Professional Documents
Culture Documents
Foundations of Analysis
Michael Senter
2
Part I
3
Chapter 1
Question 1.1.5. What is the intersection of all the closed intervals containing the open
interval (0, 1)? Justify your answer.
Let A denote the set of all sets such that (0, 1) A . The intersection of all closed
intervals is denoted A. It is defined as
A = {x : x A A A}.
In other words, we are looking for some set A such that A is a subset of every other set in
A. This set is A = {x : 0 < x < 1}. Consider any other subset C of A. If C = A, then
necessarily there must exist an element x such that x C and x / A, showing that A C
but C * A. Since x is not in every subset of A, x
/ A.
5
6 CHAPTER 1. THE REAL NUMBERS
Question 1.1.6. What is the union of all of the closed intervals contained in the open in-
terval (0, 1)? Justify your answer.
Let A be the setcontaining all sets containing (0, 1)as a subset. The union of all these
sets is denoted by A. An object x is an element of A if there exists some set C A
such that x C. We need to consider two cases: either the object x 0 or 1 x. The
case of 0 < x < 1 is trivial. For any x such that 1 x we can create a set C such that
C = {y : 0 < y < x}.
This since 1 x, it is guaranteed that C A. The case of x 0 is
analogous. Hence, A = (, ).
Problem 1.1.7. If A is a collection of subsets of a set X, formulate and prove a theorem
like Theorem 1.1.5 (from book numbering) for the intersection and union of A.
Theorem
c 1.1.7. Let A be a collection
c ofc subsets A1 , A2 , ..., An of some set X. Then
( A) = A1 A2 ... An and ( A) = A1 A2 ... Acn .
c c c c
Proof. Thisis a generalization of DeMorgans law, proved in the book. We begin with the
statement ( A)c = Ac1 Ac2 ... Acn . We can rewrite ( A)c as (A1 A2 ... An )c . We
can then sub-partition this collection of unions into a collection of two unions, as such:
( A)c = [A1 (A2 ... An )]c
Then we will refer to A2 ... An as B. We can then rewrite the above as (A1 B)c , for
which DeMorgans laws apply. Thus, we write (A1 B)c = Ac1 B c = Ac1 (A2 ... An )c .
As next step, we sub partition B into two sets, as such
(A2 ... An )c = [A2 (A3 .... An )]c
Then DeMorgans laws apply again as above, and we can write [A2 (A3 .... An )]c =
Ac2 (A3 ... An )c . Since intersections and unions are associative, we can then write
( A)c = (Ac1 (Ac2 (A3 ... An )c )) = Ac1 Ac2 (A3 ... An )c
We continue
an inductive application of DeMorgans laws as outlined above, until we see
that ( A)c = Ac1 Ac2 ... Acn
The other proof is analogous, requiring a sub-partition of the collection of intersections and
rewriting them into series of intersections of two sets to which DeMorgans laws apply.
Problem 1.1.8. Which of the following functions f : R R are one to one and which ones
are onto. Justify your answer.
(a) f (x) = x2 ; This function is neither onto, nor one-to-one. It is not onto, since there is
no x such that f (x) < 0. It is not one-to-one since f (x) = f (x) for all x R.
(b) f (x) = x3 ; This function is both one-to-one and onto. It is one-to-one since there
f (x) = f (y) for all x, y such that x = y. It is onto, as for any y R, there exists an x R
such that f (x) = y.
(c) f (x) = ex This function is one-to-one, but not onto. It is one-to-one, for f (x) = f (y) for
all x, y R such that x = y. It fails to be onto since there exists no x such that f (x) < 0
for any x R.
1.1. SETS AND FUNCTIONS 7
The above conditions are fulfilled for a function f (x) = x with A = B = [0, 10], and the
subsets E = [1, 6] and F = [1, 2] E.
Exercise 1.4.2. For each of the sets in (a), (b), (c) of the preceding exercise, find the least
upper bound of the set, if it exists.
Theorem 1.4.3. If a subset A of R is bounded above, then the set of all upper bounds for
A is a set of the form [x, ). What is x?
Proof. Let B denote the set of all upper bounds of A. By definition, a number m R is
considered an upper bound for the set A if z m for all z A. If the set A has a largest
number, then this largest number - y - will be in the set B. In that case, it is obvious that
all numbers m > y will also be upper bounds, since we assumed that x y for all x A,
and that m > y , it follows that x y < m. Therefore, the set [y , ) would be the set of
1.5. SUP AND INF 9
Exercise 1.5.2. Find the sup and inf of the following sets. Tell whether each set has a
maximum or a minimum.
(a) (2, 8]; The infimum of the set is 2 and the supremum is 8. The has a maximum, but
not a minimum.
(b) nn+2
2 +1 ; The infimum of the set is 0, and the supremum is 2. The set has a maximum, but
no minimum.
(c) {n/m : n, m Z, n < 5m }; The infimum of the set is 5, and the maximum is 5.
2 2
Seeing that 5 is not a rational number, the set has neither a maximum nor a minimum.
Exercise 1.5.3. Prove that if sup A < , then for each n N there is an element an A
such that sup A 1/n < an sup A.
Proof. This is true since we can easily construct an element an such that this equality holds.
We assume that A is defined for all m/n with m, n Z within A. In this case, we constructs
our term to be an = sup A 1/(n + 1). It is obvious that since 1/(n + 1) < 1/n, that
sup A 1/n < sup A 1/(n + 1) sup A.
Alternatively, we may also note that sup A 1/n < sup A for all n N by definition, so the
inequality holds in the trivial case of an = sup A.
Exercise 1.5.4. Prove that if sup A = , then for each n N there is an element an A
such that an > n.
Proof. Assume some set A whose supremum is +. In that case, x A, x < . Both
from the Archimedean property and from the Peano Axioms we know that for every n N,
there is a successor element n which is also in N, such that n < n . Since there @a such
that a = , and n < , this implies that an such that an = n and an A, showing that
n < an < .
Exercise 1.5.5. Formulate and prove the analog of Theorem 1.5.4 for inf.
Proof. By definition, a x if and only if x is a lower bound for A. If x is a lower bound for
A, then A is bounded below. This implies that its inf is its greatest lower bound, which is
necessarily greater than or equal to x. Conversely, if inf A x, then inf A is finite and is the
greatest lower bound for A. Since inf A x, x is also a lower bound for A. Thus, inf A x
if and only if a x for every a A.
If x > inf A, then x is not a lower bound for A, which means x > a for some a A.
Conversely, if x > a for some a A, then x > inf A, since a inf A. Thus, x > inf A if and
only if x > a for some a A.
Proof. According to the book, sup(A + B) = sup A + sup B (proof on p. 30). We can
then write sup(A + (B)) = sup A + sup(B). We then apply Theorem 1.5.7b, to rewrite
sup(B) as inf A. From this it follows that
Theorem. Let f and g be functions defined on a set containing A as a subset, and let
c R be a positive constant. Then supA cf = c supA f and inf A cf = c inf A f .
3. inf I f = 0, supI f = 1. The inf of f on I is actually a minimum, but the sup is not a
maximum.
Theorem. Let f and g be functions defined on a set containing A as a subset, and let
c R be a positive constant. Then supA (f ) = inf A f .
12 CHAPTER 1. THE REAL NUMBERS
Proof. We have a function f : A B. A number x is a lower bound for f (a) for all
a A if and only if x is an upper bound for the set f (a). Let L be the set of all lower
bounds for f (a). Then L is the set of all upper bounds for f (a). Furthermore, the
largest member of L and the smallest member of L are negatives of each other. That is,
inf f (a) = sup(f (a)), or equivalently inf f = sup(f ).
Theorem. Let f and g be functions defined on a set containing A as a subset, and let
c R be a positive constant. Then supA (f + g) supA f + supA g and inf A f + inf A g
inf A (f + g).
Proof. By definition, f (a) sup f for all a A and g(a) sup g for all a A. Therefore,
f (a) + g(a) sup f + sup g. Let c denote the supremum of f + g. We know that sup f + sup g
is an upper bound for f (a) + g(a). Since the supremum is always less than or equal to an
upper bound, we find that c sup f +sup g. This implies that sup(f +g) sup f +sup g.
Theorem. Let f and g be functions defined on a set containing A as a subset, and let
c R be a positive constant. Then sup{f (x) f (y) : x, y A} = supA f inf A f .
Proof. This appears somewhat obvious. The function f is defined on A, i.e., for every a A,
f maps to some value f (a) in some set, lets call it B. The value sup f is defined as to be
the least upper bound of f (a), i.e. @x such that f (x) > sup f for some x A. The infimum
is defined as the value such that there is no value x A such that x < inf f . The value
defined by f (x) f (y) for all x, y A is a measure of the distance between these two values.
Since sup f and inf f are defined as above, we can see that there cannot be a greater distance
between any other two points in B than the distance between sup f and inf f . Therefore,
for any collection of distances between points in B reached by f (x) for all points x A, the
supremum of this collection - namely, the largest value of this set such that no other value
is larger - cannot be any other than the distance between the supremum and the infimum of
the function itself.
Chapter 2
Sequences
(c) if |x a| < 1/2 and |y b| < 1/2, then |x + y (a + b)| < 1. We add the inequalities
and get |x a| + |y b| < 1/2 + 1/2 = 1. We can then rewrite using the triangle inequality
as above
|x 1| + |x 2| < 1,
Exercise 2.1.3. Put each of the following sequences in the form a1 , a2 , a3 , . . . , an . This
requires that you compute the first 3 terms and find an expression for the nth term.
(a) the sequence of positive odd integers; This is a sequence of the form 1, 3, 5, . . .. To
find the n th term, we express this sequence as an = 2n 1, with n N.
(b) the sequence defined inductively by a1 = 1 and an+1 = a2n ; The sequence begins
13
14 CHAPTER 2. SEQUENCES
with 1, 1/2, 1/4, . . .. The nth term will be something like an = ((1)n1 )/(2n1 ) for n N.
an
(c) the sequence defined inductively by a1 = 1 and an+1 = n+1 . This is the series
2
1, 1/3, 1/12, 1/60, . . .. The nth term is: an = (n+1)! .
Exercise 2.1.4. Find lim 1/n2 .
The larger n become, the smaller 1/n2 will become. We guess the limit to be 0. For
2
any > 0, we need an N such that whenever n > N , 1/n < . We find that this is true
2
whenever 1/ < n , or in other words - whenever 1/ < n.
Exercise 2.1.5. Find lim 2n1
3n+1
.
Exercise 2.1.9. The limit is zero. We know for any > 0, there exists N1 such that for all
n > N,
1
< /2,
n
since /2 is another positive real and we have shown the limit of 1/n as n goes to infinity is
zero in problem 4. Similarly, we may find N2 such that for all n > N2 ,
(1)n
n2 < /2,
since we found the limit as n of (1)n /n earlier to be 0, and it can be shown in much
the same way that lim (1)n /n2 = 0. Hence, for n > max{N1 , N2 }, both inequalities hold
n
and we may add the two and apply the triangle inequalityto obtain
1 (1)n 1 (1)n
+
0 + 2 < ,
n n2 n n
hence the limit is zero.
Exercise 2.1.10. Prove that lim 2n = 0. Hint: prove first that 2n n for all natural
numbers n.
Proof. We wish to show that 2n > n for all n. Proof by induction. The base case, 21 > 1
is obviously true, since 21 = 2. We assume now that 2n > n for some n. Then we wish to
check 2n+1 . But, we can rewrite this simply as 2n 21 . Let k = 2n . Since we know that k > n,
it is obvious that 2k > n + 1. Thus, 2n > n for all n N.
We note that 2n = 21n . Thus, lim 2n = lim 21n . Since 2n increases until infinity, we see that
1/2n will grow smaller and smaller, since 1/2n > 1/2n+1 for all n.
We see that for any > 0, we need to simply pick n such that 1/ < 2n . As such, the limit
is 0.
Exercise 2.1.11. Prove that if an 0 and k is any constant, then kan 0.
If an 0, this means that an < for any > 0. We multiply by k and find that kan < k.
Exercise 2.2.3. Make an educated guess as to what you think the limit is, then use the
definition of limit to prove that your guess is correct.
This last step is true because for any positive integer n, we have 3n + 2 > 0, hence 2n2 + 4n +
2 > 2n2 + n, and so dividing by n(n2 + 2n + 1) on both sides yields the desired inequality.
Hence, (
)2 2
n 2
1 < < = .
n+1 n N
Exercise 2.2.5. Make an educated guess as to what you think the limit is, then use the
definition of limit to prove that your guess is correct.
1 1
The limit is . Let A = {n N : |an | } for a given . This set is finite since
2 2
( )2
1 1
n +nn n +nn + +
2 2 2
+ n(2 + 1)
2 2
( )2
1
0 + + 2n.
2
But this last inequality is never true, since each of the terms is strictly greater than zero. Of
course, we might also have
1
n + n2 + n n2 + +n(1 2) + 2 n2 + n
2
2n + 2 0
n,
2
which is certainly only true for finite n. Hence for a given the set A is in fact finite, so the
1
limit is .
2
Exercise 2.2.6. Make an educated guess as to what you think the limit is, then use the
definition of limit to prove that your guess is correct.
We note that each term is of the form c/n or multiples thereof for some constant c. It has
already been shown that each such term tends can be made smaller than any . This also
holds for the sum.
Exercise 2.2.7. Suppose the sequence converges to a, then for all > 0, there is an N
so that when n > N , |an a| < a < an < a + . In particular, let = 1, then
certainly for all n > N , the sequence is bounded above by a + 1 and below by a 1. For
n < N, notice that the a1 , ..., aN is bounded above by e = max{|a1 |, ..., |an |} and below
by f = min{a1 , ..., aN }. Hence the entire sequence is bounded above by max{e, a + 1} and
min{f, a 1}.
We then note that we are given that |an a| < . From this we see that
Exercise 2.2.9. Does the sequence {cos(n/3)} have a limit? Justify your answer.
No. The sequence {cos(n/3)} oscillates between 1 and 1; a limit cannot converge to
two dierent values. Hence, this sequence does not have a limit.
Exercise 2.2.10. For a similar reason as in the previous problem, the sequence an =
cos(n) does not converge but the sequence |an | = | cos(n)| does converge since |an | = 1
for all n.
Exercise 2.2.11. Prove that if {an } and {bn } are sequences with |an | bn for all n and if
lim bn = 0, then lim an = 0 also.
We are given that |an | bn for all n. Therefore, we know that lim |an | lim bn . We
know that lim bn = 0. Hence we can write - equivalently - that lim |an | 0. We notice that
|an | is defined to be greater than or equal to zero. Hence we have 0 lim |an | 0, from
which it follows by the squeeze theorem (proof on p. 43 of the book) lim |an | = 0.
Exercise 2.2.12. Prove the following partial converse to Theorem 2.2.3: Suppose {an } is a
convergent sequence. If there is an N such that an c for all n > N , then lim an c. Also,
if there is an N such that b an for all n > N , then b lim an .
Note that an is bounded by c according to the premise. In this case, we can say that
an sup an c for all n. Let a = lim an . We know by definition that a sup an , and
therefore we can write that lim an sup an c.
Likewise, we can say that b is a lower bound for an such that b inf an . We know that by
definition inf an a, allowing us to write b lim an .
18 CHAPTER 2. SEQUENCES
Exercise 2.2.15. Suppose this sequence converges to( zero. Then)for all > 0, there exists
N
N such that for all n > N , |an | < . But consider n = 6
+ 1 106 . This is a multiple
10
of one million, so an . And, it is larger than N since
N N
< +1
106 106
( )
N
by definition, 106 + 1 > N. Hence we have a contradiction.
106
n2 5
lim 3 = (dividing top and bottom by n3 )
n + 2n2 + 5
1/n 5/n3
lim =
1 + 2/n + 5/n3
lim(1/n 5/n3 ) 0
= = 0.
lim(1 + 2/n + 5/n3 ) 1
Exercise 2.3.4. Let an = 1/n, then lim an = 0, and let bn = sin n, then |bn | 1. By
sin n
theorem 2.3.2, the sequence = an bn has limit zero.
n
Exercise 2.3.5. Prove Theorem 2.3.2.
Proof. We know that lim an = 0, hence we know that for all > 0, there exists an N such
that whenever n > N , |an | < . Likewise, we know that bn is bounded, such that we can
state that q bb q. We can then also write |an | < |q|
. We guess that the limit of (an )(bn )
is zero, so we write:
Thus, the lim an bn = 0, since we there is an N such that the above inequality is true whenever
we pick an n > N .
2.3. LIMIT THEOREMS 19
Exercise 2.3.6. Prove that a sequence {an } is both bounded above and bounded below if
and only if its sequence of absolute values {|an |} is bounded above.
Proof. By definition, if {|an |} is bounded above, then there exists some M such that |an | M
for all n. This is equivalent to saying M an M , which proves that {an } is bounded
above and below.
Proof. Since both an and bn have a limit, we can write |an a| < 2 and |bn b| < 2 . For
all , we have an N such that if we choose n > N , these inequalities are true. We know add
them together and find
Exercise 2.3.8. Prove that if {bn } is a sequence of positive terms and bn b > 0, then
there is a number m > 0 such that bn m for all n.
This is true by virtue of the definition of R. The statement above is equivalent to saying
that we are looking for some m such that 0 < m bn . By definition R is full, such that
between any two numbers, there are infinitely more numbers.
Exercise 2.3.9. Prove part (d) of Theorem 2.3.6. Hint: Use the previous exercise. I.e, that
if an a and bn b, an /bn a/b, if b = 0 and bn = 0 for all n.
Proof.
1 1 1 1 1 1 1 1
|an a | = |an a + a a | |an = a|| | + |a|| |
bn b bn bn bn b bn bn b
We know that {1/bn } is bounded, and hence {|1/bn |} is bounded above. We also have
|an a| 0. Therefore, |an a||1/bn | 0. Also, |a||1/bn 1/b| 0. By (b) we know that
|an a||1/bn | + |a||1/bn 1/b| 0, proving that an /bn a/b.
Exercise 2.3.10. Prove part (f) of theorem 2.3.6. Hint: use the identity:
xk y k = (x y)(xk1 + xk2 y + . . . + y k1 )
1/k 1/k
with x = an and y = a1/k , to show that an a1/k if an 0 for all n.
n a
a1/k = (an a)(a1/K1 a1/k + . . . + a1/K1 ) = (an a)bn
1/K
n + a1/K2
n
20 CHAPTER 2. SEQUENCES
where
bn = a1/K1
n + a1/K2
n a1/k + . . . + a1/K1
We know that {an } converges, and hence that {|an |} is bounded above. We choose an upper
1/k
bound m for {|an |} which satisfies that |an | m. Then bn 1/k
m
sowing that {|bn |} is
bounded above. According to theorem 2.3.2 we conclude that |an a||bn | and find from
1/k
theorem 2.3.1 that an a1/k .
Exercise 2.3.12. Prove that lim a1/n = 1. Hint: use the result of the previous exercise.
We notice that n1/n > 1 for all n N.We can therefore write that we are looking for a
solution to n1/n 1 < .We can rearrange and raise both sides to the nth power, resulting in
the equation n < (1 + )n . We can expand the right hand side using the binomial theorem:
1
n < 1 + n + n(n 1)2 + . . .
2
As long as n < 21 n(n 1)2 this inequality holds, requiring that n > 1 + 2
2
. Therefore, for
any > 0 there exists an N such that whenever n > N , |n1/n 1| < .
Exercise 2.4.4.
5 3
4 n+1 = .
Exercise 2.4.8. Prove that lim nn+3n +2
5 3
4 n+1 = .
And hence, lim nn+3n +2
Exercise 2.4.10. If the sequence is bounded, then we are done by theorem 2.4.1. Otherwise
there is no M R such that M an for all n. This is equivalent to saying any M R, there
exists N such that aN > M . Since the sequence is monotone, for all n > N , an > aN > M .
Hence it has limit equal to infinity.
Proof. If an , there exists some value of an such that an > M for any possible M R.
If we consider the sequence an (1), we see clearly that an < M for any M R. But if
that is true, then lim an = .
Exercise 2.4.12. Suppose lim bn < . Then bn is bounded by theorem 2.2.3, so there exists
M with bn M for all n. Since an bn , we have an M so lim an < , a contradiction.
Exercise 2.4.13. For all M R, given k 0, M/k R. Hence there exists N , for all
n > N , an > M/k. Since bn k, we have an bn > k M/k = M, hence lim an bn = .
Exercise 2.5.4. Prove by induction that if {nk } is an increasing sequence of natural num-
bers, then nk k for all k.
Proof. Assume the base case nk = n, which is the series 1, 2, 3, 4, 5, . . .. Since k is the counter
index, i.e. k N, it is obvious that nk = k = 1, 2, 3, 4, 5, . . .. We generalize to the n + 1 case,
i.e. nk = n + 1. In that case we have nk = n + 1 = k + 1 > k.
Exercise 2.5.5. Which of the following sequences {an } have a convergent subsequence?
Justify your answer.
(a) an = (2)n ; None of the subsequences are convergent, as they either tend to + or .
nn
(b) an = 5+(1)
2+3n
; This sequence is convergent for all n such that n mod 2 = 0, which is
the sequence starting with 0.875, 0.6428, 0.55, 0.5, 0.46875, 0.4473, 0.4318, 0.42, 0.41071, . . .
n
(c) an = 2(1) This sequence has convergent subsequences for all n such that n mod 2 = 0
and for n mod 2 = 1.
Exercise 2.5.7. For each of the following sequences, determine how many dierent limits
of subsequences there are. Justify your answer.
(a) {1 + (1)n }; This sequence is 0, 2, 0, 2, 0, 2, . . . and as such has two dierent limits: 0
and 2.
(b) {cos(n/3)}; There are two dierent limits. The first approaches 1 for the sequence of all
n where n mod 6 = 0. The second limit is attained at 1 for all n such that n mod 6 = 3.
(c) 1, 21 , 1, 12 , 31 , 1, 12 , 31 , 14 , 1. 21 , 13 , 41 , 15 , . . . The terms a1 , a3 , a6 , a10 , a1 5, . . . are convergent to 1.
Exercise 2.5.8. Does the sequence sin n have a convergent subsequence? Why?
Exercise 2.5.9. Prove that a sequence which satisfies |an+1 an | < 2n for all n is a Cauchy
sequence.
Proof. We notice that the sequences defined by the above condition are non-increasing and
covergent. We notice the following pattern:
Inductively, we see that this pattern continues for all patterns an and an + k with k N.
Now, we assume two indices m and n such that m > n. We find
0
2n ( 2k ) = 2n (2 2m+n+1 ) = 21n 21m
k=m+n+1
We want to to prove |21n 21m | 21n + 21m < . We solve the equations 21n <
2
and
21m < 2 . The solution to this is 2 log(2)
log()
< n, m
Exercise 2.6.2. Find lim inf and lim sup for the sequence an = n
2kn
1 with kn being the
largest integer k so that 2k n.
Exercise 2.6.3. Find lim inf and lim sup for the sequence 1, 12 , 1, 12 , 13 , 1, 21 , 13 , 14 , 1, . . ..
Exercise 2.6.5. If lim sup an is finite, prove that lim inf(an ) = lim sup an .
24 CHAPTER 2. SEQUENCES
Proof. By assumption, lim sup an is equal to some a, such that a an for all an {an }.
We multiply this by 1 to find the inverse sequence {an }. Then we have a an for
all an {an }. By definition, this means a = lim inf(an ). Therefore, lim inf(an ) =
lim sup an .
Exercise 2.6.8. If {an } and {bn } are non-negative sequences and {bn } converges, prove
that lim sup an bn = (lim sup an )(lim bn ).
Proof. We need to consider two cases. First, assume {an } is not bounded above. Then
lim sup an = . It then doesnt matter what we multply an with, we will always get infinity
provided that bn = 0.Then lim sup(an bn ) = lim sup an lim bn = .
We now consider case 2, where an is bounded above. By Bolzano-Weierstrass we know that
an then has at least one convergent subsequence. Let a be the subsequential limit of an , and
let M be the upper bound of an . We know then that lim sup an exists and lim sup an M .
MORE WORK NEEDED ON THIS. We note that according to the main limit theorem, if
an a and bn b, an bn ab. Thus lim sup(an bn ) = lim sup an lim bn .
Exercise 2.6.12. Which numbers do you think are subsequential limits of {sin n}
n=1 ? Can
you prove that your guess is correct?
Continuous Functions
3.1 Continuity
Exercise 3.1.1. If f is a function with domain [0, 1], what is the domain of f (x2 1)?
======= If f is a function with domain [0, 1], what is the domain of f (x2 1)?
The domain is R\{1, 1}. The function is continuous everywhere except for the points
not part of the domain.
Exercise 3.1.3. For any real x, x2 + 1 is defined and nonzero. And, 1 + x2 is continuous
on R, so by theorem 3.1.9(d), it is continuous.
Exercise 3.1.4. Show that the function f (x) = |x| is continuous on all of R.
Proof. We need to find a such that for any > 0, we have ||x| |a|| < whenever
|x a| < .
Exercise 3.1.5. Assuming sin is continous, prove that sin(x3 4x) is continuous.
Proof. We know that |sin(x)| < 1 for all x.
Exercise 3.1.6. Since f (x), g(x) are continuous, and {f (xn )} converges to f (a) and {g(xn )}
converges to g(a), when {xn } converges to a. By theorem 2.3.6(d), {f (xn )/g(xn )} converges
to f (a)/g(a) since g(a) = 0 by assumption and g(xn ) = 0 in the domain of f (x)/g(x) by
definition.
25
26 CHAPTER 3. CONTINUOUS FUNCTIONS
Exercise 3.1.10. Let f be a function with domain D and suppose f is continuous at some
point a D. Prove that, for each > 0, there is a > 0 such that
|f (x) f (y)| < whenever x, y D (a , a + )
solved in class
Exercise 3.1.11. Prove that the function (piecewise function) is not continuous at 0.
solved in class
solved in class
Exercise 3.2.2. Prove that if f is a continuous function on a closed bounded interval I and
if f (x) is never 0 for x I, then there is a number m > 0 such that f (x) m for all x I
or f (x) m for all x I.
Proof. Assume f (a) > 0. We have that f ([a, b]) = [m, M ]. We know that m = min f ,
M = max f . Lets prove that m > 0. By contradiction: assume m < 0. Value 0 is taken
(non-legible) an intermediate value [m, f (a)] which contradicts f (x) = 0. Prove for case 2 is
analogous (show that M < 0).
Exercise 3.2.3. Prove that if f is a continuous function on a closed bounded interval [a, b]
and if (x0 , y0 ) is any point in the plane, then there is a closest point to (x0 , y0 ) on the graph
of f .
Proof. Pick any point x [a, b]. Then the distance to x0 , y0 is dist ((x0 , yo ), (x, f (x))) =
1
((x x0 )2 + (y0 f (x))2 )) 2 . We must prove that this function attains its minimum value
1
in [a, b] and that if f is continuous, then ((x x0 )2 + (y0 f (x))2 )) 2 is also continuous on
[a, b]. Then distance takes its minimum value there.
Exercise 3.2.4. Find an example of a function which is continuous on a bounded (but not
closed) interval I, but is not bounded. Then find an example of a function which is contin-
uous and bounded on a bounded interval I, but does not have a maximum value.
which would not assume a maximum on such an interval would be by violating the continuity.
For example, the function
{
2x x < 1/2
f (x) =
0 x 1/2
fails to achieve its maximum on a bounded interval [0, 1]. However, it does so by having a
discontinuity at x = 1/2.
Exercise 3.2.5. Let f (x) = ex , I = [1, ). Then f is continuous on a closed (but not
bounded) interval, but is not bounded. If f (x) = 1/(1 + ex ) and I = [0, ), then f is
continuous and bounded by 1 on I which is closed but does not have a maximum.
Exercise 3.2.7. Give an example of a function defined on the interval [0, 1] which does not
take on every value between f (0) and f (1).
In other words, we are looking for a function with a discontinuity between [0, 1]. One
example would be: {
x : 0 x < 21
f (x) =
2x : 12 x 1
Exercise 3.2.8. Show that if f and g are continuous functions on the interval [a, b] such
that f (a) < g(a) and g(b) < f (b), then there is a number c (a, b) such that f (c) = g(c).
Proof. We create a function h(x) = f (x) g(x). This is continuous since it is a linear
combination of continuous functions, and it is defined on [a, b]. We know that h(a) =
f (a) g(a) < 0 and h(b) = f (b) g(b) > 0. Bt the intermediate value theorem there exists
a c such that h(c) = f (c) g(c) = 0, which implies f (c) = g(c).
Exercise 3.2.9. Let f be a continuous function from [0, 1] to [0, 1].Prove that there is a
point c [0, 1] such that f (c) = c - that is, show that f has a fixed point. Hint: Apply the
Intermediate Value Theorem to the function g(x) = f (x) x.
Proof. Let g(x) = f (x) x. Since f (x) is continuous, we know that g(x) is also continuous.
Then g(a) 0 and g(b) 0. By the intermediate value theorem we know that there exists
some x [0, 1] such that g(x) = 0, which implies that f (x) = x.
Exercise 3.2.10. Use the intermediate value theorem to prove that if n is a natural number,
then every positive number a has a positive n-th root.
Proof. We write the function f (x) = xn which is continuous on [0, ) since it is a polynomial.
We notice f (0) = 0 < a. We know that there is a number m N such that m > a which
implies f (m) = mn m > a. Thus we have f (0) < a and f (m) > a and since f is
continuous on [0, m], the intermediate value theorem states that there exists a c such that
f (c) = cn = a.
Exercise 3.2.11. Prove that a polynomial of odd degree has at least one real root.
3.3. UNIFORM CONTINUITY 29
Proof. Assume g(x) : R R is an odd degree polynomial. Then g is of the form nk=0 ak xk ,
n1an nxk N such that n mod 2 =1.n1
where ak is the k-th coecient of the polynomial We can
k
then factor g to be of the form g(x) = x (an + k=0 ak xn ). We note that limx k=0 ak xxn =
n
Exercise 3.2.12. Use the Intermediate Value Theorem to prove that f is a continuous
function on an interval [a, b] and if f (x) m for every x [a, b), then f (b) m.
Proof. Assume that m < f (b), such that f (x) m < f (b) for all x [a, b). We then know
that m = f (b) for some > 0 R. But, by properties of the real numbers, we would
also have m = f (b) < f (b) < f (b) some some , such as = 2 . But f (b) [a, b)
- contradiction: by the intermediate value theorem, since f is continuous, we know that
there exists some x such that f (x) = f (b) , and thus we require m f (b) . Hence,
f (b) m.
Exercise 3.3.2. Is the function f (x) = 1/x2 uniformly continuous on (0, +) (actual text
printed asks only about interval up to 1 )? Justify your answer.
Exercise 3.3.3. Is the function f (x) = x2 uniformly continuous on (0, +)? Justify your
answer.
No, it its not. As x we find that the distance between y, y gets bigger and bigger,
such that x, x need to be closer and closer for y, y to still be within of each other. This
means that does depend on a, so it is not uniformly continuous.
30 CHAPTER 3. CONTINUOUS FUNCTIONS
Exercise 3.3.4. Using only the definition of uniform continuity, prove that the function
x
f (x) = x+1 is uniformly continuous on [0, ).
Proof.
x y x(y + 1) + y(x + 1)
|f (x) f (y)| = | |=| |
x+1 y+1 (x + 1)(y + 1)
xy + x xy y |x y|
=| |= |x y|
(x + 1)(y + 1) (x + 1)(y + 1)
Estimate |f (x) f (y)| |x y|. Then for all > 0, = implies |x y| < = will result
in |f (x) f (y)| < .
Exercise 3.3.5. In example 3.3.8 we showed that x is uniformly continuous on [1, ).
Show that it is also uniformly continuous on [0, 1].
By theorem 3.3.4: if x is continuous on [0, 1], it is uniformly continuous there.
Exercise 3.3.6. Prove that if I and J are overlapping intervals in R(I J = and f is a
function, defined on I J, which is uniformly continuous on I and uniformly continuous on
it is also uniformly continuous on I J. Use this and the previous exercise to prove
J, then
that x is uniformly continuous on [0, +).
Proof. By assumption I J = . We shall assume that the interval I is the lower one of
the two. Then there exists an x such that x I J. Since I is uniformly continuous by
assumption, we know that [x a, x] is uniformly continous for all (x a) I. Likewise we
know that [x, x + b] is uniformly continous for all (x + b) J since J is uniformly continous
by assumption.This implies that the whole interval [x a, x + b] is uniformly continous.
To prove that x is uniformly continuous, we assume we are given some > 0. For this, we
pick = 2 . We note that | x y| | x + y|. If |x y| < = 2 , we find:
| x y|2 | x y|| x + y| = |x y| < 2
This guarantees that | x y| < , proving that x is uniformly continous on (0, ).
Exercise 3.3.7. Suppose f is not continuous on I, then since uniform continuity implies
continuity, f is not uniform continuous on I. Suppose f is continuous on I. Then f is
But a continuous function on a closed,
uniform continuous on I if f is continuouss on I.
bounded interval is bounded, but f is unbounded. This is a contradiction so f cannot be
uniformly continuous.
Exercise 3.3.8. Let f be a function defined on an interval I and suppose that there are
positive constants K and r such that
Proof. According to assumption, we find that |f (x) f (y)| K|x y|r for all x, y I.
This implies that if K|x y|r < , |f (x) f (y)| < . Thus we find that we
need to solve
K|x y| < , and find that for any given , we pick a such that = r K . Since
r
Exercise 3.3.9. Is the function f (x) = sin( x1 ) continuous on (0, 1)? Is it uniformly contin-
uous on (0, 1)? Justify your answers.
Proof. Since sin is a trigonometric function, it is continuous on its whole domain. Likewise,
sin(1/x) is continous since it is merely a composition of two elementary functions.
However, sin(1/x) is not uniformly continous. The reason for this is that the as x 0
the function oscillates between 1 and 1. Thus, a that would work at one point in the
function will can produce potentially a dierence |f (x) f (y)| = 2 for x, y suciently close
to 0. Hence, the functions is not uniformly continous.
Exercise 3.3.10. Is the function f (x) = x sin(1/x) uniformly continuous on (0, 1)? Justify
your answer.
Proof. Method 1: f (1) = sin(1). It is still uniformly continuous. limx0 f (x) = limx0 x sin(1/x) =
0. By squeeze theorem:
0 by squeeze thrm.
z }|
( ){
1
0
|{z} |x sin | |x|
x |{z}
0 0
If we define f (0) = 0, f (1) = sin(1), then f (x) becomes continuous on [0, 1]. then by
theorem 3.3.4, f is uniformly continuous on [0, 1] = f is uniformly continuous on (0, 1).
Method 2:
Then for all > 0 we have |f (x) f (y)| < if x, y (0, 2 ]. If now x, y > /3, then there
exists a > 0 such that |f (x) f (y)| < whenever |x y| < :
|f (x) f (y)| |x| + |y| < + |x| + |y x| < + + <
3 3 3
if < 3 . Then we choose = min( 3 , 0 ).
The Derivative
Exercise 4.3.1. If we apply the MVT to the intervals [1, 1], [0, 1] and [1, 0] we find
that there exist points where the derivative equals 1/2, 1, and 0.
Exercise 4.3.2. Observe that f (x) = sin x satisfies the hypotheses of theorem 4.3.9
where we let (a, b) = (, ) and so it must satisfy |f (x) f (y)| M |x y| where M is a
bound for the derivative. In this case f (x) = cos x 1 for all x, hence we may set M = 1.
Then | sin x sin y| = |f (x) f (y)| M |x y| = |x y| as desired.
Exercise 4.3.3.
Exercise 4.3.4. We know that there is some c (0, ) with f (x)f (y) = f (c)(xy) for
x, y (0, ). Then |f (x) f (y)| f (c)|x y| M |x y|. If we take the limit of both sides
as y 0, then since |f (x) f (y)| is continuous, lim |f (x) f (y)| = |f (x) f (0)| = |f (x)|.
y0
Hence |f (x)| M x.
Exercise 4.3.5.
Exercise 4.3.6. We know that f (x) = 6x2 + 6x 12 = 6(x + 2)(x 1). Hence f (x) 0
on [2, 1] and is positive elsewhere. Hence f (x) is decreasing on [2, 1] and increasing
elsewhere.
Exercise 4.3.7.
Exercise 4.3.8.
33
34 CHAPTER 4. THE DERIVATIVE
Exercise 4.3.9.
Exercise 4.3.10.
Exercise 4.3.11.
Exercise 4.3.12.
Exercise 4.3.13.
Exercise 4.3.14.
Exercise 4.3.15.
Exercise 4.3.16.
xr 1 xr 1
ln x = ,
rcr r
as desired.
Exercise 4.4.2.
Exercise 4.4.3.
Exercise 4.4.4.
Exercise 4.4.5.
ln x 1/x
lim = lim
x xr x rxr1
1
= lim = 0.
x rxr
4.4. LHOPITALS RULE 35
ln x
Exercise 4.4.7. If we write x ln x = then the limit is of the form /. By
1/x
LHopitals, the limit is equal to
1/x
lim = lim x = 0.
x0 1/x2 x0
This has numerator and denominator , hence we may apply LHopitals to get
Exercise 4.4.11.
Exercise 4.4.12.
Exercise 4.4.13.
Exercise 4.4.14.
Exercise 4.4.15.
Exercise 4.4.16.
36 CHAPTER 4. THE DERIVATIVE
Chapter 5
The Integral
and
n
k11 1 n(n 1) n1
L(f, Pn ) = = 2
= .
k=1
n n n 2 2n
Since these have the same limit as n , we know that limn U (f, Pn ) L(f, Pn ) = 0,
as desired.
Exercise 5.1.3. The base case is clearly true, so suppose that
k
k(k + 1)(2k + 1)
j2 = .
j=1
6
k+1
k(k + 1)(2k + 1) (k + 1)(k + 2)(2(k + 1) + 1)
j2 = + (k + 1)2 = ,
j=1
6 6
as desired.
37
38 CHAPTER 5. THE INTEGRAL
a (n 1)a
where Pn = {0 < < < < a} and Mk = (ka/n)2 since f (x) = x2 is increasing.
n n
Moreover, notice that the limit is a3 /3. We now find that
n ( )2
(k 1)a a
L(f, Pn ) =
k=1
n n
This gives the first inequality in the chain of inequalities. The second inequality is obtained
by the definition of supremum and infimum. The third inequality is found by noting that
Mk M .
Exercise 5.1.9. For any partition,
n
n
n
U (f, P ) = Mj (xj xj1 ) = k(b a) = mk (b a) = L(f, P ).
k=1 j=1 j=1
Hence U (f, P ) L(f, P ) = 0 < for any partition P , so by theorem 5.1.7 the integral exists.
Applying definition
n 5.1.6 and the definition of upper and lower integral, we find that the
integral equals j=1 k(b a).
5.2. EXISTENCE AND PROPERTIES OF THE INTEGRAL 39
Exercise 5.1.10. See definition 5.1.1. If P partitions [a, b] into n equal subintervals,
then xk xk1 = (a b)/n hence the dierence is
ab
n
(Mk mk ).
n k=1
and hence the derivative of the first integral is ex and of the second it is e1/x /x2 by
2 2
Exercise 5.3.6. Letting g(x) = f (x), applying (5.3.5) gives that the integral equals
b
f (b) f (a)
2 2
(f (x))2 dx.
a
1 n+1
Exercise 5.3.8. Let x t and g (t) = tn and f (t) = ln t. Then g(t) = t and
n+1
f (t) = 1/t. Both f and g are continuous on [0, ) and dierentiable on (0, ). We know
f g and gf are continuous, hence by (5.3.5) the integral equals
x x
1 n+1 1 n+1 1
t ln t t dt.
n+1 0 0 n+1 t
Simplifying this and making the necessary evaluations, the integral comes out as
1 1
xn+1 ln x xn+1 .
n+1 (n + 1)2
Exercise 5.3.10. Theorem 5.3.1 requires f (x) = x/|x| be continuous at every point in
[1, 1], but f (x) is not continuous at x = 0.
Infinite Series
Proof. We use the Ratio test to show that the sum converges on the entire real line:
1 1
(k3k )1/k = = 31 .
3 kk
41
42 CHAPTER 6. INFINITE SERIES
(1)k1
Exercise 6.4.5. Radius of convergence of k=0 k+1
(x + 2)k .
(1)k1
We have ck = k+1
. We take the k-th root:
( )1 ( )1
(1)k1 k (1)k k
(ck )1/k
= =
k+1 k+1
(1)k/k 1 1
= 1/k
= 1/k
= .
(k + 1) (k + 1) (k + 1)1/k
We then use the formula for the radius of convergence:
1 1 1
R= = 1/k
= = 1.
lim sup ck
k
lim sup (k + 1) 1
uniformly on [0, 1], and hence converges to a continuous function on this interval.
Proof. According to theorem 6.3.2, the series k=0 (1)
k+1
ak converges provided that {ak }
is a non-increasing sequence of non-negative numbers. We notice that this power series is
uniformly Cauchy, since the sequence of partial sums is uniformly Cauchy according to the
theorem. Therefore, this series converges uniformly.
Notice
thatk+1for all x [0, 1], (1)
k+1
ak xk (1)k+1 ak . Therefore, we see that the series
(1) ak xk is bounded by
k=0 k=0 (1)
k+1
ak , which by the Weierstrass M-test implies
k+1 k
that k=0 (1) ak x is uniformly convergent on [0, 1].
Exercise 6.4.12. Prove that if f (x) is the sum of a power series centered at a and with
radius of convergence R, then f is infinitely dierentiable on (a R, a + R) - that is, its
derivative of order m exists on this interval for all m N.
k=0 ck (x a) . Then, according
k
Proof. Suppose that f (x) = to theorem 6.4.12, f (x)
is dierentiable on (a R, a + R) and its derivative is f (x) = k=1 kck (x a)k1 . This
means we are left with a k 1 degree dierentiable polynomial. By induction we see that
we can then integrate this power series again. Since we take k to infinity, this means we can
take an infinite number of derivatives. Additionally, any polynomial has an infinite number
of derivatives: the kth derivative is a constant, and all others are zero.
6.5. TAYLORS FORMULA 43
Exercise 6.5.3. Use Taylors formula to estimate the error if cos(x) is approximated by
2
1 x2 on the interval [0.1, 0.1].
We estimate as follows:
x4 0.14
5 106
4! 4!
Thus we can bound the error by 5 106 .
Exercise 6.5.5. Taylors formula for f (x) = t + x with a = 0.
We begin by taking the first few derivatives of f (x) = t + x.
1
f (x) =
2 t+x
1
f (x) = 3
4 (t + x) 2
3
f (3) (x) = 5
8 (t + x) 2
15
f (4) (x) = 7
16 (t + x) 2
Then the Taylor formula for the above function for the first 5 terms, with remainder, at
a = 0 will be as follows:
f (n) (a)
f (x) = (x a)n
n=0
n!
1 1 1 3 1 15 1 4 105 1 5
t + x 3 x 2 + 5 x3 7 x + 9 x
2 t 4t 2 2 8t 2 3! 16t 2 4! 32 (c + t) 2 5!
x 1 1 3 5 7
= t + 3 x2 + 5 x
4
7 x +
5
9 x .
2 t 8t 2 16t 2 128t 2 256 (c + t) 2
Exercise 6.5.13. If g(x) = e1/x for x = 0 and g(0) = 0, show that g is infinitely dieren-
2
tiable on the entire real line but all of its derivatives at 0 are 0. Argue that this means that
g cannot be analytic at 0. Hint: Use the previous exercise to help compute the derivatives of
g at 0.
1/x2
First note that g (x) = 2 e x3 . The Taylor expansion of e1/x centered around a is
2
of the form f (k) (a) (k)
k=0 k!
(x a)k . This is a power series where ck = f k!(a) and as such is
dierentiable according to theorem 6.4.12, and infinitely so (cf. exercise 6.4.12). Likewise,
e1/x is infinitely dierentiable, and all of its dieerentials are clearly defined everywhere
2
except
n at x = 0. This is due to the fact that all n-th derivatives of g are sums of the form
1/x2 (n+k+1)
k=1 k e
a x . Therefore, analytically g(x) is not defined at 0.
1/x2
Remember however that limx0 e xn = 0. Therefore, if we define g(x) and and all of its
derivatives to be 0 at 0, g(x) is infinitely dierentiable.
Part II
Multivariable - 3220
45
Chapter 7
Exercise 7.1.6. Prove that equality holds in the Cauchy-Schwartz inequality if and only if
one of the vectors u, v is a scalar multiple of the other.
Proof. Let v be a scalar multiple of u, such that v = u. Then
Therefore, we find that |u|2 |v |2 = |u| |u|2 = |u|2 , and hence, equality holds. The proof
for the case where u is a scalar multiple of v is analogous.
47
48 CHAPTER 7. CONVERGENCE IN EUCLIDEAN SPACE
Proof. According to theorem 7.3.10, a set A Rd is closed if and only if every convergent
sequence in A converges to a point x A. Consider the sequence { n1 , n1 }. This sequence
is clearly contained in the set A = {(x, y) R2 : y > 0}. However, { n1 , n1 } (0, 0), and
(0, 0)
/ A. Therefore, the set A is not closed. Since A is not closed, A is an open set.
Exercise 7.3.4. Find the interior, closure, and boundary for the set
Exercise 7.3.6. Let A be an open set and B a closed set. If B A, prove that A\B is
open. If A B, prove that B\A is closed.
Proof. Case 1: B A. Since B is a closed set, for any sequence {xn } B, there is some
Li B such that {xn } Li . Since B A, we know that {xn }, Li A for all n, i N.
If B is a proper subset of A, then for every Li there exists some sequence {yn } such that
{yn } A\B and {yn } Li . Since Li B, we find that Li / A\B. But then not all limit
points of A\B are actually in A\B, thus showing that A\B is an open set.
Case 2: A B. Since A is an open set, Li , {xn } such that Li B, {xn } A and
{xn } Li . Consider now the set B\A. Since Li B, but Li / A, we find that Li B\A i.
Therefore, B\A is a closed set.
Hint: set m = sup{|x| : x K} and consider the open balls Bm1/n (0).
Exercise 7.4.6. Prove that the conclusion of the previous exercise also holds if we only
assume that K is a closed subset of Rd . Hint: replace K by its intersection with a suitably
large closed ball centered at y.
Exercise 7.4.9. Show that it is true that the union of any finite collection of compact sub-
sets of Rd is compact, but it is not true that the union of an infinite collection of compact
subsets is necessarily compact. Show the latter statement by finding an example of an infi-
nite union of compact sets which is not compact.
Any compact set has a finite subcoverage. Having a finite subcoverage - given an open
cover - is sucient for a set to be considered compact. The union of 2 compact sets then has
two sets of compact subcoverages. The union of subcoverages is likewise finite. Assume we
have n compact sets with finite subcoverage. The union of these n sets with an additional
compact set results in the union of a set of finite subcoverages of the union of the n sets
together with a finite subcoverage of the n + 1-th set. This is then evidently also finite.
Hence, any finite union of compact sets is also compact. However, an infinite collection is
not necessarily compact as this infinite union may lead to the existence of an infinite number
of subcoverages, which would make this union no longer compact.
Exercise 7.4.10. Prove that if A and B are compact subsets of a Rd , then A B and A B
are also compact.
Proof. Any compact set has a finite open coverage. Hence, there is a finite open cover UA of
A and an open coverage UB of B. We now consider the union UA UB . Since both coverages
are finite, the union is finite as well. This union now is a finite coverage of A B. Since
A B has a finite coverage, it is also compact.
50 CHAPTER 7. CONVERGENCE IN EUCLIDEAN SPACE
Exercise 7.5.3. A = {(x, y) R2 : 1 < |(x, y)| < 2}. This set is an open set in the form of
a disk, and as such is connected.
Exercise 7.5.5. What are the connected components of the complement of the set of inte-
gers in R?
The connected components of this set consists of all sets A = {x R : m < x < m + 1}
for all m Z.
Exercise 7.5.7. Which subsets of R are both compact and connected? Justify.
Any closed interval in R will be both compact and connected. That is any set A = {x
R : m x n} for some m, n R. It is important that this interval not contain any
holes, as it is possible to create a compact set with holes which would not be connected. For
example, the Cantor set is considered to be both compact and totally disconnected.
The function consists solely of continuous functions in R, and as such (t) is likewise
continuous (theorem 8.1.5).
x, y > 0
x, y < 0
x < 0, y > 0
or x > 0, y < 0.
The function is also continuous at zero if we approach (0, 0) along the axis from the same
quadrant.
x2 y
f (x, y) =
x4 + y 2
51
52 CHAPTER 8. FUNCTIONS ON EUCLIDEAN SPACE
show that f has a limit 0 as (x, y) (0, 0) along any straight line through the origin but
that it does not have a limit as (x, y) (0, 0) in R2 .
Assume thatwe approach along the x = 0 axis. In that case, the function has the limit 0.
Assume now we approach along some line, which has the form y = mx. Then we consider
x2 (mx) x3 m mx
lim 4 2 2
= lim 4 2 4
= lim = 0.
x,y(0,0) x + (mx ) x,y(0,0) x + m x x,y(0,0) x + m2
2
The function does not however have a limit itself. For a function to have a limit, its limit
must be path-independent. This is not the case for the function in question. Assume you
approach via the path y = mx2 . The resulting limit
x2 (mx2 ) x4 m m
lim 4 2 2
= lim 4 4
= lim
x,y(0,0) x + (mx ) x,y(0,0) x + x m x,y(0,0) 1 + m
Exercise 8.1.12. Let B1 (0) be the open unit ball in R2 . Is it true that every continuous
function f : B1 (0) R takes Cauchy sequences to Cauchy sequences?
I cant come up with a counter example; as such, I think this holds true given that we
are considering an open ball which includes the point (0, 0). If we however choose the go
towards the functions via a path of the form y = mx2 .
Exercise 8.1.14. Find a parameterized curve (t) in R2 , with parameter interval [0, ),
that begins at (1, 0), spirals inward in the counterclockwise direction, and approaches (0, 0)
as t .
We know that a continuous function maps a compact set to a compact set, and that a
continuous function also maps a connected set to a connected set. As such, I would argue
that f (K) is a compact, connected set.
Exercise 8.2.5. The image of a compact set under a continuous function is compact, hence
closed by theorem 8.2.3. Is the image of a closed set under a continuous function necessarily
closed? Prove that it is or give a counter example.
A simple counter example would be the function f : R R+ where f (x) = ex . Then the
preimage of f is closed, yet the image of f is not, as the limit point limx ex = 0 is not
part of the image set.
8.3. SEQUENCES OF FUNCTIONS 53
The sphere described above is connected. We know that any path-connected set is a
connected set. The surface of a sphere is path connected, as we can join any two points
through a continuous path function. As such, I would argue that the sphere is simply
connected, which implies that it is a connected set.
I think it is uniformly continuous on B1 (0, 0), as well as on B2 (0, 0). In both cases, we
can bound the function appropriately such that it goes to 0 without depending on x and y.
For the function to converge uniformly, both parts of the function need to individually
converge uniformly. While it is true that t/n converges uniformly on [0, 1] since it can be
bound by 1/n, it is not true that 1/(1 + nt) converges uniformly. Specifically, we find that
1 1
1 + n 1 + nt 1.
As can be seen, this bound is dependent on t, since we need n large whenever t approaches
0 to be within > 0. Hence, this part of the function fails to converge uniformly, which
results in our sequence n (t) failing to converge uniformly.
54 CHAPTER 8. FUNCTIONS ON EUCLIDEAN SPACE
Exercise 8.3.3. Does the sequence {(k 1 sin(kx), k 1 cos(ky))} converge pointwise in R2 ?
Does it converge uniformly in R2 ? Justify your answers.
{( )}
1 1
We can rewrite this sequence as ,
k csc(kx) k sec(ky)
. Clearly, the limit of this sequence
is (0, 0) for all x. As both the secant and cosecant functions arent properly bounded, I
would argue that this convergence cannot be uniform, and must be pointwise.
Exercise 8.3.8. Does the series k k
k=0 x y converge uniformly on the square
{ }
(x, y) R2 : 1 < x < 1, 1 < y < 1 ?
Justify your answer.
converge pointwise on (0, 1)? On which subsets of (0, 1) does it converge uniformly? Justify
your answers.
The series does not converge pointwise on [0, 1], since n
k=0 1 does not converge. It does
however converge pointwise on (0, 1), since k=0 k converges pointwise provided |k| < 1.
n
I would argue that the series in question converges uniformly for all rightclosed
subsets of
(0, 1), i.e. all sets of the form (0, a] where a <1. It is evident then that k=0 x n
passes the
Weierstrass M-test. We can see though that (1 x) will n
also pass this test. Assume we
pick a value x = > 0. Then we can bound the series (1 ) with some function k n
n
such that k = 1 > 1 for some > > 0. We are guaranteed such a point exists, and
hence this series passes the Weierstrass M-test on all sets of the form (0, a] with a < 1.
Exercise 8.3.12. Prove that if D is a subset of Rp and {Fn } is a sequence of functions from
D to Rq , then {Fn } fails to converge uniformly to 0 if and only if there is a sequence {xn }
in D such that the sequence of numbers {Fn (xn )} does not converge to 0.
Proof. Assume that {Fn } does not converge to 0, and that there does not exist a sequence
{xn } such that {Fn (xn )} fails to converge to 0. Then for all sequences {xi } D, we find
that {Fn (xi )} 0. This means that for all x, |Fn (x)| < for some > 0 with x D
and n N for some N . But this is the definition of convergence for{Fn }. Hence, it cannot
be that {Fn } fails to converge provided every sequence of {xn } results in {Fn (xn )} going to 0.
Now, assume that {Fn } converges to 0, but there exists a sequence {xn } D such that
Fn (xn ) does not converge to 0. But then, it is not true that |Fn (x)| < whenever x D
and n N for some N, > 0. As such, {Fn } would no longer converge.
Therefore, {Fn } converges to 0 if and only if for all sequences {xn } D, {Fn (xn )}
0.
8.4. LINEAR FUNCTIONS, MATRICES 55
Exercise 9.2.5. Find the dierential of the real-valued function f (x, y, z) = xy 2 cos(xz).
Then find the best ane approximation to f at the point (1, 1, /2).
[ ]
The dierential is df = xy 2 z sin (xz) + y 2 cos (xz) 2xy cos (xz) x2 y 2 sin (xz) .
We find the linear approximation of f at the given point as follows:
f (x, y, z) f (1, 1, /2) + fx (1, 1, /2)(x 1) + fy (1, 1, /2)(y 1) + fz (1, 1, /2)(z /2)
( ) ( )
=0+ (x 1) + 0 + z +
2 2
1
= z (x 1) = (x + 2z 2) .
2 2 2
We can see that we can now approximate f by the linear function L(x, y, z) = 21 (x + 2z 2).
Evaluating L at the point of interest, we see that L(1, 1, /2) = 0.
f (a + h) S(a + h)
lim =0
h0 h
then S(a + h) = f (a) + f (a)h.
57
58 CHAPTER 9. DIFFERENTIATION IN SEVERAL VARIABLES
f (a + h) S(a + h) = f (a + h) m a m h b.
Since we assume f (a) = S(a), we find then that b = f (a) m a. Then, we rewrite as
f (a + h) S(a + h) = f (a + h) f (a) m h. We find
f (a + h) S(a + h) f (a + h) f (a) mh
0 = lim = lim lim .
h0 h h0 h h0 h
This implies that m = f (a) and S(x) = f (a)x + f (a) m a. If we let x = a + h, we find
Proof. F is dierentiable at a if and only if there exists some function q(h) for h near 0 such
that F (a + h) F (a) = Q(h)h, and Q(0) = dF (a). Let Q be a function satisfying these
conditions. Now let B be some p q-matrix. Then the following holds:
B (F (a + h) F (a)) = B Q(h) h
B F (a + h) B F (a) = B Q(h) h
d d x d y
= +
s x s y s
d d x d y
= +
t x t y t
Exercise 9.3.8. If F (x, y) = (f1 (x, y), f2 (x, y)) is a dierentiable function from R2 to R2
and if we define G : R2 R2 by G(s, t) = F (st, s + t), find an expression for the dierential
matrix of G in terms of the partial derivatives of f1 , f2 .
Exercise 9.3.9. If (x, y, z) are the Cartesian coordinates of a point in R3 and the spherical
coordinates of the same point are r, , , then
Let u be a variable which is a dierentiable function of (x, y, z) on R3 . Find a formula for the
partial derivatives of u with respect to r, , in terms of its partial derivatives with respect
to x, y, z.
60 CHAPTER 9. DIFFERENTIATION IN SEVERAL VARIABLES
u d x d y d z
= + +
r x r y r z r
d d d
= cos() sin() + sin() sin() + cos()
x y z
u d x d y d z
= + +
x y z
d d
= r sin () sin () + r sin () cos ()
x y
u d x d y d z
= + +
x y z
d d d
= r cos () cos () + r sin () cos () r sin ()
x y z
Exercise 9.4.6. Show that the gradient at x Rp of the function g(x) = x x is the vector
2x.
Exercise 9.4.11. Find the equation of the tangent plane to the cone z = x2 + y 2 at the
point (1, 2, 5).
z z
We see that x
= 2x and y
= 2y. Then:
z z
z z0 = (x x0 ) + (y y0 )
x y
z z
z= (x x0 ) + (y y0 ) + z0
x y
z = 2x(x x0 ) + 2y(y y0 ) + z0
z = 2x(x 1) + 2y(y 2) + 5.
9.5. TAYLORS FORMULA 61
Exercise 9.4.12. Show that for each point (a, b, c) on the surface x2 + y 2 + z 2 = 1, there
is a neighborhood of (a, b, c) in which the surface may be represented as a smoothly param-
eterized 2-surface. Hence, there is a tangent plane to this surface at every point.
Each of these functions is injective and dierentiable. For any a S, we have that a is in
the image of one of these functions, so we can construct a tangent plane to thee surface at
any possible point.
Exercise 9.4.13. Find an equation for the tangent plane to the surface x2 + y 2 z 2 = 1 at
each point on the surface.
f 2f
= 2x + y, =2
x x2
f 2f
= x, =1
y y 2
f 2f
= = 1.
xy yx
62 CHAPTER 9. DIFFERENTIATION IN SEVERAL VARIABLES
We now use the known equation with these derivatives at the point a = (1, 2).
( ) ( )
f (a, b) f (a, b) 1 2 f (a, b) 2 2 f (a, b) 2 f (a, b) 2
f (a + x, b + y) f (a, b) + y + x +2 xy + y
x y 2 x2 xy y 2
1( 2 )
= 3 + (4x + y) + 2x + 2xy + y
2
1( 2 )
= 2x + 2xy + 8x + y 2 + 2y + 6 .
2
Exercise 9.5.4. Suppose U is an open convex set and f is a dierentiable real-valued
function on U . If there is a number M > 0 such that |df (x)| M for all x U , then
|f (x) f (y) M |x y|
for all x, y U .
Proof. Let x, y U . Then by Taylors theorem we have f (x) = f (y) + df (c)(x y) for some
c [y, x]. Then by algebra:
Exercise 9.5.6. Show that the following form of the Mean Value Theorem is not true: if
F : R2 R2 is a dierentiable function and a, b R2 , then there is a c on the line segment
joining a to b such that F (b) F (a) = dF (c)(b a). The problem here is that F is vector-
valued, not real valued.
If the function F is vector-valued as above, we cannot make sense of the equation. Specif-
ically, we find that dF (c)(b a) is defined as a scalar multiplication. In this case, the result
is a scalar, whereas F (b) F (a) would be a case of vector addition. A scalar cannot equal
a vector. Alternatively, we could attempt to solve this issue by redefining dF (c)(b a) as
matrix multiplication. However, then we could only have equality in the case of dF begin a
square matrix; otherwise, the dimensions wouldnt match up. As such, it appears as though
we cant make sense of the equation as is for vector valued functions.
Exercise 9.5.8. Find all points of relative maximum and relative minimum and all saddle
points for f (x, y) = 1 2x2 2xy y 2 .
( )
We begin by finding the dierential f (x, y) = f , f
x y
= (4x 2y, 2x 2y). This
is only zero at the point (0, 0). We calculate the Discriminant at this point:
D = (4)(2) (2)2 = 8 4 = 4.
2f
Since D > 0, x2
< 0 we find that we are dealing with a point of local maximum.
9.6. THE INVERSE FUNCTION THEOREM 63
Exercise 9.5.9. Find all points of relative maximum and relative minimum and all saddle
points for f (x, y) = y 3 + y 2 + x2 2xy 3y.
f 2f
= 2x 2y, =2
x x2
f 2f
= 3y 2 + 2y 2x 3, = 6y + 2
y y 2
2f 2f
= = 2
xy yx
We then find the points where f = 0. As f = (2x 2y, 3y 2 + 2y 2x 3), we find that
we have critical points at (1, 1) and (1, 1).
At the point (1, 1) we have D = 2(6(1) + 2) (2)2 = 12, and therefore we are
dealing with a saddle point here.
At the point (1, 1) we have D = 2(6 + 2) (2)2 = 4, and hence we have a point of local
minimum here.
1 ( )
x= u + 2v + u 2v
2
1 ( )
y= u + 2v u 2v
2
We calculate the dierential:
[ x ] [ ]
u
x
v
1
4 u+2v
+ 1
4 u2v
1
2 u+2v
1
2 u2v
=
y
u
y
v
1
4 u+2v
1
4 u2v
1
2 u+2v
+ 1
2 u2v
64 CHAPTER 9. DIFFERENTIATION IN SEVERAL VARIABLES
Exercise 9.6.5. Find a smooth local inverse function near (1, /2) for the function F of
Example 9.6.6.
( )
cos() r sin()
We find that dF (r, ) = . The determinant is non-zero whenever
sin() r cos(theta)
r = 0. At the point F (a) = (0, 1). We know by the inverse function theorem that dF 1 (b) =
(dF (a))1 . We now use this information and find the inverse dierential:
[ ] [ ]
1 1 cos() sin() 0 1
dF (b) = (dF (a)) = = .
r1 sin() r1 cos() 1 0
( )
We find that the inverse function is F 1 (x, y) = x2 + y 2 , tan1 (y/x) .
Exercise 9.6.8. Show by example that the result of the previous problem is not true if
U is only assumed to be connected, rather than convex. Hint: Try the function F (x, y) =
(x2 y 2 , 2xy) on R2 \{0}.
[ ]
2x 2y
Consider the function f (x, y) = (x y , 2xy). Then df =
2 2
. Then the
2y 2x
determinant of df is greater than 0, namely it is 4(x2 + y 2 ). Consider now U = R2 \{(0, 0)},
which is a connected but not a convex set. Here, df is non-singular but f is not injective
since f (1, 1) = f (1, 1).
Exercise 9.6.10. Show that the condition that dF (a) be non-singular is necessary in the
Inverse Function Theorem by showing that if a function F from a neighborhood of a in RP
to Rp is dierentiable at a and has an inverse function at a which is dierentiable at F (a),
then dF (a) is non-singular.
We are trying to show that if F is locally invertible, its dierential matrix is also invertible,
i.e. non-singular.
9.7. THE IMPLICIT FUNCTION THEOREM 65
f1 (x, y, u, v) = u2 + v 2 + x2 + y 2 ,
f2 (x, y, u, v) = xu + yv + x y.
At which points (x, y, u, v) is this matrix non-singular?
67
68 CHAPTER 10. INTEGRATION IN SEVERAL VARIABLES
Chapter 11
Vector Calculus
The solution is the function (t) = u + t(v u) where u = (1, 2), v = (3, 0), and t [0, 1].
We rewrite:
[ ] ([ ] [ ])
1 3 1
(t) = +t
2 0 2
[ ] [ ] [ ]
1 2 1 + 2t
= +t =
2 2 2 2t
Exercise 11.1.2. We find the derivative (t) = (cos(t) t sin(t), sin(t) + t cos(t)). We
proceed to find the norm
| (t)| = (cos(t) t sin(t))2 + (sin(t) + t cos(t))2 = 1 + t2 .
4
To find the length, we solve the integral 0
1 + t2 dt. We utilize a trig substitution, lettting
t = tan().
4 tan1 (4)
2
1 + t dt = sec()d
0 0
1
= ln (sec() + tan())|tan (4
)
( ( 1 )
0
) ( )
= ln sec tan (4) + 4 = ln 4 + 16 2 + 1
The last step is justified by letting sec (tan1 (4)) = sec(), and solving tan2 ()+1 = sec2 ().
69
70 CHAPTER 11. VECTOR CALCULUS