
Given f : A → R and a limit point c of A, limx→c f (x) = L if and only if
Solution
Let ϵ > 0, pick δ1 so 0 < x − c < δ1 implies |f (x) − L| < ϵ, pick δ2 so 0 < c − x < δ2 implies
|f (x) − L| < ϵ. Finally, set δ = min{δ1 , δ2 } to get |f (x) − L| < ϵ when 0 < |x − c| < δ, as
desired.
Exercise 4.6.5
Prove that the only type of discontinuity a monotone function can have is a jump disconti-
nuity.
4.6. SETS OF DISCONTINUITY 117
Solution
Let f be monotone and assume f is increasing. For some c we want to show limx→c− f (x)
and limx→c+ f (x) exist.
Let ϵ > 0 and set L = sup{f (x) : x < c}, by the definition of sup, L − ϵ is not an
upper bound for {f (x) : x < c}, hence there exists a δ1 > 0 with f (c − δ1 ) > L − ϵ, thus
0 < c−x < δ1 implies |f (x)−L| < ϵ (this is where we use the fact that f is increasing!), hence
the lower limit exists. Likewise for M = inf{f (x) : x > c} we get δ2 > 0 with 0 < x − c < δ2
implying |f (x) − M | < ϵ, hence the upper limit exists.
Putting these together, we see that f is continuous at c if and only if L = M . In other
words, the only possible discontinuity is a jump discontinuity L ̸= M .
Exercise 4.6.6
Construct a bijection between the set of jump discontinuities of a monotone function f and a
subset of Q. Conclude that Df for a monotone function f must either be finite or countable,
but not uncountable.
Solution
In 4.6.5 we showed every c ∈ Df is a jump discontinuity, i.e. both sided limits L and M
exist and L ̸= M . Pick some r ∈ (L, M ) ∩ Q and assign f (c) = r. Continue like this to
define a bijection f : Df → Q where Q ⊆ Q. Thus Df must be finite or countable.
Exercise 4.6.7
In Section 4.1 we constructed functions where the set of discontinuity was R (Dirichlet’s
function), R\{0} (modified Dirichlet function), and Q (Thomae’s function).
(a) Show that in each of the above cases we get an Fσ set as the set where the function is
discontinuous.
(b) Show that the two sets of discontinuity in Exercise 4.6.1 are Fσ sets.
Solution
(a) R is closed, so it is in Fσ , R\{0} = ∞
S
n=1SR\(−1/n, 1/n) is in Fσ since R\(−1/n, 1/n) is
closed, and finally Q is in Fσ since Q = ∞ n=1 {rn } (where rn enumerate Q, all countable
sets are Fσ sets.)
(b) Recall countable unions
S of Fσ sets are Fσ (see 3.5.2) and that open intervals are Fσ
c
sets, meaning Z = z∈Z (z, z + 1) is an Fσ set. As for {x : 0 < x ≤ 1} = (0, 1] I refer
you to 3.5.3 (b).
Exercise 4.6.8
Prove that, for a fixed α > 0, the set Dfα is closed.
Solution c
We do this by showing the complement is open. Let x ∈ Dfα , by the definition of α-
continuity there exists a δ > 0 such that y, z ∈ Vδ (x) have |f (y) − f (z)| < α. To see
c
openness notice Vδ/2 (x) ⊆ Dfα since any x′ ∈ Vδ/2 (x) is α-continuous with δ ′ = δ/2.
Exercise 4.6.9
′
If α < α′ , show that Dfα ⊆ Dfα .
118 CHAPTER 4. FUNCTIONAL LIMITS AND CONTINUITY
Solution
This is obvious. If |f (y) − f (z)| < α and α < α′ clearly |f (y) − f (z)| < α′ as well. Increasing
α only makes the condition less strict.
Exercise 4.6.10
Let α > 0 be given. Show that if f is continuous at x, then it is α-continuous at x as well.
Explain how it follows that Dfα ⊆ Df .
Solution
Let ϵ = α/2 and use continuity to get δ > 0 with 0 < |x−y| < δ implying |f (x)−f (y)| < α/2,
which shows every y, z ∈ Vδ (x) satisfies |f (y) − f (z)| < α by the triangle inequality. Thus f
is α-continuous at x.
The negation of “continuous at x implies α-continuous at x” is “not α-continuous at x
implies not continuous at x”, hence Dfα ⊆ Df .
Exercise 4.6.11
Show that if f is not continuous at x, then f is not α-continuous for some α > 0. Now
explain why this guarantees that
[∞
Df = Dfαn
n=1
where αn = 1/n.
Solution
Negating the definition of f being continuous at x, we see f is not continuous at x iff there
exists an ϵ0 > 0 such that no δ > 0 satisfies |f (x) − f (y)| < ϵ for all 0 < |x − y| < δ. Once
αn < ϵ0 (i.e. n > 1/ϵ0 ) we will have x ∈ Dfαn .
(This completes the proof that Df is an Fσ set!)
Chapter 5
The Derivative
119
120 CHAPTER 5. THE DERIVATIVE
(b) If f g and g are differentiable at zero, then (f g)/g = f is differentiable at zero provided
the quotient is well defined. However if we let g(x) = 0 then f g = 0 is differentiable
at zero regardless of f . (Note we must have g(0) = 0 otherwise f = (f g)/g would be
differentiable at zero)
We have
t′ (0) = lim t(x)/x
x→0
This limit doesn’t exist, but if we define f (x) = xt(x) then the inside is thomae’s
function and so
f ′ (0) = lim f (x)/x = lim t(x) = 0
x→0 x→0
Exercise 5.2.3
(a) Use Definition 5.2.1 to produce the proper formula for the derivative of h(x) = 1/x.
(b) Combine the result in part (a) with the Chain Rule (Theorem 5.2.5) to supply a proof
for part (iv) of Theorem 5.2.4.
(c) Supply a direct proof of Theorem 5.2.4 (iv) by algebraically manipulating the difference
quotient for (f /g) in a style similar to the proof of Theorem 5.2.4 (iii).
Solution
(a)
1/y − 1/x x−y 1 1
h′ (x) = lim = lim = lim − =− 2
y→x y−x y→x xy(y − x) y→x xy x
(b) By chain rule (1/g)′ (x) = −g ′ (x)/g(x)2 combined with product rule gives
1 f ′ (x) g ′ (x)
(f · 1/g)′ (x) = f ′ (x) · + f (x)(1/g)′ (x) = − f (x)
g(x) g(x) g(x)2
f ′ (x)g(x) − f (x)g ′ (x)
=
g(x)2
(c)
(f /g)(y) − (f /g)(x) f (y)g(x) − f (x)g(y)
(f /g)′ (x) = lim = lim
y→x y−x y→x g(x)g(y)(y − x)
5.2. DERIVATIVES AND THE INTERMEDIATE VALUE PROPERTY 121
Now the g(x)g(y) goes to g(x)2 , we just need to evaluate the derivatives in the nu-
merator. We do this by adding and subtracting f (x)g(x) similar to the proof of the
product rule, then use the functional limit theorem to finish it off
Exercise 5.2.4
Follow these steps to provide a slightly modified proof of the Chain Rule.
(b) Use this criterion for differentiability (in both directions) to prove Theorem 5.2.5.
Solution
(a) First suppose h is differentiable at a, then we can define
(
h(x)−h(a)
x−a
x ̸= a
l(x) = ′
h (a) x=a
(b) Let f be differentiable at a and g be differentiable at f (a). we will show g(f (x)) is
differentiable at a with derivative g ′ (f (a))f ′ (a).
Multiply the top and bottom by f (y) − f (a) to get
We’re almost done, the right hand side is f ′ (a) we just need to evaluate the nested
limit on the left. Define l(y) = g(y)−g(f (a))
y−f (a)
and l(a) = g ′ (f (a)) then we have a product
of continuous functions so we can use the algebraic limit theorem
f (y) − f (a)
lim l(f (y)) · = g ′ (f (a)) · f ′ (a)
y→a y−a
122 CHAPTER 5. THE DERIVATIVE
Exercise 5.2.5
Let (
xa if x > 0
fa (x) =
0 if x ≤ 0
(a) For which values of a is f continuous at zero?
(b) For which values of a is f differentiable at zero? In this case, is the derivative function
continuous?
(c) For which values of a is f twice-differentiable?
Solution
(a) All a > 0, if a = 0 we get the step function and a < 0 gives an asymptote
(b) At zero, if the derivative exists, then the one-sided limit from above
xa
lim+ = lim+ xa−1
x→0 x x→0
must exist and be equal to zero (to match the limit from below). Thus a > 1 is
necessary, and in this case the derivative will be continuous.
(c) The first derivative is axa−1 , and its derivative at zero must be zero, so a > 2.
Exercise 5.2.6
Let g be defined on an interval A, and let c ∈ A.
(a) Explain why g ′ (c) in Definition 5.2.1 could have been given by
g(c + h) − g(c)
g ′ (c) = lim .
h→0 h
(b) Assume A is open. If g is differentiable at c ∈ A, show
g(c + h) − g(c − h)
g ′ (c) = lim .
h→0 2h
Solution
(a) This is just a change of variable from the normal definition, set h = (x − c) to get
g(x) − g(c) g(c + h) − g(c)
lim = lim
x→c x−c h→0 h
(b) Some basic algebra
g(c + h) − g(c) + g(c) − g(c − h) 1 g(c + h) − g(c) g(c) − g(c − h)
lim = lim + lim
h→0 2h 2 h→0 h h→0 h
The first term is clearly g ′ (c) and the second is g ′ (c) with the substitution h = c − x
since
g(c) − g(x)
g ′ (c) = lim
x→c c−x
′
Thus the whole expression is g (c).
5.2. DERIVATIVES AND THE INTERMEDIATE VALUE PROPERTY 123
Exercise 5.2.7
Let (
xa sin(1/x) if x ̸= 0
ga (x) =
0 if x = 0
Find a particular (potentially noninteger) value for a so that
(a) ga is differentiable on R but such that ga′ is unbounded on [0, 1].
(c) ga is differentiable on R and ga′ is differentiable on R, but such that ga′′ is not continuous
at zero.
Solution
We need a > 0 to make ga continuous at zero, for differentiation notice
Thus for differentiation we need xa−1 sin(1/x) to have a limit at zero. Keep this in mind for
the problems. Also note
(a) To get unboundedness we need the xa−2 cos(1/x) term to become unbounded, so pick
a = 1.5 to satisfy a > 1 (differentiable) and a < 2 (unbounded derivative)
(b) Pick a = 2.5, It’s clear that ga′ (x) = 2.5x1.5 sin(1/x)−x0.5 cos(1/x) is continuous at zero,
but not differentiable since x0.5 cos(1/x) isn’t differentiable at zero and 2.5x1.5 sin(1/x)
is (if both terms were not differentiable they could cancel eachother out, constant
vigilance!).
(c) Pick a = 3.5, for all intents and purposes ga′ behaves like ga−2 (because the xa−2 cos(1/x)
term acts as a bottleneck) meaning ga′′ exists, but isn’t continuous since we get another
−2 to a leading to a x−0.5 term in ga′′ .
Exercise 5.2.8
Review the definition of uniform continuity (Definition 4.4.4). Given a differentiable function
f : A → R, let’s say that f is uniformly differentiable on A if, given ϵ > 0 there exists a
δ > 0 such that
f (x) − f (y) ′
x−y − f (y) < ϵ whenever 0 < |x − y| < δ.
(b) Show that if a function is uniformly differentiable on an interval A, then the derivative
must be continuous on A.
124 CHAPTER 5. THE DERIVATIVE
(c) Is there a theorem analogous to Theorem 4.4.7 for differentiation? Are functions that
are differentiable on a closed interval [a, b] necessarily uniformly differentiable?
Solution
(a) We have
2
x − y2
x − y − 2y = |(x + y) − 2y| = |x − y| < δ
Let y = (x + h) to get
|x2 + x(x + h) − 2(x + h)2 | = |x2 + x2 xh − 2x2 − 4xh − 2h2 | = |3xh + 2h2 |
Fix 0 < h < δ, since x can be as big as we want we can make |3xh + 2h2 | > ϵ. As no
fixed δ works x3 is not uniformly differentiable.
(b) We need to show that limx→y f ′ (x) = f ′ (y). Now, if we choose δ small enough such
that when |x − y| < δ, both
′
f (x) − f (x) − f (y) ϵ
<
x−y 2
and
f (x) − f (y) ′
ϵ
x−y − f (y) <
2
as desired.
(c) Consider the counterexample f (x) = x2 sin(1/x) over [0, 1] (where f (0) = 0). f is
differentiable over [0, 1] but not uniformly differentiable.
Intuitively this is because I can find xn , yn such that the slope between them becomes
unbounded, but the derivative f ′ must stay bounded. To be exact set
1 1
xn = , yn =
2πn + π/2 2πn
5.2. DERIVATIVES AND THE INTERMEDIATE VALUE PROPERTY 125
then
f (xn ) − f (yn ) ′
1 ′
xn − y n − f (xn ) = − f (xn )
x − yn
n
(2πn)(2πn + π/2) ′
= − f (xn )
π/2
= |4n(2πn + π/2) − f ′ (xn )|
Now since f ′ (xn ) = 2x sin(1/x) − cos(1/x) is bounded I can defeat any δ by picking n
large enough so that
f (xn ) − f (yn )
− f ′ (xn ) ≥ ϵ
|xn − yn | < δ and
xn − yn
Exercise 5.2.9
Decide whether each conjecture is true or false. Provide an argument for those that are true
and a counterexample for each one that is false.
(a) If f ′ exists on an interval and is not constant, then f ′ must take on some irrational
values.
(b) If f ′ exists on an open interval and there is some point c where f ′ (c) > 0, then there
exists a δ-neighborhood Vδ (c) around c in which f ′ (x) > 0 for all x ∈ Vδ (c).
Solution
(a) If f ′ is not constant there exist x, y with f ′ (x) < f ′ (y), since derivatives obey the
intermediate value proper (Theorem 5.2.7) f ′ takes on the value of every irrational
number in (f ′ (x), f ′ (y)).
(b) True if f ′ is continuous, False in general. Let f (x) = x2 sin(1/x) + x/2 so that
Notice how f ′ alternates between positive and negative for small x. We have
x2 sin(1/x) − 0
lim = lim x sin(1/x) = 0
x→0 x−0 x→0
Thus f ′ (0) = 1/2, pick any δ and I can find x ∈ Vδ (0) with f ′ (x) ≤ 0. To be explicit
define xn = 1/(2πn) so that f ′ (xn ) = −1/2 then pick n large enough so xn ∈ Vδ (0).
126 CHAPTER 5. THE DERIVATIVE
(c) A direct proof using L’Hospital’s rule: continuity of f forces the limit of the quotient
to be 0/0.
f (x) − f (0)
f ′ (0) = lim = lim f ′ (x) = L
x→0 x x→0
In general a corollary of L’Hospital’s rule is that derivatives can only have essential
discontinuities, discontinuities where limx→c f ′ (x) doesn’t exist.
Alternatively you can see this by contradiction. Derivatives obeying the IVP rules out
jump discontinuities, and removable discontinuities may be ruled out by ϵ-δing.
Exercise 5.2.10
Recall that a function f : (a, b) → R is increasing on (a, b) if f (x) ≤ f (y) whenever x < y
in (a, b). A familiar mantra from calculus is that a differentiable function is increasing if its
derivative is positive, but this statement requires some sharpening in order to be completely
accurate. Show that the function
(
x/2 + x2 sin(1/x) if x ̸= 0
g(x) =
0 if x = 0
is differentiable on R and satisfies g ′ (0) > 0. Now, prove that g is not increasing over any
open interval containing 0 .
In the next section we will see that f is indeed increasing on (a, b) if and only if f ′ (x) ≥ 0
for all x ∈ (a, b).
Solution
Already did this in 5.2.9 (b) as I came up with the same counterexample as abbott!
Exercise 5.2.11
Assume that g is differentiable on [a, b] and satisfies g ′ (a) < 0 < g ′ (b).
(a) Show that there exists a point x ∈ (a, b) where g(a) > g(x), and a point y ∈ (a, b)
where g(y) < g(b).
(b) Now complete the proof of Darboux’s Theorem started earlier.
Solution
g(x)−g(a)
(a) Since g ′ (a) = limx→a x−a
< 0 we know
g(x) − g(a)
g ′ (a) − ϵ < < g ′ (a) + ϵ when |x − a| < δ
x−a
Therefor
g(x) < g(a) + (g ′ (a) + ϵ)(x − a)
Pick ϵ small enough that g ′ (a) + ϵ < 0, and since (x − a) > 0 we have g(x) < g(a) as
desired. A similar argument works for g(y) < g(b).
(b) By EVT g obtains a minimum value over [a, b]. We just showed a and b are not minima,
therefore the minimum point c must be in the interior (a, b) which has g ′ (c) = 0 by
Theorem 5.2.6.
5.3. THE MEAN VALUE THEOREMS 127
for all x ̸= y in A
(a) Show that if f is differentiable on a closed interval [a, b] and if f ′ is continuous on [a, b],
then f is Lipschitz on [a, b].
(b) Review the definition of a contractive function in Exercise 4.3.11. If we add the as-
sumption that |f ′ (x)| < 1 on [a, b], does it follow that f is contractive on this set?
Solution
(a) Since f ′ is continuous on the compact set [a, b] we can set M such that |f ′ (x)| ≤ M
over [a, b], then pick x, y ∈ [a, b] with x < y. Apply MVT on [x, y] to get a c ∈ (x, y)
with
f (x) − f (y)
= f ′ (c)
x−y
Which implies
f (x) − f (y) ′
x − y = |f (c)| ≤ M
(b) For f to be contractive we need some c ∈ (0, 1) with |f (x) − f (y)| ≤ c|x − y|. Let x
and y be arbitrary, and consider
f (x) − f (y)
c=
x−y
By the mean value theorem, there must be some d ∈ (0, 1) where |f ′ (d)| = c. But since
x and y are arbitrary and |f ′ (d)| < 1, we must have c < 1 always, and therefore f is
contractive.
Exercise 5.3.2
Let f be differentiable on an interval A. If f ′ (x) ̸= 0 on A, show that f is one-to-one on A.
Provide an example to show that the converse statement need not be true.
Solution
Let x, y be in A with x < y, to show f (x) ̸= f (y) apply the Mean Value Theorem on [x, y]
to get c ∈ (x, y) with
f (x) − f (y)
f ′ (c) =
x−y
Now since f ′ (c) ̸= 0 we must have f (x) − f (y) ̸= 0, and thus f (x) ̸= f (y).
To see the converse is false consider how f (x) = x3 is 1-1 but has f ′ (0) = 0.
Exercise 5.3.3
Let h be a differentiable function defined on the interval [0, 3], and assume that h(0) =
1, h(1) = 2, and h(3) = 2.
(a) Argue that there exists a point d ∈ [0, 3] where h(d) = d.
Solution
(a) Consider g(x) = h(x) − x which is continuous and has g(0) = 1 and g(3) = −1, then
apply the IVT to find d ∈ (0, 3) with g(d) = 0 which implies h(d) = d.
h(0) − h(3)
h′ (c) = = 1/3
0−3
(c) We can find c ∈ (0, 3) with h′ (c) = 1/3 and a d ∈ (1, 3) with h′ (d) = 0. So by
Darboux’s theorem there exists a point x ∈ (c, d) with h′ (x) = 1/4.
Exercise 5.3.4
Let f be differentiable on an interval A containing zero, and assume (xn ) is a sequence in A
with (xn ) → 0 and xn ̸= 0.
(a) If f (xn ) = 0 for all n ∈ N, show f (0) = 0 and f ′ (0) = 0.
5.3. THE MEAN VALUE THEOREMS 129
(b) Add the assumption that f is twice-differentiable at zero and show that f ′′ (0) = 0 as
well.
Solution
(a) Since f ′ (0) exists and f (xn ) = 0 for all n we have
f (xn )
f ′ (0) = lim =0
xn
(b) By the mean value theorem over [0, xn ] there exists a cn ∈ (0, xn ) such that
f (xn )
f ′ (cn ) =
xn
Then like in (a)
f ′ (cn )
f ′′ (0) = lim =0
cn
Exercise 5.3.5
(a) Supply the details for the proof of Cauchy’s Generalized Mean Value Theorem (Theo-
rem 5.3.5).
(b) Give a graphical interpretation of the Generalized Mean Value Theorem analogous to
the one given for the Mean Value Theorem at the beginning of Section 5.3. (Consider
f and g as parametric equations for a curve.)
Solution
(a) Let h(x) = [f (b)−f (a)]g(x)−[g(b)−g(a)]f (x) and apply the MVT to h to get c ∈ (a, b)
with
h(b) − h(a) [f (b) − f (a)][g(b) − g(a)] − [g(b) − g(a)][f (b) − f (a)]
h′ (c) = = =0
b−a b−a
Thus we have
h′ (c) = [f (b) − f (a)]g ′ (c) − [g(b) − g(a)]f ′ (c) = 0
Completing the proof.
(b) Let h : [0, a] → R be twice differentiable, h′ (0) = h(0) = 0 and |h′′ (x)| ≤ M for all
x ∈ [0, a]. Show |h(x)| ≤ M x2 /2 for all x ∈ [0, a].
130 CHAPTER 5. THE DERIVATIVE
(c) Conjecture and prove an analogous result for a function that is differentiable three
times on [0, a].
Solution
(a) For x ∈ [0, a], apply MVT to find a c ∈ [0, x] with
g(x)
g ′ (c) = =⇒ g(x) = g ′ (c)x =⇒ |g(x)| ≤ M x
x
(b) This is a special case of the theorem that if f (0) = g(0) = 0 and f ′ (x) ≤ g ′ (x) for
all x ∈ [0, a] then f (x) ≤ g(x). To prove this note how letting h(x) = g(x) − f (x)
changes the statement into h′ (x) ≥ 0 implying h(x) ≥ 0. Which is true since MVT to
get c ∈ [0, x] implies h′ (c) = h(x)/x ≥ 0 thus h(x) ≥ 0.
Now returning to |h′′ (x)| ≤ M x2 /2 apply the above result to both cases in the inequal-
ity
−M x2 /2 ≤ h′′ (x) ≤ M x2 /2 =⇒ −M x ≤ h′ (x) ≤ M x
=⇒ −M ≤ h(x) ≤ M
=⇒ |h(x)| ≤ M
Which proves |h(x)| ≤ M .
(c) I conjecture |f (x)| ≤ x3 /6 when f (0) = f ′ (0) = f ′′ (0) = 0. The proof is the same as
(b), except we differentiate one more time.
Exercise 5.3.7
A fixed point of a function f is a value x where f (x) = x. Show that if f is differentiable on
an interval with f ′ (x) ̸= 1, then f can have at most one fixed point.
Solution
Suppose for contradiction that x, y are fixed points of f with x < y, then apply MVT on
[x, y] to get
f (x) − f (y) x−y
f ′ (c) = = =1
x−y x−y
But we know f ′ (c) ̸= 1, therefore finding more than one fixed point is impossible.
Exercise 5.3.8
Assume f is continuous on an interval containing zero and differentiable for all x ̸= 0. If
limx→0 f ′ (x) = L, show f ′ (0) exists and equals L.
Solution
Using L’Hospital’s rule: Let g(x) = f (x)−f (0) (and note that they have the same derivatives
and are both continuous), then
g(x) g ′ (x)
f ′ (0) = g ′ (0) = lim = lim =L
x→0 x x→0 1
(A modified function is necessary to ensure limx→0 g(x) = 0.)
5.3. THE MEAN VALUE THEOREMS 131
Exercise 5.3.9
Assume f and g are as described in Theorem 5.3.6, but now add the assumption that f and
g are differentiable at a, and f ′ and g ′ are continuous at a with g ′ (a) ̸= 0. Find a short proof
for the 0/0 case of L’Hopital’s Rule under this stronger hypothesis.
Solution
Let (xn ) be a sequence approaching a and apply MVT on [xn , a] to find cn , dn ∈ (xn , a) with
f (xn ) g(xn )
f ′ (cn ) = and g ′ (dn ) =
xn − a xn − a
Meaning
f (xn ) f ′ (cn )/(xn − a) f ′ (cn )
lim = lim ′ = lim ′
g(xn ) g (dn )/(xn − a) g (dn )
The continuity of f ′ and g ′ combined with g ′ (a) ̸= 0 implies the limit exists
f ′ (cn ) f ′ (a)
lim ′ = ′ =L
g (dn ) g (a)
Since we showed lim f (xn )/g(xn ) = L for all sequences (xn ) the Sequential Criterion for
Functional Limits (Theorem 4.2.3) implies
f (x)
lim =L
x→a g(x)
Exercise 5.3.10
2 2
Let f (x) = x sin (1/x4 ) e−1/x and g(x) = e−1/x . Using the familiar properties of these
functions, compute the limit as x approaches zero of f (x), g(x), f (x)/g(x), and f ′ (x)/g ′ (x).
Explain why the results are surprising but not in conflict with the content of Theorem 5.3.6.
1
Solution
f (x)
lim f (x) = 0, lim g(x) = 0, lim = 0,
x→0 x→0 x→0 g(x)
To compute the last limit we need to find f ′ and g ′ using derivative rules. Let h(x) =
x sin(1/x4 ) for bookkeeping.
d −x−2 −2
g ′ (x) = e = e−x (2x−3 ) = g(x)(2x−3 )
dx
h (x) = sin(x−4 ) + x cos(x−4 )(−4x−5 ) = sin(x−4 ) − 4x−4 cos(x−4 )
′
d
f ′ (x) = h(x)g(x)
dx
= h′ (x)g(x) + g(x)(2x−3 )h(x)
= g(x)(h′ (x) + 2x−3 h(x))
132 CHAPTER 5. THE DERIVATIVE
L’Hopital’s rule for 0/0 would apply if f ′ /g ′ existed. But when it doesn’t exist f /g may still
exist. Put another way, the converse of L’Hopital’s rule does not hold.
Exercise 5.3.11
(a) Use the Generalized Mean Value Theorem to furnish a proof of the 0/0 case of L’Hopital’s
Rule (Theorem 5.3.6).
(b) If we keep the first part of the hypothesis of Theorem 5.3.6 the same, but we assume
that
f ′ (x)
lim ′ =∞
x→a g (x)
Solution
(a) Let f, g be continuous functions with f (a) = g(a) = 0 and f ′ (x) ̸= 0, g ′ (x) ̸= 0 around
a, suppose
f ′ (x)
lim ′ =L
x→a g (x)
f (x)
We would like to show limx→a g(x)
= L. Choose ϵ > 0 then let δ > 0 be such that
′
f (x)
g ′ (x) − L < ϵ
Let x ∈ (a, a + δ) and apply the generalized mean value theorem on (a, x) to get a
c ∈ (a, x) with
f (x) − f (a) f ′ (c)
= ′
g(x) − g(a) g (c)
Subtract L from both sides and take absolute values, (and use f (a) = g(a) = 0) to get
f (x)
g(x) − L < ϵ
5.3. THE MEAN VALUE THEOREMS 133
We could do the same process starting from x ∈ (a−δ, a) as well, thus, for all 0 < |x| < δ
we have
f (x)
g(x) − L<ϵ
f (x)
Implying limx→a g(x)
= L as desired.
An interesting thing to note is that the same δ works for both f ′ (x)/g ′ (x) and for
f (x)/g(x). In other words, f (x)/g(x) converges to L at least as fast as f ′ (x)/g ′ (x)
does.
(b) Choose M > 0 and let δ > 0 be such that 0 < |x − a| < δ implies f ′ (x)/g ′ (x) > M .
Let x ∈ (a, a + δ) be arbitrary, then apply MVT on (a, x) to get c ∈ (a, x) with
f (x) f ′ (c)
= ′
g(x) g (c)
f (x)
>M
g(x)
for all x ∈ (a, a+δ), but again, we could just as easily apply this reasoning to (a−δ, a).
So in general, all x with 0 < |x − a| < δ satisfy
f (x)
>M
g(x)
Exercise 5.3.12
If f is twice differentiable on an open interval containing a and f ′′ is continuous at a, show
f (a + h) − 2f (a) + f (a − h)
lim = f ′′ (a).
h→0 h2
(Compare this to Exercise 5.2.6(b).)
Solution
Let ϵ > 0 and choose δ1 > 0 so every |h| < δ1 has
f (a + h) − f (a) ′
− f (a) < ϵ
h
Choose δ2 > 0 so every 0 < |x − a| < δ2 has |f ′′ (x) − f ′′ (a)| < ϵ (this is where we use the
continuity of f ′′ at a.) and set δ = min{δ1 , δ2 }.
Without loss of generality assume h > 0, apply MVT on (a, a + h) to get c ∈ (a, a + h)
with
f (a + h) − f (a)
= f ′ (c)
h
134 CHAPTER 5. THE DERIVATIVE
f (a) − f (a − h)
= f ′ (d)
h
Meaning for this specific h our estimate for f ′′ (a) is
< ϵ.
1
hn (x) = h(2n x)
2n
as n gets larger.
Solution
(1/2)h(2x) looks like h(x) shrunk down by a factor of 2 - both the amplitude and the period
of the sawtooth are halved. Similarly, hn (x) looks like h(x) shrunk down by a factor of 2n .
Exercise 5.4.2
Fix x ∈ R. Argue that the series
∞
X 1
n
h (2n x)
n=0
2
converges and thus g(x) is properly defined.
Solution
hn (x) ≤ 1/2n and we know ∞ n
P
n=0 1/2 converges, so by the comparison test (Theorem 2.7.4)
g(x) is properly defined.
5.4. A CONTINUOUS NOWHERE-DIFFERENTIABLE FUNCTION 135
Exercise 5.4.3
Taking the continuity of h(x) as given, reference the proper theorems from Chapter 4 that
imply that the finite sum
m
X 1
gm (x) = n
h (2n x)
n=0
2
is continuous on R.
Solution
The Composition of Continuous Functions (Theorem 4.3.9) ensure h(2n x) is continuous,
and the various parts of the Algebraic Continuity Theorem (Theorem 4.3.2) ensure their
combinations to form gm (x) is continuous.
Exercise 5.4.4
As the graph in Figure 5.7 suggests, the structure of g(x) is quite intricate. Answer the
following questions, assuming that g(x) is indeed continuous.
(a) How do we know g attains a maximum value M on [0, 2]? What is this value?
(b) Let D be the set of points in [0, 2] where g attains its maximum. That is D = {x ∈
[0, 2] : g(x) = M }. Find one point in D.
(c) Is D finite, countable, or uncountable?
Solution
(a) [0, 2] is a compact set, so by the Extreme Value Theorem g must have a maximum.
Since the infinite series defining g converges (and converges absolutely, for that matter),
we are free to use associativity to analyze it. Group the terms in pairs, so that f0 (x) =
h0 (x) + h1 (x) and in general fn (x) = h2n (x) + h2n+1 (x). Note that over [0, 2],
2x
x ≤ 1/2
f0 (x) = 1 1/2 ≤ x ≤ 3/2
−2x + 4 3/2 ≤ x
and in particular, f0 (x) reaches a maximum of 1 over [1/2, 3/2], an interval of length
1. Now, f1 (x) looks like a repeated f0 (x) scaled down a factor of 4, and therefore has
a period of 2/4 = 1/2 and a maximum value of 1/4 over an interval of length 1/4.
Since the period of f1 (x) is less than half the length of the interval [1/2, 3/2], there
must be one cycle of f1 (x) fully within [1/2, 3/2], and therefore the maximum value of
f0 (x)+f1 (x) is 1+1/4. (One cycle is when the function starts at 0, goes to a maximum
and plateaus, then comes back down to 0.)
A
Pnscaling argument between fn and fn+1 can then be used to show that max g(x) >
1
k=0 4k for all n ∈ N. However,
∞ ∞
X X 1
g(x) = fk (x) ≤ = 4/3
k=0 k=0
4k
(b) For this we’ll need to track the intervals where fn (x) reaches its maximum more care-
fully. Note that the endpoints of each cycle in fn (x) are also endpoints of cycles in
fn+1 (x).
Some computation shows that if one of the cycles of fn (x) reaches its maximum over the
interval [an , an + bn ], then there will be two cycles in fn+1 (x) which cover [an , an + bn /2]
and [an +bn /2, an +bn ], leading to fn+1 (x) having maximum intervals in [an +bn /8, an +
3bn /8] and [an + 5bn /8, an + 7bn /8].
To find a point in D we can repeatedly only consider the lower maximum interval at
each iteration of fn (x). Defining a0 = 1/2, b0 = 1, and an+1 = an + bn /8, bn+1 = bn /4,
clearly bn = 1/4n . an is (nearly) a geometric series, with
1 1 4n − 1 1 4n + 5
an = + , a n + b n = +
2 6 4n 2 6
Both an and an + bn approach 1/2 + 1/6 = 2/3, so we might conjecture 2/3 ∈ D.
Indeed, 2/3 is in every [an , an + bn ], and therefore 2/3 is a point in D.
(c) D is uncountable; we will use an argument similar to showing that the Cantor set is
uncountable - by mapping all sequences xn of infinite 0s and 1s to a unique point in
D. Construct a sequence of intervals In as such: If xn = 0 then take the first cycle of
fn (x) which is in In−1 and define In to be the maximum interval of fn (x) in that cycle;
if xn = 1 then take the second cycle. (For completeness, define I0 = [1/2, 3/2].) By
the Nested Interval Property the infinite intersection of these intervals yields a point
in D, while clearly each unique sequence xn will map to a unique point in D. Since
the set of all xn is uncountable, so too is D.
Exercise 5.4.5
Show that
g (xm ) − g(0)
=m+1
xm − 0
and use this to prove that g ′ (0) does not exist.
Solution
Note that hn (xm ) = xm for 0 ≤ n ≤ m and hn (xm ) = 0 for n > m, therefore g(xm ) =
(m + 1)xm . Clearly g(0) = 0 , so
g (xm ) − g(0)
=m+1
xm − 0
Exercise 5.4.6
(a) Modify the previous argument to show that g ′ (1) does not exist. Show that g ′ (1/2)
does not exist.
(b) Show that g ′ (x) does not exist for any rational number of the form x = p/2k where
p ∈ Z and k ∈ N ∪ {0}.
Solution
(a) Let xm = 1 + 1/2m with m ≥ 0. Then h0 (xm ) = 1 − 1/2m , hn (xm ) = 1/2m for
1 ≤ n ≤ m, and hn (xm ) = 0 for n > m. g(1) = 1, so
and for the same reason as in Exercise 5.4.5, g ′ (1) does not exist.
h0 (x) is differentiable at 1/2, so we can instead consider whether (g − h0 )(x) is differ-
entiable at 1/2. But since hn (x) = 2hn+1 (x/2), g(x) = 2(g − h0 )(x/2). Since g ′ (1) does
not exist, both g − h0 and g are not differentiable at 1/2.
(b) Note that hn (x) is only non-differentiable at points of the form q/2n where q ∈ Z.
Express x in lowest form, so that p is odd, and consider
k−1
X ∞
X
i(x) = g(x) − hn (x) = hk (x)
n=0 n=k
which is differentiable at x if and only iff g is as well. Since ha+b (x) = ha (2b x)/2b ,
∞
1 X k 1 k 1
i(x) = h0 (2 x) = g(2 x) = g(p)
2k n=0 2k 2k
We’ve shown that g is not differentiable at 0 or 1, and since g is periodic it’s easy to
show it’s not differentiable at any p ∈ Z, completing the proof.
Exercise 5.4.7
(a) First prove the following general lemma: Let f be defined on an open interval J and
assume f is differentiable at a ∈ J. If (an ) and (bn ) are sequences satisfying an < a < bn
and lim an = lim bn = a, show
f (bn ) − f (an )
f ′ (a) = lim .
n→∞ b n − an
(b) Now use this lemma to show that g ′ (x) does not exist.
138 CHAPTER 5. THE DERIVATIVE
Solution
(a) Keeping in mind the Sequential Criterion for Functional Limits (Theorem 4.2.3),
Note also that hn (k) = 0 when k is of the form p/2n−1 with p ∈ Z. This fact, combined
with how we chose xn and yn so that [xn+1 , yn+1 ] ⊂ [xn , yn ] means we can use the above
constant for each term in g as it appears.
n+1
g(yn+1 ) − g(xn+1 ) X hk (yn+1 ) − hk (xn+1 )
=
yn+1 − xn+1 k=0
yn+1 − xn+1
n
hn+1 (yn+1 ) − hn+1 (xn+1 ) X hk (yn+1 ) − hk (xn+1 )
= +
yn+1 − xn+1 k=0
yn+1 − xn+1
n
X hk (yn ) − hk (xn )
= ±1 +
k=0
yn − xn
g(yn ) − g(xn )
= ±1
y n − xn
This implies that
g(yn ) − g(xn )
lim
n→∞ y n − xn
does not exist, since the difference between consecutive elements does not converge to
zero, and therefore by our lemma in part (a), g is not differentiable at x.
Exercise 5.4.8
Review the argument for the nondifferentiability of g(x) atPnondyadic points. Does the
argument still work if we replacePg(x) with the summation ∞ n n
n=0 (1/2 )hn (3 x)? Does the
∞ n n
argument work for the function n=0 (1/3 )hn (2 x)?
5.4. A CONTINUOUS NOWHERE-DIFFERENTIABLE FUNCTION 139
Solution
The critical part to showing that g(x) is not differentiable at nondyadic points was showing
that
hn+1 (yn+1 ) − hn+1 (xn+1 )
yn+1 − xn+1
does
P∞ not converge to zero, preventing the limit defining the derivative to exist. For the case
n n
n=0 (1/2 )hn (3 x), the above term would diverge to infinity, since yn − xn would decrease
141
142 CHAPTER 6. SEQUENCES AND SERIES OF FUNCTIONS
Meaning for all ϵ > 0, setting N > 1/ϵ implies every n ≥ N has |fn (x) − f (x)| ≤
1/N < ϵ for every x ∈ (1, ∞).
Exercise 6.2.2
(a) Define a sequence of functions on R by
(
1 if x = 1, 21 , 13 , . . . , n1
fn (x) =
0 otherwise
In each case, explain how the results are consistent with the content of the Continuous
Limit Theorem (Theorem 6.2.6).
Solution
(a) Each fn is continuous at zero, but f is not continuous at zero meaning (by Theorem
6.2.6) that fn does not converge to f uniformly.
(b) Each gn is continuous at zero, and the pointwise limit g is also continuous at zero.
Since we aren’t contradicting 6.2.6 the convergence may or may not be uniform.
The definitions show |g(x) − gn (x)| < 1/n for all x (max is at x = 1/(n + 1)). Setting
N > 1/ϵ gives (for all n ≥ N and for all x ∈ R)
(c) Each hn is continuous at zero, and so is the pointwise limit h. 6.2.6 doesn’t apply so
we’ll have to check if the convergence is uniform. Notice that if xn = 1/n then
For all n, meaning no matter how big n is, we can’t make |h − hn | < 1/2 for all x
implying hn does not converge to h uniformly.
6.2. UNIFORM CONVERGENCE OF A SEQUENCE OF FUNCTIONS 143
Exercise 6.2.3
For each n ∈ N and x ∈ [0, ∞), let
(
x 1 if x ≥ 1/n
gn (x) = and hn (x) =
1 + xn nx if 0 ≤ x < 1/n
Answer the following questions for the sequences (gn ) and (hn );
(b) Explain how we know that the convergence cannot be uniform on [0, ∞).
(c) Choose a smaller set over which the convergence is uniform and supply an argument
to show that this is indeed the case.
Solution
(a)
x if x ∈ [0, 1)
(
1 if x > 0
lim gn (x) = 21 if x = 1 and lim hn (x) =
0 if x = 0
0 if x ∈ (1, ∞)
(b) They can’t converge uniformly since it would contradict Theorem 6.2.6 as both gn and
hn are continuous but the limit functions are not.
(c) Over [1, ∞) we have hn (x) = h(x) = 1 for all n, thus |hn (x) − h(x)| = 0 for all
x ∈ [1, ∞) so hn converges uniformly.
Now for gn . Let t ∈ [0, 1), I claim gn (x) → x uniformly over [0, t) since
x − x(1 + xn ) xn+1 n+1
x
1 + xn − x =
= < ϵ ∀x
1 + xn < t
1 + xn
Exercise 6.2.4
Review Exercise 5.2.8 which includes the definition for a uniformly differentiable function.
Use the results discussed in Section 2 to show that if f is uniformly differentiable, then f ′ is
continuous.
Solution
The definition of f being uniformly differentiable tells us: for every ϵ > 0 there exists a δ > 0
such that
f (x) − f (y) ′
x−y − f (y) < ϵ whenever 0 < |x − y| < δ
We can use this to show continuity of f ′ via a triangle inequality and exploiting the symmetry
in x, y.
′ ′
′ f (x) − f (y) f (x) − f (y) ′
|f (x) − f (y)| < f (x) −
+ − f (y) < ϵ
x−y x−y
144 CHAPTER 6. SEQUENCES AND SERIES OF FUNCTIONS
f (x) − f (yn )
fn′ (x) =
x − yn
Each fn′ (x) is continuous, and uniform differentiability implies that fn′ uniformly converges
to f ′ (x); hence f ′ (x) is continuous by Theorem 6.2.6.
Exercise 6.2.5
Using the Cauchy Criterion for convergent sequences of real numbers (Theorem 2.6.4), supply
a proof for Theorem 6.2.5 (Cauchy Criterion for Uniform Convergence). (First, define a
candidate for f (x), and then argue that fn → f uniformly.)
Solution
In the forward direction, suppose (fn ) converges uniformly to f and set N large enough that
n ≥ N has |fn (x) − f (x)| < ϵ/2 for all x, then use the triangle inequality (where m ≥ N as
well)
|fn (x) − fm (x)| ≤ |fn (x) − f (x)| + |f (x) − fm (x)| < ϵ/2 + ϵ/2 = ϵ.
In the reverse direction, suppose we can find an N so that every n, m ≥ N has |fn (x) −
fm (x)| < ϵ/2 for all x. Fix x and apply Theorem 2.6.4 to conclude the sequence (fn (x))
converges to some limit L, and define f (x) = L. Doing this for all x gives us the pointwise
limit f . Now we show (fn ) → f uniformly using the fact that |fn (x) − fm (x)| < ϵ/2 for all
x. Let n ≥ N , notice that for all m
For any x we can choose m large enough to ensure |fm (x) − f (x)| < ϵ/2 (pointwise con-
vergence), and since the inequality is for all m this implies |fn (x) − f (x)| ≤ ϵ for all x as
desired.
Exercise 6.2.6
Assume fn → f on a set A. Theorem 6.2.6 is an example of a typical type of question
which asks whether a trait possessed by each fn is inherited by the limit function. Provide
an example to show that all of the following propositions are false if the convergence is only
assumed to be pointwise on A. Then go back and decide which are true under the stronger
hypothesis of uniform convergence.
(c) If each fn has a finite number of discontinuities, then f has a finite number of discon-
tinuities.
(d) If each fn has fewer than M discontinuities (where M ∈ N is fixed), then f has fewer
than M discontinuities.
(e) If each fn has at most a countable number of discontinuities, then f has at most a
countable number of discontinuities.
Solution
(a) False pointwise when
(
1 1 if x = 0
fn (x) = and f (x) =
1 + nx2 0 otherwise
|f (x) − f (y)| ≤ |f (x) − fn (x)| + |fn (x) − fn (y)| + |fn (y) − f (y)|
By uniform convergence to f , we can fix n large enough so that the first and last terms
are both less than ϵ/3. Then since fn is uniformly continuous, we can choose δ so
that |x − y| < δ implies the middle term is less than ϵ/3, and that |f (x) − f (y)| < ϵ,
implying f is uniformly continuous.
and (
1/m if x = 1/m, m ∈ N
f (x) =
0 otherwise
The reason (fn ) → f uniformly is
( !
1/m if x = 1/m and m > n, m ∈ N
|f (x) − fn (x)| = < 1/(n + 1)
0 otherwise
Each fn has exactly n point discontinuities, but f has countably many. Thus this is a
counterexample.
Intuitively fn adds on finer and finer details of f , which is why it converges uniformly.
But discontinuities can be as small/detailed as we want without screwing up uniform
convergence.
(d) False pointwise since we can have each (fn ) continuous (zero discontinuities) but have
f not be continuous (see (a) for an example).
Now suppose (fn ) → f uniformly. Let x0 be a discontinuity of f , meaning there
exists an ϵ0 such that |f (x0 ) − f (x)| > ϵ0 no matter how small |x − x0 | is. I’d like to
show x0 is a discontinuity of fn for some n, i.e. that there exists an ϵ′0 > 0 such that
|fn (x0 ) − fn (x)| > ϵ′0 no matter how small |x − x0 | is.
Pick ϵ < ϵ0 /2 and set N large enough that n ≥ N implies |fn (x) − f (x)| < ϵ. Applying
the three way triangle inequality gives
Letting ϵ′0 = ϵ0 − 2ϵ > 0 we see |fn (x0 ) − fn (x)| > ϵ′0 meaning fn is not continuous at
x0 for all n ≥ N .
Now given discontinuities of f , applying the above process multiple times shows that
eventually fn will have every discontinuity of f , but fn has at most M discontinuities,
implying f has at most M discontinuities.
(e) False pointwise when (using a modified version of Thomae’s function for fn )
1 if x = 0
(
n 1 if x ∈ Q
fn (x) = n+q if x = p/q (in lowest terms) and f (x) =
0 if x ∈
/Q
0 otherwise
6.2. UNIFORM CONVERGENCE OF A SEQUENCE OF FUNCTIONS 147
which given a fixed n, becomes arbitrarily big as x goes to infinity. Hence (fn ) does not
converge uniformly.
Exercise 6.2.8
Let (gn ) be a sequence of continuous functions that converges uniformly to g on a compact
set K. If g(x) ̸= 0 on K, show (1/gn ) converges uniformly on K to 1/g.
Solution
Let’s examine the difference
g(x) − gn (x) 1
|1/g(x) − 1/gn (x)| =
= |g(x) − gn (x)|
g(x)gn (x) g(x)gn (x)
Exercise 6.2.9
Assume (fn ) and (gn ) are uniformly convergent sequences of functions.
(b) Give an example to show that the product (fn gn ) may not converge uniformly.
(c) Prove that if there exists an M > 0 such that |fn | ≤ M and |gn | ≤ M for all n ∈ N,
then (fn gn ) does converge uniformly.
Solution
(a) Obvious by the triangle inequality and Cauchy Criterion
(b) Let fn (x) = x = f (x) and gn (x) = x + 1/n. Suppose n, m ≥ N for some N , Cauchy
gives us
|fn gn − fm gm | = |x(1/n − 1/m)|
Making x large makes the error blow up regardless of how big N is, thus fn gn does not
converge uniformly.
(c) By the triangle inequality (same trick as for the product rule)
After setting N big enough that n, m ≥ N implies |fn − fm | < M/2ϵ and |gn − gm | <
M/2ϵ.
Exercise 6.2.10
This exercise and the next explore partial converses of the Continuous Limit Theorem (The-
orem 6.2.6). Assume fn → f pointwise on [a, b] and the limit function f is continuous on
[a, b]. If each fn is increasing (but not necessarily continuous), show fn → f uniformly.
Solution
f is continuous and therefore will map [a, b] to a closed interval [c, d]. By the Order Limit
Theorem, we also know that f must be increasing - just compare the sequence limn→∞ fn (x1 )
to limn→∞ fn (x2 ) when x2 > x1 .
The key implication of fn and f being increasing is that there is a bound to how fast
the error |f − fn | can grow with respect to x, which is proportional to how fast f grows.
Intuitively, if we have f > fn and want the error to grow as fast as possible with respect to
x, all we can do is hold fn constant. Alternatively, if we have f < fn , fn can’t go too far
above f since f needs to catch up eventually.
More formally, for ϵ > 0, define y1 , y2 , y3 , . . . , yn to evenly split up [c, d] into intervals of
size at most ϵ1 = ϵ/5. (In other words, y1 = c, yn = d, yk+2 − yk+1 = yk+1 − yk < ϵ1 .) Since
f is increasing, we can define x1 , . . . , xn so that f (xk ) = yk .
6.2. UNIFORM CONVERGENCE OF A SEQUENCE OF FUNCTIONS 149
Since fn → f , we can find Mk so that mk > Mk implies |yk − fmk (xk )| < ϵ1 . Let
M = max{M1 , M2 , . . . , Mn }. Now, let m > M be arbitrary. Keeping in mind that fm is
increasing, we can bound fm (xi+1 ) − fm (xi ) by
fm (xi+1 ) − fm (xi ) = |fm (xi+1 ) − fm (xi )|
≤ |fm (xi+1 ) − yi+1 | + |yi+1 − yi | + |yi − fm (xi )|
< ϵ1 + ϵ1 + ϵ1 = 3ϵ1
Now consider |f (x) − fm (x)| x ∈ [xi , xi+1 ] with i arbitrary. Since f is increasing,
yi ≤ f (x) ≤ yi+1 , so f (x) − yi < ϵ1 . Similarly fm (x) − fm (xi ) ≤ fm (xi+1 ) − fm (xi ) < 3ϵ1 .
Finally, we have
|f (x) − fm (x)| = |f (x) − yi | + |yi − fm (xi )| + |fm (x) − fm (xi )|
< ϵ1 + ϵ1 + 3ϵ1 = ϵ
Since m and i were arbitrary, this completes the proof.
Exercise 6.2.11 (Dini’s Theorem)
Assume fn → f pointwise on a compact set K and assume that for each x ∈ K the sequence
fn (x) is increasing. Follow these steps to show that if fn and f are continuous on K, then
the convergence is uniform.
(a) Set gn = f − fn and translate the preceding hypothesis into statements about the
sequence (gn ).
(b) Let ϵ > 0 be arbitrary, and define Kn = {x ∈ K : gn (x) ≥ ϵ} . Argue that K1 ⊇ K2 ⊇
K3 ⊇ · · · , and use this observation to finish the argument.
Solution
(a) We want (gn ) → 0 uniformly, where gn is continuous, gn (x) is decreasing and gn (x) → 0.
(b) x ∈ Kn+1 by definition has gn+1 (x) ≥ ϵ, since (gn (x)) is decreasing we must also have
gn (x) ≥ ϵ. Thus Kn+1 ⊆ Kn . Now, if each Kn ̸= ∅ T the nested compact set property
(Theorem 3.3.5) would imply there exists an x0 ∈ ∞ n=1 Kn . But this is impossible
because gn (x0 ) → 0 implies eventually gn (x0 ) < ϵ. Therefore since the compact set
property doesn’t apply, there must exist an N with KN = ∅, implying every n ≥ N
has (by subsets) Kn = ∅ and thus |gn | < ϵ.
Exercise 6.2.12 (Cantor Function)
Review the construction of the Cantor set C ⊆ [0, 1] from Section 3.1. This exercise makes
use of results and notation from this discussion.
(a) Define f0 (x) = x for all x ∈ [0, 1]. Now, let
(3/2)x
for 0 ≤ x ≤ 1/3
f1 (x) = 1/2 for 1/3 < x < 2/3
(3/2)x − 1/2 for 2/3 ≤ x ≤ 1
Sketch f0 and f1 over [0, 1] and observe that f1 is continuous, increasing, and constant
on the middle third (1/3, 2/3) = [0, 1]\C1 .
150 CHAPTER 6. SEQUENCES AND SERIES OF FUNCTIONS
(b) Construct f2 by imitating this process of flattening out the middle third of each non-
constant segment of f1 . Specifically, let
(1/2)f1 (3x)
for 0 ≤ x ≤ 1/3
f2 (x) = f1 (x) for 1/3 < x < 2/3
(1/2)f1 (3x − 2) + 1/2 for 2/3 ≤ x ≤ 1
If we continue this process, show that the resulting sequence (fn ) converges uniformly
on [0, 1].
(c) Let f = lim fn . Prove that f is a continuous, increasing function on [0, 1] with f (0) = 0
and f (1) = 1 that satisfies f ′ (x) = 0 for all x in the open set [0, 1]\C. Recall that the
“length” of the Cantor set C is 0 . Somehow, f manages to increase from 0 to 1 while
remaining constant on a set of “length 1.”
Solution
(a) f0 in red, f1 in blue:
(b) Define
(1/2)fn (3x)
for 0 ≤ x ≤ 1/3
fn+1 (x) = fn (x) for 1/3 < x < 2/3
(1/2)fn (3x − 2) + 1/2 for 2/3 ≤ x ≤ 1
We will aim to use the Cauchy Criterion to show fn uniformly converges. I first prove
through induction that |fn (x) − fn−1 (x)| < 1/2n . The base case of n = 1 is trivial by
the definitions of f1 and f0 .
Now first assume |fn (x) − fn−1 (x)| < 1/2n ; our goal is to show |fn+1 (x) − fn (x)| <
1/2n+1 . Note this is obviously true over (1/3, 2/3) since fn+1 (x) = fn (x) in this interval.
Consider x ∈ [0, 1/3], then
1 1 1 1 1 1
|fn+1 (x) − fn (x)| = fn (3x) − fn−1 (3x) = |fn (3x) − fn−1 (3x)| < =
2 2 2 2 2n 2n+1
(c) fn (0) = 0 and fn (1) = 1 can easily be proven with induction, and imply that each
fn is continuous. Since each fn is continuous, by the Continuous Limit Theorem f
is continuous. One of the sub-results from Exercise 6.2.10 was that all fn increasing
implies f is increasing, which can be applied here to show f is increasing. Since fn (0)
and fn (1) are constant with respect to n, f (0) = 0 and f (1) = 1.
Observe that for any n, fn+1 (x) = fn (x) for x ∈ [0, 1]\Cn . To prove this formally, we
can use induction. By definition fn+1 = fn (x) over (1/3, 2/3). Focusing on [0, 1/3], we
have
1
fn (3x) = fn (x) over [0, 1/3]\Cn
2
1 1
⇐⇒ fn (3x) = fn−1 (3x) over [0, 1/3]\ (Cn−1 /3)
2 2
⇐⇒fn (x) = fn−1 (x) over [0, 1]\Cn−1
Exercise 6.2.13
Recall that the Bolzano-Weierstrass Theorem (Theorem 2.5.5) states that every bounded
sequence of real numbers has a convergent subsequence. An analogous statement for bounded
sequences of functions is not true in general, but under stronger hypotheses several different
conclusions are possible. One avenue is to assume the common domain for all of the functions
in the sequence is countable. (Another is explored in the next two exercises.) Let A =
{x1 , x2 , x3 , . . .} be a countable set. For each n ∈ N, let fn be defined on A and assume there
exists an M > 0 such that |fn (x)| ≤ M for all n ∈ N and x ∈ A. Follow these steps to show
that there exists a subsequence of (fn ) that converges pointwise on A.
(a) Why does the sequence of real numbers fn (x1 ) necessarily contain a convergent sub-
sequence (fnk ) ? To indicate that the subsequence of functions (fnk ) is generated by
considering the values of the functions at x1 , we will use the notation fnN = f1,k .
(b) Now, explain why the sequence f1,k (x2 ) contains a convergent subsequence.
(c) Carefully construct a nested family of subsequences (fm,k ), and show how this can be
used to produce a single subsequence of (fn ) that converges at every point of A.
Solution
(a) Since fn is bounded, (fn (x1 )) is a bounded sequence, so by Bolzano-Weierstrauss
(fn (x1 )) has a convergent subsequence which uses fnk .
(c) Let f2,k represent the subsequence of functions generated in part (b), and keep in mind
that this subsequence of functions continues to converge pointwise at x1 . We can then
repeat this process, finding a subsequence of functions while considering f2,k (x3 ), and
repeat for every element in A.
If A were finite, we could simply take the last sequence and we would be done. But
since A is countably infinite, there is no last element; we need to be a bit more careful.
Instead of picking a particular subsequence, define a new subsequence of functions (gn ),
with gn = fn,n (where fa,b is the b’th element of the sequence of functions fa,k ). For
any fixed q, the sequence gn (xq ) is guarenteed to converge, since after q elements the
sequence consists solely of elements from fq,k which converges at xq .
Exercise 6.2.14
A sequence of functions (fn ) defined on a set E ⊆ R is called equicontinuous if for every
ϵ > 0 there exists a δ > 0 such that |fn (x) − fn (y)| < ϵ for all n ∈ N and |x − y| < δ in E.
(a) What is the difference between saying that a sequence of functions (fn ) is equicontinu-
ous and just asserting that each fn in the sequence is individually uniformly continuous?
(b) Give a qualitative explanation for why the sequence gn (x) = xn is not equicontinuous
on [0, 1]. Is each gn uniformly continuous on [0, 1] ?
Solution
(a) For equicontinuous functions the same δ works for every function in the sequence, as
opposed to individually being uniformly continuous where δ depends on n.
(b) Not equicontinuous since as n increases we need δ to be smaller, hence δ cannot be writ-
ten independent of n. Each gn is uniformly continuous however (since gn is continuous
on the compact set [0, 1]).
Exercise 6.2.15 (Arzela-Ascoli Theorem)
For each n ∈ N, let fn be a function defined on [0, 1]. If (fn ) is bounded on [0, 1]—that
is, there exists an M > 0 such that |fn (x)| ≤ M for all n ∈ N and x ∈ [0, 1]—and if the
collection of functions (fn ) is equicontinuous (Exercise 6.2.14), follow these steps to show
that (fn ) contains a uniformly convergent subsequence.
(a) Use Exercise 6.2.13 to produce a subsequence (fnk ) that converges at every rational
point in [0, 1]. To simplify the notation, set gk = fnk . It remains to show that (gk )
converges uniformly on all of [0, 1].
(b) Let ϵ > 0. By equicontinuity, there exists a δ > 0 such that
ϵ
|gk (x) − gk (y)| <
3
for all |x − y| < δ and k ∈ N. Using this δ, let r1 , r2 , . . . , rm be a finite collection of
rational points with the property that the union of the neighborhoods Vδ (ri ) contains
[0, 1]. Explain why there must exist an N ∈ N such that
ϵ
|gs (ri ) − gt (ri )| <
3
6.3. UNIFORM CONVERGENCE AND DIFFERENTIATION 153
for all s, t ≥ N and ri in the finite subset of [0, 1] just described. Why does having the
set {r1 , r2 , . . . , rm } be finite matter?
(c) Finish the argument by showing that, for an arbitrary x ∈ [0, 1],
|gs (x) − gt (x)| < ϵ
for all s, t ≥ N .
Solution
(a) ...is this actually a question? The rational numbers in [0, 1] are countable, so the results
from Exercise 6.2.13 can be applied.
(b) ri is a rational number, so gn (ri ) is a Cauchy sequence, and we can find Ni for each ri
where s, t > Ni ensures |gs (ri ) − gt (ri )| < ϵ/3. Then just have N = max{N1 , N2 , . . . , Nm }.
We need {r1 , r2 , . . . , rm } to be finite so that the final operation of taking the maximum
of all Ni is valid.
(c) We have
|gn (x) − gm (x)| ≤ |gn (x) − gn (a)| + |gn (a) − gm (a)| + |gm (a) − gm (x)|
where a is some rational number. The first and last terms can be made less than ϵ/3
by continuity of gn and gm and by choosing a close enough to x, while the middle term
can be made less than ϵ/3 from part (b).
Exercise 6.3.2
Consider the sequence of functions
r
1
hn (x) = x2 + .
n
(a) Compute the pointwise limit of (hn ) and then prove that the convergence is uniform
on R.
(b) Note that each hn is differentiable. Show g(x) = lim h′n (x) exists for all x, and explain
how we can be certain that the convergence is not uniform on any neighborhood of
zero.
Solution √ √ √ √
(a) As n → ∞, hn (x) → x2 = |x|. Now, recall that a + b ≥ a + b (I think
this has been√proved earlier,
√ but if not, this is easily shown by squaring both sides);
√
alternatively a + b − a ≤ b, and so
s √ r
1 1 1
x2 + − x2 ≤ =√
n n n
(b)
x
h′n (x) = p
2
x + 1/n
which converges to
1
x>0
g(x) = 0 x=0
−1 x < 0
Define h(x) to be the pointwise limit of (hn ). If the convergence to g was uniform, that
would imply that gn = h′ (x) (by the Differentiable Limit Theorem). But from part
(a) h is not differentiable at 0, therefore the convergence to g cannot be uniform in a
neighborhood around 0.
Exercise 6.3.3
Consider the sequence of functions
x
fn (x) = .
1 + nx2
(a) Find the points on R where each fn (x) attains its maximum and minimum value. Use
this to prove (fn ) converges uniformly on R. What is the limit function?
(b) Let f = lim fn . Compute fn′ (x) and find all the values of x for which f ′ (x) = lim fn′ (x).
6.3. UNIFORM CONVERGENCE AND DIFFERENTIATION 155
Solution
(a) fn is differentiable on R, so by the Interior Extremum Theorem the maximum and
minimum values will appear where fn′ (x) = 0. We have
1 − nx2
fn′ (x) =
(1 + nx2 )2
√
which is zero at x = ±1/ n. Plugging these values back into fn (x) we get that
|fn (x)| ≤ 2√1 n . Clearly this forces fn to converge uniformly to 0.
Exercise 6.3.4
Let
sin(nx)
hn (x) = √ .
n
Show that hn → 0 uniformly on R but that the sequence of derivatives (h′n ) diverges for
every x ∈ R.
Solution
|sin(x)| ≤ 1 and so |hn (x)| < √1n which shows that hn → 0 uniformly on R. h′n (x) =
n cos(nx) √ √
√
n
= n cos(nx). Intuitively this diverges because of the unbounded n factor, but to
prove it formally requires some thought. We want to show that for any fixed real numbers
x and M we can find some n where h′n (x) ≥ M (this will show that h′n (x) is unbounded
and thus diverges). First let N1 > 4M 2 , then we just need to find some n ≥ N1 so that
|nx − Zπ| ≤ π/3 for some integer Z; this would cause |cos(nx)| ≥ 1/2 and thus h′n (x) ≥ M .
Express N1 x = 2πP + k where P is some integer and 0 ≤ k < 2π. (This next bit
is reminiscent of arithmetic modulo 2π.) Now if k ∈ [0, π/3] or [2π/3, 4π/3] or [5π/3, 2π]
we’re done. Otherwise k ∈ (π/3, 2π/3) or (4π/3, 5π/3); so consider 2N1 x = 4πP + 2k, with
2k ∈ (2π/3, 4π/3) or (8π/3, 10π/3); both of these cases will have 2N1 x within π/3 of a
multiple of π, hence h′n (x) diverges for all x.
Exercise 6.3.5
Let
nx + x2
gn (x) =
2n
and set g(x) = lim gn (x). Show that g is differentiable in two ways:
(a) Compute g(x) by algebraically taking the limit as n → ∞ and then find g ′ (x).
(b) Compute gn′ (x) for each n ∈ N and show that the sequence of derivatives (gn′ ) converges
uniformly on every interval [−M, M ]. Use Theorem 6.3.3 to conclude g ′ (x) = lim gn′ (x).
156 CHAPTER 6. SEQUENCES AND SERIES OF FUNCTIONS
(c) Repeat parts (a) and (b) for the sequence fn (x) = (nx2 + 1) /(2n + x).
Solution
(a) By inspection g(x) = x/2 and g ′ (x) = 1/2.
(b) gn′ (x) = 1/2 + nx which approaches 12 as n → ∞. Now gn′ − 12 = nx is bounded by
M
n
which goes to 0 and is not dependent on x, and therefore (gn′ ) converges uniformly
over [−M, M ].
4n2 x + nx2 − 1
fn′ (x) = 2
4n + 4nx + x2
which approaches x as n → ∞. With some algebra we have
3
x − 3nx2 − 1 M 3 + 3nM 2 + 1
′
|fn (x) − x| = 2 ≤
4n + 4nx + x2 4n2 − 4M n
Exercise 6.3.6
Provide an example or explain why the request is impossible. Let’s take the domain of the
functions to be all of R.
(b) A sequence (fn ) of differentiable functions such that (fn′ ) converges uniformly but the
original sequence (fn ) does not converge for any x ∈ R.
(c) A sequence (fn ) of differentiable functions such that both (fn ) and (fn′ ) converge uni-
formly but f = lim fn is not differentiable at some point.
Solution
(a) Let g(x) be the continuous but nowhere-differentiable function defined in section 5.4.
Then since g(x) is bounded, fn (x) = g(x)/n clearly converges uniformly to 0, and is
thus such a sequence.
(b) Let (
1 n is odd
fn (x) =
0 n is even
Clearly fn does not converge anyewhere, but fn′ (x) converges to 0.
(c) Not possible - since fn converges uniformly, it must converge at at least one point, and
since fn′ converges uniformly, we can apply Theorem 6.3.3 to show that f is differen-
tiable everywhere.
6.4. SERIES OF FUNCTIONS 157
Exercise 6.3.7
Use the Mean Value Theorem to supply a proof for Theorem 6.3.2. To get started, observe
that the triangle inequality implies that, for any x ∈ [a, b] and m, n ∈ N,
|fn (x) − fm (x)| ≤ |(fn (x) − fm (x)) − (fn (x0 ) − fm (x0 ))| + |fn (x0 ) − fm (x0 )| .
Solution
Take any ϵ > 0; we want to show that there is some N so that for n, m > N , |fn (x)−fm (x)| <
ϵ (to use the Cauchy Criterion for Uniform Convergence).
Let hn,m (x) = fn (x) − fm (x), and note that h′n,m converges uniformly to 0 as n, m go to
infinity (as a consequence of (fn′ ) converging
uniformly).
More formally, for any ϵ1 > 0 we
have some N1 for which if n, m > N1 then h′n,m (x) < ϵ1 ∀x ∈ [a, b].
By the Mean Value Theorem,
|(fn (x) − fm (x)) − (fn (x0 ) − fm (x0 ))| = |x − x0 | h′n,m (x1 ) ≤ (b − a)ϵ1
Solution
Let ϵ > 0. Since ∞
P
n=1 Mn converges, by the Cauchy Criterion for Series there must be some
N where if n > m > N then ni=m+1 Mi < ϵ. Then
P
n
X n
X n
X
fi (x) ≤ |fi (x)| ≤ Mi < ϵ
i=m+1 i=m+1 i=m+1
Exercise 6.4.2
Decide whether each proposition is true or false, providing a short justification or counterex-
ample as appropriate.
(a) If ∞
P
n=1 gn converges uniformly, then (gn ) converges uniformly to zero.
158 CHAPTER 6. SEQUENCES AND SERIES OF FUNCTIONS
(c) If ∞
P
n=1 fn converges uniformly
P∞ on A, then there exist constants Mn such that |fn (x)| ≤
Mn for all x ∈ A and n=1 Mn converges.
Solution
(a) True: applying the Cauchy Criterion with n = m + 1 we have that |gn (x)| < ϵ for any
ϵ > 0, therefore (gn ) → 0.
(b) True:
X n Xn Xn X n
fi (x) = fi (x) ≤ gi (x) = gi (x) < ϵ
i=m+1 i=m+1 i=m+1 i=m+1
P
and therefore (fi ) converges uniformly.
(c) False: Consider the following sequence of functions, defined over [0, 1):
(
2−i 2−i (j − 1) ≤ x < 2−i j
gi,j (x) =
0 otherwise
with i ≥ 1 and j an integer ranging from 1 to 2i inclusive. Each gi,j (x) consists of a
pulse of height and width 2−i , at disjoint locations for each i. Let fn (x) be obtained
by iterating through each g1,j , then through each g2,j , then through each g3,j , and so
on.
P∞
n=1 fn converges to 1 because
X2i
gi,k = 2−i
k=1
, and uniform convergence is achieved when we include all of the gi,j for a given gi . On
the other hand, the upper bound (and therefore minimum value of the constant Mn )
for each gi,j is 2−i , with
X2i
max gi,k (x) = 1
k=1
P∞
which implies that n=1 Mn will not converge.
Exercise 6.4.3
(a) Show that
∞
X cos (2n x)
g(x) =
n=0
2n
is continuous on all of R.
(b) The function g was cited in Section 5.4 as an example of a continuous nowhere differ-
entiable function. What happens if we try to use Theorem 6.4.3 to explore whether g
is differentiable?
6.4. SERIES OF FUNCTIONS 159
Solution n x)
(a) Define gn (x) = cos(2 2n
and Mn = 2−n > |gn (x)|. By the Weierstrass M-test, g(x)
converges uniformly on R. Since each gn (x) is continuous and g(x) converges uniformly,
g(x) must also be continuous.
therefore the series does not converge. For |x| < 1, |hn (x)| ≤ x2n , which forms a geometric
series in x2 , which converges, so g(x) converges by the Order Limit Theorem.
Note that for any 0 ≤ a < 1, |h(x)| ≤ a2n = Mn over [−a, a], and thus by the Weierstrass
M-test g(x) uniformly converges over [−a, a] and is thus continuous over this interval. This
last statement is equivalent to saying g(x) is continuous over (−1, 1), which is also the set
where g(x) is well defined.
Exercise 6.4.5
(a) Prove that
∞
X xn x2 x3 x4
h(x) = =x+ + + + ···
n=1
n2 4 9 16
is continuous on [−1, 1].
Solution
(a) For x ∈ [−1, 1], we have n
x
≤ 1 = Mn
n2 n2
P 1
and since n2
converges (Example 2.4.4), h converges uniformly and is therefore
continuous.
160 CHAPTER 6. SEQUENCES AND SERIES OF FUNCTIONS
(b) Given a fixed x0 , we can consider the interval (−a, a) ⊂ [−1, 1) where −1 < −a <
n xn
|x0 | < a < 1. Then by setting Mn = an we will have Mn > n0 in a neighbourhood
around x0 , allowing us to show via the M-Test that f is continuous at x0 .
Exercise 6.4.6
Let
1 1 1 1 1
f (x) = − + − + − ··· .
x x+1 x+2 x+3 x+4
Show f is defined for all x > 0. Is f continuous on (0, ∞) ? How about differentiable?
Solution
f (x) converges for any x > 0 by the Alternating Series Test. Since f converges we are free
to use associativity to group the terms in pairs, from which we get
1 1 1 1
f (x) = − + − + ···
x x+1 x+2 x+3
1 1
= 2 + 2
+ ···
x + x (x + 2) + (x + 2)
1 1
< 2+ + ···
x (x + 2)2
1
Temporarily skipping the first term, we can use the Weierstrass M-Test with Mn = (2n) 2 , to
show that f (x) − 1/x converges uniformly and is therefore continuous for x > 0. Since 1/x2
2
by the Differentiable Limit Theorem. Since f (x) = g(x) + 1/x and both 1/x and g(x) are
differentiable over (0, ∞), we have that f is differentiable as well.
Exercise 6.4.7
Let
∞
X sin(kx)
f (x) =
k=1
k3
6.4. SERIES OF FUNCTIONS 161
(a) Show that f (x) is differentiable and that the derivative f ′ (x) is contimuous.
Solution
sin(nx)
(a) Let fn (x) = n3
. We have
cos(nx)
|fn′ (x)| = ≤ 1
n2 n2
and so ∞ ′
P
n fn (x) converges uniformly by the Weierstrass M-Test. We also have f (x)
converging at x = 0 (since every term is P zero), so by the differentiable limit theorem
we have f (x) differentiable with f (x) = ∞
′ ′
n=1 fn (x). Since this converges uniformly,
f ′ (x) is continuous.
sin(kx)
(b) Probably not easily - trying the same trick leaves us with trying to bound k with
Mn converges, but Mn = 1/k doesn’t work as ∞
P P
Mn where k=1 1/k diverges.
Exercise 6.4.8
Consider the function
∞
X sin(x/k)
f (x) = .
k=1
k
Where is f defined? Continuous? Differentiable? Twice-differentiable?
Solution
We can use the inequality |sin x| ≤ |x| to show that f (x) converges uniformly by the Weier-
strass M-Test over any interval (−a, a), with Mn = a/k 2 . Therefore f is defined and is
continuous over all real numbers.
The derivative of each term is
cos(x/k)
k2
which can easily be shown to converge uniformly; hence by the Differentiable Limit Theorem
we have that f (x) is differentiable as well. A similar argument shows that f is also twice-
differentiable.
Exercise 6.4.9
Let
∞
X 1
h(x) =
n=1
x2 + n2
Solution
1
(a) Use the M-Test with Mn = n2
.
162 CHAPTER 6. SEQUENCES AND SERIES OF FUNCTIONS
x2 x3 x4 x5
g(x) = x − + − + − ··· .
2 3 4 5
(a) Is g defined on (−1, 1) ? Is it continuous on this set? Is g defined on (−1, 1] ? Is
it continuous on this set? What happens on [−1, 1] ? Can the power series for g(x)
possibly converge for any other points |x| > 1 ? Explain.
Solution
(a) g(1) converges by the Alternating Series Test, so the radius of convergence is at least 1,
and g must be defined on at least (−1, 1]. Theorem 6.5.1 and Abel’s Theorem together
indicate that indicate that g converges absolutely on (−1, 1] as well. Thus, since each
term is continuous, g(x) is continuous on (−1, 1].
g is not defined at −1 since g(−1) would otherwise be
∞
X −1
n=1
n
6.5. POWER SERIES 163
which diverges.
g cannot converge at any point |x| > 1 because if it did, that would imply the radius of
convergence is strictly larger than 1, and thus g would need to converge at −1, which
it doesn’t.
(b) g ′ (x) is at least defined on (−1, 1), by Theorem 6.5.7, with the derivative given by
∞
X 1
g ′ (x) = (−x)n =
n=0
x+1
g ′ (x) cannot be defined at x ≤ −1 since g isn’t even defined there. To show that g ′ (1)
is defined and is also given by this formula requires a bit more care, since the infinite
sum does not actually converge for 1. We return to the definition of the derivative:
P∞ (−1)n+1 P∞ (−1)n+1 n ∞
′ − n=1 n x X (−1)n+1 1 − xn
g (1) = lim n=1 n = lim
x→1 1−x x→1
n
n 1−x
∞
1 X (−1)n+1
= lim (1 − xn )
x→1 1 − x n
n
With some algebra, we can show that this converges by the alternating series test,
n
keeping in mind that we can assume x ∈ (0, 1). We have 1−x
n
< n1 → 0, so we just
n n+1
need to show 1−x
n
≥ 1−x
n+1
:
1 − xn 1 − xn+1
≥ ⇐⇒ (1 − xn )(n + 1) ≥ n − nxn+1
n n+1
⇐⇒ n − nxn + 1 − xn ≥ n − nxn+1
⇐⇒ 1 − xn ≥ nxn (1 − x)
n−1 n−1
1 − xn X i X n
⇐⇒ = x ≥ x = nxn
1−x i=0 i=0
Now we know that g ′ (1) exists. We can show that g ′ (1) = 1+11 1
= 0.5 by noting that x+1 is
′
strictly decreasing on [0, 1), so in order for the derivative g (x) to maintain the intermediate
value property, g ′ (1) = 0.5.
Exercise 6.5.2
an xn has the given prop-
P
Find suitable coefficients (an ) so that the resulting power series
erties, or explain why such a request is impossible.
(c) Converges absolutely for all x ∈ [−1, 1] and diverges off of this set.
The first term is finite, while the second term converges by the original assumption.
This shows that F (x) is defined on −R, R, at which point we can use Theorem 6.5.7
to conclude F ′ (x) = f (x).
(b) From Corollary 5.3.4, g(x) = F (x) + k for some constant k; k gets folded into the
constant term of the power series.
Exercise 6.5.5
(a) If s satisfies 0 < s < 1, show nsn−1 is bounded for all n ≥ 1.
(b) Given an arbitrary x ∈ (−R, R), pick t to satisfy |x| < t < R. Use this start to
construct a proof for Theorem 6.5.6.
Solution
1
(a) Note first that all nsn−1 > 0, and that for n + 1 > N > 1−s (with N ∈ N, we can
n n−1
rearrange for s to have n+1 > s. This implies that ns > (n+1)sn ; thus the sequence
in n must be bounded by the maximum of the first N terms.
(b) Choose s satisfying
P∞ |s| = t andPwith s having the same sign as x. As a preliminary,
note that n=0 an s n−1
= (1/s) ∞ n
n=0 an s converges. We have
∞
X ∞
X x n−1 ∞
X
n−1 n−1
nan x = an s n ≤M an sn−1
n=0 n=0
s n=0
where, denoting p = x/s (with 0 < p < 1), M is an upper bound for npn−1 . This
completes the proof.
Exercise 6.5.6
Previous work on geometric series (Example 2.7.5) justifies the formula
1
= 1 + x + x2 + x3 + x4 + · · · , for all |x| < 1.
1−x
P∞
Use the results about power series proved in this section to find values for n=1 n/2n and
P ∞ 2 n
n=1 n /2 . The discussion in Section 6.1 may be helpful.
Solution
Let an = 1; we have
∞
X 1
an xn =
n=0
1−x
with a radius of convergence of 1. By Theorem 6.5.6 we can differentiate this termwise, to
get
∞ ∞ ∞ ∞ ∞
X
n−1
X
n
X
n
X 1 X 1
nan x = (n + 1)x = x + nxn = + nxn =
n=1 n=0 n=0 n=1
1 − x n=1 (1 − x)2
∞
X x
nxn =
n=1
(1 − x)2
166 CHAPTER 6. SEQUENCES AND SERIES OF FUNCTIONS
P∞
Substituting x = 1/2 we have n=1 n/2n = 2. We can differentiate the series again to get
∞ ∞ ∞ ∞
X
2 n−1
X
2 n
X
n
X 2
(n + n)x = n x +3 nx + 2 xn =
n=1 n=0 n=0 n=0
(1 − x)3
P∞
Substituting x = 1/2 we have n=1 n2 /2n = 6.
Exercise 6.5.7
an xn be a power series with an ̸= 0, and assume
P
Let
an+1
L = lim
n→∞ an
exists.
(a) Show that if L ̸= 0, then the series converges for all x in (−1/L, 1/L). (The advice in
Exercise 2.7.9 may be helpful.)
(b) Show that if L = 0, then the series converges for all x ∈ R.
(c) Show that (a) and (b) continue to hold if L is replaced by the limit.
′
a k+1
L = lim sn where sn = sup :k≥n .
n→∞ ak
(General properties of the limit superior are discussed in Exercise 2.4.7.)
Solution
(a) Let bn = an xn . If |x| < L, we have
bn+1 an+1 x
lim = lim = lim |Lx|
n→∞ bn n→∞ an n→∞
and thus by the ratio test, if |Lx| < 1 then the series ∞ n
P
n=1 an x converges. This
implies a radius of convergence of 1/L if L ̸= 0.
(b) By the same logic, if L = 0 then |Lx| < 1 regardless of x and the series converges
∀x ∈ R.
(c) Since (sn ) converges to L′ , for any ϵ > 0 we have that |ak+1 /ak | < M = L′ + ϵ once
k > N for some N ∈ N. Therefore by the ratio test and similar logic to above, the
radius of convergence is at least 1/M ; since ϵ is arbitrary, this is effectively a radius of
convergence of 1/L, so (a) and (b) continue to hold.
Exercise 6.5.8
(a) Show that power series representations are unique. If we have
∞
X ∞
X
n
an x = bn x n
n=0 n=0
Use Abel’s Theorem and the result in Exercise 2.8.7 to establish this result.
Solution
By Abel’s Theorem we have uniform convergence of the series defining f , g, and h over the
compact set [0, 1]; therefore each of these functions is continuous and bounded over this set.
We can thus conclude that for x ∈ [0, 1],
N X N N
! N !
X X X
ai xi bj xj = lim an x n bn xn = f (x)g(x)
lim
N →∞ N →∞
i=0 j=0 n=0 n=0
P
Since
P dn converges, limn→∞ |dn+1 /dn | ≤ 1 (otherwise dn would not be bounded). But
since dn only converges conditionally, limn→∞ |dn+1 /dn | = 1 (if it were less than 1, then
we could use the Ratio Test to prove absolute convergence). We therefore have absolute
convergence of the series defining h(x) for |x| < 1 by the Ratio Test.
From the work in Section 2.8, because we have absolute convergence, informally we have
a lot of leeway in how to evaluate the double summations when |x| < 1. In particular,
N X
X N ∞ X
X ∞ ∞
X
i j i j
dn xn = h(x)
lim ai x bj x = ai x bj x =
N →∞
i=0 j=0 i=0 j=0 n=0
We now have the equality f (x)g(x) = h(x), for |x| < 1. h(x) is a power series, and Abel’s
Theorem implies h(x) is continuous over [0, 1]. We also have continuity of f (x)g(x) over
P [0, 1];
thus by taking limits we have that h(1) = f (1)g(1), and we thus have that AB = dn .
168 CHAPTER 6. SEQUENCES AND SERIES OF FUNCTIONS
Exercise 6.5.10
Let g(x) = ∞ n
P
n=0 bn x converge on (−R, R), and assume (xn ) → 0 with xn ̸= 0. If g (xn ) = 0
for all n ∈ N, show that g(x) must be identically zero on all of (−R, R).
Solution
Let f (n) denote the n’th derivative of f (with f (0) = f ). The intermediate claims we make
along the way are:
2. Any function f with a bounded derivative over an interval containing 0 with some
sequence (xn ) → 0 satisfying f (xn ) = 0, will also satisfy f (0) = 0.
For claim 1, we apply the Mean Value Theorem to get some yn between xn and xn+1 with
′
f (yn ) = 0; we thus have |yn | ≤ max{xn , xn+1 } and therefore (yn ) → 0.
For claim 2, suppose that f (0) = ϵ ̸= 0, and |f ′ (x)| < M . Now since (xn ) → 0 we can
find some xi satisfying |xi | < ϵ/M . By the Mean Value Theorem, we then have that
′
ϵ
|f (c)| =
=M
ϵ/M
for some c, violating the assumption that f ′ (x) is bounded. Hence f (0) = 0.
For claim 3, we differentiate termwise n times and note that all terms that still have x
will evaluate to 0. We thus have
and thus an = 0.
From claim 1, we have by indution that every g (i) has some sequence (xi,n ) satisfying
limn→∞ xi,n → 0 and g (i) (xi,n ) = 0. Now since each of g (n) is bounded (by continuity over
the compact set [−R/2, R/2], for example), each of g (n) also has a bounded derivative, and
thus we can apply claim 2 to get that g (n) (0) = 0 for all n. Finally claim 3 implies that
bn = 0 for all n, and hence g(x) must be identically 0 over (−R, R).
ExerciseP6.5.11
A series ∞n=0 an is said to be Abel-summable to L if the power series
∞
X
f (x) = an x n
n=0
(a) Show that any series that converges to a limit L is also Abel-summable to L.
Solution
(a) If a series ∞
P P∞ n
n=0 an converges to L then by Abel’s Theorem the power series n=0 an x
converges uniformly over [0, 1], and is therefore continuous over this interval. Hence
by continuity the series is Abel-summable to L.
(b) The relevant power series here is f (x) = ∞ n
P
n=0 (−x) which has the closed-form ex-
1
pression 1+x for |x| < 1, and limx→1− f (x) evaluates to 1/2.
(b) Since
x d −1/8
2 2
=
(1 + 4x ) dx 1 + 4x2
We can use the geometric series then differentiate
∞ ∞
−1/8 X
2 n
X (−1)n+1 (4x2 )n
= (−1/8) (−4x ) =
1 − (−4x2 ) n=0 n=0
8
170 CHAPTER 6. SEQUENCES AND SERIES OF FUNCTIONS
Every |x| < 1/2 converges by the ratio test, meaning the P
right-hand series converges
for all x ∈ (−1/2, 1/2). Checking endpoints x = 1/2 gives (−1)n+1 n/2 which clearly
doesn’t converge, similarly x = −1/2 doesn’t converge.
Finally, we must show our differentiated series converges to the right thing. Let fm =
m ′
n+1
xn(42 )n−1 and f (x) = x/(1 + 4x2 ). We have (fm ) → f ′ uniformly (by
P
n=0 (−1)
construction via geometric series), and since (fm ) and f agree at x = 0 (fm (0) → 0
and f (0) = 0) we must have (fm ) → f uniformly.
X (−1)n xn ∞
1 1
log(1 + x) = x − x2 + x3 − · · · =
2 3 n=1
n
Exercise 6.6.3
Derive the formula for the Taylor coefficients given in Theorem 6.6.2.
Solution
We are given f (x) = ∞ n
P
n=0 an x defined on some nontrivial interval (−R, R). a0 = f (0) by
definition, now applying Theorem 6.5.7 we get
∞
X
′
f (x) = an nxn−1
n=1
and f ′ (0) = a1 · 1 by definition of f ′ , continue like this applying 6.5.7 each time to get
f (n) (0)
an =
n!
Exercise 6.6.4
Explain how Lagrange’s Remainder Theorem can be modified to prove
1 1 1 1 1
1− + − + − + · · · = log(2)
2 3 4 5 6
6.6. TAYLOR SERIES 171
Solution
(−1)n xn
Let EN (x) = log(1 + x) − ∞
P
n=1 n
. The proof presented of Lagrange’s Remainder
Theorem assumed that x > 0 to simplify notation, and found a value of c ∈ (0, x), with the
proof for x < 0 being implicit. But we can just ignore x < 0 to state that given x ∈ (0, R)
there exists 0 < c < x satisfying
f (N +1) (c) N +1
EN (x) = x
(N + 1)!
Now, f (N +1) (c) = N !(1 + c)−(N +1) , so for c < x = 1 we have
1
|EN (x)| ≤
N +1
which converges to 0 as N → ∞. Hence the Taylor series is equal to log(1 + x) over at least
(0, 1), and we can extend this equality to (0, 1] since both log(1 + x) and the Taylor series
are continuous at 1 (we established the latter in Exercise 6.5.1). Plugging x = 1 into both
equations leaves us with the desired equality.
Exercise 6.6.5
(a) Generate the Taylor coefficients for the exponential function f (x) = ex , and then prove
that the corresponding Taylor series converges uniformly to ex on any interval of the
form [−R, R].
(c) Use a substitution to generate the series for e−x , and then informally calculate ex · e−x
by multiplying together the two series and collecting common powers of x.
Solution
(a) f (x) = ex = ∞ n (n)
P
n=0 x /n! from f (0) = 1
n n n n
X n! k
X n k
X n X n
(n!)dn = (−1) = (−1) = −
k=0
k!(n − k)! k=0
k k=0, k even
k k=0, k odd
k
The first term is the total number of ways to choose an even number of elements from
a set of size n, while the second term is the number of ways to choose an odd number of
elements. For n odd, these two terms must be equal, since for every unique subset with
172 CHAPTER 6. SEQUENCES AND SERIES OF FUNCTIONS
an even number of elements, we get a unique subset with an odd number of elements
by taking the set complement. Hence for n odd we must have dn = 0.
It takes a bit more work for n even. Call this set N and divide it into disjoint subsets
A containing the first n − 1 elements and B containing the remaining element. Let
P P
even
be the number of ways to choose an even number of elements from P and odd
be defined similarly. Then
N A B A B A A
= + = +
odd odd even even odd odd even
while
N A B A B A A N
= + = + =
even even even odd odd even odd odd
and we once again have dn = 0. (Incidentally, this trick does not work for n = 0 since
it relies on being able to remove an element from N to form A.)
Putting everything together we have ex · e−x = d0 = 1 as expected.
Exercise 6.6.6
Review the proof that g ′ (0) = 0 for the function
( 2
e−1/x for x ̸= 0
g(x) =
0 for x = 0
(a) Compute g ′ (x) for x ̸= 0. Then use the definition of the derivative to find g ′′ (0).
(b) Compute g ′′ (x) and g ′′′ (x) for x ̸= 0. Use these observations and invent whatever
notation is needed to give a general description for the nth derivative g (n) (x) at points
different from zero.
(c) Construct a general argument for why g (n) (0) = 0 for all n ∈ N.
Solution
−2
(a) g ′ (x) = 2x−3 e−x . Repeatedly using L’Hospital’s Rule,
−2
2x−3 e−x 2x−4 −4x−5 x3 −2x−3 x3
g ′′ (0) = lim = lim 2 = 2 lim 2 = 4 lim 2 = 0
x→0 x x→0 e1/x x→0 −2e1/x x→0 −2e1/x
(b) Explicitly computing things sounds rather tedious, so let’s skip to the general form. We
2
claim that g (n) (x) for x ̸= 0 is of the form Pn (x)e−1/x where Pn (x) is some polynomial in
2
x−1 . We can prove this inductively, by noting g (n+1) (x) = [Pn′ (x)+Pn (x)(−2x−3 )]e−1/x .
Differentiating a polynomial in x−1 with respect to x only increases the powers, and
polynomials are closed under addition and multiplication, so clearly the term in square
brackets continues to be a polynomial in x−1 .
6.6. TAYLOR SERIES 173
It’s easy to show that this is equal to 0, since every time we apply L’Hospital’s rule, the
denominator doesn’t change (and continues to go to infinity) while the exponent in the
numerator increases by 2 each time, eventually becoming non-negative and preventing
the numerator from also going to infinity.
Exercise 6.6.7
Find an example of each of the following or explain why no such function exists.
(a) An infinitely differentiable function g(x) on all of R with a Taylor series that converges
to g(x) only for x ∈ (−1, 1).
(b) An infinitely differentiable function h(x) with the same Taylor series as sin(x) but such
that h(x) ̸= sin(x) for all x ̸= 0.
(c) An infinitely differentiable function f (x) on all of R with a Taylor series that converges
to f (x) if and only if x ≤ 0.
Solution
(a) g(x) = 1/(1 + x2 ). We already have that the Taylor series for this is ∞ n 2
P
n=0 (−1) x n,
and that it converges to g(x) for |x| < 1, but it does not converge at all at ±1.
(b) Let a(x) be the counterexample function introduced at the end of this section. Then
set h(x) = sin(x) + a(x); since the Taylor series of a(x) is identically 0, h(x) has the
same Taylor series as sin(x).
(c) (
0 x≤0
f (x) = 2
e−1/x x>0
has Taylor series identically 0.
Exercise 6.6.8
Here is a weaker form of Lagrange’s Remainder Theorem whose proof is arguably more
illuminating than the one for the stronger result.
(a) First establish a lemma: If g and h are differentiable on [0, x] with g(0) = h(0) and
g ′ (t) ≤ h′ (t) for all t ∈ [0, x], then g(t) ≤ h(t) for all t ∈ [0, x].
(b) Let f, SN , and EN be as Theorem 6.6.3, and take 0 < x < R. If f (N +1) (t) ≤ M for
all t ∈ [0, x], show
M xN +1
|EN (x)| ≤
(N + 1)!
174 CHAPTER 6. SEQUENCES AND SERIES OF FUNCTIONS
Solution
(a) Let e(t) = h(t) − g(t). If g(t) > h(t) for some t ∈ [0, x] then we could apply the
Mean Value Theorem on e(t) between 0 and t to find c ∈ (0, t) ⊆ [0, x] with e′ (c) =
h′ (t) − g ′ (t) < 0, a contradiction since h′ (t) ≥ g ′ (t).
(N +1) (N +1) (N +1)
(b) Since SN = 0, f (N +1) (t) = E (N +1) (t). We then have EN (t) ≤ EN ≤ M,
M xN +1
and by repeated application of the lemma in part (a) we have EN (x) ≤ (N +1)!
. A
M xN +1
similar argument holds for EN (x) ≥ − (N +1)! , and we can combine these succiently as
M xN +1
|EN (x)| ≤
(N + 1)!
Let EN (x, a) = f (x)−SN (x, a). Now fix x ̸= 0 in (−R, R) and consider EN (x, a) as a function
of a.
′ −f (N +1) (a)
EN (x, a) = (x − a)N .
N!
(c) Show
f (N +1) (c)
EN (x) = EN (x, 0) = (x − c)N x
N!
for some c between 0 and x. This is Cauchy’s form of the remainder for Taylor series
centered at the origin.
Solution
(a)
EN (x, x) = f (x) − SN (x, x) = f (x) − c0 = f (x) − f (x) = 0
(b)
N N
′ ′
X f (n) (a) n−1
X f (n+1) (a)
EN (x, a) = −SN (x, a) = (x − a) − (x − a)n
n=1
(n − 1)! n=0
n!
N −1 N
X f (n+1) (a) X f (n+1) (a) −f (N +1) (a)
= (x − a)n − (x − a)n = (x − a)N
n=0
(n)! n=0
n! N !
6.6. TAYLOR SERIES 175
so we know that the Taylor series at least converges to something for x ∈ [0, 1).
Lagrange’s Remainder Theorem gives us
Q q
N +1 1 N +1 1
i=1 (2i − 1) 1−c 1−c
xN +1
EN (x) =
2N +1 (N + 1)!
for some c ∈ (0, x). For x = 1/2 and c < x:
N +1
! N +1 r !
Y 2i − 1 1 1 1
|EN (x)| ≤
i=1
i 2 − 2c 1 − c 2 +1
N
r ! r
1 1 1
≤ 2N +1 dN +1 N +1
= dN +1
1−c 2 1−c
where d = 1/(2 − 2c) < 1; this shows EN converges to 0 over [0, 1/2].
Writing EN in product notation,
N +1
Y (2i − 1)x
EN (x) =
i=1
2i(1 − c)
176 CHAPTER 6. SEQUENCES AND SERIES OF FUNCTIONS
x
If x > 1/2, then it’s possible for 1−c
> 1. Then for some I, for i > I we have
x 2i
> >1
1−c 2i − 1
and beyond that point, the terms in the product begin increasing, with the product as
a whole growing exponentially and diverging.
(b) Plugging in Cauchy’s Remainder Theorem,
N
! r
Y 2i − 1 (2N + 1)(x − c)N x 1
EN (x) = N N +1
i=1
i 2 (1 − c) 1−c
r
x 1
≤ (2N + 1) (d)N
1−c 1−c
where d = x−c
1−c
< 1. The first term is linear in N , the second is exponentially decaying
in N , and the last two terms are constant, so the behaviour is dominated by exponential
decay and EN (x) converges to 0.
(b) Find the fourth degree polynomial that interpolates g(x) = |x| at the points x =
−1, −1/2, 0, 1/2, and 1 . Add a sketch of this polynomial to the graph from (a).
Solution
(a) p(x) = x2
(b) p(x) = 37 x2 − 43 x4
Exercise 6.7.4 √
Show that f (x) = 1 − x has Taylor series coefficients an where a0 = 1 and
−1 · 3 · 5 · · · (2n − 3)
an =
2 · 4 · 6 · · · 2n
for n ≥ 1.
Solution
For n ≥ 1:
QN −1
(2i − 1) 2n−1
(n)
= − i=1 n
f (1 − x)− 2
2
(n)
QN −1
f (0) (2i − 1) −1 · 3 · 5 · · · (2n − 3)
an = = − Q i=1 Q =
n! N N 2 · 4 · 6 · · · 2n
i=1 2 i=1 i
Exercise 6.7.5
(a) Follow the advice in Exercise 6.6.9 to prove the Cauchy form of the remainder:
f (N +1) (c)
EN (x) = (x − c)N x
N!
for some c between 0 and x.
(b) Use this result to prove equation (1) is valid for all x ∈ (−1, 1).
Solution
(a) See solution to Exercise 6.6.9
(b) !
1 N N
−x(1 − x)− 2 Y 2i − 1 x−c
EN (x) =
2 i=1
2i 1−c
For 0 < c < x < 1, the first term is constant, the second term is less than 1, and the
last term converges to 0.
Exercise 6.7.6
(a) Let
1 · 3 · 5 · · · (2n − 1)
cn =
2 · 4 · 6 · · · 2n
for n ≥ 1. Show cn < √ 2 .
2n+1
178 CHAPTER 6. SEQUENCES AND SERIES OF FUNCTIONS
(c) Carefully explain how this verifies that equation (1) holds for all x ∈ [−1, 1]
Solution
(a) We can show this by induction, if a bit inelegantly. The base case is trivial matter of
q
2n+1 2 2n+1
computation. For the inductive case, we want to show cn+1 = cn 2n+2 ≤ √2n+1 2n+3
. If
q
we work through the algebra of the claim 2n+1
2n+1
< 2n+12n+3
, we find that this is equivalent
to
8n3 + 20n2 + 14n + 3 < 8n3 + 20n2 + 16n + 4
which is clearly true for n ≥ 1, and the inductive step is done.
(b) |an | = cn /(2n − 1) < (2n−1)2√2n+1 ≤ 2 · (2n − 1)2/3 which implies absolute convergence
by comparison against an appropriate geometric series.
√
(c) (b) implies that the Taylor series of 1 − x converges absolutely at 1. With Theorem
6.5.2, the Taylor series converges uniformly over√[−1, 1] and is therefore continuous.
We also have that the Taylor series converges to 1 − x for x ∈ (−1, 1), which is also
continuous. Therefore taking limits as both functions approach ±1 gets us that they
are equal over [−1, 1].
Exercise 6.7.7 √
(a) Use the fact that |a| = a2 to prove that, given ϵ > 0, there exists a polynomial q(x)
satisfying
||x| − q(x)| < ϵ
for all x ∈ [−1, 1].
Solution √
(a) Let the polynomial p(x) be the partial sum of the Taylor series of 1 − x which satisfies
√ 2
p(x) − 1 − x < ϵ, and let q(x) = p(1 − x ). We then have
p
q(x) − 1 − (1 − x2 ) = ||x| − q(x)| < ϵ
as desired.
(b) Let c = max{|a|, |b|}, and let p(x) satisfy ||x| − p(x)| < ϵ/c. Then
x x ϵ
− p <
c c c
x
|x| − c · p <ϵ
c
for x ∈ [−c, c] ⊇ [a, b], so we can use the polynomial c · p( xc ).
6.7. THE WEIERSTRAUSS APPROXIMATION THEOREM 179
Exercise 6.7.8
(a) Fix a ∈ [−1, 1] and sketch
1
ha (x) = (|x − a| + (x − a))
2
over [−1, 1]. Note that ha is polygonal and satisfies ha (x) = 0 for all x ∈ [−1, a].
(b) Explain why we know ha (x) can be uniformly approximated with a polynomial on
[−1, 1].
(c) Let ϕ be a polygonal function that is linear on each subinterval of the partition
(d) Complete the proof of WAT for the interval [−1, 1], and then generalize to an arbitrary
interval [a, b].
Solution
(a) Left as an application for your favourite graphing calculator
(d) Fix ϵ > 0. For a function f continuous over [−1, 1], apporixmate it uniformly within
ϵ/2 with a polygonal function ϕ(x), and approximate ϕ(x) uniformly within ϵ/2 with
a polynomial. The triangle inequality ensures that this polynomial uniformly approx-
imates f within ϵ. To generalize over [a, b] the same technique in Exercise 6.6.7 of
scaling x and f can be used.
Exercise 6.7.9
(a) Find a counterexample which shows that WAT is not true if we replace the closed
interval [a, b] with the open interval (a, b).
(b) What happens if we replace [a, b] with the closed set [a, ∞). Does the theorem still
hold?
Solution
(a) 1/x over (0, 1), since 1/x is unbounded while any approximating polynomial must be
bounded over (0, 1).
(b) ex , since exponentials grow faster than polynomials, and therefore the difference be-
tween ex and any polynomial is unbounded over [0, ∞).
180 CHAPTER 6. SEQUENCES AND SERIES OF FUNCTIONS
Exercise 6.7.10
Is there a countable subset of polynomials C with the property that every continuous function
on [a, b] can be uniformly approximated by polynomials from C ?
Solution
Yes - we will approach this by adapting the logic used to prove WAT.
We start by choosing some sequence of polynomials which converge uniformly to |x|.
The set of all polynomials that appear in this sequence is countable; denote this set by A.
For a fixed a ∈ [−1, 1], we can turn this into a sequence of polynomials approaching ha as
described in Exercise 6.7.8; denote this countable set of polynomials Aa .
The rationals in [−1, 1] are countable, and therefore the union of all A⊣ for rational a is
also countable; denote this set as B. For a similar reason, the set of all polynomials of the
form a · p(x/b) where a, b are rational numbers and p ∈ B is also countable; denote this set
D.
Because rationals are dense in R, it’s easy to adapt the proof of Theorem 6.7.3 (Exercise
6.7.2) to only work with polygonal functions whose segment endpoints are rational (i.e. both
the endpoint and the value of the function at the endpoint are rational). Let P denote the
set of polygonal functions over [−1, 1] with rational segment endpoints only, and Pn be the
elements in P with n segments. Any element in Pn can be uniformly approximated by the
sum of n elements from D plus some rational constant c; call the set of polynomials
S∞ that can
be generated this way Dn . Since n is finite, Dn is also countable, and so is i=1 Di , meaning
we have a countable set of polynomials which can uniformly approximate any function from
P.
This implies any continuous function can be uniformly apprxoimated over [−1, 1] by a
countable set of polynomials. The scaling trick used to extend the original proof of WAT can
be used here (except limited to scaling factors of rational numbers) to complete the proof.
Exercise 6.7.11
Assume that f has a continuous derivative on [a, b]. Show that there exists a polynomial
p(x) such that
|f (x) − p(x)| < ϵ and |f ′ (x) − p′ (x)| < ϵ
for all x ∈ [a, b].
Solution
Let q(x) be a polynomial which uniformly approximates f ′ (x) to within (b−a)ϵ
, and p(x) to
be the antiderivative of q(x) which passes through (a, f (a)). Let e(x) = f (x) − p(x) be the
error between f and p, and note that |e′ (x)| < b−a
ϵ
.
By the Mean Value Theorem, for any x ∈ [a, b],
ϵ
|e(x) − e(a)| = |(x − a)e′ (c)| < (b − a) =ϵ
(b − a)
Solution
If U (f ) < L(f, P ) then since U (f ) = inf{U (f, P ) : P ∈ P} there must also be some P1 with
U (f, P1 ) < L(f, P ) which contradicts Lemma 7.2.4.
Similarly, if U (f ) < L(f ) then there must be some P where U (f ) < L(f, P ) we’ve just
shown to be impossible.
Exercise 7.2.2
Consider f (x) = 1/x over the interval [1, 4]. Let P be the partition consisting of the points
{1, 3/2, 2, 4}.
(b) What happens to the value of U (f, P ) − L(f, P ) when we add the point 3 to the
partition?
(c) Find a partition P ′ of [1, 4] for which U (f, P ′ ) − L (f, P ′ ) < 2/5.
Solution
TODO
Rb
and in this case a
f = limn→∞ U (f, Pn ) = limn→∞ L (f, Pn ).
181
182 CHAPTER 7. THE RIEMANN INTEGRAL
(b) For each n, let Pn be the partition of [0, 1] into n equal subintervals. Find formulas for
U (f, Pn ) and L (f, Pn ) if f (x) = x. The formula 1 + 2 + 3+ · · · + n = n(n + 1)/2 will
be useful.
Solution
(a) ( =⇒ ) If f is integrable, then we can choose PU n to satisfy U (f, PU n ) − U (f ) < 1/n,
PLn to satisfy L(f ) − L(f, PLn ) < 1/n, and Pn to be the common refimenemt of PU n
Rb
and PLn ; it’s easy to show in this case that limn→∞ U (f, Pn ) = U (f ) = a f = L(f ) =
limn→∞ L(f, Pn ).
( ⇐= ) Consider limn→∞ U (f, Pn ) − L(f ) ≤ U (f, Pn ) − L(f, Pn ). By the Squeeze
Theorem, U (f, Pn ) − L(f ) approaches 0, and therefore limn→∞ U (f, Pn ) = L(f ). A
similar argument shows that limn→∞ L(f, Pn ) = U (f ). Applying the Algebraic Limit
Rb
Theorem gets us that L(f ) − U (f ) = 0, or that L(f ) = U (f ) = a f by definition, with
this being equal to limn→∞ U (f, Pn ) = limn→∞ L(f, Pn ).
(b) Let [ani , bni ] be the i’th partition of Pn (indexing from 0), with ani = i/n and bni =
(i + 1)/n. It is easy to see that
n−1 n−1
X bni X i+1 1 1
U (f, Pn ) = = = +
i=0
n i=0
n2 2 2n
and
n−1 n−1
X ani X i 1 1
L(f, Pn ) = = 2
= −
i=0
n i=0
n 2 2n
R1
(c) U (f, Pn ) and L(f, Pn ) both approach 1/2 as n → ∞ and therefore 0
f = 1/2.
Exercise 7.2.4
Let g be bounded on [a, b] and assume there exists a partition P with L(g, P ) = U (g, P ).
Rb
Describe g. Is it integrable? If so, what is the value of a g ?
Solution
TODO
Exercise 7.2.5
Assume that, for each n, fn is an integrable function on [a, b]. If (fn ) → f uniformly on
[a, b], prove that f is also integrable on this set. (We will see that this conclusion does not
necessarily follow if the convergence is pointwise.)
Solution
Since both U (f, Pn ) and U (f ) are defined in terms of supremums (and likewise for L and
infimums), it’s useful to have the following lemma: For all n ∈ N, let An be a set of real
numbers with supremum sn and infimum in , and let B be a set with supremum s and infimum
7.2. THE DEFINITION OF THE RIEMANN INTEGRAL 183
. Because (fn ) → f uniformly, we can apply our lemma to claim that as n → ∞, un,c,d →
uc,d . Since the interval [c, d] is arbitrary, we can apply this to each interval in Pm to get
U (fn , Pm ) → U (f, Pm ).
This, plus using our lemma again, implies that U (fn ) → U (f ). A similar argument shows
L(fn ) → L(f ). But since U (fn ) = L(fn ), we must have U (f ) = L(f ) and therefore f is
integrable on [a, b].
Exercise 7.2.6
A tagged partition (P, {ck }) is one where in addition to a partition P we choose a sampling
point ck in each of the subintervals [xk−1 , xk ]. The corresponding Riemann sum,
n
X
R(f, P ) = f (ck ) ∆xk ,
k=1
|R(f, P ) − A| < ϵ.
Show that if f satisfies Riemann’s definition above, then f is integrable in the sense of
Definition 7.2.7. (The full equivalence of these two characterizations of integrability is proved
in Section 8.1.)
Solution
Let ϵ > 0. By Riemann’s definition we can easily form a partition Pn , and tag it to form Pu
with the sampling point close to the supremum, so that
Similarly we can retag the partition to form Pl with the sampling point close to the infimum,
to get
|R(f, Pl ) − A| < ϵ/4 and |L(f, Pn ) − R(f, Pl )| < ϵ/4
Applying the Triangle Inequality we get U (f, Pn ) − L(f, Pn ) < ϵ, so by Theorem 7.2.8 f is
integrable.
Exercise 7.2.7
Let f : [a, b] → R be increasing on the set [a, b] (i.e., f (x) ≤ f (y) whenever x < y). Show
that f is integrable on [a, b].
Solution
Since f is increasing, let Pn be comprised of the n + 1 evenly spaced points {x0 , · · · , xn },
with ∆xk = xk − xk−1 = ∆n (where ∆n is a constant for a fixed n). Then
n
X n
X
U (f, P ) = f (xi )(xi − xi−1 ) = ∆n f (xi )
i=1 i=1
and
n−1
X n−1
X
L(f, P ) = f (xi )(xi+1 − xi ) = ∆n f (xi )
i=0 i=0
Therefore U (f, P ) − L(f, P ) = ∆n (f (b) − f (a)). For a given ϵ > 0 if we set ∆n < ϵ/(f (b) −
f (a)) then we can use Theorem 7.2.8 to conclude f is integrable.
(b) Let ϵ > 0, and consider the set of points Dϵ/2 = {x ∈ [0, 1] : t(x) ≥ ϵ/2}. How big is
Dϵ/2 ?
(c) To complete the argument, explain how to construct a partition Pϵ of [0, 1] so that
U (t, Pϵ ) < ϵ.
Solution
(a) Because the irrationals are dense, for any interval there must be some point in that
interval where t(x) = 0, which is also the minimum possible value of t; hence L(t, P ) =
0.
(b) This is the size of the set of rational numbers which, when expressed in lowest terms
as m/n, have n ≤ 2/ϵ. This must be finite; let the size of Dϵ/2 be denoted as |Dϵ/2 |.
(c) Informally, we’ll set up the partition to “isolate” each point in Dϵ/2 . Define the partition
as
Pϵ = {d + k/2 : d ∈ Dϵ/2 } ∪ {d − k/2 : d ∈ Dϵ/2 } ∪ {0, 1}
where k = ϵ/(2|Dϵ/2 |). (If some of these isolating intervals overlap, then decrease k
until they don’t; this can be formalized by requring k be the minimum of what it is
currently and the smallest distance between two points in Dϵ/2 .) Then
where the first term comes from all partition intervals that aren’t isolating an element
in Dϵ/2 , and the second term comes from the elements in Dϵ/2 .
Exercise 7.3.3
Let (
1 if x = 1/n for some n ∈ N
f (x) =
0 otherwise.
R1
Show that f is integrable on [0, 1] and compute 0 f .
Solution
TODO
Exercise 7.3.4
Let f and g be functions defined on (possibly different) closed intervals, and assume the
range of f is contained in the domain of g so that the composition g ◦ f is properly defined.
(a) Show, by example, that it is not the case that if f and g are integrable, then g ◦ f is
integrable. Now decide on the validity of each of the following conjectures, supplying
a proof or counterexample as appropriate.
186 CHAPTER 7. THE RIEMANN INTEGRAL
Solution
(a) Let f (x) be Thomae’s function (which we’ve shown in Exercise 7.3.2 to be integrable),
and define g over the interval [0, 1] as
(
0 x=0
g(x) =
1 x>0
By Theorem 7.3.2, g is integrable. But for any rational number q, f (q) > 0, and so
g ◦ f is just Dirichlet’s function, which is not integrable.
(b) False. Note: the following counterexample requires results from Section 7.6.
Let c(x) be the Cantor Function, defined in Exercise 6.2.12, and consider d(x) =
c(x) + x, defined over [0, 1]. Recall that c(x) is increasing and continuous; therefore
d(x) is strictly increasing and continuous. Therefore, since d is one-to-one, d−1 is
defined, continuous, and increasing over [0, 2].
Notably, c maps the Cantor set C (which has measure zero) to a set of positive measure,
since c maps [0, 1]\C to the dyadic points in [0, 1], which are countable and thus have
measure zero. This behaviour is also true of d. To prove this a bit formally, assume
for contradiction that d(C) = {d(x) : x ∈ C} has measure zero, and find a collection
of open intervals with total length less than ϵ whose union contains d(C). For a
representative interval (p1 , p2 ) = (d(x1 ), d(x2 )), let d(S) be the set of points in d(C)
which are covered by (p1 , p2 ), where S be the points in C which map to d(S). But note
that then (c(x1 ), c(x2 )) ⊇ c(S), and that
Repeating this for every interval leaves us with a collection of intervals covering c(C)
whose total length is less than ϵ, a contradiction since c(C) has positive measure.
Finally, let f = d−1 and g be the indicator function of C; that is,
(
1 x∈C
g(x) =
0 x∈ /C
We show later in Exercise 7.3.9 that g is integrable and its set of discontinuities is C.
But I claim that g ◦ f is discontinuous on d(C), a set of positive measure. This claim
would imply that g ◦ f is not integrable.
Note that f −1 = d is continuous. If g ◦ f was continuous on any point x ∈ d(C), then
g = (g ◦ f ) ◦ f −1 should be continuous on d−1 (x) ∈ C, but g is discontinuous over C.
Thus a contradiction and g ◦ f is discontinuous over d(C).
Exercise 7.3.5
Provide an example or give a reason why the request is impossible.
(a) A sequence (fn ) → f pointwise, where each fn has at most a finite number of discon-
tinuities but f is not integrable.
(b) A sequence (gn ) → g uniformly where each gn has at most a finite number of disconti-
nuities and g is not integrable.
(c) A sequence (hn ) → h uniformly where each hn is not integrable but h is integrable.
Solution
(a) Let (rn ) be an enumeration of the rational numbers in [0, 1], let Rn be the set of ri
where i ≤ n, and let (
1 x ∈ Rn
fn (x) =
0 otherwise
Then fn has n discontinuities, and (fn ) approaches Dirichlet’s function pointwise.
Exercise 7.3.6
Let {r1 , r2 , r3 , . . .} be an enumeration of all the rationals in [0, 1], and define
(
1 if x = rn
gn (x) =
0 otherwise.
P∞
(a) Is G(x) = n=1 gn (x) integrable on [0, 1] ?
P∞
(b) Is F (x) = n=1 gn (x)/n integrable on [0, 1] ?
Solution
(a) G(x) is Dirichlet’s function and is not integrable
(b) The same approach as that used in Exercise 7.3.2 can be used; in particular the set of
points {x ∈ [0, 1] : F (x) ≥ ϵ/2} is finite.
Exercise 7.3.7
Assume f : [a, b] → R is integrable.
(a) Show that if g satisfies g(x) = f (x) for all but a finite number of points in [a, b], then
g is integrable as well.
(b) Find an example to show that g may fail to be integrable if it differs from f at a
countable number of points.
Solution
TODO
188 CHAPTER 7. THE RIEMANN INTEGRAL
Exercise 7.3.8
As in Exercise 7.3.6, let {r1 , r2 , r3 , . . .} be an enumeration of the rationals in [0, 1], but this
time define
1 if rn < x ≤ 1
hn (x) = .
0 if 0 ≤ x ≤ rn
P∞ n
Show H(x) = n=1 hn (x)/2 is integrable on [0, 1] even though it has discontinuities at
every rational point.
SolutionP
Hn (x) = ni=1 hi (x)/2n is integrable and (Hn ) uniformly approaches H.
(a) Let f be bounded on [a, b]. Show that if the set of discontinuous points of f has content
zero, then f is integrable.
(c) Content zero sets do not have to be finite. They do not have to be countable. Show
that the Cantor set C defined in Section 3.1 has content zero.
Solution
(a) Let ϵ > 0, let M be the difference between the maximum and minimum values of
fP. Let A be the set of discontinuous points of f , and choose {O1 , . . . , ON } so that
|On | < ϵ/(2M ). Assume without loss of generality that none of Oi overlap; if there
was an overlapping pair then of intervals then they can be merged into a single interval.
Start a partition P with the endpoints of each Oi , and note that the sum of (Mk −
mk )∆xk over the subintervals from each Oi is less than ϵ/2. Moreover over the remain-
ing subintervals (of which there are finitely many) f is continuous and thus integrable,
and so it is possible to refine P so that over the remaining subintervals the sum of
(Mk − mk )∆xk is less than ϵ/2.
Putting these together gets us that f is integrable.
(b) Let this finite set A have N elements, then take Vδ (x) where x ∈ A and δ = ϵ/N .
7.4. PROPERTIES OF THE INTEGRAL 189
(c) Refer to Section 3.1 for the definition of Cn . Let n be large enough so that the total
length of the intervals in Cn is less than ϵ/2. Let Cn,i refer to the i’th interval in Cn ,
and define Oi so that Cn,i ⊂ Oi and |Oi | − |Cn,i | = δ, where δ < ϵ/2n+1 is small enough
Pn Pn
so that 2i=1 |Oi | − 2i=1 |Cn,i | < ϵ/2.
(d) C is closed and its complement is open; therefore the set of discontinuities of h is C.
Parts (a) and (c) imply h is integrable.
Exercises 3.5.8 and 3.5.9 show that the complement of C is dense, so for any partition
P , L(h, P ) = 0. Thus since h is integrable, the value of the integral must be 0.
(b) Show that if f is integrable on the interval [a, b], then |f | is also integrable on this
interval.
R R
b b
(c) Provide the details for the argument that in this case we have a f ≤ a |f |.
Solution
(a) If f (x) ≥ 0 then M ′ = M, m′ = m, and the result is trivial. Similarly if f (x) ≤ 0
then M ′ = −m, m′ = −M . Finally if f (x) has both positive and negative values then
M ≥ 0, m ≤ 0, M ′ = max{M, −m}, m′ ≥ 0, M − m = M + (−m) ≥ M ′ ≥ M ′ − m′
(b) Given part (a), if we have a partition P so that U (f, P ) − L(f, P ) < ϵ, then we also
have U (|f |, P ) − L(|f |, P ) ≤ U (f, P ) − L(f, P )ϵ.
Rb Rb Rb Rb Rb
(c) f ≤ |f |,
R so a
f ≤ a
|f |. Similarly a
f ≥ a
−|f | = − a
|f |, which together means
b
R b
a f ≤ a |f |
Exercise 7.4.2
(a) Let g(x) = x3 , and classify each of the following as positive, negative, or zero.
Z −1 Z 1 Z 0 Z 1 Z −2 Z 1
(i) g+ g (ii) g+ g (iii) g+ g.
0 0 1 0 1 0
(b) Show that if b ≤ a ≤ c and f is integrable on the interval [b, c], then it is still the case
Rb Rc Rb
that a f = a f + c f .
190 CHAPTER 7. THE RIEMANN INTEGRAL
Solution
TODO
Exercise 7.4.3
Decide which of the following conjectures is true and supply a short proof. For those that
are not true, give a counterexample.
(b) Assume g is integrable and g(x) ≥ 0 on [a, b]. If g(x) > 0 for an infinite number of
Rb
points x ∈ [a, b], then a g > 0.
(c) If g is continuous on [a, b] and g(x) ≥ 0 with g(y0 ) > 0 for at least one point y0 ∈ [a, b],
Rb
then a g > 0.
Solution
(a) Letting d be Dirichlet’s function, let f (x) = d(x) − 1/2 (not integrable), with |f | = 1/2
(integrable)
(c) Let ϵ = g(y0 )/2. Since g is continuous there must be some δ > 0 so that over Vδ (y0 ),g >
y0 /2. Then let (
g(y0 )/2 x ∈ Vδ (y0 )
h(x) =
0 otherwise
Rb Rb
then since g ≥ h, a g ≥ a h = δg(y 2
0)
> 0.
Exercise 7.4.4 Rb
Show that if f (x) > 0 for all x ∈ [a, b] and f is integrable, then a f > 0.
Solution
I claim that there exists some non-empty interval I = [c, d] ⊆ [a, b] and some ϵ > 0 such that
for all non-empty subintervals J ⊆ I, there exists x ∈ J with f (x) ≥ ϵ. Note that once we
Rb Rb
prove this claim, we can readily show a f ≥ a g = ϵ(d − c), where
(
ϵ x∈I
g(x) =
0 otherwise
We’ll prove this claim by contradiction. Suppose that for all non-empty intervals I ⊆ [a, b]
and for all ϵ > 0, there exists a non-empty subinterval of J ⊆ I where f (x) < ϵ ∀x ∈ J.
1/n and In ⊆ In−1 satisfying f (x) < ϵn ∀x ∈ In .
Now let I0 = [a, b], and for n > 0, let ϵn = T
Now by the Nested Interval Property, ∃s ∈ ∞ n=1 In . Now, s satisfies f (s) < 1/n for all n,
implying f (s) ≤ 0 — a contradiction since f (x) > 0 for all x ∈ [a, b].
Exercise 7.4.5
Let f and g be integrable functions on [a, b].
7.4. PROPERTIES OF THE INTEGRAL 191
U (f + g, P ) ≤ U (f, P ) + U (g, P ).
Provide a specific example where the inequality is strict. What does the corresponding
inequality for lower sums look like?
(b) Review the proof of Theorem 7.4.2 (ii), and provide an argument for part (i) of this
theorem.
Solution
TODO
Exercise 7.4.6
Although not part of Theorem 7.4.2, it is true that the product of integrable functions is
integrable. Provide the details for each step in the following proof of this fact:
(a) If f satisfies |f (x)| ≤ M on [a, b], show
(f (x))2 − (f (y))2 ≤ 2M |f (x) − f (y)|
(c) Now show that if f and g are integrable, then f g is integrable. (Consider (f + g)2 .)
Solution
(a)
f (x)2 − f (y)2 = |f (x) + f (y)| |f (x) − f (y)| ≤ 2M |f (x) − f (y)|
(b) Define the set Si,j = [(i − 1)/j, i/j] for j > i > 0, and the sequence of sets Ri
enumerating through all Si,j ; specifically R1 = S1,1 , R2 = S1,2 , R3 = S2,2 , R4 = S1,3 ,
and so on. Let (
1 x ∈ Rn
gn (x) =
0 otherwise
Each gn looks like a “pulse”, and as n increases and we increase the j index of Si,j , the
R1
pulse gets increasingly narrower; in particular, if gn is based on Si,j then 0 gn = 1/j.
R1
This ensures limn→∞ 0 gn = 0. However, as we go through the i index, we slide the
pulse over all values in [0, 1]. A bit more formally, for all x ∈ [0, 1] and for all j > 0,
x ∈ Si,j for some i. This ensures that for all x ∈ [0, 1], gn (x) = 1 infinitely many times,
preventing gn (x) → 0.
Exercise 7.4.8
For each n ∈ N, let
1/2n if 1/2n < x ≤ 1
hn (x) = ,
0 if 0 ≤ x ≤ 1/2n
P∞ R1
and set H(x) = n=1 hn (x). Show H is integrable and compute 0
H.
Solution
R1
h = (1/2)n − (1/4)n . Hn (x) = ni=1 hi (x) is integrable, and (Hn ) → H uniformly, so
P
0 i
an
R 1 application of the geometric series formula and the Integrable Limit Theorem gives us
0
H = 2/3.
Exercise 7.4.9
Let gn and g be uniformly bounded on [0, 1], meaning that there exists a single M > 0
satisfying |g(x)| ≤ M and |gn (x)| ≤ M for all n ∈ N and x ∈ [0, 1]. Assume gn → g
pointwise on [0, 1] and uniformly on any setR of the Rform [0, α], where 0 < α < 1. If all the
1 1
functions are integrable, show that limn→∞ 0 gn = 0 g.
Solution
TODO
Exercise 7.4.10
Assume g is integrable on [0, 1] and continuous at 0 . Show
Z 1
lim g (xn ) dx = g(0)
n→∞ 0
7.5. THE FUNDAMENTAL THEOREM OF CALCULUS 193
Solution
Let |g| < M and let M > ϵ > 0. Define the partition P = {0, 1 − ϵ/(4M ), 1}. Since g is
continuous, we can find N so that for n > N ,
ϵ n ϵ
g 1 − − g(0)<
4M 4
Thus,
ϵ ϵ Mϵ
U (g(xn ), P ) ≤ g(0) + 1− +
4 4M 4M
ϵ 2M ϵ ϵ
≤ g(0) + + + = g(0) + ϵ
4 4M 4
A similar argument can be made that L(g(xn ), P ) ≥ g(0) − ϵ.Assuming that g(xn ) is inte-
grable, this implies that Z 1
n
g(0) − g(x )dx<ϵ
0
completing the proof.
Exercise 7.4.11
Review the original definition of integrability in Section 7.2, and in particular the definition
of the upper integral U (f ). One reasonable suggestion might be to bypass the complications
introduced in Definition 7.2.7 and simply define the integral to be the value of U (f ). Then
every bounded function is integrable! Although tempting, proceeding in this way has some
significant drawbacks. Show by example that several of the properties in Theorem 7.4.2
no
R b longer hold if we replace our current definition of integrability with the proposal that
a
f = U (f ) for every bounded function f .
Solution
TODO
Solution
TODO
Exercise 7.5.2
Decide whether each statement is true or false, providing a short justification for each con-
clusion.
194 CHAPTER 7. THE RIEMANN INTEGRAL
Solution
(a) False, e.g. h(x) = x2 sin(1/x) as explored in section 5.1.
(b) True; if g is continuous then it Ris also integrable, and by the Fundamental Theorem of
x
Calculus we have that G(x) = a g is differentiable over [a, b] with G′ = g.
(c) False;Rconsider Thomae’s function t(x). From R xExercise 7.3.2 we have that t is integrable
1
with 0 t = 0; it’s also easy to show that 0 t = 0. Then H(x) = 0 is differentiable
over [0, 1], but t(x) is not continuous over [0, 1].
Exercise 7.5.3
The hypothesis in Theorem 7.5.1 (i) that F ′ (x) = f (x) for all x ∈ [a, b] is slightly stronger
than it needs to be. Carefully read the proof and state exactly what needs to be assumed
with regard to the relationship between f and F for the proof to be valid.
Solution
We use the Mean Value Theorem to ensure that for all subintervals of a particular partition
[xk−1 , xk ], we have F (xk ) − F (xk−1 ) = F ′ (tk )(xk − xk−1 ). But the MVT does not require
that F is differentiable at the endpoints, so we can relax the requirement slightly and have
F ′ (x) = f (x) over the open interval (a, b).
Exercise 7.5.4 Rx
Show that if f : [a, b] → R is continuous and a f = 0 for all x ∈ [a, b], then f (x) = 0
everywhere on [a, b]. Provide an example to show that this conclusion does not follow if f is
not continuous.
Solution Rx
Since f is continuous, by Theorem 7.5.1 (ii), letting F (x) = a f = 0, we have f (x) =
F ′ (x) = 0. If f is not continuous, then this does not hold (e.g. Thomae’s function over
[0, 1])
Exercise 7.5.5
The Fundamental Theorem of Calculus can be used to supply a shorter argument for Theo-
rem 6.3.1 under the additional assumption that the sequence of derivatives is continuous.
Assume fn → f pointwise and fn′ → g uniformly on [a, b]. Assuming each fn′ is continuous,
we can apply Theorem 7.5.1 (i) to get
Z x
fn′ = fn (x) − fn (a)
a
Solution Rx
Since
Rx fn′ → g uniformly, g is continuous. The Integrable Limit Theorem tells us a fn′ →
a
g = f (x) − f (a), and Theorem 7.5.1 (ii) tells us G(x) = f (x) − f (a) satisfies
(b) Explain how the result in Exercise 7.4.6 can be used to slightly weaken the hypothesis
in part (a).
Solution
TODO
Exercise 7.5.7
Use part (ii) of Theorem 7.5.1 to construct another proof of part (i) of Theorem
Rx 7.5.1 under
the stronger hypothesis that f is continuous. (To get started, set G(x) = a f .)
Solution
TODO
Solution
(a) L(1) = 0. Since 1/t is continuous, by Theorem 7.5.1 (ii) L′ (x) = 1/t.
(b) Differenting L(xy) with respect to x leaves us with 1/x, so by Theorem 7.5.1 (i) we
have
Z x
1
dt = L(x) − L(1) = L(x) = L(xy) − L(y) =⇒ L(xy) = L(x) + L(y)
1 t
(e)
γ2n − γn = h2n − hn − L(2n) + L(n) = h2n − hn − L(2)
If we pair every element of hn with every other element of h2n we get
2n
X (−1)i+1
γ2n − γn + L(2) =
i=1
i
Since (γn ) converges, the left side converges to L(2), giving us the desired identity.
Exercise 7.5.9
Given a function f on [a, b], define the total variation of f to be
( n )
X
V f = sup |f (xk ) − f (xk−1 )| ,
k=1
(b) Use the Mean Value Theorem to establish the reverse inequality and conclude that
Rb
V f = a |f ′ |.
Solution
(a) For any subinterval [c, d] ⊆ [a, b],
d d
Z Z
′ ′
|f (x)| ≥ f (x) = |f (d) − f (c)|
c c
Rb
and hence V f ≤ a
|f ′ |.
(b) Let ϵ > 0. Since |f ′ | is uniformly continuous over the closed interval [a, b], we can
create a partition P with n elements so that each subinterval has length less than δ,
where |x − y| < δ =⇒ ||f ′ (x)| − |f ′ (y)|| < b−a
ϵ
.
Within a representative subinterval [xk−1 , xk ] of P , by the Mean Value Theorem ∃x̂k ∈
[xk−1 , xk ] satisfying
f (xk ) − f (xk−1 )
f ′ (x̂k ) =
xk − xk−1
Letting Mk be the supremum of |f ′ | over [xk−1 , xk ], note that
n n
′
X X f (xk ) − f (xk−1 )
U (|f | , P ) − |f (xk ) − f (xk−1 )| = Mk −
(xk − xk−1 )
k=1 k=1
x k − x k−1
n
X
= (Mk − |f ′ (x̂k )|) (xk − xk−1 )
k=1
n
ϵ X
≤ xk − xk−1 = ϵ
b − a k−1
Rb Rb Rb
Since U (|f ′ |, P ) ≥ a
|f ′ |, we have V f ≥ a
|f ′ | − ϵ, or simply V f ≥ a
|f ′ |.
It’s worth noting that we can’t use argument with L(|f ′ |, P ) for the ≤ inequality
because of the sup in V f .
(a) Why are we sure f is the derivative of some function? How about (f ◦ g)g ′ ?
(b) Prove the change-of-variable formula
Z b Z g(b)
′
f (g(x))g (x)dx = f (t)dt.
a g(a)
Solution
TODO
Exercise 7.5.11
Assume f is integrable on [a, b] and has a ”jump discontinuity” at c ∈ (a, b). This means
that both one-sided limits exist as x approaches c from the left and from the right, but that
where we’ve essentially shifted f afterR x x > c to line up with x < c, so that g is
continuous at c. As a result, G(x) = a g is differentiable at c.
Rx Rx
Now if F = a f was also differentiable at c,Rthen we would also have G − F = a g − f
x
differentiable at c. But actually evaluating a g − f leaves us with
Z x (
0 x≤c
g−f =
a (l − r)(x − c) x > c
(a) First argue that L(t, P ) = 0 for any partition P of [0, 1].
(b) Let ϵ > 0, and consider the set of points Dϵ/2 = {x ∈ [0, 1] : t(x) ≥ ϵ/2}. How big is
Dϵ/2 ?
(c) To complete the argument, explain how to construct a partition Pϵ of [0, 1] so that
U (t, Pϵ ) < ϵ.
Solution
See Exercise 7.3.2
Exercise 7.6.2
We first met the Cantor set C in Section 3.1. We have since learned that C is a compact,
uncountable subset of the interval [0, 1].
Define (
1 if x ∈ C
h(x) =
0 if x ∈
/C
(a) Show h has discontinuities at each point of C and is continuous at every point of the
complement of C. Thus, h is not continuous on an uncountably infinite set.
Solution
See Exercise 7.3.9 (d)
Exercise 7.6.3
Show that any countable set has measure zero.
Solution
Enumerate the elements of any countable set A as {a1 , a2 , · · · }. Given ϵ > 0, define the
ϵ ϵ
collection of open intervals On with Oi = (ai − 2i+1 , ai + 2i+1 ).
Exercise 7.6.4
Prove that the Cantor set has measure zero.
200 CHAPTER 7. THE RIEMANN INTEGRAL
Solution
Exercise 7.3.9 (c) showed that the Cantor set has content zero, and any set with content
zero clearly has measure zero.
Exercise 7.6.5
Show that if two sets A and B each have measure zero, then A ∪ B has measure zero as well.
In addition, discuss the proof of the stronger statement that the countable union of sets of
measure zero also has measure zero.
Solution
Let ϵ > 0, and define OA,n and OB,n so that
∞ ∞ ∞ ∞
[ [ X ϵ X ϵ
A⊆ OA,n , B ⊆ OB,n , |OA,n | ≤ , |OB,n | ≤
n=1 n=1 n=1
2 n=1 2
Then let On be the union of the sets in OA,n and OB,n ; On satisifies the conditions necessary
to show A ∪ B has measure zero.
This argument is essentially identical for a countable union of sets of measure zero. Some
key points:
• The countable union of countable sets is also countable.
• Since |OAm ,n | ≥ 0, the infinte sums involved all converge absolutely, so we can use the
results in Section 2.8 to safely sum the lengths of intervals over an enumeration of the
final collection of sets On .
Exercise 7.6.6
′
If α < α′ , show that Dα ⊆ Dα .
Solution
See Exercise 4.6.9.
Exercise 7.6.7
(a) Let α > 0 be given. Show that if f is continuous at x ∈ [a, b], then it is α-continuous
at x as well. Explain how it follows that Dα ⊆ D.
(b) Show that if f is not continuous at x, then f is not α-continuous for some α > 0. Now,
explain why this guarentees that
∞
[
D= D αn where αn = 1/n
n=1
Solution
(a) See Exercise 4.6.10
Exercise 7.6.8
Prove that for a fixed α > 0, the set Dα is closed.
Solution
See Exercise 4.6.8
Exercise 7.6.9
Show that there exists a finite collection of disjoint open intervals {G1 , G2 , . . . , GN } whose
union contains Dα and that satisfies
N
X ϵ
|Gn | <
n=1
4M
Solution
Given D has measure zero, and since Dα ⊆ D, we also know that P∞ Dα has measure zero. We
ϵ
can thus construct a countable open cover {H1 , . . . } such that n=1 |Gn | < 4M .
α
Since D is closed, we can find a finite subcover {I1 , . . . , IP }. Finally, since this is a finite
set, we can merge any overlapping intervals (which can only decrease the total length of the
intervals), leaving us with the desired finite collection of disjoint open intervals.
Exercise 7.6.10
Let K be what remains of the interval [a, b] after the open intervals Gn are all removed; that
is, K = [a, b]\ N
S
n=1 Gn . Argue that f is uniformly α-continuous on K.
Solution
Note that since each Gn is open, and the union of any collection of open sets is open, N
S
n=1 Gn
SN c
is open, its complement is closed, and K = [a, b] ∩ ( n=1 Gn ) is closed. K is also bounded
so K is compact.
By how we defined Gn , f is α-continuous pointwise on K. Taking it as fact that α-
continuity on a compact set implies uniform α-continuity, since K is compact, we have that
f is uniformly α-continuous on K.
Exercise 7.6.11
Finish the proof in this direction by explaining how to construct a partition Pϵ of [a, b] such
that U (f, Pϵ ) − L(f, Pϵ ) ≤ ϵ.
Solution
ϵ
Let α = 2(b−a) , and |f | < M . Begin by defining the collection of open intervals G =
{G1 , . . . , GN } with the properties described in Exercise 7.6.9. Define P1 as the partition
containing all of the endpoints of G.
Consider the collection K of subintervals which do not contain elements of Dα ; by Exercise
7.6.10, f is uniformly α-continuous over all sets in K. Therefore, there exists a δ > 0 so that
for x, y ∈ K, |f (x) − f (y)| < α. Define P2 as the partition consisting of evenly spaced points
such that each subinterval has length at most δ, and define Pϵ as the common refinement of
P1 and P2 .
202 CHAPTER 7. THE RIEMANN INTEGRAL
Let X X
U (f, Pϵ ) − L(f, Pϵ ) = (Mk − mk )∆xk + (Mk − mk )∆xk
k∈S1 k∈S2
where S1 includes all of the intervals containing elements of Dα and S2 includes all of the
other subintervals. Now
X X ϵ
(Mk − mk )∆xk < 2M ∆xk <
k∈S k∈S
2
1 1
and X ϵ X ϵ ϵ
(Mk − mk )∆xk < ∆xk ≤ (b − a) =
k∈S2
2(b − a) k∈S 2(b − a) 2
2
Exercise 7.6.12
(a) Prove that Dα has measure zero. Point out that it is possible to choose a cover for Dα
that consists of a finite number of open intervals.
Solution
(a) Consider the collection of subintervals S in Pϵ where Mk − mk > α. Now, for any
d ∈ Dα , if d ∈
/ Pϵ then clearly d lies in a subinterval in S.
Now, the sums of the lengths of elements of S cannot be greater than ϵ (otherwise
U (f, Pϵ ) − L(f, Pϵ ) ≥ αϵ). So we can easily create a collection of open intervals O1 ,
whose length is no more than 2ϵ, whose union is a superset of the union of all elements
of S. A value greater than ϵ is needed because subintervals are closed intervals, so we
need to go a little beyond the endpoint in each case.
Now, all elements of Dα appear in some set in O1 , except possibly for elements in Pϵ .
However, there are only finitely many elements of Pϵ , so we can just have a second
collection of open intervals O2 with total length ϵ covering all elements of Pϵ . The
union of O1 and O2 give us a collection of open intervals, whose length is no more than
3ϵ, and whose union contains all elements of Dα . Since ϵ was arbitrary, this argument
can readily be extended to show Dα has measure zero.
(b) We know the countable union of sets of measure zero also has measure zero, and
Exercise 7.6.7 (b) constructs D as the countable union of many Dαn , each of which has
measure zero; hence D has measure zero.
Exercise 7.6.13
Show that if f and g are integralbe on [a, b], then so is the product f g.
Show that if g is integrable on [a, b] and f is continuous on the range of g, then the compo-
sition f ◦ g is integrable on [a, b].
7.6. LEBESGUE’S CRITERION FOR RIEMANN INTEGRABILITY 203
Solution
(a) The set of discontinuities of f g is the union of the set of discontinuities of f and that
of g. Since f and g are both integrable, these are both sets with measure zero, and
hence their union is also measure zero; thus f g is integrable.
(b) Since f is continuous, f ◦ g and g share the same sets of discontinuity; therefore g is
integrable implies f ◦ g is integrable.
Exercise 7.6.14
(a) Find g ′ (0).
(b) Use the standard rules of differentiation to compute g ′ (x) for x ̸= 0.
(c) Explain why, for every δ > 0, g ′ (x) attains every value between 1 and −1 as x ranges
over the set −δ, δ. Conclude that g ′ is not continuous at x = 0.
Solution
See discussion in Section 5.1.
Exercise 7.6.15
(a) If c ∈ C, what is limn→∞ fn (c)?
(b) Why does limn→∞ fn (x) exist for x ∈
/ C?
Solution
(a) By construction fn (x) = 0 for x ∈ Cn , and by definition c ∈ C =⇒ c ∈ Cn ∀n, so
limn→∞ fn (c) = 0.
(b) If x ∈
/ C then for some N , x ∈ / Cn ∀n ≥ N . Now by construction fn = fN over Cnc , so
for n ≥ N , fn (x) = fN (x) is constant and therefore limn→∞ fn (x) = fN (x).
Exercise 7.6.16
(a) Explain why f ′ (x) exists for all x ∈
/ C.
(b) If c ∈ C, argue that |f (x)| ≤ (x − c)2 for all x ∈ [0, 1]. Show how this implies f ′ (c) = 0.
(c) Give a careful argument for while f ′ (x) fails to be continuous on C. Remember that C
contains many points besides the endpoints of the intervals that make up C1 , C2 , C3 , . . ..
Solution
(a) As noted in Exercise 7.6.15 (b), if x ∈
/ C then fn (x) = fN (x) for some N and all n ≥ N .
Since [0, 1]\C is open, there must be some [a, b] ⊆ [0, 1]\C with x ∈ [a, b]; clearly (fn )
converges uniformly in [a, b] so by the Differentiable Limit Theorem f ′ (x) exists and is
equal to fN′ (x).
(b) Note that for c ∈ Cn , we have |fn (x)| ≤ (x − c)2 . Given c ∈ C, for all n, c ∈ Cn and
so |fn (x)| ≤ (x − c)2 . Thus by the Algebraic Limit Theorem |f (x)| ≤ (x − c)2 . Now,
f (x) − f (c) f (x)
f ′ (c) = lim =
x→c x−c x−c
And given |f (x)| ≤ (x − c)2 , we can use the Squeeze Theorem to find f ′ (c) = 0.
204 CHAPTER 7. THE RIEMANN INTEGRAL
(c) Although C contains many points other than the interval endpoints of Cn , every point
in C is “close” to some interval endpoint. More formally, if c ∈ C, ∀ϵ > 0 ∃x satisfying
|x − c| < ϵ, where x is an endpoint of one of the intervals in Cn . To see this, simply
take n large enough that 3−n (the length of each interval in Cn ) is less than ϵ; then
since c ∈ Cn (and specifically, is in one of the intervals that make up Cn ) there must
be an endpoint of an interval in Cn which is less than ϵ away from c.
To show that f ′ is discontinuous at any c ∈ C, we need to show that for some fixed
ϵ > 0, ∀δ > 0 ∃x ∈ Vδ (c) where |f ′ (x) − f ′ (c)| = |f ′ (x)| > ϵ. Take ϵ = 1/2, and let p
be an interval endpoint in Cn where |c − p| < δ/2. We also have, using Exercise 7.6.14
(c), that over Vδ/2 (p), |fn′ (p)| > ϵ at some x, and it’s clear that x ∈ / Cn . Therefore
′ ′
|f (x)| = |fn (x)| > ϵ, and by the Triangle Inequality x ∈ Vδ (c).
Exercise 7.6.17
Why is f ′ (x) Riemann-integrable on [0, 1]?
Solution
f ′ (x) is discontinuous only at c ∈ C, and C has measure zero.
Exercise 7.6.18
Show that, under these circumstances, the sum of the lengths of the intervals making up
each Cn no longer tends to zero as n → ∞. What is this limit?
Solution
∞ ∞ n−1
X
n−1 1 1X 2 1
2 = =
n=1
3n+1 9 n=1 3 3
Exercise 7.6.19
As a final gesture, provide the example advertised in Exercise 7.6.13 of an integrable function
f and a continuous function g where the composition f ◦ g is properly defined but not
integrable.
Solution
The solution to Exercise 7.3.4 satisfies this request (although f and g have swapped names).
Chapter 8
Additional Topics
205
206 CHAPTER 8. ADDITIONAL TOPICS
R(f, P ) = U (f, P )
(b) If f is not continuous, it may not be possible to find tags for which R(f, P ) = U (f, P ).
Show, however, that given an arbitrary ϵ > 0, it is possible to pick tags for P so that
Solution
(a) Each subinterval is closed, and since f is continuous, the image of each subinterval
under f (the set of points which f maps the subinterval to) is also closed, and thus
contains its supremum and infimum; this allows us to pick tags so that R(f, P ) =
U (f, P ).
Solution
Given arbitrary ϵ > 0, find a δ and a corresponding δ-fine partition P so that for any set of
tags {ck }, we have
ϵ
|R(f, P ) − A| <
4
Then pick tags c1 so that R(f, (P, c1 )) − L(f, P ) < ϵ/4 and tags c2 so that U (f, P ) −
R(f, (P, c2 )) < ϵ/4. For conciseness let L = L(f, P ), U = U (f, P ), R1 = R(f, (P, c1 )),
R2 = R(f, (P, c2 )). Then
U − L ≤ U − R1 + |R1 − A| + |A − R2 | + R2 − L < ϵ
showing f is Riemann-integrable.
Exercise 8.1.6
Consider the interval [0, 1].
(a) If δ(x) = 1/9, find a δ(x)-fine tagged partition of [0, 1]. Does the choice of tags matter
in this case?
8.1. THE GENERALIZED RIEMANN INTEGRAL 207
(b) Let (
1/4 if x = 0
δ(x) =
x/3 if 0 < x ≤ 1
Construct a δ(x)-fine tagged partition of [0, 1].
Solution
(a) P = {n/10 : n ∈ N, 0 ≤ n ≤ 10}. Choice of tags does not matter since δ(x) is constant.
Exercise 8.1.7
Finish the proof of Theorem 8.1.5.
Solution
If both halves have a tag which can satisfy δ(ci ) > (b − a)/2, then we’re done. Otherwise,
bisect any halves which do not have such a tag. Repeat, identifying subintervals that have
candidate tags and bisecting the subintervals that don’t. To show this process ends, by
contradiction assume that it doesn’t and there is an infinite sequence of nested subintervals
(Ii ) which do not have a candidate tag, and by construction the length of these subintervals
goes to zero. By the Nested Interval Property there must be some c in the intersection of
all of the subintervals, but δ(c) > 0, meaning that at some point c would have been a valid
tag for some In - a contradiction.
Exercise 8.1.8
Finish the argument.
Solution
Consider an arbitrary ϵ > 0. We have that with some gauge δ1 (x), we have |R(f, P ) − A1 | <
ϵ/2 for all P which are δ1 (x)-fine. Similarly we have δ2 (x) where |R(f, P ) − A2 | < ϵ/2. Let
δ(x) = min{δ1 (x), δ2 (x)} and note that δ is also a gauge, and that any partition P which is
δ(x)-fine is also δ1 (x)-fine and δ2 (x)-fine. Find a partition P which is δ(x)-fine.
We then have |R(f, P ) − A1 | < ϵ/2 and |A2 − R(f, P )| < ϵ/2, so |A2 − A1 | < ϵ, for
arbitrary ϵ > 0. Therefore A1 = A2 .
Exercise 8.1.9 Rb
Explain why every function that is Riemann-integrable with a f = A must also have gen-
eralized Riemann integral A.
Solution
For a given ϵ > 0, we can simply use the gauge δ(x) = δ, at which point the Riemann integral
and generalized Riemann integral are equivalent.
Exercise 8.1.10
Show that if (P, {ck }nk=1 ) is a δ(x)-fine tagged partition, then R(g, P ) < ϵ.
208 CHAPTER 8. ADDITIONAL TOPICS
Solution
Subintervals with irrational tags do not contribute to R(g, P ), so the maximum value of
R(g, PP) is the sum of the lengths of the subintervals with rational tags. This can be no more
than x∈Q δ(x) = ϵ. To show strict inequality, note that any subinterval must contain more
than one rational number, and so some of the subinterval lengths are “missing”.
Exercise 8.1.11
Show that
n
X
F (b) − F (a) = [F (xk ) − F (xk−1 )]
k=1
Solution
This is a telescoping sum, with the right hand side becoming F (xn ) − F (x0 ) = F (b) − F (a).
Exercise 8.1.12
For each c ∈ [a, b], explain why there exists a δ(c) > 0 (a δ > 0 depending on c) such that
F (x) − F (c)
− f (c) <ϵ
x−c
evaluate at x = xk−1 , c = ck and multiply both sides by |ck − xk−1 |, noting that
ck ≥ xk−1 .
8.2. METRIC SPACES AND THE BAIRE CATEGORY THEOREM 209
(b) Add the two inequalities in part (a) together and apply the Triangle Inequality. Then
summing over k from 1 to n gets us
ϵ(xn − x0 ) = ϵ(b − a)
Xn
> |F (xk ) − F (xk−1 ) − f (ck )(xk − xk−1 )|
k=1
≥ |F (b) − F (a) − R(f, P )|
To prove the theorem we need |F (b) − F (a) − R(f, P )| < ϵ, which can be readily
obtained by instead constructing the gauge δ(c) for ϵ/(b − a).
Exercise 8.1.14
(a) Why are we sure that f and (F ◦ g)′ have generalized Riemann integrals?
(b)
Z b Z b Z g(b)
′ ′
(f ◦ g) · g = (F ◦ g) = (F ◦ g)(b) − (F ◦ g)(a) = F (g(b)) − F (g(a)) = f
a a g(a)
(b) It’s fairly easy to see that the first two properties are met. To demonstrate the triangle
inequality, note
and
d(x, z) + d(z, y) ≥ |x2 − z2 | + |z2 − y2 | ≥ |x2 − y2 |
therefore d(x, z) + d(z, y) is greater than or equal to both possible values of d(x, y).
210 CHAPTER 8. ADDITIONAL TOPICS
(c) Property (i) is not met; take x = (0, 1) and y = (1, 0).
Exercise 8.2.2
Let C[0, 1] be the collection of continuous functions on the closed interval [0, 1]. Decide
which of the following are metrics on C[0, 1].
Solution
(a) The first two properties are trivial. For the triangle inequality:
Solution
Property (i) and (ii) are trivial. For property (iii), there are two cases - either x = y or
x ̸= y. If x = y then certainly ρ(x, z) + ρ(z, y) ≥ 0 by property (i). If x ̸= y then at least
one of z ̸= y and z ̸= x is true, so
Exercise 8.2.4
Show that a convergent sequence is Cauchy.
Solution
Given (xn ) → x, choose N large enough that for n ≥ N , d(xn , x) < ϵ/2. Then for m, n ≥ N ,
d(xm − xn ) ≤ d(xm − x) + d(x, xn ) < ϵ
Exercise 8.2.5
(a) Consider R2 with the discrete metric ρ(x, y) examined in Exercise 8.2.3. What do
Cauchy sequences look like in this space? Is R2 complete with respect to this metric?
8.2. METRIC SPACES AND THE BAIRE CATEGORY THEOREM 211
(b) Show that C[0, 1] is complete with respect to the metric in Exercise 8.2.2 (a).
(c) Define C 1 [0, 1] to be the collection of differentiable functions on [0, 1] whose derivatives
are also continuous. Is C 1 [0, 1] complete with respect to the metric defined in Exercise
8.2.2 (a)?
Solution
(a) In order to reach any ϵ < 1, after a certain point in any Cauchy sequence, all elements
need to be identical, with the sequence converging to this identical value. Therefore
R2 (and any set) is complete with respect to the discrete metric.
(b) By Theorem 6.2.5 (Cauchy Criterion for Uniform Convergence), any (fn ) which is a
Cauchy sequence, uniformly converges to some function f . Since uniform convergence
preserves continuity, f is also continuous and thus in C[0, 1].
(c) Recall that Theorem 6.2.5 is an if-and-only-if statement, so any uniformly convergent
sequence of functions is a Cauchy sequence. With that in mind, Exercise 6.3.2 is an
example of a sequence of functions which converges uniformly to a function which is
not differentiable. Thus C 1 [0, 1] is not complete with respect to this metric.
Exercise 8.2.6
Which of these functions from C[0, 1] to R (with the usual metric) are continuous?
R1
(a) g(f ) = 0 f k, where k is some fixed function in C[0, 1].
(b) g(f ) = f (1/2).
(c) g(f ) = f (1/2), but this time with respect to the metric on C[0, 1] from Exercise 8.2.2
(c).
Solution
(a) Continuous. Let M be a bound on |k|, and for any ϵ > 0, let δ = ϵ/M . Then for any
function h satisfying d(f, h) < δ, we have
Z 1 Z 1 Z 1 Z 1 Z 1
fk − hk = (f − h)k ≤
|(f − h)| |k| < δM = ϵ
0 0 0 0 0
(b) Continuous. Let δ = ϵ, and note that |f (1/2) − h(1/2)| ≤ d(f, h) < δ = ϵ
(c) Not continuous. Let f = 0, and for any δ > 0, define
(
1/2 x ∈ Vδ (1/2)
hδ (x) =
0 otherwise
Clearly for any δ, d(hδ , f ) = δ. But (h − f )(1/2) = 1/2, so any ϵ < 1/2 cannot be
achieved.
Exercise 8.2.7
Describe the ϵ-neighborhoods in R2 for each of the different metrics described in Exercise
8.2.1. How about for the discrete metric?
212 CHAPTER 8. ADDITIONAL TOPICS
Solution
(a) Since this is the Euclidean distance, a ϵ-neighborhood is a circle (not including its
boundary) with radius ϵ.
Exercise 8.2.8
Let (X, d) be a metric space.
(a) Verify that a typical ϵ-neighborhood Vϵ (x) is an open set. Is the set
Cϵ (x) = {y ∈ X : d(x, y) ≤ ϵ}
a closed set?
(b) Show that a set E ⊆ X is open if and only if its complement is closed.
Solution
(a) For any point y ∈ Vϵ (x), define a = ϵ−d(x, y); by the triangle inequality Va (y) ⊆ Vϵ (x).
Consider any limit point y of Cϵ (x). For any a > 0, ∃z ∈ Cϵ where
(b) The proof is identical to that of Theorem 3.2.13, which is this statement in the special
case for R with the usual metric.
Exercise 8.2.9
(a) Show that the set Y = {f ∈ C[0, 1] : ∥f ∥∞ ≤ 1} is closed in C[0, 1].
(b) Is the set T = {f ∈ C[0, 1] : f (0) = 0} open, closed, or neither in C[0, 1]?
Solution
(a) Y c = {f ∈ C[0, 1] : ∥f ∥∞ > 1}. Let f ∈ Y c and ∥f ∥∞ = 1 + ϵ. Then if g ∈ Vϵ/2 (f ),
we have ∥g∥∞ + ∥f − g∥∞ ≥ ∥f ∥, implying ∥g∥∞ ≥ 1 + ϵ/2 and so Y c is open and Y
is closed.
(b) T is not open; the function f (x) = 0 is in T , but gϵ (x) = ϵ/2 ∈ Vϵ (f ). T is closed; let
f be a limit point of T . We have that for any ϵ > 0, ∃h so that ∥f − h∥∞ < ϵ, with
h ∈ T . Then |f (0)| < ϵ, which implies f (0) = 0 and hence f ∈ T .
Exercise 8.2.10
(a) Supply a definition for bounded subsets of a metric space X, d.
(b) Show that if K is a compact subset of the metric space X, c, then K is closed and
bounded.
8.2. METRIC SPACES AND THE BAIRE CATEGORY THEOREM 213
(c) Show that Y ⊆ C[0, 1] from Exercise 8.2.9 (a) is closed and bounded but not compact.
Solution
A subset S is bounded if there exists some x ∈ X and some finite bound M such that
S ⊆ VM (x). Equivalently, the ∃x ∈ X can be replaced with ∀x ∈ X, since given x0 and a
bound M , x1 ∈ X and M + d(x0 , x1 ) will also work.
Suppose K is not bounded, and let k0 be some element in K. Then construct the sequence
(kn ) satisfying d(k0 , kn ) > n. Clearly no subsequence of (kn ) can converge, and therefore K
is not compact. Thus, if K is compact then K must be bounded.
Let k ∈ X be a limit point of K, and construct the sequence (kn ) ∈ K satisfying
d(kn , k) < 1/n. Let (kn′ ) be a convergent subsequence of (kn ); given that d(kn′ , k) < 1/n we
must have that (kn′ ) → k, and hence k ∈ K and K is closed.
We showed Y is closed in 8.2.9 (a), and the definition of Y shows it lies within V2 (0) and is
hence bounded. Now consider the sequence fn where
(
n x ≤ 1/n
fn (x) =
1 x > 1/n
(fn ) does not converge to a function in C[0, 1]; in particular, it converges pointwise to
(
0 x=0
f (x) =
1 x>0
It’s clear intuitively that (fn ) cannot have a convergent subsequence. For formality’s sake,
let (gm ) be some subsequence of (fn ). For any m, let gm be the a’th element of (fn ) (i.e.
gm = fa , and identify gp = fb where b > 2a. Note that gm (1/(2a)) = 1/2 and gp (1/(2a)) = 1,
hence d(gm , gp ) > 1/2. Since m was arbitrary, (gm ) cannot be a Cauchy sequence and thus
cannot be convergent.
Exercise 8.2.11
(a) Show that E is closed if and only if E = E. Show that E is open if and only if E ◦ = E.
c
(b) Show that E = (E c )◦ , and similarly that (E ◦ )c = E c .
Solution
The proof is identical to that for Exercise 3.2.14, which is the same question in the context
of R but whose solution does not use any features unique to R.
Exercise 8.2.12
(a) Show
Vϵ (x) ⊆ {y ∈ X : d(x, y) ≤ ϵ},
in an arbitrary metric space (X, d).
214 CHAPTER 8. ADDITIONAL TOPICS
(b) To keep things from sounding too familiar, find an example of a specific metric space
where
Vϵ (x) ̸= {y ∈ X : d(x, y) ≤ ϵ}.
Solution
(a) Let y be a limit point of Vϵ (x). For all a > 0, we have some z ∈ Vϵ (x) where d(z, y) < a,
and thus
d(x, y) ≤ d(x, z) + d(z, y) < ϵ + a
which implies d(x, y) ≤ ϵ and hence
Vϵ (x) ⊆ {y ∈ X : d(x, y) ≤ ϵ}
(b) Consider the metric space (R, ρ) where ρ is the discrete metric, and consider V1 (0) =
{0}. V1 (0) = {0} but {y ∈ R : ρ(x, y) ≤ 1} = R.
Exercise 8.2.13
If E is a subset of a metric space (X, d), show that E is nowhere-dense in X if and only if
c
E is dense in X.
Solution
o o c
E = ∅ ⇐⇒ (E )c = X ⇐⇒ E = X
Exercise 8.2.14
(a) Give the details for why we know there exists a point x2 ∈ Vϵ1 (x1 ) ∩ O2 and an ϵ2 > 0
satisfying ϵ2 < ϵ1 /2 with Vϵ2 (x2 ) contained in O2 and
(b) Proceed along this line and use the completeness of (X, d) to produce a single point
x ∈ On for every n ∈ N.
Solution
(a) It’s clear from the definition that, for a dense set S, any point x ∈ X, and any ϵ > 0,
Vϵ (x) ∩ S ̸= ∅. Since O2 is dense, there must be some x2 ∈ Vϵ1 /2 (x1 ) ∩ O2 . Now
since O2 is open, we have that for some ϵa , Vϵa (x2 ) ⊆ O2 . We can also find ϵb so that
Vϵb (x2 ) ⊆ Vϵ1 (x1 ). To do this, recall from Exercise 8.2.12 (a) that if x ∈ Vϵb (x2 ), then
d(x, x2 ) ≤ ϵb . By choosing any ϵb < ϵ/2, we have
and hence x ∈ Vϵ1 (x1 ). Finally, set ϵ2 = min{ϵa , ϵb }, so that ϵ2 satisifies both properties.
(b) We can repeat this process to define ϵn > 0 and xn for n > 1, satisfying ϵn < ϵn−1 /2,
Vϵn (xn ) ⊆ On , and Vϵn (xn ) ⊆ Vϵn−1 .
Note that (ϵn ) → 0, and that ∞
P
i=n ϵi < 2ϵn . These two facts imply that (xn ) is a
Cauchy sequence, and by completeness (xn ) converges to some x ∈ X.
8.2. METRIC SPACES AND THE BAIRE CATEGORY THEOREM 215
Theorem 3.2.5 is still true for arbitrary metric spaces, and the proof can be reused
(replacing the absolute value function with a general distance metric). Let N ∈ N be
arbitrary. Noting that for n ≥ N + 1, xn ∈ VϵN +1 (xN +1 ), we have x ∈ VϵN +1 ⊆ VϵN .
Hence, x ∈ On for any n ∈ N.
Exercise 8.2.15
Complete the proof of the theorem.
Solution
Let (X, d) be a complete metric space. Let (Cn ) be a countable collection of nowhere-dense
c
sets. Note that since Cn is nowhere-dense, Cn = (Cnc )o is dense and open.
∞
!c ∞ ∞ ∞
[ \ \ [
Cn = Cnc ⊇ (Cnc )o ̸= ∅ =⇒ X ̸= Cn
i=1 i=1 i=1 i=1
Exercise 8.2.16
Show that if f ∈ C[0, 1] is differentiable at a point x ∈ [0, 1], then f ∈ Am,n for some pair
m, n ∈ N.
Solution
Let f be differentiable at x, and let ϵ = 1. Then there exists δ > 0 such that |x − t| < δ,
f (x) − f (t)
− |f ′ (x)| ≤ f (x) − f (t) − f ′ (x) < ϵ = 1
x−t x−t
(b) Let ϵ > 0 be arbitrary. f is continuous and (xkl ) → x, so for some K, for kl > K1 ,
|f (xkl ) − f (x)| < ϵ/2. We also have (fk ) uniformly converging to f , so for some K2 and
kl > K2 , |f (xkl ) − fkl (xkl )| < ϵ/2. Thus if kl > max{K1 , K2 } then |f (x) − fkl (xkl )| <
ϵ.
(c) For conciseness, I’ll abuse notation a bit and redefine (xk ) to be the convergent subse-
quence (xkl ) in part (a) and (b). Let y ∈ V1/m (x). Let ϵ > 0 be arbitrary. Choose a xk
“close” to x (we’ll specify exactly how close in a moment). We have
|f (y) − f (x)| ≤ |f (x) − f (xk )| + |f (xk ) − fk (xk )| + |fk (xk ) − fk (y)| + |fk (y) − f (y)|
216 CHAPTER 8. ADDITIONAL TOPICS
The first term on the right can be made small by continuity on f ; for an arbitrary ϵ1
we can choose xk ∈ Vδ1 (x) to ensure |f (x) − f (xk )| < ϵ1 . The second and fourth terms
can be made small because of uniform convergence of (fk ); we can identify δ2 so that
xk ∈ Vδ2 (x) implies |fk − f | < ϵ2 for arbitrary ϵ2 > 0. For the third term, first note
that if δ3 < 1/m − |y − x|, then y ∈ Vδ3 (xk ). Now since fk ∈ Am,n ,
(c) Let M be a bound on |p′ | (for where p′ is defined); i.e. a bound on the maximum
slope of p. Then let h zig-zag between ±1 with slope greater than 2(M + n)/ϵ. More
formally, let K > 2(M + n)/ϵ, define i(x) = xK/2 , and set
(
2 (i(x) − ⌊i(x)⌋) − 1 ⌊i(x)⌋ even
h=
1 − 2 (i(x) − ⌊(i(x)⌋)) ⌊i(x)⌋ odd
(⌊y⌋ denotes the floor function, the largest integer z with z ≤ y.) It’s clear that g is a
polygonal function with each line segment has slope greater than n, although proving
this formally requires some case work which is not a good use of our time. Thus,
g∈/ Am,n regardless of m.
Solution
See Exercise 7.5.6 (a).
Exercise 8.3.3
(a) Using the simple identity sinn (x) = sinn−1 (x) sin(x) and the previous exercise, derive
the recurrence relation
n−1
bn = bn−2
n
for all n ≥ 2.
218 CHAPTER 8. ADDITIONAL TOPICS
(b) Use this relation to generate the first three even terms and the first three odd terms
of the sequence (bn ).
(c) Write a general expression for b2n and b2n+1 .
Solution
(a) Apply integration-by-parts with h = sinn−1 (x) and k = − cos(x):
Z π
2
π π
sinn−1 (x) sin(x) = sinn−1 − sinn−1 0 (cos 0)
− cos
0 2 2
Z π
2
+ (n − 1) sinn−2 (x) cos2 (x)
0
Z π
2
sinn−2 (x) 1 − sin2 (x) = (n − 1)bn−2 − (n − 1)bn
bn = (n − 1)
0
nbn = (n − 1)bn−2
n−1
bn = bn−2
n
(b) Evens: b2 = 41 π, b4 = 3
16
π, b6 = 5
32
π. Odds: b1 = 1, b3 = 23 , b5 = 8
15
(c) For n ≥ 1,
n n
π Y 2i − 1 Y 2i
b2n = and b2n+1 =
2 i=1 2i i=1
2i + 1
Exercise 8.3.4
Show
b2n
= 1,
lim
b2n+1
n→∞
In other words, (yn ) grows at least exponentially, and therefore must not be bounded.
We now show that for arbitrary ϵ > 0, we can find N large enough so that for n ≥ N ,
Z π Z π
2 2
n+1
sin x ≥ (1 − ϵ) sinn x
0 0
1
√
where we’ll choose ϵ1 to satisfy 1+ϵ1
= 1 − ϵ. This ensures that
π π
√
Z Z
2 2
n
sin x > 1−ϵ sinn x
a 0
and Z π Z π
2 2
n+1
sin x ≥ (1 − ϵ) sinn x.
0 0
We can convert this to indicate that for large enough n, bn /bn+1 ≤ 1 + ϵ, which together
with our earlier observation that bn > bn+1 lets us conclude
bn
lim =1
n→∞ bn+1
On the other hand, plugging in the formulas for b2n and b2n+1 in Exercise 8.3.3 (c) into
b2n
lim
n→∞ b2n+1
leaves us with Wallis’s product.
Exercise 8.3.5
Derive the following alternative form of Wallis’s product formula:
√ 22n (n!)2
π = lim √
n→∞ (2n)! n
220 CHAPTER 8. ADDITIONAL TOPICS
Solution
∞
π Y (2i)2 2n (n!)2
= = lim Q n Qn
2 i=1
(2i − 1)(2i + 1) n→∞ (
i=1 (2i − 1)) ( i=1 (2i + 1))
2
22n (n!)2 (2n n!) (2n n!(2n + 1)) 22n (n!)2
1 2n
= lim = lim
n→∞ ((2n + 1)!)2 n→∞ (2n)! 2n 2n + 1
!
2n 2 2
2 (n!)2 22n (n!)2
1 2n
π = lim lim = lim √
n→∞ (2n)! n n→∞ 2n + 1 n→∞ (2n)! n
√ 22n (n!)2
π = lim √
n→∞ (2n)! n
Exercise 8.3.6
√
Show that 1/ 1 − x has Taylor expansion ∞ n
P
n=0 cn x , where c0 = 1 and
(2n)! 1 · 3 · 5 · · · (2n − 1)
cn = =
22n (n!)2 2 · 4 · 6 · · · 2n
for n ≥ 1.
Solution
√
c0 = 1/ 1 − 0 = 1, and we computed in Exercise 6.6.10 (a) that
Qn
i=1 (2i − 1) (2n − 1)! (2n)!
cn = = = 2n
2n n! (2n n!) (2n−1 (n− 1)!) 2 (n!)2
Exercise 8.3.7
Show that lim cn = 0 but ∞
P
n=0 cn diverges.
Solution
We showed in Exercise 2.7.10 (b) that lim cn = 0. We can show that ∞
P
n=0 cn diverges by
comparison against the harmonic series (hn ), which we modify for convenience to define
hn = 1/(n + 1). We have c0 ≥ h0 , and note that
2n − 1 n hn
cn /cn−1 = ≥ =
2n n+1 hn−1
for n ≥ 1, so by induction cn ≥ hn .
Exercise 8.3.8
Using the expression for EN (x) from Lagrange’s Remainder Theorem, show that equation
(4) is valid for all |x| < 1/2. What goes wrong when we try to use this method to prove (4)
for x ∈ (1/2, 1)?
Solution
See Exercise 6.6.10 (a).
8.3. EULER’S SUM 221
Exercise 8.3.9
(a) Show Z x
f (x) = f (0) + f ′ (t)dt .
0
n+1
(c) In general, with h = f (n+1) and k(t) = − (x−t) n+1
,
Z x n+1 Z x
(n+1) n (n+1) (n+1) x 1
f (t)(x−t) = f (x)·0+f (0)· + f (n+2) (t)(x−t)n+1 dt
0 n + 1 n + 1 0
Z x x
f (n+1) n+1
Z
1 1
f (n+1) (t)(x − t)n = x + f (n+2) (t)(x − t)n+1 dt
n! 0 (n + 1)! (n + 1)! 0
Armed with this, we can show inductively that, for any N ∈ N,
x N
f (i) (x)
Z
1 (N +1) N
X
f (x) = f (t)(x − t) dt + xi = EN (x) + SN (x)
N! 0 i=0
i!
Exercise 8.3.10 √
(a) Make a rough sketch of 1/ 1 − x and S2 (x) over the interval −1, 1, and compute E2 (x)
for x = 1/2, 3/4, and 8/9.
(b) For a general x satisfying −1 < x < 1, show
2
15 x x − t
Z
1
E2 (x) = dt
16 0 1−t (1 − t)3/2
is valid, and use this to find an overestimate for |E2 (x)| that no longer involves an
integral. Note that this estimate will necessarily depend on x. Confirm that things are
going well by checking that this overestimate is infact larger than |E2 (x)| at the three
computed values from part (a).
222 CHAPTER 8. ADDITIONAL TOPICS
Solution
(a)
1 3
S2 (x) = 1 + x + x2
2 8
A plot is best made using your favourite graphing calculator.
x x x 2
15 (x − t)2
x−t
Z Z Z
1 (3) 2 1 15 1
E2 (x) = f (t)(x − t) = =
2 0 2 0 8 (1 − t)7/2 16 0 1−t (1 − t)3/2
(c) We have that |t| ≤ |x| < 1 and t having the same sign as x, so
x − t
1 − t ≤ |x| ⇐⇒ |x − t| ≤ |x(1 − t)|
For x ≥ t > 0,
|x − t| ≤ |x(1 − t)| ⇐⇒ x − t ≤ x − xt ⇐⇒ t ≥ xt
|x − t| ≤ |x(1 − t)| ⇐⇒ t − x ≤ xt − x ⇐⇒ t ≤ xt
15 x 2 15x2 x 2
Z Z
1 −3/2 15x 1 − √ 1
|E2 (x)| ≤ |x| dt= (1 − t) dt =
16 0 (1 − t)3/2 16 0 8 1 − x
so Z !
N +1
1 x Y 2i − 1 1
|EN (x)| = dt
N! 0 2(1 − x) (1 − t)1/2
i=1
Z N
!
1 x Y (2i − 1)(x − t) 2N + 1
= dt
N! 0 2(i − t) 2(1 − t)3/2
Zi=1 !
N N
2N + 1 x Y 2i − 1
x−t 3/2
= (1 − t) dt
2 2i 1−t
0
i=1
Z x
2N + 1
≤ |x|N −3/2
(1 − t) dt
2
0
1
= |x|N (2N + 1) 1 − √
1 − x
For a fixed x, with |x| < 1, as N increases |x|N will go to 0 exponentially, faster than
2N + 1 increases, and therefore limn→∞ EN (x) = 0 for x ∈ (−1, 1).
Exercise 8.3.11 √
Assuming that the derivative of arcsin(x) is indeed 1/ 1 − x2 , supply the justification that
allows us to conclude
∞
X cn
arcsin(x) = x2n+1 for all |x| < 1
n=0
2n + 1
Solution
Term-by-term antidifferentiation is a valid operation with the expected results, as proven in
Exercise 6.5.4.
Exercise 8.3.12
Our work thus far shows that the Taylor series in (5) is valid for all |x| < 1, but note that
arcsin(x) is continuous for all |x| ≤ 1. Carefully explain why the series in (5) converges
uniformly to arcsin(x) on the closed interval [−1, 1].
Solution
If we can show that the power series converges absolutely for x = 1, then by Theorem 6.5.2
the series uniformly converges over [−1, 1], and is therefore continuous. Taking limits on
P c±1
both sides approaching would lead us to conclude that the equality is valid for |x| ≤ 1.
We now show that n
2n+1
converges absolutely.
By the Cauchy Condensation Test (Theorem 2.4.6), this series converges if
∞ ∞ ∞
X
n c2n X 2n 1X
2 = c2 n ≤ c2n
n=0
2 · 2n + 1 n=0 2 · 2n + 1 2 n=0
(a) Without looking ahead, decide if there is a natural way to define 0#. How about
(−2)#? Conjecture a reasonable value for 72 #.
(b) Now prove n# = 12 n(n + 1) for all n ∈ N, and revisit part (a).
Solution
(a) Noting that N # = (N − 1)# + N and 1# = 1, we could have 0# = 0, −1# = 0,
and −2# = 1. 7/2# could just be defined to be the result from linearly interpolating
between 3# = 6 and 4# = 10 to get 8.
Exercise 8.4.2
Verify that the series converges absolutely for all x ∈ R, that E(x) is differentiable on R,
and E ′ (x) = E(x).
Solution
Note that
∞ n
X x
= E(|x|)
n!
n=0
for some finite constant K. The infinite series left over is a geometric series which converges.
Term-by-term differentiation is safe to apply on power series which converge (Theorem
6.5.6), and it’s clear that E ′ (x) = E(x) when applying termwise differentiation.
Exercise 8.4.3
(a) Use the results of Exercise 2.8.7 and the binomial formula to show that E(x + y) =
E(x)E(y) for all x, y ∈ R.
(b) Show that E(0) = 1 , E(−x) = 1/E(x), and E(x) > 0 for all x ∈ R.
226 CHAPTER 8. ADDITIONAL TOPICS
Solution
(a)
∞ ∞ n ∞ X n i
(x + y)n y n−i
X X 1 X n! i n−i
X x
E(x + y) = = xy =
n=0
n! n=0
n! i=0 i!(n − i)! n=0 i=0
i! (n − i)!
∞ X ∞ ∞
! ∞
!
xi yj xi yi
X X X
= = = E(x)E(y)
i=0 j=0
i! j! i=0
i! i=0
i!
(b) For E(0), all terms for n ≥ 1 become 0, so E(0) = 1. We showed somewhat informally
in Exercise 6.6.5(c) that E(−x)E(x) = 1 by collecting common P terms. However,
Exercise 2.8.7 lets us conclude that E(−x)E(x) does in fact equal ∞ n=0 dn where dn
is defined as in Exercise 6.6.5(c). Finally, it’s clear that E(x) > 0 for x ≥ 0 simply
because all terms are positive; then E(−x) = 1/E(x) > 0 and so E(x) > 0 for x < 0
as well.
Exercise 8.4.4 √ m
Define e = E(1). Show E(n) = en and E(m/n) = ( n e) for all m, n ∈ Z.
Solution
We have for n ≥ 1 that
n
! n
X Y
E(n) = E 1 = E(1) = en
i=1 i=1
Finally for n > 0, e−n = 1/en = 1/E(n) = E(−n), so en = E(n) for all n ∈ Z.
√ √
By definition n e is the unique positive number which satisfies ( n e)n = e. E(1/n)
satisfies this equality, since
n
! n
X 1 Y
E(1) = E = E(1/n)
i=1
n i=1
√
so n
e = E(1/n). Finally
m
! m
X 1 Y √
n
m
E(m/n) = E = E(1/n) = e
i=1
n i=1
Exercise 8.4.5
Show limx→∞ xn e−x = 0 for all n = 0, 1, 2, . . ..
Solution
Note that for a fixed n, ∀K ∈ R,we can find N > 0 so that whenever x ≥ N , ex x−n > K.
We do this by noting
E(x) xn+1 x
n
> n
=
x (n + 1)!x (n + 1)!
8.4. INVENTING THE FACTORIAL FUNCTION 227
(c) Fix y > 0 and differentiate log(xy) with respect to x. Conclude that
(d) For t > 0 and n ∈ N, tn has the usual interpretation as t · t · · · t (n times). Show that
(b) For convenience of notation let f (x) = ex and g(x) = f −1 (x) = log x. We have, from
Exercise 5.2.12,
1 1 1 1
(log x)′ = g ′ (x) = = = log x =
f ′ (g(x) f (g(x) e x
(c)
y 1
(log(xy))′ = = = (log x)′
xy x
and therefore log(xy) = log x + C for some constant C. Now
Exercise 8.4.7 √
(a) Show tm/n = ( n t)m for all m, n ∈ N.
(b) Noting that tx > 0, take log of both sides of the definition of tx .
(c)
′
(tx )′ = ex log t = ex log t log t = tx log t
Exercise 8.4.8
Inspired by the fact that 0! = 1 and 1! = 1, let h(x) satisfy
(i) h(x) = 1 for all 0 ≤ x ≤ 1, and
(a) Find a formula for h(x) on [1, 2], [2, 3], and [n, n + 1] for arbitrary n ∈ N.
(b) Now do the same for [−1, 0], [−2, −1], and [−n, −n + 1].
(c) This is an exercise best left for your favourite graphing utility.
Exercise 8.4.9 R∞
(a) Show that the improper integral a f converges if and only if, for all ϵ > 0 there exists
M > a such that whenever d > c ≥ M it follows that
Z d
f < ϵ.
c
R a+n
(In one direction it will be useful to consider the sequence an = a f .)
R∞ R∞
(b) Show that if 0 ≤ f ≤ g and a g converges then a f converges.
(c) Part (a) is a Cauchy criterion, and part (b) is a comparison test. State and prove an
absolute convergence test for improper integrals.
8.4. INVENTING THE FACTORIAL FUNCTION 229
Solution Rb
(a) In the forward direction, assume that limb→∞ a f exists. Note that for any c,
Z ∞ Z c Z b Z c Z a Z ∞
f = lim f+ f− f= f+ f
c b→∞ a c a c a
Now if d > c ≥ M ,
∞ ∞ ∞ ∞
Z d Z Z Z Z
ϵ
f = f− f ≤ f + f < 2 = ϵ
c
c d c d 2
R
d
In the reverse direction, define Mn large enough so that d > c ≥ Mn implies c f <
RM
1/n and satisfying Mn > a + n. Define the sequence xn = a n f , and note that (xn )
is a Cauchy sequence and converges to some limit x. Thus for any ϵ > 0, we can find
N1 so that for n > N1 , Z Mn
ϵ
x − f <
a 2
and set N > max{2/ϵ, N1 } so that for b > c = MN ,
Z b Z b Z MN
ϵ
f =
f − f <
2
MN a a
Exercise 8.4.10
(a) Use the properties of et previously discussed to show
Z ∞
e−t dt = 1.
0
(b) Show Z ∞
1
= e−αt dt, for all α > 0.
α 0
Solution
(a) The antiderivative of e−t is −e−t , so
Z b
lim e−t dt = lim = lim −e−b + e0 = 1
b→∞ 0 b→∞ b→∞
Exercise 8.4.12
Assume the function f (x, t) is continuous on the rectangle D = {(x, t) : a ≤ x ≤ b, c ≤ t ≤
d}. Explain why the function
Z d
F (x) = f (x, t)dt
c
is properly defined for all x ∈ [a, b].
Solution
For a fixed x, f (x, t) is continuous with respect to t. To see this, we need to show ∀t0 ∈
[c, d], ϵ > 0, we can find δ so that |t − t0 | < δ implies |f (x, t) − f (x, t0 )| < ϵ. But since
|t − t0 | = ∥(x, t) − (x, t0 )∥
we can just use the continuity of f over D to conclude f (x, t) is continuous with respect to
Rd
t, and therefore c f (x, t)dt is properly defined.
8.4. INVENTING THE FACTORIAL FUNCTION 231
Exercise 8.4.13
Prove Theorem 8.4.5.
Solution
Since f (x, t) is uniformly continuous, for any ϵ > 0, we can find δ so that |f (x, t) − f (x0 , t)| <
ϵ
d−c
whenever ∥(x, t) − (x0 , t)∥ = |x − x0 | < δ. Then when |x − x0 | < δ,
Z d Z d
ϵ
|F (x) − F (x0 )| ≤ |f (x, t) − f (x0 , t)| dt < dt = ϵ
c c d−c
showing F (x) is continuous.
Exercise 8.4.14
Finish the proof of Theorem 8.4.6.
Solution
Note that
d
F (z) − F (x) f (z, t) − f (x, t)
Z
= dt
z−x c z−x
so choosing δ so that 0 < |z − x| < δ implies
fx (x, t) − f (z, t) − f (x, t) < ϵ
z−x d−c
we have
Z d Z d Z d
F (z) − F (x) f (z, t) − f (x, t) ϵ
− fx (x, t)dt =
− fx (x, t) <
d − c = ϵ
z−x c c z − x c
as desired.
Exercise 8.4.15 R∞
(a) Show that the improper integral 0 e−xt dt converges uniformly to 1/x on the set
[1/2, ∞).
(b) Is the convergence uniform on 0, ∞?
Solution
(a) From our earlier work, we have that
Z ∞ d
e−xd
Z
−xt
e dt − e−xt = ≤ 2e−d/2
0 0 x
−d/2
Notably
R ∞ −xt this is not dependent on x, and since limd→∞ 2e = 0, we have that
0
e dt converges uniformly.
(b) For any fixed d,
e−xd
=∞lim+
x→0 x
implying that the convergence cannot be uniform on (0, ∞).
232 CHAPTER 8. ADDITIONAL TOPICS
Exercise 8.4.16
Prove the following analogue of the Weierstrauss
R∞ M-Test for improper
R ∞integrals: If f (x, t)
satisfies |f (x, t)| ≤ g(t) for all x ∈ A and a g(t)dt converges, then a f (x, t)dt converges
uniformly on A.
Solution
We want, ∀ϵ > 0, to find M > a so
Z ∞ Z b
f (x, t)dt − f (x, t)dt<ϵ
a a
Solution R c+n
Let Fn (x) = c f (x, t)dt and note that Fn (x) is continuous. So F (x) = limn→∞ Fn (x)
must also be continuous, and since [a, b] is compact F (x) is also uniformly continuous over
[a, b].
Exercise 8.4.18
Prove Theorem 8.4.9.
Solution R
c+n Rd
Let gn (x) = c f (x, t)dt. From Theorem 8.4.6, gRn′ (x) = c fx (x, t)dt. We have (gn (x)) →
∞
F (x) pointwise, and (gn′ ) converges uniformly to c fx (x, t)dt. Therefore by the Differen-
tiable Limit Theorem (Theorem 6.3.1),
Z ∞
′
F (x) = fx (x, t)dt
c
Exercise 8.4.19
(a) Although we verified it directly, show how to use the theorems in this section to give
a second justification for the formula
Z ∞
1
= te−αt dt, for all α > 0.
α2 0
Solution R∞
(a) We would like to use Theorem 8.4.9, with f (α, t) = e−αt and F (α) = 0 f (α, t)dt. We
already have that F (α) = 1/α exists for α
R∞> 0. So, we need to show that f (α, t) and
fα (α, t) = −teαt are continuous and that 0 fα (α, t)dt converges uniformly.
Let’s first show that e−αt is continuous. For any fixed α0 and t0 , we want ∀ϵ > 0, that
whenever ∥(α, t) − (α0 , t0 )∥ < δ > 0, that
−αt
− e−α0 t0 < ϵ
e
δ1 δ1
Now if |t0 − t| < 2|α0 |
= δt , |t0 − t| < 1, and |α0 − α| < 2(|t0 |+1)
= δα , then
We can ensure both of these conditions by setting δ = min{δt , δα , 1}, thus showing
f (α, t) is continuous.
It’s fairly easy to show that if g(x) is continuous, so is g(x, y) = g(x). Additionally, a
careful re-reading of the proofs for Theorem 4.2.3 (Sequential Criterion for Functional
Limits) and Exercise 4.2.1 (Algebraic Limit Theorem for Functional Limits) show that
they can be reused for complete metric spaces such as R2 with the usual metric. Thus
Theorem 4.3.4 (Algebraic Continuity Theorem) can also be generalized to R2 , and we
can conclude for any x ∈ R, the function g(α, t) = tx e−αt is continuous. Thus, both f
and fα are continuous.
R∞
Now we need to show 0 fα (α, t)dt converges uniformly when Rα ≥ p > 0. To make
∞
part (b) easier we’ll prove the more general statement that 0 tx e−αt dt R ∞converges
uniformly, for any x ∈ R with x ≥ 0. As a prerequisite, note that if a g(x, y)dy
R∞ Rb
converges uniformly, then so does b g(x, y)dy, provided a g(x, y)dy is well defined.
We can adapt the solution to Exercise 8.4.5 to show that for any x ∈ R, α > 0,
lim
R ∞t→∞ tx e−αt = 0. Thus, for any fixed x ≥ 0, we can find some M so that g(α) =
M
tx e−αt dt converges uniformly, by Exercise 8.4.16 (variant ofR the Weierstrauss M-
∞
Test) and comparison against e−αt/2 . This in turn shows that 0 tx e−αt dt converges
uniformly.
Finally, before applying Theorem 8.4.9 to get our desired result, we need to define the
domain D. For a given α > 0, we want D to contain all points of the form α, t, so
define D = (x, t) : α/2 ≤ x ≤ α + 1, 0 ≤ t.
(b) This is just a repeated application of Theorem 8.4.9 on the formula proved in part (a);
the solution to part (a) shows that all of the prerequisites to Theorem 8.4.9 are always
met in this process.
Exercise 8.4.20
(a) Show that x! is an infinitely differentiable function on (0, ∞) and produce a formula
for the nth derivative. In particular show that (x!)′′ > 0.
234 CHAPTER 8. ADDITIONAL TOPICS
(b) Use the integration-by-parts formula employed earlier to show that x! satisfies the
functional equation
(x + 1)! = (x + 1)x!
Solution
(a) Before continuing, note that tx currently isn’t defined at t = 0; this can easily be solved
by defining 0x = 0.
We show that x! is infinitely differentiable at x0 > 0 by repeatedly applying Theorem
8.4.9, over the domain D = {(x, t) : x0 /2 ≤ x ≤ x0 + 1, 0 ≤ t}. Denote the n’th
derivative of tx e−t with respect to x as fn (x, t) = tx e−t logn t (except at t = 0, where
f (x, t) = 0). The work in Exercise 8.4.19 (a) means that to show fn is continuous over
D, we only need to show that tx is continuous. Note that for any fixed (x1 , t1 ) ∈ D,
Since g(t) = tx1 is continuous, we can require |t1 − t| < δt to ensure the first term is
less than ϵ/2. Since tx is bounded for finite t and x, let M1 be an upper bound for |tx |
over V1 (x1 , t1 ) ∩ D. Let M2 be an upper bound for | log t| over Vt1 /2 (t1 ).
We can find δ1 so that when |(x1 − x) log t| < δ1 , e(x1 −x) log t − 1 < ϵ . Then requrir-
2M1
δ1
ing |x1 − x| < M2
= δx guarentees that
ϵ
|tx | tx1 −x − 1 ≤ M1 e(x1 −x) log t − 1 <
2
Putting it all together,
t1
∥(x, t) − (x0 , t0 )∥ < min δt , δx , 1, =⇒ |tx − tx0 0 | < ϵ
2
proving tx is continuous.
R∞
We also need to show that R0 fn (x, t)dt converges uniformly. Now, since x > log x ≥
∞
0 for x ≥ 1, we can show 1 fn (x, t)dt converges uniformly by comparison against
R ∞ x+n −t R1
1
t e dt. We also have f n (x, t) continuous in t over [0, 1] so f (x, t)dt is well
0 n
defined.
Thus, repeatedly applying Theorem 8.4.9 lets us conclude that
Z ∞
(n)
(x!) = tx e−t (log t)n dt
0
Note that for (x!)′′ , the term under the integral is always positive, so (x!)′′ > 0.
(b) Z ∞ Z ∞
x −t x 0 x −t
x! = t e dt = 0 e − lim t e + xtx−1 et dt = x(x − 1)!
0 t→∞ 0
Exercise 8.4.21
(a) Use the convexity of log(f (x)) and the three intervals [n − 1, n], [n, n + x], and [n, n + 1]
to show
x log(n) ≤ log(f (n + x)) − log(n!) ≤ x log(n + 1)
nx n!
1
0 ≤ log(f (x)) − log ≤ x log(1 + )
(x + 1)(x + 2) · · · (x + n) n
nx n!
f (x) = lim , for all x ∈ (0, 1].
n→∞ (x + 1)(x + 2) · · · (x + n)
Solution
(a) Since the slopes are increasing between these three intervals,
n
! !
Y
0 ≤ log(f (x)) − log(n!) − log (x + i) + log (nx ) ≤ x log(n + 1) − x log(n)
i=1
x
n n! n+1 1
0 ≤ log(f (x)) − log Qn ≤ x log = x log n +
i=1 (x + i) n n
236 CHAPTER 8. ADDITIONAL TOPICS
1
limn→∞ log 1 + n
= 0 so by the Squeeze Theorem,
nx n!
log(f (x)) − lim log Qn =0
i=1 (x + i)
n→∞
We only need to show this equation holds for x = 0. We have f (0) = 1 by property (i), and
n0 n! n!
lim Qn = lim = 1 = f (0)
i=1 i n!
n→∞ n→∞
Exercise 8.4.22
x
(a) Where does g(x) = x!(−x)!
equal zero? What other familiar function has the same set
of roots?
2
(b) The function e−x provides the raw material
R ∞for−xthe all-important
√ Gaussian bell curve
2
from probability, where it is known that −∞ e dx = π. Use this fact (and some
standard integration techniques) to evaluate (1/2)!.
(c) Now use (a) and (b) to conjecture a striking relationship between the factorial function
and a well-known function from trigonometry.
Solution
x sin(πx)
=
x!(−x)! π
Exercise 8.4.23
As a parting shot, use the value for (1/2)! and the Gauss product formula in equation (9) to
derive the famous product formula for π discovered by John Wallis in the 1650s:
π 2·2 4·4 6·6 2n · 2n
= lim ···
2 n→∞ 1 · 3 3·5 5·7 (2n − 1)(2n + 1)
8.5. FOURIER SERIES 237
Solution
√ n n
π √ Y i √ Y 2i
= lim n = lim n
2 n→∞
i=1
i + 1/2 n→∞ i=1
2i + 1
n
! n ! n
! n !
π Y 2i Y 2i Y 2i Y 2i
= lim 2n = lim
2 n→∞ i=1
2i + 1 i=1
2i + 1 n→∞
i=1
2i − 1 i=1
2i + 1
n
Y (2i)2
= lim
n→∞
i=1
(2i − 1)(2i + 1)
(b) Explain why any finite sum of functions of the form given in part (a) would also satisfy
(1), (2), and (3).
Solution
(a)
∂ 2u 2 ∂ 2u
= −b n n sin(nx) cos(nt) =
∂x2 ∂t2
u(0, t) = bn sin(0) cos(nt) = 0
u(π, t) = bn sin(nπ) cos(nt) = 0
(Note that the above equation is no longer true if n ∈
/ N.)
∂u
(x, 0) = −bn sin(nx) sin(0) = 0
∂t
(b) The differential equation itself and the boundary conditions are linear; that is, if u1 and
u2 both satisfy equations (1) through (3), then so does c1 u1 + c2 u2 for any c1 , c2 ∈ R.
Exercise 8.5.2
Using trigonometric identities when necessary, verify the following integrals.
(a) For all n ∈ N, Z π Z π
cos(nx)dx = 0 and sin(nx)dx = 0
−π −π
For m ̸= n,
Z π Z π
cos(mx) cos(nx)dx = 0 and sin(mx) sin(nx)dx = 0
−π −π
Solution
(a) Setting u = nx,
Z π
1 nπ
Z
1
cos(nx)dx = cos(u)du = (sin(nπ) − sin(−nπ)) = 0
−π n −nπ n
Z π
1 nπ −1
Z
sin(nx)dx = sin(u)du = (cos(nπ) − cos(−nπ)) = 0
−π n −nπ n
(c) If m = n: Z π Z π
cos(nx) sin(nx)dx = sin(2nx)dx = 0
−π −π
If m ̸= n:
Z π
sin(nπ) sin(mπ) − sin(−nπ) sin(−mπ)
cos(nx) sin(mx)dx =
−π m
Z π
n
− cos(nx) sin(mx)dx
m −π
n π
Z
=− cos(nx) sin(mx)dx
m −π
n − cos(nπ) cos(mπ) + cos(−nπ) cos(−mπ)
=−
m n
π
n2
Z
+ 2 sin(nx) cos(mx)dx
m −π
n2 π
Z
= 2 sin(nx) cos(mx)dx
m −π
and
π π
n2
Z Z
sin(mx) sin(nx)dx = 2 sin(mx) sin(nx) = 0
−π m −π
Exercise 8.5.3
Derive the formulas
Z π Z π
1 1
am = f (x) cos(mx)dx and bm = f (x) sin(mx)dx
π −π π −π
for all m ≥ 1.
Solution
" ∞
#
Z π Z π X
f (x) cos(mx)dx = a0 cos(mx) + an cos(nx) cos(mx) + bn sin(nx) cos(mx) dx
−π −π n=1
∞ Z
X π
= [an cos(nx) cos(mx) + bn sin(nx) cos(mx)] dx
n=1 −π
Z π
= am cos2 (mx) = am π
−π
" ∞
#
Z π Z π X
f (x) sin(mx)dx = a0 sin(mx) + an cos(nx) sin(mx) + bn sin(nx) sin(mx) dx
−π −π n=1
∞ Z
X π
= [an cos(nx) sin(mx) + bn sin(nx) sin(mx)] dx
n=1 −π
Z π
= bm sin2 (mx) = bm π
−π
Exercise 8.5.4
(a) Referring to the previous example, explain why we can be sure that the convergence
of the partial sums to f (x) is not uniform on any interval containing 0.
(b) Repeat the computations of Example 8.5.1 for the function g(x) = |x| and examine
graphs for some partial sums. This time, make use of the fact that g is even (g(x) =
g(−x)) to simplify the calculations. By just looking at the coefficients, how do we
know this series converges uniformly to something?
240 CHAPTER 8. ADDITIONAL TOPICS
(c) Use graphs to collect some empirical evidence regarding the question of term-by-term
differentiation in our two examples to this point. Is it possible to conclude convergence
or divergence of either differentiated series by looking at the resulting coefficients? The-
orem 6.4.3 is about the legitimacy of term-by-term differentiation. Can it be applied
to either of these examples?
Solution
(a) f (x) is not continuous at 0, and each of the partial sums is continuous. Uniform
convergence would imply that the function which the partial sums converge to must
be continuous.
(b) Z π Z π
1 1 π
a0 = |x| dx = 2 xdx =
2π −π 2π 0 2
For n ≥ 1,
1 π 2 π
Z Z
an = |x| cos(nx)dx = x cos(nx)dx
π −π π 0
2
π 1 Z π
2
= x sin(x) − sin(nx)dx = 2 (cos(nπ) − 1)
π 0 n 0 nπ
(
4
− n2 π n odd
=
0 n even
1 π
Z Z π Z 0
1
bn = |x| sin(nx)dx = x sin(nx)dx − −x sin nx
π −π π 0 −π
Z π Z 0
1
= x sin(nx)dx + x sin(nx)dx = 0
π 0 π
We get
∞
−4 X 1
g(x) = cos((2n + 1)x)
π n=0 (2n + 1)2
Noting that the series of non-zero coefficients converges absolutely, we can use the
Weierstrauss M-Test with ∞
X
|an |
n=0
(c) Taking the termwise derivative of the series representation of g(x) leaves us with the
series representation of f (x), which is promising since g ′ (x) = f (x) where g ′ (x) is
defined. But convergence is not immediately clear looking at the coefficients.
Taking the termwise derivative of the series representation of f (x) leaves us with
∞
4X
cos((2n + 1)x)
π n=0
8.5. FOURIER SERIES 241
which looks like it should diverge, although proving this is difficult. As a specific
example, though, at x = π/3 the partial sums will cycle between three different values
and fail to converge.
Exercise 8.5.5
Explain why h is uniformly continuous on R.
Solution
h is continuous on the compact set [−π, π] and therefore uniformly continuous over this set,
and thus for any ϵ > 0 we can find δ so that |x − x0 | < δ implies |h(x) − h(x0 )| < ϵ/2,
at least if x and x0 are both in [−π, π] or both in the same “copy” of h. If they are not,
however, then there must be some k = nπ, with n odd, separating them, with |x − k| < δ
and |k − x0 | < δ. Then
Exercise 8.5.6
R
b
Show that a h(x) sin(nx)dx < ϵ/n, and use this fact to complete the proof.
Solution
Slight change to the premise - we’ll require that |h(x) − h(y)| < ϵ/(2π) when |x − y| < δ.
Define ∆h(x) satisfying h(x) = h( a+b2
) + ∆h(x) and note that by uniform continuity,
Rb
|∆h(x)| < ϵ/2π. Note also that a sin(nx)dx = 0.
Z b Z b Z b Z b
a+b
h(x) sin(nx)dx = h
sin(nx)dx + ∆h(x)dx ≤ |∆h(x)| dx
a 2 a a a
ϵ 2π
< = ϵ/n
2π n
Now, subdivide [−π, π] into n subintervals, each of size 2π/n, and apply the above result to
each interval; this shows that Z π
h(x) sin(nx) < ϵ
−π
Exercise 8.5.7
(a) First, argue why the integral involving qx (u) tends to zero as N → ∞.
(b) The first integral is a little more subtle because the function px (u) has the sin(u/2)
term in the denominator. Use the fact that f is differentiable at x (and a familiar limit
from calculus) to prove that the first integral goes to zero as well.
242 CHAPTER 8. ADDITIONAL TOPICS
Solution
(a) This is a direct result of the Riemann-Lebesgue Lemma (Theorem 8.5.2).
(b) We would like to show that px (u) is continuous, and the only place this is not au-
tomatically true is when sin(u/2) = 0; this only occurs when u = 0. Strictly, px (u)
isn’t even defined here, but if limu→0 px (u) is well defined, then we can simply define
px (0) = limu→0 px (u) and be on our merry way. We have
f (x + u) − f (x) u u
′ 2 u
lim px (u) = lim cos = f (x) cos = 2f ′ (x)
sin u2 cos u2
u→0 u→0 u 2 2
where we have used L’Hospital’s rule in the second term. We can now conclude that
px (u) is effectively continuous, for the purposes of applying the Riemann-Lebesgue
Lemma, and thus the integral goes to zero.
Exercise 8.5.8
Prove that if a sequence of real numbers (xn ) converges, then the arithmetic means
x1 + x2 + x3 + · · · + xn
yn =
n
also converges to the same limit. Give an example to show that it is possible for the sequence
of means (yn ) to converge even if the original sequence (xn ) does not.
Solution
See Exercise 2.3.11.
Exercise 8.5.9
Use the previous identity to show that
" #2
1/2 + D1 (θ) + D2 (θ) + · · · + DN (θ) 1 sin (N + 1) 2θ
=
sin 2θ
N +1 2(N + 1)
Solution
We need one more trigonometric identity, for any a, b:
sin(a + b) sin(a − b) = (sin x cos b + sin b cos a) (sin a cos b − sin b cos a)
= sin2 a cos2 b − sinb cosa
= sin2 a 1 − sin2 b − sin2 b 1 − sin2 a
Assume that θ ̸= 2nπ for integer n; otherwise the right side isn’t defined. What follows is a
8.5. FOURIER SERIES 243
lot of algebra:
N N
1 X 1 1 X θ
+ Di (θ) = + sin iθ +
2 i=1 2 2 sin 2θ i=1 2
N
1 1 X θ θ
= + sin iθ cos + sin cos iθ
2 2 sin 2θ i=1 2 2
" N
! N
!!#
1 θX θ 1 X 1
= θ
cos sin iθ + sin + cos iθ +
2 sin 2 2 i=1 2 2 i=1 4
cos 2θ sin N2θ sin (N + 1) 2θ sin 2θ sin (2N2+1)θ
1
= θ θ
+ θ θ
+
2 sin 2 sin 2 2 sin 2 2 sin 2 4
θ Nθ θ
1 cos 2 sin 2 sin (N + 1) 2
= +
4 2 sin 2θ sin 2θ
sin 2θ sin N2θ cos (N +1)θ
2
+ cos N2θ sin (N +1)θ
2
+
2 sin 2θ 2 sin 2θ
1 1 Nθ (N + 1)θ θ θ (N + 1)θ
= + sin sin cos + sin cos
4 4 sin2 2θ 2 2 2 2 2
(N + 1)θ θ Nθ θ Nθ
+ sin cos sin + sin cos
2 2 2 2 2
1 1 Nθ (N + 2)θ 2 (N + 1)θ
= + sin sin + sin
4 4 sin2 2θ 2 2 2
1 Nθ (N + 2)θ 2 θ 2 (N + 1)θ
= sin sin + sin + sin
4 sin2 2θ 2 2 2 2
1 (N + 1)θ
= 2 θ
2 sin2
4 sin 2 2
sin2 (N +1)θ
2
=
2 sin2 2θ
!2
1 sin (N +1)θ
2
=
2 sin 2θ
which is the desired identity, except missing a factor of 1/(N + 1) on both sides.
Exercise 8.5.10
(a) Show that Z π
1
σN (x) = f (u + x)FN (u)du
π −π
(b) Graph the function FN (u) for several values of N . Where is FN large, and where is it
close to zero? Compare this function to the Dirichlet kernel DN (u). Now, prove that
FN → 0 uniformly on any set of the form {u : |u| ≥ δ}, where δ > 0 is fixed (and u is
restricted to the interval (−π, π].)
244 CHAPTER 8. ADDITIONAL TOPICS
Rπ
(c) Prove that −π
FN (u)du = π.
(d) To finish the proof of Fejér’s Theorem, first choose a δ > 0 so that
|u| < δ implies |f (x + u) − f (x)| < ϵ.
Set up a single integral that represents the difference σN (x) − f (x) and divide this
integral into sets where |u| ≤ δ and |u| ≥ δ. Explain why it is possible to make each
of these integrals sufficiently small, independently of the choice of x.
Solution
(a) Note that D0 (u) = 1/2. We have
N
1 X
σN (x) = Sn (x)
N + 1 n=0
N
1 X1 π
Z
= f (u + x)DN (u)du
N + 1 n=0 π −π
N
!
1 π
Z
1 1 X
= f (u + x) + DN (u) du
π −π N +1 2 n=1
1 π
Z
= f (u + x)FN (u)du
π −π
(b) FN (u) is large when u is close to zero, and close to zero everywhere else. In contrast,
DN (u) continues to oscillate with large amplitude away from u = 0. To show FN → 0
uniformly when |u| ≥ δ, note that under this condition,
!
u
1 sin (N + 1) 2 1
|FN (u)| = ≤
2(N + 1) sin u2 2(N + 1) sin 2δ
(d)
1 π 1 π
Z Z
σN (x) − f (x) = f (u + x)FN (u)du − f (x)FN (u)du
π −π π −π
1 π
Z
= (f (u + x) − f (x)) FN (u)du
π −π
For the set where |u| ≤ δ,
Z π Z π
1 ϵ
|σN (x) − f (x)| ≤ |f (u + x) − f (x)| FN (u)du ≤ FN (u) = ϵ
π −π π −π
8.6. A CONSTRUCTION OF R FROM Q 245
and since ϵ is chosen freely, this part of the integral can be made arbitrarily small.
For the set where |u| ≥ δ, let M be a bound on f (x). Then
1 π 2M π
Z Z
|σN (x) − f (x)| ≤ |f (u + x) − f (x)| FN (u)du ≤ FN (u)du
π −π π −π
Exercise 8.5.11
(a) Use the fact that the Taylor series for sin(x) and cos(x) converge uniformly on any
compact set to prove WAT under the added assumption that [a, b] is [0, π].
(b) Show how the case for an arbitrary interval [a, b] follows from this one.
Solution
(a) For any function f , Fejér’s Theorem implies we can construct a function of the form
N1
X N
X
g(x) = k0 + ki sin(ci x) + ki cos(ci x)
i=1 i=N1 +1
satisfying |g(x) − f (x)| < ϵ/2 over x ∈ [0, π], and uniform convergence of the Taylor
series of sin x and cos x imply we can find a set of polynomials Pi satisfying
ϵ
|Pi (x) − ki sin(ci x)| <
2N
for 1 ≤ i ≤ N1 and
ϵ
|Pi (x) − ki cos(ci x)| <
2N
for N1 + 1 ≤ i ≤ N Then the polynomial P (x) = k0 + N
P
i=1 Pi (x) satisfies
ϵ
|P (x) − g(x)| < and |P (x) − f (x)| < ϵ
2
π
(b) For x ∈ [a, b], apply the variable substitution y = b−a
(x − a) and use part (a) on y.
(b) S = {t ∈ Q : t ≤ 2}
(d) U = {t ∈ Q : t2 ≤ 2 or t < 0}
Solution
(a) Cr contains r − 1 and does not contain r, so (c1) is satisfied. If p ∈ Cr and q < p, then
q < p < r and thus q ∈ Cr , so (c2) is satisfied. Also, p < p+r
2
< r so p+r
2
∈ Cr and (c3)
is satisfied.
(b) Not a cut, S has the maximum p = 2. There are no elements in S that can be greater
than 2.
(c) Is a cut. 0 ∈ T and 2 ∈ / T so (c1) is satisfied. Let r ∈ T and q < r. If q < 0 then q ∈ T
trivially. Otherwise, r > q ≥ 0 implies 2 > r2 > q 2 and therefore q ∈ T , showing (c2)
is satisfied. Finally, to show (c3), let r ∈ T with r ≥ 1. (If r < 1 then we can trivially
identify r < 1 ∈ T to confirm (c3).) Let a = 2 − r2 , and note 1 ≥ a > 0. Consider the
rational number a 2 2 a a2
q= r+ =r + + >r
4r 2 4r
2
It is easy to show a4r < a2 , implying q < r2 + a = 2 and thus q ∈ 2, and thus r is not a
maximum and (c3) is true.
(d) Is a cut. The only difference from part (c) is that we cannot immediately claim by
definition that a > 0; instead the definition of U only implies a ≥ 0. However, Section
1.1. provides a proof that a ̸= 0; therefore we can maintain a > 0 and reuse the rest
of the logic.
Exercise 8.6.2
Let A be a cut. Show that if r ∈ A and s ∈
/ A, then r < s.
Solution
If r ≥ s then by (c2) s would be in A, a contradiction.
Exercise 8.6.3
Using the usual definitions of addition and multiplication, determine which of these proper-
ties are possessed by N, Z, and Q, respectively.
Solution
N possesses properties (f1), (f2), and (f5). The natural numbers do not contain 0 or any
additive inverses. Z additionally satisfies (f3) but still does not satisfy (f4) for multiplicative
inverses (e.g. 2−1 is not in Z). Q possesses all properties; in particular if q ∈ Q then so are
−q and 1/q.
Exercise 8.6.4
Show that this defines an ordering on R by verifying properties (o1), (o2), and (o3) from
Definition 8.6.5.
8.6. A CONSTRUCTION OF R FROM Q 247
Solution
To prove property (o1), assume A ⊈ B. By definition, this means there is some a ∈ A with
a∈/ B. From Exercise 8.6.2 this means ∀b ∈ B, b < a. Then by property (c2) b ∈ A; hence
B ⊆ A.
(o2) is a direct result of the definition of set equality, and (o3) is true because of transivity
of the set inclusion relationship.
Exercise 8.6.5
(a) Show that (c1) and (c3) also hold for A + B. Conclude that A + B is a cut.
Solution
(a) For (c1), we can find a ∈ A, a′ ∈/ A, b ∈ B, and b′ ∈ / B. Then a + b ∈ A + B so
A + B ̸= ∅. Also, since a < a and b < b′ , a + b ̸= a′ + b′ and therefore a′ + b′ ∈
′
/ A + B.
For (c3), let arbitrary a + b ∈ A + B with a ∈ A and b ∈ B, and let a < a′ ∈ A and
b < b′ ∈ B. Then a + b < a′ + b′ ∈ A + B.
(b) Let arbitrary a + b ∈ A + B; then b + a ∈ B + A; hence (f1) holds for addition. Let
arbitrary (a + b) + c ∈ (A + B) + C, then a + (b + c) ∈ A + (B + C); hence (f2) holds
for addition.
(c) Let Y ⊆ Z, and let x+y ∈ X +Y with x ∈ X and y ∈ Y . Then y ∈ Z so x+y ∈ X +Z,
implying X + Y ⊆ X + Z.
Exercise 8.6.6
(a) Prove that −A defines a cut.
(c) If a ∈ A and r ∈ −A, show a + r ∈ O. This shoes A + (−A) ⊆ O. Now, finish the
proof of property (f4) for addition in Definition 8.6.4.
248 CHAPTER 8. ADDITIONAL TOPICS
Solution
(a) Let t ̸= A and r < −t. Then −r > t so r ∈ −A and −A ̸= ∅. Let a ∈ A. Then
∀t ̸= A, t > a; therefore −a ∈ / −A and A ̸= Q, proving (c1). To prove (c2), let
arbitrary r ∈ −A and let q < r. Then −q > −r > t for some t ∈ / A; hence q ∈ −A. To
prove (c3), let r ∈ −A, and let t ∈
/ A satisfy t < −r. Let q = (t − r)/2, and note that
t < q < −r, so −q ∈ −A. Moreover −q > r, so (c3) is proved.
(b) This would fail on the cut A = {t ∈ Q : t < −2}, since then −A = {t ∈ Q : t ≤ 2}
which from Exercise 8.6.1 (b) is not a cut.
(c) Since −r ∈ −A, then we have some t ∈ / A where −r > t > a. Then 0 > a + r implying
a + r ∈ O. Now suppose o ∈ O; we would like to find a ∈ A and r ∈ −A satisfying
o ≤ a + r; this would imply O ⊆ A + (−A) and thus O = A + (−A). Since o ∈ Q and
o < 0, let o = −p/q where p, q ∈ N. We can prove the following lemma: for any cut A
and n ∈ N, we can find z ∈ Z where
z z+1
∈ A and ∈
/A
n n
To do this, start with a ∈ A and a′ ∈
/ A, and find N, M ∈ Z satisfying
N M
< a and > a′
n n
Clearly Nn ∈ A and Mn
/ A. If we count upwards from i = N , at some point ni is going
∈
to transition from being in A to not being in A, so that by the time we reach i = M
we have ni ∈/ A. This transition point implies the existence of z.
n n+1
Now, let a = 2q
∈ A with 2q
/ A. Then r = − n+2
∈ 2q
∈
/ A, and
−2 1 −p
a+r = =− ≥ =o
2q q q
completing the proof.
Exercise 8.6.7
(a) Show that AB is a cut and that property (o5) holds.
(b) Propose a good candidate for the multiplicative identity (1) on R and show that this
works for all cuts A ≥ O.
(c) Show the distributive property (f5) holds for non-negative cuts.
Solution
(a) For (c1): −1 ∈ AB so AB ̸= ∅. If a′ ∈ / A and b′ ∈
/ B, then a′ b′ ≥ 0, and for any
′ ′ ′
a ∈ A, b ∈ B, ab < ab < a b ̸= ab so AB ̸= Q proving (c1).
For (c2): suppose r ∈ AB and let q < r. There are two cases to consider: r < 0 and
r = ab. If r < 0, then q < r < 0 must be in AB. If r = ab, let’s assume q ≥ 0 (the
case q < 0 is trivial), implying ab > 0. Then
q q
q= a (b) = b
ab b
8.6. A CONSTRUCTION OF R FROM Q 249
q q q
where since ab
< 1 and is defined, b
< a and b
∈ A. Hence q ∈ AB, proving property
(c2).
For (c3): suppose r ∈ AB. If r < 0 then r < r/2 < 0 is also in AB. If r = 0 then that
implies 0 ∈ A and 0 ∈ B, which means we can find a ∈ A and b ∈ B where a, b > 0
and so r < ab ∈ AB. If r > 0 then r = ab, and we can find a < a′ ∈ A and b < b′ ∈ B
with ab < a′ b′ ∈ AB, proving (c3) in all cases.
For (o5): Note that the definition of AB is of the form K ∪ O with K some set, so
O ⊆ AB trivially.
(b) Let the multiplicative identity I = p ∈ Q : p < 1. 1/2 ∈ I and 1/2 ∈
/ O so I ̸= 0. Let
A ≥ O and let a ∈ A, and let a < a′ ∈ A. Then aa′ < 1 so aa′ ∈ I, and a = aa′ a′ ∈ I
implying A ⊆ IA.
Now let r ∈ IA. If r < 0 then since A ≥ O, r ∈ A. If r ≥ 0, then we have r = ia with
i ∈ I and a ∈ A, with a ≥ 0. Then since i < 1, we have ia < a ∈ A and so r ∈ A.
Therefore A = IA.
(c) If a ∈ X(Y + Z), then either a < 0 or a = x(y + z), for some x ∈ X, y ∈ Y , and
z ∈ Z. If a < 0, since XY + XZ ≥ O, we have a ∈ XY + XZ. If a = x(y + z), we
have a = xy + xz ∈ XY + XZ, so either way X(Y + Z) ⊆ XY + XZ.
Now suppose a ∈ XY + XZ. Then a = r1 + r2 where r1 ∈ XY and r2 ∈ XZ. We
have three cases to consider: r1 , r2 ≥ 0, r1 , r2 < 0, and r1 < 0 while r2 ≥ 0. (The case
of r1 > 0 and r2 < 0 is equivalent to the last case, with Y and Z flipped, which is an
acceptable operation since Y + Z = Z + Y .)
If r1 and r2 are both negative, then a < 0 and a ∈ X(Y + Z) trivially. If r1 ≥ 0 and
r2 ≥ 0, then we have a = x1 y + x2 z where x1 , x2 ∈ X, y ∈ Y , z ∈ Z, and these four
numbers are all non-negative. Let x = max{x1 , x2 }, and note that a ≤ xy + xz =
x(y + z) ∈ X(Y + Z); therefore a ∈ X(Y + Z).
If r1 < 0 while r2 ≥ 0, then a = r1 + xz where 0 ≤ x ∈ X and 0 ≤ z ∈ Z. If x = 0
then 0 > a ∈ X(Y + Z), so let’s take x > 0. If a < 0 then clearly a ∈ X(Y + Z).
Otherwise, define y = rx1 < 0 and note y ∈ Y . We have a = xy + xz = x(y + z) ≥ 0,
so y + z ≥ 0 and a ∈ X(Y + Z).
Exercise 8.6.8
Let A ⊆ R be nonempty and bounded above, and let S be the union of all A ∈ A.
(a) First, prove that S ∈ R by showing that it is a cut.
(b) Now, show that S is the least upper bound for A.
Solution
(a) For (c1): Let A ∈ A, and let a ∈ A. Then a ∈ S, so S ̸= ∅. Now let B be a bound
on A, and let b′ ∈
/ B. Then for any s ∈ S, we have s ∈ A ∈ A for some A. Thus since
A ≤ B, s ∈ B and so b′ > s, so b′ ∈
/ S and S ̸= Q.
For (c2) and (c3): Consider arbitrary s ∈ S, with s ∈ A ∈ A. Then for any q < s we
must have q ∈ A =⇒ q ∈ S, so (c2) is satisfied. Also, we can find s < a ∈ A, with
a ∈ A, so (c3) is satisfied.
250 CHAPTER 8. ADDITIONAL TOPICS
(b) Let B be an upper bound for A. Now for any s ∈ S, with s ∈ A ∈ A, we have A ⊆ B
so s ∈ B. Therefore S ⊆ B and S is the least upper bound for A.
Exercise 8.6.9
Consider the collection of so-called “rational” cuts of the form
Cr = {t ∈ Q : t < r}
(a) Show that Cr + Cs = Cr+s for all r, s ∈ Q. Verify Cr Cs = Crs for the case when
r, s ≥ 0.
Solution
(a) Let a + b ∈ Cr + Cs satisfying a ∈ Cr and b ∈ Cs . Then a + b < r + s so a + b ∈ Cr+s .
Now let c ∈ Cr+s . Let δ = r + s − c > 0. Then d = r − δ/2 ∈ Cr , e = s − δ/2 ∈ Cs ,
and c = d + e ∈ Cr + Cs . Hence Cr + Cs = Cr+s .
Let x ∈ Cr Cs . Since rs ≥ 0, if x < 0 then x ∈ Crs . If x ≥ 0, x = ab where a ∈ Cr and
b ∈ Cs . Then a < r and b < s, implying ab < rs and x ∈ Crs .
Now let c ∈ Crs . If c < 0 then c ∈ Cr Cs trivially; otherwise note rs > 0 and define
q
q = c+rs
2
, so 0 ≤ c < q < rs. Define 0 ≤ δ1 = qc < 1 and δ2 = rs < 1. Then δ1 r ∈ Cr ,
δ2 s ∈ Cs , and c = δ1 rδ2 s ∈ Crs . Therefore Cr Cs = Crs .