Real Analysis and Multivariable Calculus: Graduate Level

Problems and Solutions

Igor Yanovsky

1
Real Analysis and Multivariable Calculus Igor Yanovsky, 2005 2

Disclaimer: This handbook is intended to assist graduate students with qualifying

examination preparation. Please be aware, however, that the handbook might contain,
and almost certainly contains, typos as well as incorrect or inaccurate solutions. I can
not be made responsible for any inaccuracies contained in this handbook.
Real Analysis and Multivariable Calculus Igor Yanovsky, 2005 3

Contents
1 Countability 5

2 Unions, Intersections, and Topology of Sets 7

3 Sequences and Series 9

4 Notes 13
4.1 Least Upper Bound Property . . . . . . . . . . . . . . . . . . . . . . . . 13

5 Completeness 14

6 Compactness 16

7 Continuity 17
7.1 Continuity and Compactness . . . . . . . . . . . . . . . . . . . . . . . . 18

8 Sequences and Series of Functions 19

8.1 Pointwise and Uniform Convergence . . . . . . . . . . . . . . . . . . . . 19
8.2 Normed Vector Spaces . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19
8.3 Equicontinuity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21
8.3.1 Arzela-Ascoli Theorem . . . . . . . . . . . . . . . . . . . . . . . . 21

9 Connectedness 21
9.1 Relative Topology . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21
9.2 Connectedness . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21
9.3 Path Connectedness . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 23

10 Baire Category Theorem 24

11 Integration 26
11.1 Riemann Integral . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 26
11.2 Existence of Riemann Integral . . . . . . . . . . . . . . . . . . . . . . . . 27
11.3 Fundamental Theorem of Calculus . . . . . . . . . . . . . . . . . . . . . 27

12 Differentiation 30
12.1 R → R . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 30
12.1.1 The Derivative of a Real Function . . . . . . . . . . . . . . . . . 30
12.1.2 Rolle’s Theorem . . . . . . . . . . . . . . . . . . . . . . . . . . . 30
12.1.3 Mean Value Theorem . . . . . . . . . . . . . . . . . . . . . . . . 30
12.2 R → Rm . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 31
12.3 Rn → Rm . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 31
12.3.1 Chain Rule . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 34
12.3.2 Mean Value Theorem . . . . . . . . . . . . . . . . . . . . . . . . 35
∂ ∂f ∂ ∂f
12.3.3 ∂x ( ∂y ) = ∂y ( ∂x ) . . . . . . . . . . . . . . . . . . . . . . . . . . . 36
12.4 Taylor’s Theorem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 37
12.5 Lagrange Multipliers . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 40
Real Analysis and Multivariable Calculus Igor Yanovsky, 2005 4

13 Successive Approximations and Implicit Functions 41

13.1 Contraction Mappings . . . . . . . . . . . . . . . . . . . . . . . . . . . . 41
13.2 Inverse Function Theorem . . . . . . . . . . . . . . . . . . . . . . . . . . 41
13.3 Implicit Function Theorem . . . . . . . . . . . . . . . . . . . . . . . . . 44
13.4 Differentiation Under Integral Sign . . . . . . . . . . . . . . . . . . . . . 46
Real Analysis and Multivariable Calculus Igor Yanovsky, 2005 5

1 Countability
The number of elements in S is the cardinality of S.
S and T have the same cardinality (S ' T ) if there exists a bijection f : S → T .
card S ≤ card T if ∃ injective1 f : S → T .
card S ≥ card T if ∃ surjective2 f : S → T .
S is countable if S is finite, or S ' N.

Theorem. S, T 6= φ. ∃ injection f : S ,→ T ⇔ ∃ surjection g : T ³ S.

Theorem. Q is countable.

Proof. Need to show that there is a bijection f : N → Q.

Since N ⊆ Q, card N ≤ card Q, and therefore, ∃f that is injective.
To show card N ≥ card Q, construct the following map. The set of all rational numbers
can be displayed in a grid with rows i = 1, 2, 3, . . . and columns j = 0, −1, 1, −2, 2, −3, 3, . . ..
Each aij , the ij’th entry in a table, would be represented as ji . Starting from a11 , and
assigning it n = 1, move from each subsequent row diagonally left-down, updating n.
This would give a map g : N → Q, which will count all fractions, some of them more
than once. Therefore, card N ≥ card Q, and so g is surjective. Thus, card N = card Q,
and Q is countable.

Theorem. R is not countable.

Proof. It is enough to prove that [0, 1) ⊂ R is not countable. Suppose that the set of all
real numbers between 0 and 1 is countable. Then we can list the decimal representations
of these numbers (use the infinite expansions) as follows:
a1 = 0.a11 a12 a13 . . . a1n . . .
a2 = 0.a21 a22 a23 . . . a2n . . .
a3 = 0.a31 a32 a33 . . . a3n . . .
and so on. We derive a contradiction by showing there is a number x between 0 and
1 that is not on the list. For each positive integer j, we will choose jth digit after the
decimal to be different than ajj :
x = 0.x1 x2 x3 . . . xn . . ., where xj = 1 if ajj 6= 1, and xj = 2 if ajj = 1.
For each integer j, x differs in the jth position from the jth number on the list, and
therefore cannot be that number. Therefore, x cannot be on the list. This means the
list as we chose is not a bijection, and so the set of all real numbers is uncountable.
(Need to worry about not allowing 9 tails in decimal expansion: 0.399 . . . = 0.400 . . .).

1
injective = 1-1: f (s1 ) = f (s2 ) ⇒ s1 = s2 .
2
surjective = onto: ∀t ∈ T, ∃s ∈ S, s.t. f (s) = t.
Real Analysis and Multivariable Calculus Igor Yanovsky, 2005 6

Problem (F’01, #4). The set of all sequences whose elements are the digits 0 and 1
is not countable.

Let S be the set of all binary sequences. We want to show that there does not exist
a one-to-one mapping from the set N onto the set S.

Proof. 1) Let A be a countable subset of S, and let A consist of the sequences s1 , s2 , . . ..

We construct the sequence s as follows. If the nth digit in sn is 1, let the nth digit of
s be 0, and vice versa. Then the sequence s differs from every member of A in at least
/ A. However, s ∈ S, so that A is a proper3 subset of S.
one place; thus s ∈
Thus, every countable subset of S is a proper subset of S, and therefore, S is not
countable.

Proof. 2) Suppose there exists a f : N → S that is injective. We can always exhibit an

injective map f : N → S by always picking a different sequence from the set of sequences
that are already listed. (One way to do that is to choose a binary representation for
each n ∈ N).
Suppose f : N → S is surjective. Then, all sequences in S could be listed as
s1 , s2 , . . .. We construct the sequence s as follows. If the nth digit in sn is 1, let the
nth digit of s be 0, and vice versa. Then the sequence s differs from every member of
the list. Therefore, s is not on the list, and our assumption about f being surjective is
false. Thus, there does not exist f : N → S surjective.

Theorem. card (A) < card (P (A))4 .

Proof. card (A) ≤ card (P (A)), since A can be injectively mapped to the set of one-
element sets of A, which is a subset of P (A).
We need to show there is no onto map between A and P (A). So we would like to find a
thing in P (A) which is not reached by f . In other words, we want to describe a subset
of A which cannot be of the form f (a) for any a ∈ A.
Suppose |A| = |P (A)|. Then there is a 1-1 correspondence f : A → P (A). We
obtain a contradiction to the fact that f is onto by exhibiting a subset X of A such
that X 6= f (a) for any a ∈ A.
For every a ∈ A, either a ∈ f (a), or a ∈/ f (a). Let X = {a ∈ A : a ∈
/ f (a)}.
Consider a ∈ A. If a ∈ f (a), then a ∈ / X, so f (a) 6= X.
If a ∈
/ f (a), than a ∈ X, so f (a) 6= X. Therefore, X 6= f (a), ∀a ∈ A, a contradiction.
Therefore, card (A) < card (P (A)).

Theorem. Suppose that f : [0, 1] → R is an increasing function. Show that f can have
at most a countable number of discontinuities.

Proof. Let E = {x ∈ [0, 1] : f is discontinuous at x}. Given any x ∈ E, we know that

limt→x− f (t) < limt→x+ f (t)
and, using this fact, we choose r(x) ∈ Q such that limt→x− f (t) < r(x) < limt→x+ f (t).
⇒ We have defined a 1 − 1 function r : E → Q.
3
A is a proper subset of B if every element of A is an element of B, and there is an element of B
which is not in A.
4
Power set of a set S is the set whose elements are all possible subsets of S, i.e.
S = {1, 2}, P (S) = 2S = {φ, {1}, {2}, {1, 2}}. |P (A)| = 2n , if |A| = n.
Real Analysis and Multivariable Calculus Igor Yanovsky, 2005 7

2 Unions, Intersections, and Topology of Sets

Theorem. Let Eα be a collection of sets. Then
[ \
( Eα )c = (Eαc ).
α α
S T S
Proof. Let A = ( Eα )c and B = ( Eαc ). If x T∈ A, then x ∈/ Eα , hence x ∈ / Eα for
any α, hence x ∈ Eαc for every α, so that x ∈ Eαc . Thus A ⊂ B. S
Conversely, ifSx ∈ B, then x ∈ Eαc for every α, hence x ∈
/ Eα for any α, hence x ∈
/ Eα ,
so that x ∈ ( Eα )c . Thus B ⊂ A.

Theorem. S
a) For any collection Gα of open sets, Tα Gα is open.
b) For any collection Fα of closed sets, α Fα is closed. T
c) For any finite collection G1 , . . . , Gn of open sets, Sni=1 Gi is open.
d) For any finite collection F1 , . . . , Fn of closed sets, ni=1 Fi is closed.
S
Proof. a) Put G = α Gα . If x ∈ G, then x ∈ Gα for some α. Since x is an interior
point of Gα , x is also an interior point of G, and G is open.
b) By theorem above,
[ \ \ [
( Fα )c = (Fαc ) ⇒ ( Fα )c = (Fαc ), (2.1)
α α α α
T
and Fαc is open. Hence a) implies that the right equation of (2.1) is open so that α Fα
is closed. T
c) Put H = ni=1 Gi . For any x ∈ H, there exists neighborhoods Nri of x, such that
Nri ⊂ Gi (i = 1, . . . , n). Put r = min(r1 , . . . , rn ). Then Nr (x) ⊂ Gi for i = 1, . . . , n, so
that Nr (x) ⊂ H, and H is open. S T
d) By taking complements, d) follows from c): ( ni=1 Fi )c = ni=1 (Fic ).

Theorem.
T S ⊂ M ⊆ X. S open relative to M ⇔ ∃ open U ⊂ X such that
S = U M.

Proof. ⇒ S open relative to M . To each x ∈ S, ∃rx , |x − y| < rx , y ∈ M ⇒ y ∈ S.

Define
[
U= Nrx (x) ⇒ U ⊂ X open.
x∈S
T T
It is clear T
that S ⊂ U M . By T
our choice of Nrx (x), we have Nrx (x) M ⊂ S, ∀x ∈ S
so that U M ⊂ S ⇒ S = U MT.
⇐ If U is open T in X and S = U M , every x ∈ S has a neighborhood Nrx (x) ⊂ U .
Then Nrx (x) M ⊂ S so that S is open relative to M .
Real Analysis and Multivariable Calculus Igor Yanovsky, 2005 8

Theorem. K ⊂ Y ⊂ X. K compact relative to X ⇔ K compact relative to Y .

Proof. ⇒ Suppose K is compactSrelative to X, and let {Vα } be a collection of sets,

open relative to Y , such
T that K ⊂ α Vα . By the above theorem, ∃Uα open relative to
X, such that Vα = Y Uα , ∀α; and since K is compact relative to X, we have
[ [
K ⊂ Uα1 · · · Uαn }
[ [
Since K ⊂ Y ⇒ K ⊂ Vα1 ··· Vαn . ~
⇐ Suppose K is compact relative T to Y . Let {Uα } be a collection of open subsets of X
which covers K, and put Vα = Y Uα . Then ~ will hold for some choice of α1 , . . . , αn ;
and since Vα ⊂ Uα , ~ implies }.

Theorem. Compact subsets of metric spaces are closed.

Proof. Let K ⊂ X be a compact subset. We prove K c is open.

Suppose x ∈ X, x ∈ / K. If y ∈ K, let Vy and Wy be neighborhoods S of xS and y,
respectively, of radius less than 12 d(x, y). Since K is compact, K ⊂ Wy1 · · · Wyn =
W. T T
If V = Vy1 · · · Vyn , then V is a neighborhood of x which does not intersect W .
Hence V ⊂ K c , so that x is an interior point of K c .

Theorem. Closed subsets of compact sets are compact.

Proof. Suppose F ⊂ K ⊂ X, F is closed (relative to X), and K is compact. Let {Vα }

be an open cover of F . If F c is adjoined to {Vα }, we obtain an open cover Ω of K.
Since K is compact, there is finite subcollection Φ of Ω which covers K, and hence F .
If F c is a member of Φ, we may remove it from Φ and still retain an open cover of F .
We have thus shown that a finite subcollection of {Vα } covers F .
T
Corollary. If F is closed and K is compact, then F K is compact.
T T
Proof. K is closed (by a theorem Tabove), and thus F K is closed. Since F K ⊂ K,
the above theorem shows that F K is compact.

Theorem. If {Kα } is a collection of compact subsets of a metric space

T X such that the
intersection of every finite subcollection of {Kα } is nonempty, then Kα is nonempty.

Proof. Fix a member K1 of {Kα }. Assume that no point of K1 belongs to every

Kα . SThenS the sets Kαc form an open cover
T of KT1 ; and
T since K1 is compact, K1 ⊂
Kαc 1 · · · Kαc n . But this means that K1 Kα1 · · · Kαn is empty. ⇒ contra-
diction.
Real Analysis and Multivariable Calculus Igor Yanovsky, 2005 9

3 Sequences and Series

A sequence {pn } converges to p ∈ X if:
∀² > 0, ∃N : ∀n ≥ N |pn − p| < ² ⇔ limn→∞ pn = p.
A sequence {pn } is Cauchy if:
∀² > 0, ∃N : ∀n, m ≥ N |pn − pm | < ².
Cauchy criterion: P A sequence converges in Rk ⇔ itP is a Cauchy sequence.
m
Cauchy criterion:
P a n converges ⇔ ∀² > 0, ∃N
P : | k=n ak | ≤ ² if m ≥ n ≥ N .
The series an is said to converge absolutely if |an | converges.
P
Problem (S’03, #2). If a1 , a2 , a3 , . . . is a sequence of real numbers with ∞ j=1 |aj | <
PN P P
∞, then limN →∞ j=1 aj exists. ( an converges absolutely ⇒ an converges).
P
Proof. If sn = nj=1 aj is a partial sum, then for m ≤ n we have
n
X m
X n
X n
X
|sn − sm | = | aj − aj | = | aj | ≤ |aj |
j=1 j=1 j=m j=m
P P
Since |aj | converges, given ² > 0, ∃N , s.t. m, n ≥ N (m ≤ n), then nj=m |aj | < ².
P
Thus {sn } is a Cauchy sequence in R, and converges. ⇒ limN →∞ N j=1 aj exists.

Problem (F’01, #2). Let N denote the positive integers, let an = (−1)n n1 , and let
α be any real number. Prove there is a one-to-one and onto mapping σ : N → N such
that
∞
X
aσ(n) = α.
n=1

Proof. an = (−1)n n1 ⇒ a1 = −1, a2 = 12 , a3 = − 13 , a4 = 14 , . . ..

a
P2n > 0, a2n−1 < 0, P n = 1, 2, . . . P
1 P∞ 1
(positive terms) = ∞ n=1 a2n =
∞ 1
n=1 2n = 2 n=1 n diverges, and limn→∞ 2n =
1

0.
P P∞ P
(negative terms) = n=1 a2n−1 = − ∞ 1
n=1 2n−1 diverges by comparison with
P∞ 1 1
− n=1 2n , and limn→∞ 2n−1 = 0.
Claim:
P∞ ∀α ∈ R, there is a one-to-one and onto mapping σ : N → N such that
n=1 aσ(n) = α, where σ(n) is the rearrangement of indices of the original series.
Given α, choose positive terms in sequential order until their sum exceeds α. At this
switch point, choose negative terms until their sum is less than α. Repeat the process.
Note: This process never stops because no matter how many positive and negative
terms are taken, there are still infinitely many both positive and negative terms left;
the sum of positive terms is ∞, the sum of negative terms is −∞. P
Let the sum of terms at the N th step be denoted by SN , SN = N n=1 aσ(n) . At switch
point, |α − SN | is bounded by the size of the term added:

|α − SN | → 0, N →∞

All terms {an } will eventually be added to the£ sum (σ : N ³ N is surjective (onto)) ¤ at
different steps (σ : N ,→ N is injective (1-1)). σ : {1, 2, 3, . . .} → {n1 , n2 , n3 , . . .} .
Real Analysis and Multivariable Calculus Igor Yanovsky, 2005 10

P p
Root Test. P Given an , put α = lim supn→∞ n |an |. Then
a) if α < 1, P an converges;
b) if α > 1, an diverges;
c) if α = 1, the test gives no information.
P
Ratio Test. The series an
a) converges if lim supn→∞ | an+1an | < 1,
b) diverges if | an+1
an | ≥ 1 for all n ≥ n0 , where n0 is some fixed integer.

Alternating Series. Suppose

(a) |a1 | ≥ |a2 | ≥ . . . ;
(b) a2m−1 ≥ 0, a2m ≤ 0, m = 1, 2, 3, . . . ;
(c) limPn→ an = 0.
Then an converges.

Geometric Series. |x| < 1 :

∞
X 1
xk = .
1−x
k=0

P roof : Sn = 1 + x + x2 + · · · + xn ~
2 3 n+1
xSn = x + x + x + · · · + x }
n+1
~ − } = (1 − x)Sn = 1 − x
Xn
1 − xn+1
⇒ Sn = xk =
1−x
k=0
Real Analysis and Multivariable Calculus Igor Yanovsky, 2005 11

Problem (F’02, #4). By integrating the series

1
= 1 − x2 + x4 − x6 + x8 · · ·
1 + x2
π 1 1 1 1
prove that = 1− + − + ··· .
4 3 5 7 9
Justify carefully all the steps (especially taking the limit as x → 1 from below).

Proof. Geometric Series:

1
= 1 − x2 + x4 − x6 + x8 · · · ; |x| < 1.
1 + x2
Z
dx x3 x5 x7 x9
= tan−1 (x) = x − + − + ··· .
1 + x2 3 5 7 9
Z 1
dx π π
2
= [tan−1 (x)]10 = tan−1 1 − tan−1 0 = − 0 = .
0 1+x 4 4
Z 1 Z 1−²
dx dx
2
= lim
0 1+x ²→0 0 1 + x2
S
Z 1−² Z 1−² z }| { Z 1−²
dx 2 4
= [1 − x + x − · · · ] dx = lim SN (x)dx
0 1 + x2 0 | {z } 0
converges uniformly for |x|≤1−²
Z b Z b
||fn − f ||∞ → 0 ⇒ fn (x)dx → f (x)dx
a a
¯Z b Z b ¯ Z b Z b
¯ ¯
¯ f (x)dx − fn (x)dx¯¯ ≤ |f (x) − fn (x)|dx ≤ ||f − fn ||∞ dx = (b − a)||f − fn ||∞ .
¯
a a a a
N
X
S is a uniform limit of SN = (−1)n x2n .
n=1
¯X∞ ¯ ∞
X ∞
X (1 − ²)2(N +1)
¯ ¯
|S(x) − SN (x)| = ¯ (−1)n x2n ¯, ∀x ∈ [0, 1 − ²] ≤ |x|2n ≤ (1 − ²)2n = → 0.
(1 − ²)2
N +1 N +1 N +1
X (−1)n x2n+1
Above calculations show tan−1 x = , |x| < 1.
2n + 1
Alternating series test ⇒ right side converges.
π X (−1)n 1 1 1 1
= = 1 − + − + ··· .
4 2n + 1 3 5 7 9
Real Analysis and Multivariable Calculus Igor Yanovsky, 2005 12

Logarithm. |x| < 1 :

∞
X
1 2 3
= 1 − x + x − x + ··· = (−1)n xn
1+x
n=0
Z x ∞
X
dt x2 x3 x4 xn
log(1 + x) = = x− + − + ··· = (−1)n+1
0 1+t 2 3 4 n
n=1

log(1 + x) is not
P∞ valid for |x| > 1, or x < −1. Claim: If x = 1, the series converges
to log 2. Proof: (−1)n−1 xn is uniformly convergent for x ∈ [0, 1], since the sum
n=1 n
of any number of consecutive terms starting with the nth has absolute value at most
xn 1
n ≤ n , since for 0 < x < 1 we have alternating series.

Binomial Series. |x| < 1 :

∞
X µ ¶ µ ¶
α α(α − 1) · · · (α − n + 1) n α 2 α
(1 + x) = 1+ x = 1 + αx + x + ··· + xn + · · ·
n! 2 n
n=1
α(α − 1) 2 α(α − 1)(α − 2) 3
= 1 + αx + x + x + ···
2! 3!
µ ¶n
1
Problem (F’03, #3). The sequence a1 , a2 , . . . with an = 1 + n converges as
n → ∞.

Proof. By Binomial Series Theorem with α = n, x = n1 , we get:

µ ¶
1 n 1 n(n − 1) 1 n(n − 1)(n − 2) 1 n(n − 1)(n − 2) · · · 1 1
1+ = 1+n· + · 2+ · 3 + ··· + · n
n n 2! n 3! n n! n
n
X 1
1 1 1
≤ 1 + 1 + + + ··· + = → e, as n → ∞.
2! 3! n! k!
k=0
Real Analysis and Multivariable Calculus Igor Yanovsky, 2005 13

4 Notes
4.1 Least Upper Bound Property
An ordered set S is said to have the least upper bound5 property if:
E ⊂ S, E is not empty, and E is bounded above, then sup E exists in S.
Completeness axiom: If E is a nonempty subset of R that is bounded above, then
E has a least upper bound.

Problem (F’02, #2). Show why the Least Upper Bound Property (every set bounded
above has a least upper bound) implies the Cauchy Completeness Property (every Cauchy
sequence has a limit) of the real numbers.

Proof. Suppose {xn } Cauchy. The problem is to show that {xn } converges.
We first show that {xn } is bounded. Fix ² > 0 and let N be such that |xn − xm | < ²
if n, m > N . Then for any fixed n > N , the entire sequence is contained in the closed
ball of center xn and radius max{d(xn , x1 ), d(xn , x2 ), . . . , d(xn , xN ), ²}. Thus {xn } is
bounded.
Define zn = sup{xk }k≥n . Since {xn } is bounded, each zn is a finite real number and is
bounded above in absolute value by M . If m > n, then zm is obtained by taking the
sup of a smaller set than is zn ; hence {zn } is decreasing. By the greatest lower bound
property, Z = {zn |n ∈ N} has an infimum. Let x = inf Z. We claim that xn → x.6
For each ² > 0 there is a corresponding integer N such that x ≤ zN ≤ x+². Since {xn } is
Cauchy, by taking a larger N if necessary, we know that k ≥ N ⇒ xk ∈ [xN −², xN +²].
It follows that zN ∈ [xN − ², xN + ²]. Hence for k ≥ N ,
|xk − x| ≤ |xk − xN | + |xN − zN | + |zN − x| ≤ ² + ² + ² = 3².

5
least upper bound of E ≡ sup E.
6
Idea: Since {xn } is Cauchy, the terms of this sequence would approach one another. {zn } also
approaches {xn }. Since zn → x, {zn } approaches x. It follows that {xn } approaches x.
Real Analysis and Multivariable Calculus Igor Yanovsky, 2005 14

5 Completeness
A metric space X is complete if every Cauchy sequence of elements of X converges to
an element of X.

Lemma. A convergent sequence is a Cauchy sequence.

Proof. xn → x means ∀² > 0, ∃N , such that ∀n ≥ N, |x − xn | < ². Hence

|xn − xm | ≤ |xn − x| + |x − xm | ≤ 2²

when n, m ≥ N . Thus, xn is a Cauchy sequence.

Lemma. If xn is Cauchy, then xn is bounded.

Proof. If the sequence is x1 , x2 , x3 , . . ., ² > 0 and N is such that |xn − xm | < ² if

n, m > N , then for any fixed m > N , the entire sequence is contained in the closed ball
of center xm and radius max{d(xm , x1 ), d(xm , x2 ), . . . , d(xm , xN ), ²}.

Lemma. If xn is Cauchy and xnk → x, then xn → x.

Proof. Let ² > 0. Since xn is Cauchy ⇒ choose k > 0 so large that |xn − xm | < ²
whenever n, m ≥ k. Since xnk → x ⇒ choose l > 0 so large (i.e. nl large) that
|xnj − x| < ² whenever j ≥ l. Set N = max(k, nl ). If m, nj > N , then
|xm − x| ≤ |xm − xnj | + |xnj − x| < ² + ² = 2².

Theorem. [a, b] is complete.

Proof. Let xn be a Cauchy sequence in [a, b]. Let xnk be a monotone subsequence.
Since a ≤ xnk ≤ b, xnk converges (by the Least Upper Bound). ⇒ xnk → c. Since
[a, b] is closed, c ∈ [a, b]. ⇒ Any xn that is Cauchy in [a, b], converges in [a, b]. ⇒
[a, b] is complete.

The above theorem is a specific case of the following Lemma:

Theorem. Let xn be a Cauchy sequence in a compact metric space X. Then xn con-

verges to some point of X.

Proof. Since X is (sequentially) compact, then for any sequence xn ∈ X, there is a

subsequence xnk → c, c ∈ X. Using the above theorem (xn Cauchy and xnk → c ⇒
xn → c), we see that xn → c ∈ X.

Theorem. R is complete.

Proof. Let xn be a Cauchy sequence in R. xn is bounded (by the Lemma above). ⇒

{xn } ⊆ [a, b], and see above.

A direct consequence of the above theorem is the following: In Rk , every Cauchy

sequence converges.

Theorem. [0, 1) is not complete.

Real Analysis and Multivariable Calculus Igor Yanovsky, 2005 15

Bolzano-Weierstrass. Every bounded, infinite subset S ⊂ R has a limit point.

Proof. If I0 is a closed interval containing S, denote by I1 one of the closed half-intervals

of I0 that contains infinitely many points of S. Continuing in this way, we define a
nestedTsequence of intervals {In }, each of which contains infinitely many points of S.
If c = ∞n=1 In , then it is clear that c is a limit point of S.

Lemma. Every bounded sequence of R has a convergent subsequence.

Proof. If a sequence xn contains only finitely many distinct points, the conclusion is
trivial and obvious. Otherwise we are dealing with a bounded infinite set, to which
the Bolzano-Weierstrass theorem applies, giving us a limit point x. If, for each integer
k ≥ 1, xnk is a point of the sequence such that |xnk − x| ≤ 1/k, then it is clear that
xnk is a convergent subsequence.

Theorem. A closed subspace Y of a complete metrix space X is complete.

Proof. Let yn be a Cauchy sequence in Y . Then yn is also a Cauchy sequence in X.

Since X is complete, ∃ x ∈ X such that yn → x in X. Since Y is closed, x ∈ Y .
Consequently, yn → x in Y .

Theorem. A complete subspace Y of a metric space X is closed in X.

Proof. Suppose x ∈ X is a limit point of Y . ∃yn in Y that converges to x. yn is a

Cauchy sequence in X; hence it is also a Cauchy sequence in Y . Since Y is complete,
yn → y ∈ Y . Since limits of sequences are unique, y = x and x belongs to Y . Hence Y
is closed.

Theorem. V is a normed space. If we have the following implication

∞
X ∞
X
||vn || < ∞ ⇒ vn < ∞,
n=1 n=1

then V is complete (V is a Banach space).

Proof. Say vn is Cauchy. Choose nk such that m, n ≥ nk ⇒ ||vm − vn || < 2−k . We

may assume that n1 < n2P < . . .. P∞
−k ∞
||vnk+1 −vnk || < 2 ⇒ k=1 ||vnk+1 −vnk || < ∞. It follows that n=1 (vnk+1 −vnk ) <
∞.
K
X
s ← sK = (vnk+1 −vnk ) = vnK+1 − vn1 ⇒ vnk+1 converges ⇒ vn converges.
k=1
Real Analysis and Multivariable Calculus Igor Yanovsky, 2005 16

6 Compactness
M is (sequentially) compact if for any sequence xn ∈ M , there is a subsequence
xnk → c, c ∈ M . S
M is (topologically) compact if any open cover of M , M ⊆ Gα , Gα open, contains
a finite subcover.
Problem (W’02, #2). [a, b] is compact.
Proof. Let xn be a sequence in [a, b]. Let xnk be a monotone subsequence ⇒ a ≤
xnk ≤ b ⇒ xnk → c. Since [a, b] is closed, c ∈ [a, b]. ⇒ [a, b] is (sequentially)
compact.
Lemma. If M is compact, every open cover of M has a countable subcover.
Theorem. If M is sequentially compact, then it is topologically compact.
S S
Proof. Say that M ⊆ G1 G2S · · · has a countable subcover. Need to show that there
is a finite subcover, i.e. M ⊆ nk=1 Gk for some n.
Suppose that fails for every n; then for every n = 1, 2, . . ., there would exist
n
[
xn ∈ M \ Gk .
k=1

That sequence would have a convergent subsequence {xnk }. Let x be its limit, xnk → x.
Then x would be contained in Gm for some m, and thus
xnk ∈ Gm
S S
for all nk sufficiently large, which is impossible for nk > m (since xnk ∈ M \G1 ··· Gnk ).
We have reached a contradiction. So there must be a finite subcovering.
Problem (S’02, #3). If M is topologically compact, then it is sequentially com-
pact.
Proof. Let xn ∈ M and E be the range of {xn }. If E is finite, then there is x ∈ E and
a sequence {ni }, with n1 < n2 < . . ., such that
xn1 = xn2 = · · · = x
The subsequence {xni } converges to x.
If E is infinite, E has a limit point x in M (as an infinite subset of a compact
set). Every neighborhood of x contains infinitely many points of M . For each k, B 1 (x)
k
contains infinitely many xn ’s. Select one and call it xnk , such that, nk > nk−1 > . . ..
We have a subsequence {xnk } so that d(x, xnk ) < k1 → 0. ⇒ xnk → x.
Problem (F’02, #1). LetTK be a compact subset and F be a closed subset in the
metric space X. Suppose K F = φ. Prove that
0 < inf{d(x, y) : x ∈ K, y ∈ F }.
Proof. Given x ∈ K, x ∈ / F , dx = d(x, F )T> 0. Then, the ball centered at that x with
radius dx /2, i.e. B dx (x), satisfies B dx (x) F = φ. Since x was taken arbitrary, this is
2 2
true ∀x
S ∈ K, x ∈
/ F .
K ⊂ x∈K B dx (x). Since K is compact, ∃x1 , . . . , xn ∈ K, n < ∞, such that
S 2 T
K ⊂ nk=1 B dx (xk ), and B dxk (xk ) F = φ. Since mink {dxk } > 0, we have
2 2
0 < inf{d(x, y) : x ∈ K, y ∈ F }.
Real Analysis and Multivariable Calculus Igor Yanovsky, 2005 17

7 Continuity
Limits of Functions: limx→p f (x) = q if:
∀² > 0, ∃ δ such that ∀x ∈ E 0 < |x − p| < δ ⇒ |f (x) − q| < ².
A function f is continuous at p: limx→p f (x) = f (p) if:
∀² > 0, ∃ δ such that ∀x ∈ E |x − p| < δ ⇒ |f (x) − f (p)| < ².
Negation: f is not continuous at p if:
∃ ² > 0, ∀δ such that ∃x ∈ E |x − p| < δ ⇒ |f (x) − f (p)| > ².
f is uniformly continuous on X if:
∀² > 0, ∃ δ such that ∀x, z ∈ X |x − z| < δ ⇒ |f (x) − f (z)| < ².
Negation: f is not uniformly continuous on X if:
∃ ² > 0, ∀δ such that ∃x, z ∈ X |x − z| < δ ⇒ |f (x) − f (z)| > ².
Examples: f (x) = x1 on (0, 1] and f (x) = x2 on [1, ∞) are not uniformly continuous.
Theorem. f : X → Y is continuous ⇔ f −1 (V ) is open in X for every open set V in
Y.
Proof. 7 ⇒ Suppose f is continuous on X. Let V be an open set in Y . We have
to show that every point of f −1 (V ) is an interior point of f −1 (V ). Let x ∈ f −1 (V ).
Choose ² such that B² (f (x)) ⊂ V . Since f is continuous8 , ∃ δ > 0 such that f (Bδ (x)) ⊂
B² (f (x)) ⊂ V . Hence, Bδ (x) ⊆ f −1 (V ). Since f −1 (V ) contains an open ball about
each of its points, f −1 (V ) is open.
⇐ Suppose f −1 (V ) is open in X for every open set V in Y . Let x ∈ X and let
² > 0. Then f −1 (B² (f (x))) is open in X. Hence, ∃ δ such that Bδ (x) ⊂ f −1 (B² (f (x))).
Applying f , we obtain f (Bδ (x)) ⊂ B² (f (x)), and so f is continuous.
Problem (S’02, #4; S’03, #1). A function f : (0, 1) → R is the restriction to (0, 1)
of a continuous function F : [0, 1] → R ⇔ f is uniformly continuous on (0, 1).
Proof. ⇐ We show that if f : (0, 1) → R is uniformly continuous, then there is a
continuous F : [0, 1] → R with F (x) = f (x) for all x ∈ (0, 1).
Let xn be a sequence in (0, 1) converging to 0. Since f is uniformly continuous,
given ² > 0, ∃δ, s.t. |x − y| < δ ⇒ |f (x) − f (y)| < ². Therefore, we have
|f (xn ) − f (xm )| < ²
for n, m large enough. f (xn ) is a Cauchy sequence, so it converges to some ξ. Define
F (0) = limn→∞ f (xn ) = ξ. We want to show that this limit is well defined. Let yn
be another sequence, s.t. yn → 0, so f (yn ) is Cauchy by the same argument. Since
the sequence (f (x1 ), f (y1 ), f (x2 ), f (y2 ), . . .) is Cauchy by still the same argument, and
that there is a subsequence f (xn ) → ξ, then the entire sequence converges to ξ. Thus,
F (0) = limn→∞ f (xn ) = ξ is well defined.
By the same set of arguments, F (1) = η. The function F : [0, 1] → R given by



f (x) for x ∈ (0, 1),
F (x) = ξ for x = 0,


η for x = 1.
is the unique continuous extension of f to [0, 1].
⇒ F : [0, 1] → R is continuous, and [0, 1] is compact. Therefore, F is uniformly
continuous on [0, 1]. Thus, f = F |(0,1) : (0, 1) → R is uniformly continuous.
7
Gamelin, Green, p. 26; Edwards, p. 51.
8
f continuous at x ⇒ ∀² > 0, ∃ δ such that z ∈ Bδ (x) ⇒ f (z) ∈ B² (f (x)), or f (Bδ (x)) ⊂ B² (f (x)).
Real Analysis and Multivariable Calculus Igor Yanovsky, 2005 18

7.1 Continuity and Compactness

Theorem. Let f : X → Y is continuous, where X is compact. Then f (X) is compact.

Proof. 1) Let {Vα } be an open cover of f (X), (Vα ⊂ Y ). Since f is continuous, f −1 (Vα )
is open. Since X is compact, there are finitely many α1 , . . . , αn , such that
[ [
X ⊂ f −1 (Vα1 ) . . . f −1 (Vαn ).

Since f (f −1 (E)) ⊂ E for every E ⊂ Y , then

[ [
f (X) ⊂ Vα1 . . . Vαn .

⇒ f (X) is compact.

Proof. 2) Let {yn } be a sequence in the image of f . Thus we can find xn ∈ X, such
that yn = f (xn ). Since X is compact the sequence {xn } has a convergent subsequence
{xnk } with limit s ∈ X. Since f is continuous,

lim ynk = lim f (xnk ) = f (s)

k→∞ k→∞

Hence, the given sequence {yn } has a convergent subsequence which converges in f (X).
⇒ f (X) is compact.

Problem (F’01, #1). Let K ⊂ R be compact and f (x) continuous on K. Then f

has a maximum on K (i.e. there exists x0 ∈ K, such that f (x) ≤ f (x0 ) for all x ∈ K).

Proof. By theorem above, the image f (K) is closed and bounded. Let b be its least
upper bound. Then b is adherent to f (K). Since f (K) is closed ⇒ b ∈ f (K), that is
∃x0 ∈ K, such that b = f (x0 ), and thus f (x0 ) ≥ f (x), ∀x ∈ K.

Theorem. Let f : [a, b] → R be continuous function. Then:

1) f is bounded;
2) f assumes its max and min values;
3) f (a) < p < f (b) ⇒ ∃x : f (x) = p.

Proof. 1) f is continuous ⇒ f ([a, b]) is compact ⇒ f ([a, b]) is closed and bounded.
2) φ 6= f ([a, b]) ≤ M . Let M0 = sup f ([a, b]) ⇒ M0 ∈ closure(f ([a, b])) = f ([a, b]) ⇒
∃x0 ∈ [a, b] : f (x0 ) = M0 .
3) [a, b] is connected ⇒ f ([a, b]) is connected ⇒ f ([a, b]) is an interval.

Theorem. f : X → Y continuous and X compact. Then f is uniformly continuous.

Proof. 9 Suppose that f is not uniformly continuous. Then there exist ² > 0 and
(setting δ = 1/k in the definition) points xk , zk ∈ X 10 such that |xk − zk | < 1/k while
|f (xk ) − f (zk )| ≥ ². Passing to a subsequence, we can assume that xk → x ∈ X.11
Since |xk − zk | → 0, we also obtain zk → x. Since f is continuous, f (xk ) → f (x)
and f (zk ) → f (x), so that |f (xk ) − f (zk )| ≤ |f (xk ) − f (x)| + |f (x) − f (zk )| → 0, a
contradiction.

9
Gamelin, Green, p. 26-27; Rudin, p. 91.
10
See the technique of negation in the beginning of the section.
11
Since {xk } is Cauchy, and the convergent subsequence can be constructed, xk → x.
Real Analysis and Multivariable Calculus Igor Yanovsky, 2005 19

8 Sequences and Series of Functions

8.1 Pointwise and Uniform Convergence
Suppose {fn }, n = 1, 2, 3, . . ., is a sequence of functions defined on a set E, and suppose
that the sequence of numbers {fn (x)} converges ∀x ∈ E. Define a function f by

f (x) = lim fn (x), (x ∈ E).

n→∞

{fn } converges pointwise to f on E.

A sequence of functions {fn } converges uniformly on E to f if ∀² > 0, ∃N , such

that ∀n ≥ N ,

|fn (x) − f (x)| ≤ ², ∀x ∈ E.

Example: Consider f (x) = xn on [0, 1] ⇒ convergent, but not uniformly convergent

on [0, 1].

Problem (F’01, #3). If {fn } is a sequence of continuous functions on E, and if

fn → f uniformly on E, then f is continuous on E.

Proof. Fix ² > 0. Since fn → f uniformly, choose N , s.t. ∀n ≥ N

²
|fn (x) − f (x)| < , ∀x ∈ E
3
Since fn is continuous at p, choose δ, s.t. x ∈ E, |x − p| < δ then
²
|fn (x) − fn (p)| < , n≥N
3
Thus, |f (x) − f (p)| ≤ |f (x) − fn (x)| + |fn (x) − fn (p)| + |fn (p) − f (p)| < 3² + 3² + 3² = ².
Hence, given any ², ∃δ, s.t. |x − p| < δ ⇒ |f (x) − f (p)| < ², and f is continuous at
p.

8.2 Normed Vector Spaces

Problem (F’03, #7). C 0 [a, b] with the metric d(f, g) ≡ supx∈[0,1] |f (x) − g(x)|
= ||f − g||∞ is complete.

Proof. Let {ϕn }∞ 0

1 be a Cauchy sequence of elements of C [a, b]. Given ² > 0, choose
N such that m, n ≥ N ⇒
²
||ϕm − ϕn ||∞ < (sup norm).
2
Then, in particular, ||ϕm (x) − ϕn (x)||∞ < ²/2 for each x ∈ [a, b]. Therefore {ϕn (x)}∞
1
is a Cauchy sequence of real numbers, and hence converges to some real number ϕ(x).
It remains to show that the sequence of functions {ϕn } converges uniformly to ϕ; if so,
it will imply that ϕ is continuous on [a, b], i.e. ϕ ∈ C 0 [a, b].
Claim: For n ≥ N , (N same as above, n fixed), |ϕ(x) − ϕn (x)| < ² for all x ∈ [a, b].
To see this, choose m ≥ N sufficiently large (depending on x) s.t. |ϕ(x)−ϕm (x)| < ²/2.
⇒ |ϕ(x)−ϕn (x)| ≤ |ϕ(x)−ϕm (x)|+|ϕm (x)−ϕn (x)| < ²/2+||ϕm −ϕn || < ²/2+²/2 = ².
Since x ∈ [a, b] was arbitrary, it follows that ||ϕn − ϕ|| < ² as desired.
Real Analysis and Multivariable Calculus Igor Yanovsky, 2005 20

Theorem. Let {fn } ∈ C 1 [a, b], fn → f pointwise, fn0 → g uniformly.

Then fn → f uniformly, and f is differentiable, with f 0 = g.

Proof. By the Fundamental Theorem of Calculus, we have

Z x
fn (x) = fn (a) + fn0 ∀n, ∀x ∈ [a, b].
a

From this and the Uniform Convergence and Integration theorem, we obtain
Z x
f (x) = lim fn (x) = lim fn (a) + lim fn0 ,
n→∞ n→∞ n→∞ a
Z x
f (x) = f (a) + g.
a
Another application of the Fundamental Theorem yields f 0 = g as desired.
To see that convergence of fn → f is uniform, note that
¯Z x Z x ¯ Z x
¯ 0 ¯
|fn (x) − f (x)| = ¯ fn − g ¯ + |fn (a) − f (a)| ≤ |fn0 − g| + |fn (a) − f (a)|
a a a
≤ (b − a)||fn0 − g||∞ + |fn (a) − f (a)|.
The uniform convergence of fn therefore follows from that of fn0 .

Theorem. C 1 [a, b], with the C 1 -norm defined by

||ϕ|| = max |ϕ(x)| + max |ϕ0 (x)|,

x∈[a,b] x∈[a,b]

is complete.

Proof. Let {ϕn }∞ 1

1 be a Cauchy sequence of elements of C [a, b]. Since

max |ϕm (x) − ϕn (x)| ≤ ||ϕm − ϕn || (C 1 − norm),

x∈[a,b]

ϕn is a uniformly Cauchy sequence of continuous functions. Thus ϕn → ϕ ∈ C 0 [a, b]

uniformly. Similarly, since

max |ϕ0m (x) − ϕ0n (x)| ≤ ||ϕm − ϕn || (C 1 − norm),

x∈[a,b]

ϕ0n → ψ ∈ C 0 [a, b] uniformly. By the above theorem, ϕ is differentiable with ϕ0 = ψ,

so ϕ ∈ C 1 [a, b]. Since
max |ϕn (x) − ϕ(x)| = ||ϕn − ϕ||∞ (sup norm), and
x∈[a,b]

max |ϕ0n (x) − ϕ0 (x)| = ||ϕ0n − ϕ0 ||∞ (sup norm)

x∈[a,b]

the uniform convergence of ϕn and ϕ0n implies that ϕn → ϕ with respect to the C 1 -norm
of C 1 [a, b]. Thus every Cauchy sequence in C 1 [a, b] converges.
Real Analysis and Multivariable Calculus Igor Yanovsky, 2005 21

8.3 Equicontinuity
A family F of functions f defined on a set E ⊂ X is equicontinuous on E if:
∀² > 0, ∃ δ, such that |x − y| < δ, x, y ∈ E, f ∈ F ⇒ |f (x) − f (y)| < ².

8.3.1 Arzela-Ascoli Theorem

Suppose {fn (x)}∞ n=1 is uniformly bounded and equicontinuous sequence of func-
tions defined on a compact subset K of X. Then {fn } is precompact, i.e. the closure
of {fn } is compact, i.e. {fn } contains a uniformly convergent subsequence, i.e.
{fn } contains a subsequence {fnk } that converges uniformly on K to a function f ∈ X.

9 Connectedness
9.1 Relative Topology
Define the neighborhood of a point in Rn as N² (x) = {y : |x − y| < ²}. Consider the
subset of Rn , M ⊆ Rn . If all we are interested in are just points in M , it would be more
natural to define a neighborhood of a point x ∈ M as NM,² = {y ∈ M : |x − y| < ²}.
Thus, the relative neighborhood is just a restriction of the neighborhood in Rn to M .
Relative interior points and relative boundary points of a set, as well as a relative open
set and a relative closed set, can be defined accordingly.
Alternative definitions: T
S ⊆ M (⊆ Rn ) is open relative to M if there is an open set U in Rn such that S = U M .
S ⊆ M n n
T(⊆ R ) is closed relative to M if there is a closed set V in R such that
S = V M.
Example: A set S = [1, 4) is open T relative to M = [1, 10] ⊆ R since for the open set
U = (0, 4) in R, we have S = U M . S
Example: A set S = [1, 3] is open relative
T to M = [1, 3] [4, 6] ⊆ R since for the open
set U = (0, 4) in R, we have S = U M .

9.2 Connectedness
X is connected if it cannot be expressed as a disjoint12 union of two nonempty subsets
that are both open and closed.
S i.e. T
M is connected if M = A B, such Fthat A, B open and A B = φ, then A or B is
empty; or, M is connected if M = A B, A, B open ⇒ A = φ or B = φ.
Fact: X is connected ⇔ X and φ are the only subsets which are clopen.
SX is disconnected
T if there are closed and open subsets A and B of X such that
A B = X, A B = φ, A 6= φ, B 6= φ.
Another way of phrasing: X is disconnected if there is a closed and open U ⊂ X,
such that U 6= φ and U 6= X. If there is such a U , then the complement V = X \ U
of U is also both closed and open and X is the disjoint union of the nonempty sets U
and V .
A subset of a space is a connected subset if it is connected in the relative topology.
12 T
A B = φ, then A and B are disjoint.
Real Analysis and Multivariable Calculus Igor Yanovsky, 2005 22

Problem (S’02, #1). The closed interval [a, b] is connected.

S T
Proof. Let [a, b] = G H, s.t. G H = φ. Let b ∈ H. Then claim: G = φ. If not, let
c = sup G. Since G is closed, c ∈ G. Since G is open13 , B² (c) ⊆ G, i.e. [c, c + ²) ⊂ G.
That contradicts c = sup G. Thus G = φ.

Note: Since [a, b) and (a, b) can be expressed as the union of an increasing sequence
of compact intervals, these are also connected.
T S
Theorem. Let Sα ⊆ M , Sα connected. Suppose Sα 6= φ. Then Sα is connected.
S F S
Proof. Let S T
= Sα = G H, T G,F H areTopen in Sα .
Choose x0 ∈T Sα . Sα = (ST α G) (Sα H). Assume S x0 ∈TG. Since Sα is connected
S
and x0 ∈ Sα G, we get Sα H = φ, ∀α. Therefore, ( Sα ) H = φ. Since H ⊆ Sα
⇒ H = φ. Therefore, S is connected.

Corollary. R is connected.
S T
Proof. Let R = n∈N [−n, n], 0 ∈ [−n, n]. Therefore, R is the union of connected
subsets. By the theorem above, R is connected.

Theorem. Let f : M → N is continuous and M is connected. Then f (M ) is connected.

F S
Proof. Say f (M ) = G H, G, H 6= φ. G, H open. Then M = f −1 (G) f −1 (H),
where f −1 (G) and f −1 (H) are both open and nonempty. Contradicts connectedness of
M.

Theorem. a, b ∈ I, and a < c < b, then c ∈ I, i.e. I is an interval ⇔ I ⊆ R is

connected.

Proof. ⇒ Assume I an interval.

S = [a,S b]; S = (a, b], a ≥ T
−∞; S = [a, b), b ≤ ∞; S = (a, b), a ≥ −∞, b ≤ ∞.
[a, b) = n≥n0 [a, b − n ]. a ∈ [a, b − n1 ].
1
T F T
⇐ Say I is not an interval. ∃a < c < b, a, b ∈ I, c ∈
/ I. I = ((−∞, c) I) ((c, ∞) I),
i.e. I is not connected.

Problem (W’02, #3). The open unit ball in R2 , {(x, y) ∈ R2 : x2 + y 2 < 1} is

connected.

Proof. Let fθ (t) = t(cos θ, sin θ), −1 < t < 1. We have fθ : (−1, 1) → (t cos θ, t sin θ).
Since fθ is continuous and (−1, 1) is connected, fθ ((−1,
S 1)) is connected. The unit ball
can be expressed as {(x, y) ∈ R2 : x2 + y 2 < 1} = 0≤θ<π S fθ (t). We know that the
origin is contained in the intersection of fθ ’s. Therefore, 0≤θ<π fθ (t) is connected by
the theorem above.

13 S T c
If M =SG H, G H = φ and G, H are open, then G is closed and open, S since G = H .
MT= [0, 1] [2, 3] is not connected because if G = [0, 1], H = [2, 3], M = G H, G, H are clopen in M ,
G H = φ, and G, H 6= φ.
Real Analysis and Multivariable Calculus Igor Yanovsky, 2005 23

9.3 Path Connectedness

A path in X from x0 to x1 is a continuous function γ : [0, 1] → X, such that γ(0) = x0
and γ(1) = x1 .
X is path-connected if, for every pair of points x0 and x1 in X, there is a path γ
from x0 to x1 .

Theorem. A path-connected space is connected.

Proof. Fix x0 ∈ X. For each x ∈ X, let γx : [0, 1] → X be a path from x0 to x. By

theorems above, i.e.
(1) X connected and f : X → Y continuous ⇒ f (X) connected;
(2) any interval in R is connected;
each
S γx ([0, 1]) is a connected subset of X. Each γx ([0, 1]) contains x0 and X =
{γx ([0, 1]) : x ∈ X}, so that the theorem above shows that X is connected.

Theorem. An open subset of Rn is connected ⇔ it is path-connected.

Problem. Any subinterval of R (closed, open, or semiopen) is path-connected.

Proof. If a, b belong to an interval (of any kind), then γ(t) = (1 − t)a + tb, 0 ≤ t ≤ 1,
defines a path from a to b in the interval.

Problem. If X is path-connected and f : X → Y is a map, then f (X) is path-

connected.

Proof. If p = f (x) and q = f (y), and γ is a path in X from p to q, then f ◦ γ is a path

in f (X) from x to y.
Real Analysis and Multivariable Calculus Igor Yanovsky, 2005 24

10 Baire Category Theorem

A subset T ⊂ X is dense in X if T = X, i.e. every point of X is a limit point of T , or
a point of T , or both.
A subset Y ⊂ X is nowhere dense if Y has no interior points, i.e. int(Y ) = φ.
Y is nowhere dense ⇔ X \ Y is a dense open subset of X.
The interior of E is the largest open set in E, i.e. the set of all interior points of E.
Q=T R, (R \ Q) = R. T
[0, 1] Q is not closed; (0, 1) Q is not open in R.
Baire’s Category Theorem. T Let {Un }∞
n=1 be a sequence of dense open subsets of a
complete metric space X. Then ∞ n=1 n is also dense in X. (Any countable intersec-
U
tion of dense open sets in a complete metric space is dense.)
Corollary. LetS{En }∞n=1 be a sequence of nowhere dense subsets of a complete metric
∞
space X. Then n=1 En has empty interior. (In a complete metric space, no nonempty
open subset can be expressed as a union of countable collection of nowhere dense sets.)
Proof. T
We apply the Baire Category Theorem to the dense open sets Un = X \ E n .
Then, ∞ n=1 Un is dense in X.
∞
\ ∞
\ ∞
[
Un = (X \ E n ) = X \ En
n=1 n=1 n=1
S∞ S∞ S∞
Therefore, X \ n=1 E n is dense, ⇒ n=1 En is nowhere dense ⇒ n=1 En has
empty interior.
A subset of X is of the first category (i.e. Q) if it is the countable union of
nowhere dense subsets. A subset (i.e. I, R) that is not of the first category is said to
be of the second category.
S
S ⊆ R is Fσ set if S = T∞
n=1 Fn , Fn closed.
S ⊆ R is Gδ set if S = ∞ n=1 Gn , Gn open.
Problem (W’02, #4). The set of irrational numbers I in R is not the countable
union of closed sets (not an Fσ set).
S
Proof. Suppose I = Fn , where each Fn is closed.
[ [
R= Fn {q}
n∈N q∈Q
Thus R can be expressed as the countable union of closed sets. By (corollary to) the
Baire Category Theorem, since R is a nonempty open subset, one of these closed sets
has a nonempty interior. It cannot be one of q’s, and since any nonempty interval
contains rational numbers, it cannot be one of Fn ’s. Contradiction.
Problem (S’02, #2; F’02, #3). The set Q of rational numbers is not the countable
intersection of open sets of R (not a Gδ set).
Show that there is a subset of R which is not the countable intersection of open subsets.
T
Proof. We take complements in the preceding theorem. Suppose Q = Gn , where
each Gn open. Then,
\ [
I=R\Q=R\( Gn ) = (R \ Gn ),
n∈N n∈N

and R \ Gn is closed. Thus, I is a Fσ set, which contradicts the previous theorem.

Real Analysis and Multivariable Calculus Igor Yanovsky, 2005 25

Problem (S’03, #3). Find S ⊂ R such that both (i) and (ii) hold for S:
(i) S is not the countable union of closed sets (not Fσ );
(ii) S is not the countable intersection of open sets (not Gδ ).
S
Proof.
S Let A ⊆ [0, 1] not
S FσS, B ⊆ [2, 3] not Gδ ⇒ A B is neither Fσ nor Gδ .
If A B is Fσ , say A B = Fn ⇒
|{z}
closed
[ \ [ \ [ \
A=A B [0, 1] = Fn [0, 1] = (Fn [0, 1]) ≡ Fσ set ⇒ contradiction.
| {z }
closed
S S T
If A B is Gδ , say A B= Gn ⇒
|{z}
open

[ \ 3 7 \ \ 3 7 \ \ 3 7
B=A B ( , )= Gn ( , ) = (Gn ( , )) ≡ Gδ set ⇒ contradiction.
2 2 2 2 | {z2 2 }
open

Problem. Q is not open, is not closed, but is the countable union of closed sets (Fσ
set).

Proof. Since any neighborhood (q − ², q + ²) of a rational q contains irrationals, Q has

no inner points. ⇒ Q is not open. Since every irrational number i is the limit of a
sequence of rationals ⇒ Q is not closed. Since every one-point-set {x} ⊂ R is closed
and Q is countable, say (qn ) is a sequence of all rational numbers, we find that
[
Q= {qn }
n∈N

is the countable union of closed sets, i.e. Q is Fσ .

Problem. The set of isolated points of a countable complete metric space X forms a
dense subset of X.

Proof. For each point x ∈ X that is not an isolated point of X, define Ux = X \ {x}.
Each such Ux is open and dense in X, and the intersection of the Ux ’s consists precisely
of the isolated points of X. By the Baire Category Theorem, the intersection of the
Ux ’s is dense in X.

Problem. Suppose that F is a subset of the first category in a metric space X and E
is a subset of F . Prove that E is of the first category in X. Show by an example that
E may not be of the first category in the metric space F .
S S T
Proof.
T If F = Fn , where each Fn is nowhere dense, then E = (E Fn ),2 and each
E Fn is nowhere dense. For example, note the R is of first category in R , but R is
not of first category in itself.

Problem. Any countable union of sets of the first category in X is again of the first
category in X.

Proof. A countable union of countable unions is a countable union.

Real Analysis and Multivariable Calculus Igor Yanovsky, 2005 26

Problem. a) If a, b ∈ R satisfy a < b, then there exists a rational number q ∈ (a, b).
b) The set Q of rational numbers is dense in R.

Problem. The set of irrational numbers is dense in R.

Proof. If i is any irrational number, and if q is rational, then q + i/n is irrational, and
q + i/n → q.

Problem. Regard the rational numbers Q as a subspace of R. Does the metric space
Q have any isolated points?

Proof. The rationals have no isolated points. This does not contradict the above, i.e.
”The set of isolated points of countable complete metric space X forms dense subset of
X”
because the rationals are not complete.

Problem. Every open subset of R is a union of disjoint open intervals (finite, semi-
infinite, or infinite).

Proof. For each x ∈ U , let Ix be the union of all open intervals containing x that are
contained in U . Show that each Ix is an open interval (possibly infinite or semi-infinite),
any two Ix ’s either coincide or are disjoint, and the union of the Ix ’s is U .

11 Integration
11.1 Riemann Integral
Let [a, b] be a given interval. A partition P of [a, b] is a finite set of points x0 , . . . , xn :

a = x0 < x1 < . . . < xn−1 < xn = b, 4xi = xi − xi−1 .

A Riemann sum for f correspoding to the given partition is

n
X 0 0
S(P, f ) = f (xi )4xi , xi−1 ≤ xi ≤ xi .
i=1

Definition: f is Riemann integrable on [a, b], if ∃A ∈ R such that:

∀² > 0, ∃ δ > 0 such that whenever S is a Riemann sum for f corresponding to any
partition of [a, b] with max(4xi ) < δ ⇒ |S − A| < ².
In
R b this case A is called the Riemann integral of f between a and b and is denoted as
a f dx.
Alternative Definition: f is Riemann integrable on [a, b] if:
∀² > 0, ∃P such that U (P, f ) − L(P, f ) < ².

If f is bounded, there exist m and M , such that m ≤ f (x) ≤ M, a ≤ x ≤ b. Hence,

for every P ,

m(b − a) ≤ S(P, f ) ≤ M (b − a),

so that S(P, f ) is bounded. This shows that Riemann sums are defined for every
bounded function f .
Real Analysis and Multivariable Calculus Igor Yanovsky, 2005 27

11.2 Existence of Riemann Integral

Theorem. f is integrable on [a, b] ⇔ ∀² > 0, ∃ δ > 0 such that S1 (P, f ), S2 (P, f ),
P with max(4xi ) < δ, then |S1 − S2 | < ².
Proof. ⇒ Suppose f is integrable on [a, b]. ∀² > 0, ∃ δ > 0 such that S(P, f ), P with
Rb
max(4xi ) < δ, then |S − a f (x)dx| < ²/2. For such S1 and S2 ,
¯ Z b Z b ¯
¯¡ ¢ ¡ ¢¯ ² ²
|S1 − S2 | = ¯¯ S1 − f (x)dx − S2 − f (x)dx ¯¯ < + = ².
a a 2 2
⇐ ∀² > 0, ∃ δ > 0 such that S1 (P, f ), S2 (P, F ), P with max(4xi ) < δ ⇒ |S1 −S2 | < ².
For n = 1, 2, . . ., choose S (n) (P, f ), P with max(4xi ) < 1/n. Then,
∀² > 0, ∃ N > 0 (δ = 1/N ), such that |S (n) − S (m) | < ², n, m ≥ N ⇒ S (n) is a Cauchy
sequence of real numbers ⇒ S (n) converges to some A ∈ R ⇒ |S (N ) −A| < ², 1/N < δ.
Thus for any S(P, f ), P with max(4xi ) < δ, we have
|S − A| ≤ |S − S (N ) | + |S (N ) − A| < 2².

Theorem. If f is continuous on [a, b] then f is integrable on [a, b].

Proof. Since f is uniformly continuous on [a, b], ∀² > 0, ∃ δ > 0 such that ∀x, z ∈
[a, b], |x − z| < δ ⇒ |f (x) − f (z)| < ² ~. If P is any partition of [a, b] with
max(4xi ) < δ, then ~ implies that Mi − mi ≤ ², i = 1, . . . , n, and therefore
n
X n
X
U (P, f ) − L(P, f ) = (Mi − mi )4xi ≤ ² 4xi = ²(b − a).
i=1 i=1

Thus, ∀² > 0, ∃P such that |U (P, f ) − L(P, f )| < C² ⇒ f is integrable.

11.3 Fundamental Theorem of Calculus

Theorem. If f : [a, b] → R is continuous, and F : [a, b] → R is defined by
Z x
F (x) = f (t)dt,
a

then F is differentiable and F 0 = f .

Rx
Proof. Since f is continuous, F (x) = a f (t)dt is defined for all x ∈ [a, b]. We have to
show that for any fixed x0 ∈ [a, b]
F (x) − F (x0 )
lim = f (x0 ).
x→x0 x − x0
For any x ∈ [a, b], x 6= x0 , we have
¯ ¯ ¯R x R ¯ ¯R x ¯
¯ F (x) − F (x0 ) ¯ ¯ a f (t)dt − ax0 f (t)dt ¯ ¯ x0 f (t)dt ¯
¯ − f (x0 )¯¯ = ¯¯ − f (x0 )¯¯ = ¯¯ − f (x0 )¯¯
¯ x − x0 x − x0 x − x0
¯R x Rx ¯ ¯R x ¯ Rx
¯ f (t)dt f (x0 )dt ¯ ¯ x0 (f (t) − f (x0 ))dt ¯ |f (t) − f (x0 )|dt
= ¯¯ x0 − x0 ¯=¯ ¯ ≤ x0 ⇒ ~
x − x0 x − x0 ¯ ¯ x − x0 ¯ |x − x0 |
Since f is continuous at x0 , given ² > 0, ∃δ, such that x ∈ [a, b], |x − x0 | < δ ⇒
|f (x) − f (x0 )| < ². Then for any t between x and x0 , we have |f (t) − f (x0 )| < ².
Rx
² dt
⇒ ~ < x0 = ².
|x − x0 |
Thus F 0 (x0 ) = f (x0 ).
Real Analysis and Multivariable Calculus Igor Yanovsky, 2005 28

Corollary. If f : [a, b] → R is continuous and F 0 = f on [a, b], then

Z b
f (t)dt = F (b) − F (a).
a
d
Rx Rx
Proof.
Rx Since dx ( a f (t)dt−F (x)) = f (x)−f (x) = 0, a f (t)dt−F
Ra (x) is constant. Thus
a f (t)dt = F (x) + c, for
Rx
some c ∈ R. In particular, 0 = a f (t)dt = F (a) + c, so that
Rb
c = −F (a). Therefore, a f (t)dt = F (x) − F (a). Hence, a f (t)dt = F (b) − F (a).
Integration by Parts. Suppose f and g are differentiable functions on [a, b], f 0 , g 0 ∈
R. Then
Z b Z b
0
f (x)g (x)dx = f (b)g(b) − f (a)g(a) − f 0 (x)g(x)dx.
a a
Proof. Let h(x) = f (x)g(x) and apply the Fundamental Theorem of Calculus to h and
its derivative.
Z b
h0 (x)dx = h(b) − h(a)
a
Z b
(f 0 (x)g(x) + f (x)g 0 (x))dx = f (b)g(b) − f (a)g(a).
a
Note that h0 ∈ R.
Rb
Mean Value Theorem for Integrals. If f : [a, b] → R is continuous, then a f (x)dx =
f (c)(b − a) for some c ∈ [a, b].
Proof. Since f is continuous, by the Fundamental Theorem of Calculus, R b there is a
function F : [a, b] → R such that F 0 (x) = f (x) for x ∈ (a, b), and a f (x)dx =
F (b) − F (a).
By the Mean Value Theorem for Differentiation, ∃c ∈ (a, b) such that F (b) − F (a) =
F 0 (c)(b − a). Thus,
Z b
f (x)dx |{z} = F 0 (c)(b − a) |{z}
= F (b) − F (a) |{z} = f (c)(b − a).
a
FTC MV T FTC
Rb
Thus, ∃c ∈ (a, b) such that a f (x)dx = f (c)(b − a).
Generalized Mean Value Theorem for Integrals. If f, g : [a, b] → R are continu-
ous and g(x) > 0 for all x ∈ [a, b], then there exists c ∈ [a, b] such that
Z b Z b
f (x)g(x)dx = f (c) g(x)dx.
a a
Rb
Proof. Since g(x) > 0 for all x ∈ [a, b] and since g is continuous, a g(x) > 0.
Suppose f (x) attains its maximum M at x2 and minimum m at x1 .
Then m = f (x1 ) ≤ f (x) ≤ f (x2 ) = M for x ∈ [a, b], and
Z b Z b Z b
m g(x)dx ≤ f (x)g(x)dx ≤ M g(x)dx,
a a a
and hence
Rb
a f (x)g(x)dx
m≤ Rb ≤ M.
a g(x)dx
Rb
f (x)g(x)dx
Since f is continuous on compact [a, b], ∃c such that f (c) = a
Rb .
a g(x)dx
Real Analysis and Multivariable Calculus Igor Yanovsky, 2005 29

Uniform Convergence and Integration. Let {fn } be a sequence of continuous func-

tions on [a, b] and fn → f uniformly on [a, b]. Then
Z b Z b
lim fn (x) dx = lim fn (x)dx.
n→∞ n→∞
a | {z } a
f (x)

Proof. Let f = limn→∞ fn . Since each fn is continuous and fn → f uniformly ⇒

f is continuous. In particular, f is integrable on [a, b]. By the definition of uniform
convergence, ∀² > 0, ∃N > 0 such that n > N , |f (x) − fn (x)| < ²/(b − a), ∀x ∈ [a, b].
Thus,
² ²
− ≤ f (x) − fn (x) ≤ , ∀x ∈ [a, b] ⇒
b−a b−a
Z b
⇒ −² ≤ (f (x) − fn (x))dx ≤ ²
a
¯Z b Z b ¯
¯ ¯
or ¯ f (x)dx − fn (x)dx¯¯ ≤ ².
¯
a a
Rb Rb
The last inequality holds for all n > N , and therefore, limn→∞ a fn (x)dx = a f (x)dx.

Uniform Convergence and Differentiation. Let {fn } be a sequence of differentiable

0
functions on [a, b] such that {fn (x0 )} converges for some x0 ∈ [a, b]. If fn → f 0
uniformly, then fn → f uniformly on [a, b], and
³ ´0 0
lim fn = lim fn
n→∞ n→∞

Proof. See the section on “Sequences and Series of Functions: Normed Vector Spaces”
where the weaker statement is proved, i.e. {fn } ∈ C 1 , fn → f pointwise on [a, b].
Real Analysis and Multivariable Calculus Igor Yanovsky, 2005 30

12 Differentiation
12.1 R→R
12.1.1 The Derivative of a Real Function
Let f : (a, b) → R, x0 ∈ (a, b). f is differentiable at x0 if

f (x) − f (x0 )
f 0 (x0 ) = lim
x→x0 x − x0

exists. f 0 is the derivative of f .

Theorem. f : (a, b) → R. f is differentiable at x0 ⇒ f is continuous at x0 .

Proof.

f (x) − f (x0 )
lim (f (x) − f (x0 )) = lim · lim (x − x0 ) = f 0 (x0 ) · 0 = 0.
x→x0 x→x0 x − x0 x→x0

Since limx→x0 f (x) = f (x0 ), f is continuous at x0 .

Lemma. If f 0 (c) > 0, then f is locally strictly increasing at c, i.e., ∃δ > 0 such that:
c − δ < x < c ⇒ f (x) < f (c),
c < x < c + δ ⇒ f (c) < f (x).
f (x)−f (c) f (x)−f (c)
Proof. limx→c x−c > 0 ⇒ ∃δ > 0: x−c > 0 whenever 0 < |x − c| < δ.
f (x)−f (c)
Thus since f (x) − f (c) = (x
− c), (x 6= c), we have
x−c
c − δ < x < c ⇒ f (x) − f (c) < 0,
c < x < c + δ ⇒ f (x) − f (c) > 0.

Corollary. If f has a max (or a min) at c ∈ (a, b), i.e. f (x) ≤ f (c) (or f (x) ≥ f (c))
for all x, then f 0 (c) = 0.

Proof. Say f 0 (c) 6= 0. Say f 0 (c) > 0. Then x > c ⇒ f (x) ≤ f (c) (since f has a max
at c), contradicting the lemma above. Proofs of other conditions are similar.

12.1.2 Rolle’s Theorem

Theorem. Let f be continuous on [a, b], differentiable on (a, b), and f (a) = f (b). Then
∃c ∈ (a, b) : f 0 (c) = 0.

Proof. Let M = sup{f (x) : x ∈ [a, b]}, m = inf{f (x) : x ∈ [a, b]}.
Then m ≤ f (a) = f (b) ≤ M . If M = m ⇒ f (x) = f (a), ∀x ⇒ f 0 (c) = 0, ∀c ∈ (a, b).
Say f (a) = f (b) < M ~. Then choose c ∈ [a, b] : f (c) = M . From ~, c ∈ (a, b). We
have from corollary, f 0 (c) = 0.
Similarly, say m < f (a) = f (b) }. Then choose c ∈ [a, b] : f (c) = m. From },
c ∈ (a, b). We have from corollary, f 0 (c) = 0.

12.1.3 Mean Value Theorem

Theorem. Let f be continuous on [a, b], differentiable on (a, b). Then ∃c ∈ (a, b):

f (b) − f (a)
f 0 (c) = .
b−a
Real Analysis and Multivariable Calculus Igor Yanovsky, 2005 31

f (b)−f (a)
Proof. Let g(x) = f (x) − b−a (x − a). Then g(a) = g(b) = f (a). By Rolle’s
Theorem, ∃c ∈ (a, b), such that 0 = g 0 (c) = f 0 (c) − f (b)−f
b−a
(a)
.

Corollary. Let f be continuous on [a, b], differentiable on (a, b).

a) f 0 (x) = 0, ∀x ∈ (a, b) ⇒ f is constant.
b) f 0 (x) > 0, ∀x ∈ (a, b) ⇒ f is strictly increasing.

Proof. b) a ≤ x1 < x2 ≤ b. By mean value theorem, f (x2 ) − f (x1 ) = f 0 (c)(x2 − x1 )

for some c ∈ (x1 , x2 ). Therefore, f (x2 ) − f (x1 ) > 0 for all such x1 , x2 . Proof of a) is
similar.

12.2 R → Rm
f : R → Rm is differentiable at c ∈ R if there exists a linear map L : R → Rm such
that
||f (c + h) − f (c) − L(h)||
lim = 0.
h→0 ||h||

f10 (c)
 .. 
in which case L is defined by L = dfc = f 0 (c) =  . .
0
fm (c)
The linear mapping dfc : R → Rm is called the differential of f at c. The matrix of
the linear mapping f 0 (c) is the derivative. The differential is the linear mapping whose
matrix is the derivative.

12.3 Rn → Rm
Let U ⊆ Rn open, c ∈ U . f : U → Rm is differentiable at c if there exists a linear
map L : Rn → Rm such that

||f (c + h) − f (c) − L(h)||

lim = 0.
h→0 ||h||

in which case L is defined by

 
∂f1 ∂f1
∂x1 ··· ∂xn
 
L = dfc = f 0 (c) =  · ··· · 
∂fm ∂fm
∂x1 ··· ∂xn x=c

Let v ∈ Rn , c ∈ G ⊆ Rn . The directional derivative with respect to v of f at c is

f (c + tv) − f (c)
Dv f (c) = lim
t→0 t
provided that the limit exists. In particular, the partial derivatives of f at c are

∂f f (c + tej ) − f (c) f (c1 , . . . , cj + t, . . . , cn ) − f (c1 , . . . , cj , . . . , cn )

(c) = Dej f (c) = lim = lim
∂xj t→0 t t→0 t
Real Analysis and Multivariable Calculus Igor Yanovsky, 2005 32

Theorem. Let f be differentiable at c (i.e., dfc : Rn → Rm is defined). Then:

a)14 The directional derivative Dv f (c) exists ∀v ∈ Rn , and Dv f (c) = dfc (v) ( =
f 0 (c)v).
∂fi
b)15 The matrix f 0 (c) of dfc is dfc = f 0 (c) = [ ∂x j
(c)].

Proof. a) Since f is differentiable, and letting h = tv,

||f (c + tv) − f (c) − dfc (tv)||

lim = 0,
t→0 ||tv||

1 h ¯¯ f (c + tv) − f (c) ¯¯i

¯¯ ¯¯
lim ¯¯ − dfc (v)¯¯ = 0,
||v|| t→0 t
f (c + tv) − f (c)
lim = dfc (v).
t→0 t
i.e. Dv f (c) exists, and equals to dfc (v).
     
↑ ↑ ↑ ↑ ↑ ↑
   ∂f ∂f
Proof. b) dfc = dfc (e1 ) · · · dfc (en ) = De1 f (c) · · · Den f (c)  =  ∂x1 (c) · · · ∂xn (c)
=
↓ ↓ ↓ ↓ ↓ ↓
h i
∂fi
∂xj (c) m×n
.

f : Rn → Rm is said to be continuously differentiable at c if the partial deriva-

∂fi
tives ∂x j
exist on B² (c), and are continuous at c.
14
Edwards, Theorem 2.1
15
Edwards, Theorem 2.4
Real Analysis and Multivariable Calculus Igor Yanovsky, 2005 33

∂fi
Theorem16 . If ∂xj exist and are continuous on G, then at each c ∈ G,
h ∂f i ∂fi
i
dfc = f 0 (c) = (c) . (i.e. continuous ⇒ f dif f erentiable).
∂xj ∂xj

Proof. Since f : Rn → Rm is differentiable at c iff each of its component functions

f1 , . . . , fm is, we may assume m = 1; f : U → R. Given h = (h1 , . . . , hn ), let
h0 = (0, . . . , 0), hj = (h1 , . . . , hj , 0, . . . , 0), j = 1, . . . , n. We have
n
X
f (c + h) − f (c) = [f (c + hj ) − f (c + hj−1 )].
j=1

f (c + hj ) − f (c + hj−1 ) = f (c1 + h1 , . . . , cj−1 + hj−1 , cj + hj , cj+1 , . . . , cn )

− f (c1 + h1 , . . . , cj−1 + hj−1 , cj , cj+1 , . . . , cn )
∂f
= (c1 + h1 , . . . , cj−1 + hj−1 , cj + t, cj+1 , . . . , cn ) · hj
∂xj
for some 0 ≤ t ≤ hj , by mean-value theorem. Thus
∂f
f (c + hj ) − f (c + hj−1 ) = (dj ) · hj , for some dj , ||dj − c|| ≤ h.
∂xj
n
X ∂f
⇒ f (c + h) − f (c) = (dj ) · hj
∂xj
j=1
 
h h1 i
 
Also considering f 0 (c) · h = ∂x ∂f
1
∂f
(c) · · · ∂x n
(c)  ... , we have
hn
¯ Pn £ ∂f ¤ ¯
¯ ∂f ¯
|f (c + h) − f (c) − f 0 (c) · h)| j=1 ∂xj (dj ) − ∂xj (c) hj
lim = lim
h→0 ||h|| h→0 ||h||
n ¯
X ¯ ∂f ∂f ¯
¯
≤ lim ¯ (dj ) − (c)¯ = 0,
h→0 ∂xj ∂xj
j=1
∂f
since each dj → c as h → 0, and each ∂xj is continuous at c.

16
Edwards, Theorem 2.5
Real Analysis and Multivariable Calculus Igor Yanovsky, 2005 34

12.3.1 Chain Rule

Theorem. U ⊂ Rn , V ⊂ Rm . If the mappings F : U → Rm and G : V → Rk are
differentiable at a ∈ U and F(a) ∈ V respectively, then their composition H = G ◦ F is
differentiable at a, and

dHa = dGF (a) ◦ dFa (composition of linear mappings)

In terms of derivatives

H 0 (a) = G0 (F (a)) · F 0 (a) (matrix multiplication)

The differential of the composition is the composition of the differentials;

the derivative of the composition is the product of the derivatives.

Proof. We must show that

H(a + h) − H(a) − dGF (a) ◦ dFa (h)

lim = 0. Define
h→0 ||h||

F (a + h) − F (a) − dFa (h)

ϕ(h) = ⇒ F (a + h)−F (a) = dFa (h)+||h||ϕ̃(h) and
||h||
(12.1)

G(F (a) + k) − G(F (a)) − dGF (a) (k)

ψ(k) = ⇒ G(F (a)+k)−G(F (a)) = dGF (a) (k)+||k||ψ̃(k)
||k||
(12.2)

The fact that F and G are differentiable at a and F (a), respectively, implies that

lim ϕ(h) = lim ψ(k) = 0. Then

h→0 k→0

H(a + h) − H(a) = G(F (a + h)) − G(F (a)) = G(F (a) + (F (a + h) − F (a))) − G(F (a))
= [k=F (a+h)−F (a) in (12.2)]= dGF (a) (F (a + h) − F (a)) + ||F (a + h) − F (a)|| · ψ̃(F (a + h) − F (a))
¯¯ ¯¯
= (12.1) = dGF (a) (dFa (h) + ||h||ϕ(h)) + ¯¯dFa (h) + ||h||ϕ(h)¯¯ · ψ̃(F (a + h) − F (a))
¯¯ h ¯¯
¯¯ ¯¯
= dGF (a) ◦ dFa (h) + ||h||dGF (a) (ϕ(h)) + ||h||¯¯dFa ( ) + ϕ(h)¯¯ · ψ̃(F (a + h) − F (a))
||h||

H(a + h) − H(a) − dGF (a) ◦ dFa (h) ¯¯ h ¯¯

¯¯ ¯¯
⇒ = dGF (a) (ϕ(h)) + ¯¯dFa ( ) + ϕ(h)¯¯ · ψ̃(F (a + h) − F (a))
||h|| ||h||
dGF (a) is linear ⇒ continuous, and limh→0 ϕ(h) = 0 ⇒ limh→0 dGF (a) (ϕ(h)) = 0.
Since F is continuous at a and limk→0 ψ(k) = 0 ⇒ limh→0 ψ̃(F (a + h) − F (a)) = 0.
dFa is linear ⇒ continuous ⇒ bounded ⇒ ∃M, ||dFa (x)|| ≤ M ||x||.
Therefore, the limit of the entire expression above → 0 ⇒ dHa = dGF (a) ◦ dFa .
Real Analysis and Multivariable Calculus Igor Yanovsky, 2005 35

Theorem. Let U ⊆ Rn be open and connected. F : U → Rm . F 0 (x) = 0 for all x ∈ U

⇔ F is constant.

Proof. Since F is constant ⇔ each of its component functions is constant,

and the matrix F 0 (x) = 0 ⇔ each of its rows is 0, we may assume F = f : U → R.
Suppose f 0 (x) = ∇f (x) = 0, ∀x ∈ U .
Given a and b ∈ U , let γ : R → U be a differentiable mapping with γ(0) = a, γ(1) = b.
R →γ Rn →f R
If g = f ◦ γ : R → R, then

g 0 (t) = ∇f (γ(t)) · γ 0 (t) = 0

∀t ∈ R ⇒ g is constant on [0, 1], so f (a) = f (γ(0)) = g(0) = g(1) = f (γ(1)) =

f (b).

Example 1. w = w(x, y), x = x(r, θ), y = y(r, θ)

£ ¤ h i· ∂x ∂x ¸ h i £ ¤
· ∂x ∂x ¸−1
∂w ∂w ∂w ∂w ∂r ∂θ ∂w ∂w ∂w ∂w ∂r ∂θ
∂r ∂θ = ∂x ∂y ∂y ∂y ⇒ ∂x ∂y = ∂r ∂θ ∂y ∂y
∂r ∂θ ∂r ∂θ

∂f
Example 2. f (x, y, z) = 0, ∂z 6= 0 ⇒ z = h(x, y), f (x, y, h(x, y)) ≡ 0.
For example, can solve for ∂z
∂x :
∂
0 = ∂x f (x, y, h(x, y)) = ∂f ∂x ∂f ∂y
∂x ∂x + ∂y ∂x +
∂f ∂z
∂z ∂x =
∂f
∂x + ∂f ∂z
∂z ∂x .
∂z ∂f ∂f
⇒ ∂x = − ∂x / ∂z .
∂y ∂x ∂ ∂
Similarly, can solve for ∂z and ∂y , from ∂z f (x, y(x, z), z) and ∂y f (x(y, z), y, z),
∂x ∂y ∂z
respectively, and show ∂y ∂z ∂x = −1.

12.3.2 Mean Value Theorem

Theorem. Let f : U ⊆ Rn → R be differentiable, and a and b ∈ U , such that
[a, b] ⊆ U . Then ∃c ∈ (a, b), such that

f (b) − f (a) = ∇f (c) · (b − a).

Proof. Define γ : [0, 1] → [a, b] as γ(t) = a + t(b − a) = (1 − t)a + tb, t ∈ [0, 1].
Then γ 0 (t) = b − a. Let g = f ◦ γ. R →γ Rn →f R
Since g : [0, 1] → R, then by single-variable MVT, ∃ξ ∈ [0, 1], such that g(1) − g(0) =
g 0 (ξ). If c = γ(ξ), then

f (b) − f (a) = f (γ(1)) − f (γ(0)) = g(1) − g(0) = g 0 (ξ) =

=Chain Rule = ∇f (γ(ξ)) · γ 0 (ξ) = ∇f (c) · (b − a).

Real Analysis and Multivariable Calculus Igor Yanovsky, 2005 36

∂ ∂f ∂ ∂f
12.3.3 ∂x ( ∂y ) = ∂y ( ∂x )

Theorem. Let f : U ⊆ Rn → R. If f, ∂f ∂f
∂x , ∂y exist and are continuous on U and
∂2f ∂2f
∂x∂y , ∂y∂x exist on U and are continuous at a, then

∂ ³ ∂f ´ ∂ ³ ∂f ´
= on U.
∂x ∂y ∂y ∂x

Proof. Consider Sh (x, y) = f (x + h, y + h) − f (x + h, y) − f (x, y + h) + f (x, y).

Let g(x, y) = f (x + h, y) − f (x, y). Then,
Sh (x, y) = f (x + h, y + h) − f (x + h, y) − f (x, y + h) + f (x, y) = g(x, y + h) − g(x, y) =
∂g £ ∂f ∂f ¤
= M V T = h (x, y + βh) = h (x + h, y + βh) − (x, y + βh) =
∂y ∂y ∂y
¡
∂ ∂f ¢
= M V T = h2 (x + αh, y + βh)
∂x ∂y
where 0 < α, β < 1.
Let r(x, y) = f (x, y + h) − f (x, y). Then,
Sh (x, y) = f (x + h, y + h) − f (x + h, y) − f (x, y + h) + f (x, y) = r(x + h, y) − r(x, y) =
∂r £ ∂f ∂f ¤
= M V T = h (x + α0 h, y) = h (x + α0 h, y + h) − (x + α0 h, y) =
∂x ∂x ∂x
∂ ¡ ∂f ¢
= M V T = h2 (x + α0 h, y + β 0 h)
∂y ∂x
where 0 < α0 , β 0 < 1. ¡ ¢ ¡ ¢
∂ ∂f ∂ ∂f 0 0
For each small enough h > 0, ∂x ∂y (x + αh, y + βh) = ∂y ∂x (x + α h, y + β h).
Since the
¡ mixed
¢ partial¡ derivatives
¢ are continuous at a = (x, y), let h → 0 ⇒
∂ ∂f ∂ ∂f
⇒ ∂x ∂y (x, y) = ∂y ∂x (x, y).
Real Analysis and Multivariable Calculus Igor Yanovsky, 2005 37

12.4 Taylor’s Theorem

n-th degree Taylor polynomial of f at a; (h = x − a)

f (n) (a) n
Pn (h) = f (a) + f 0 (a)h + · · · + h
n!
Mean Value Theorem (R → R, Revisited). f : [a, b] → R. Suppose that f 0 exists
on [a, b]. h = b − a. Then ∃ξ between a and b such that

R0 (h) = f 0 (ξ)h

f (a + h) = f (a) + f 0 (ξ)h = P0 (h) + R0 (h)

Proof. Need to show: R0 (h) = f 0 (ξ)h. For t ∈ [0, h], define

R0 (t) = f (a + t) − P0 (t) = f (a + t) − f (a). So, R00 (t) = f 0 (a + t).
Define ϕ : [0, h] → R by

R0 (h)
ϕ(t) = R0 (t) − t ⇒ ϕ(0) = ϕ(h) = 0
h
⇒ By Rolle’s theorem, ∃c ∈ (0, h) such that

R0 (h) R0 (h)
0 = ϕ0 (c) = R00 (c) − = f 0 (a + c) −
h h
⇒ For ξ = a + c, R0 (h) = f 0 (ξ)h.

Taylor’s Theorem (R → R). f : [a, b] → R. Suppose that f (n+1) exists on [a, b].
h = b − a. Then ∃ξ between a and b such that

f (n+1) (ξ) n+1

Rn (h) = h .
(n + 1)!

f (n) (a) n f (n+1) (ξ) n+1

f (a + h) = f (a) + f 0 (a)h + · · · + h + h = Pn (h) + Rn (h).
n! (n + 1)!
f (n+1) (ξ) n+1
Proof. Need to show: Rn (h) = (n+1)! h . For t ∈ [0, h], define Rn (t) = f (a + t) −
Pn (t), and note that

Rn (0) = Rn0 (0) = · · · = Rn(n) (0) = 0 (12.3)

since the first n derivatives of Pn (t) at 0 agree with those of f at a. Also,

Rn(n+1) (t) = f (n+1) (a + t) (12.4)

(n+1)
since Pn (t) ≡ 0 because Pn (t) is a polynomial of degree n.
Define ϕ : [0, h] → R by

Rn (h) n+1
ϕ(t) = Rn (t) − t ⇒ ϕ(0) = ϕ(h) = 0
hn+1
⇒ By Rolle’s theorem, ∃c1 ∈ (0, h) such that ϕ0 (c1 ) = 0.
It follows from (12.3) and (12.4) that

ϕ(0) = ϕ0 (0) = · · · = ϕ(n) (0) = 0 (12.5)

Real Analysis and Multivariable Calculus Igor Yanovsky, 2005 38

Rn (h)
ϕ(n+1) (t) = f (n+1) (a + t) − (n + 1)! (12.6)
hn+1
Therefore, we can apply Rolle’s theorem to ϕ0 on [0, c1 ] to obtain c2 ∈ (0, c1 ) such that
ϕ00 (c2 ) = 0.
By (12.5), ϕ00 satisfies the hypothesis of Rolle’s theorem on [0, c2 ], so we can continue
in this way. After n + 1 applications of Rolle’s theorem, we obtain cn+1 ∈ (0, h)
f (n+1) (ξ) n+1
such that ϕ(n+1) (cn+1 ) = 0. From (12.6) we obtain Rn (h) = (n+1)! h where
ξ = a + cn+1 .

Problem (F’03, #5). Assume f : R2 → R is a function such that all partial deriva-
tives of order 3 exist and are continuous. Write down (explicitly in terms of partial
derivatives of f ) a quadratic polynomial P (x, y) in x and y such that
3
|f (x, y) − P (x, y)| ≤ C(x2 + y 2 ) 2

for all (x, y) in some small neighborhood of (0, 0), where C is a number that may depend
on f but not on x and y. Then prove the above estimate.

Proof. Taylor expand f (x, y) around (0, 0):

1£ 2 ¤
f (x, y) = f (0, 0) + xfx (0, 0) + yfy (0, 0) + x fxx (0, 0) + 2xyfxy (0, 0) + y 2 fyy (0, 0)
2!
1£ 3 ¤
+ x fxxx (0, 0) + 3x yfxxy (0, 0) + 3xy 2 fxyy (0, 0) + y 3 fyyy (0, 0) + O(x4 )
2
3!
¯1£ ¤¯¯
¯
|f (x, y) − P2 (x, y)| = ¯ x3 fxxx (ξ, η) + 3x2 yfxxy (ξ, η) + 3xy 2 fxyy (ξ, η) + y 3 fyyy (ξ, η) ¯
3!
3
Note that |x3 |, |x2 y|, |xy 2 |, |y 3 | ≤ (x2 + y 2 ) 2 . Also, since 3rd order partial derivatives
are continuous, ∃C1 ∈ R s.t. max{fxxx , 3fxxy , 3fxyy , fyyy } < C41 in some nbd of (0, 0).
Thus,
3 3
|f (x, y) − P2 (x, y)| ≤ | 3!1 (x2 + y 2 ) 2 C1 | ≤ C(x2 + y 2 ) 2 .
Real Analysis and Multivariable Calculus Igor Yanovsky, 2005 39

Problem (F’02, #5). Suppose f : R2 → R has partial derivatives at every point

bounded by A > 0.
(a) Show that there is an M such that
1
|f ((x, y)) − f ((x0 , y0 ))| ≤ M ((x − x0 )2 + (y − y0 )2 ) 2 ~

(b) What is the smallest value of M (in terms of A) for which this always works?
(c) Give an example where that value of M makes the inequality an equality.

Proof. (a) Since | ∂f ∂f

∂x | ≤ A, | ∂y | ≤ A, by the Mean Value Theorem,
|f (x, y) − f (x0 , y)| ≤ A|x − x0 |
|f (x, y) − f (x, y0 )| ≤ A|y − y0 |
|f (x, y) − f (x0 , y0 )| ≤ |f (x, y) − f (x0 , y)| + |f (x0 , y) − f (x0 , y0 )|
√ 1
≤ A(|x − x0 | + |y − y0 |) ≤ A 2(|x − x0 |2 + |y − y0 |2 ) 2
√
(b) This always works for M = A 2.
(c) If |x − x0 | = |y − y0 |, then we have an equality in ~, since then
√ 1
A(|x − x0 | + |y − y0 |) ≤ A 2(|x − x0 |2 + |y − y0 |2 ) 2
√ 1
⇒ 2A|x − x0 | ≤ A 2(2|x − x0 |2 ) 2
⇒ 2A|x − x0 | ≤ 2A|x − x0 |

Problem (F’03, #2). Let f : R → R be infinitely differentiable function. Assume

that ∀x ∈ [0, 1], ∃m > 0, such that f (m) (x) 6= 0.
Prove that ∃M such that the following stronger statement holds:
∀x ∈ [0, 1], ∃m > 0, m ≤ M such that f (m) (x) 6= 0.

Proof. There are uncountably many xα ’s in [0, 1], and for each xα ∈ [0, 1], ∃mα > 0
such that f (mα ) 6= 0 for [xα − ²α , xα + ²α ], for some ²α > 0 (since f (mα ) is continuous).
Let ² = minα (²α ). Partition [0, 1] into n = 1/² subintervals (since ² > 0, n < ∞):

0 < ² = x0 < x1 < . . . < xn = 1 − ² < 1,

such that x0 = ², xi = x0 + i², i = 1, . . . , n. Thus [0, 1] is covered by finitely many

overlapping intervals [xi − ², xi + ²]. For each xi , i = 1, . . . , n, ∃mi > 0 such that
f (mi ) (xi ) 6= 0 on [xi − ², xi + ²]. Take M = max0≤i≤n (mi ). Thus, ∀x ∈ [0, 1], ∃m >
0, m ≤ M such that f (m) (x) 6= 0.
Real Analysis and Multivariable Calculus Igor Yanovsky, 2005 40

Problem (S’03, #4). Consider the following equation for a function F (x, y) on R2 :

∂2F ∂2F
= ~
∂x2 ∂y 2

(a) Show that if a function F has the form F (x, y) = f (x+y)+g(x−y) where f : R → R
and g : R → R are twice differentiable, then F satisfies the equation ~.
(b) Show that if F (x, y) = ax2 + bxy + cy 2 , a, b, c ∈ R, satisfies ~ then
F (x, y) = f (x + y) + g(x − y) for some polynomials f and g in one variable.

Proof. (a) Let ξ(x, y) = x + y, η(x, y) = x − y, so F (x, y) = f (ξ(x, y)) + g(η(x, y)). By
Chain Rule,
∂F df ∂ξ dg ∂η df dg df dg
= · + · = ·1+ ·1= + ,
∂x dξ ∂x dη ∂x dξ dη dξ dη
µ ¶
∂2F ∂ df dg 2
d f ∂ξ 2
d g ∂η d2 f d2 g
= + = 2 · + 2· = + ,
∂x2 ∂x dξ dη dξ ∂x dη ∂x dξ 2 dη 2
∂F df dg ∂2F d2 f d2 g ∂2F ∂2F
and similarly = − , and = + , and thus , = .
∂y dξ dη ∂x2 dξ 2 dη 2 ∂x2 ∂y 2
(b) Suppose F (x, y) = ax2 + bxy + cy 2 , a, b, c ∈ R satisfies ~, then
∂F ∂2F ∂F ∂2F
= 2ax + by ⇒ = 2a, = 2cy + bx ⇒ = 2c ⇒ a = c.
∂x ∂x2 ∂y ∂y 2
(x + y)2 + (x − y)2 (x + y)2 − (x − y)2
F (x, y) = ax2 + bxy + ay 2 = a(x2 + y 2 ) + bxy = a +b .
2 4

12.5 Lagrange Multipliers

Theorem. Let f and g be C 1 on R2 . Suppose that f attains its maximum or minimum
value on the zero set S of g at the point p where ∇g(p) 6= 0. Then

∇f (p) = λ∇g(p)

for some number λ.

Problem (S’03, #5). Consider the function F (x, y) = ax2 + 2bxy + cy 2 on the set
A = {(x, y) : x2 + y 2 = 1}.
(a) Show that F has a maximum and minimum on A.
(b) Use Lagrange multipliers to show that if the maximum of F µ on A occurs
¶ at a point
a b
(x0 , y0 ), then the vector (x0 , y0 ) is an eigenvector of the matrix .
b c

Proof. (a) Since F is continuous and the circle is closed and bounded, F attains both
a maximum and minimum values on the unit circle g(x, y) = x2 + y 2 − 1 = 0.
(b) Applying the above theorem, we obtain the three equations (for x, y, λ)

2ax + 2by = 2λx, 2bx + 2cy = 2λy, x2 + y 2 = 1.

µ ¶µ ¶ µ ¶ µ ¶
a b x x x0
=λ , and is an eigenvector of this matrix.
b c y y y0
Real Analysis and Multivariable Calculus Igor Yanovsky, 2005 41

13 Successive Approximations and Implicit Functions

13.1 Contraction Mappings
C ⊆ Rn . The mapping ϕ : C → C is a contraction mapping with contraction
constant k < 1 if

|ϕ(x) − ϕ(y)| ≤ k|x − y| ∀x, y ∈ C.

Contraction Lemma. If M is complete and ϕ : M → M is a contraction mapping

with k < 1, then ϕ has a unique fixed point x∗ .

Proof. |xn+1 − xn | = |ϕ(xn ) − ϕ(xn−1 )| ≤ k|xn − xn−1 | ≤ k n |x1 − x0 |.

If m > n > 0, then
|xm − xn | ≤ |xm − xm−1 | + · · · + |xn+1 − xn | ≤ (k m−1 + · · · + k n )|x1 − x0 |
kn
≤ k n |x1 − x0 |(1 + k + k 2 + · · · ) ≤ |x1 − x0 |.
1−k
Thus the sequence {xn } is a Cauchy sequence, and therefore converges to a point x∗ .
Letting m → ∞, we get
kn
|x∗ − xn | ≤ |x1 − x0 |.
1−k
Since ϕ is contraction, it is continuous. Therefore,
ϕ(x∗ ) = lim ϕ(xn ) = lim xn+1 = x∗ .
n→∞ n→∞

Ifx∗∗ were another fixed point of ϕ, we would have |x∗ − x∗∗ | = |ϕ(x∗ ) − ϕ(x∗∗ )| ≤
k|x∗ − x∗∗ |. Since k < 1, it follows that x∗ = x∗∗ , so x∗ is the unique fixed point of
ϕ.

13.2 Inverse Function Theorem

Lemma. 17 Suppose f : Rn → Rn is C 1 , 0 ∈ W , f (0) = 0, f 0 (0) = I. Suppose
||f 0 (x) − I|| ≤ ², ∀x ∈ Br . Then

B(1−²)r ⊂ f (Br ) ⊂ B(1+²)r . (13.1)

T
If U = Br f −1 (B(1−²)r ), then f : U → B(1−²)r is bijection, and the inverse mapping
g : V → U is differentiable at 0.

Proof. ||f 0 (x) − I|| ≤ ² < 1, ∀x ∈ Br . Apply MVT18 to g(x) = f (x) − x. a, b ∈ Br ,

then

||(f (b)−b)−(f (a)−a)|| = ||g(b)−g(a)|| ≤ ||g 0 (ξ)||||b−a|| = ||f 0 (ξ)−I||||b−a|| ≤ ²||b−a||

(13.2)

(1 − ²)||b − a|| ≤ ||f (b) − f (a)|| ≤ (1 + ²)||b − a|| (13.3)

The left-hand inequality shows that f is 1-1 on Br . The right-hand inequality (with
a = 0) shows f (Br ) ⊂ B(1+²)r .
17
Problem F’01, # 6.
18
f : U ⊂ Rn → Rm is C 1 ⇒ ||f (b) − f (a)|| ≤ ||b − a|| maxx∈L ||f 0 (x)||.
Real Analysis and Multivariable Calculus Igor Yanovsky, 2005 42

To show B(1−²)r ⊂ f (Br ), we use contraction mapping theorem. Given y ∈ B(1−²)r ,

define ϕ : Rn → Rn by

ϕ(x) = x − f (x) + y.

We want to show that ϕ is a contraction mapping of Br ; its unique fixed point will
then be the desired point x ∈ Br such that f (x) = y.
To see that ϕ maps Br into itself:

||ϕ(x)|| ≤ ||f (x)−x||+||y|| ≤ (by (13.2) with a = 0) ≤ ²||x||+(1−²)r ≤ ²r+(1−²)r = r,

so if x ∈ Br , then ϕ(x) ∈ Br . Thus, ϕ(Br ) ⊆ Br .

To see ϕ : Br → Br is a contraction mapping, note that

||ϕ(b) − ϕ(a)|| = ||f (b) − f (a) − (b − a)|| ≤ ²||b − a||

Thus, ϕ has a unique fixed point

T −1x∗ , ϕ(x∗ ) = x∗ , such that f (x∗ ) = y. From the
statement of theorem, U = Br f (B(1−²)r ). f is a bijection of U onto B(1−²)r .
It remains to show that g : V → U is differentiable at 0, where g(0) = 0. Need to show

||g(h) − g(0) − h|| ||g(h) − h||

lim = lim = 0.
|h|→0 ||h|| |h|→0 ||h||

This will prove that g 0 (0) = I. Applying (13.2) with a = 0, b = g(h), h = f (b), we
get
² ²
||g(h) − h|| ≤ ²||b|| ≤ by (13.3) ≤ ||f (b)|| = ||h||
1−² 1−²
||g(h)−h||
Therefore, lim|h|→0 ||h|| = 0, with g 0 (0) = I.

Theorem. 19 Suppose f : W ⊂ Rn → Rn is C 1 , a ∈ W , b = f (a), and the matrix

f 0 (a) is nonsingular.20 Then there exist21 open sets U ⊂ W of a and V of b, such that
f maps U bijectively onto V . ( ∃ 1-1 C 1 mapping g : V → W such that

g(f (x)) = x f or x ∈ U,
f (g(y)) = y f or y ∈ V. )

Also, for all y ∈ V (y = f (x)), g = f −1 satisfies g 0 (y) = g 0 (f (x)) = f 0 (x)−1 .

Proof. Fix a ∈ U and let b = f (a). Put T = f 0 (a), a matrix / linear map. Define

f˜(h) = T −1 (f (a + h) − b) = T −1 (f (a + h) − f (a))

Note f˜(0) = T −1 · (0) = 0;

f˜0 (0) = T −1 f 0 (a) = T −1 T = I. Thus, by previous Lemma, ∃U0 open, 0 ∈ U0 , such
that f˜ : U0 → V0 is a bijection, 0 ∈ V0 .
f˜ maps U0 bijectively onto V0 containing 0. Lets express f in terms of f˜.
f (a + h) − f (a) = T f˜(h). Let x = a + h:
f (x) − f (a) = T f˜(x − a),
19
f is locally one-to-one in E ≡ each point x ∈ E has a neighborhood in which f is 1-1.
20 ∂fi
Jacobian of f = | det f 0 (a)| = | ∂xj
(a)| 6= 0.
21 1
i.e. A C map f : W → V is locally invertible at a ≡ there exist open sets U ⊂ W of a and V of
b = f (a), and a C 1 map g : V → U such that f and g are inverse to each other.
Real Analysis and Multivariable Calculus Igor Yanovsky, 2005 43

f (x) = T f˜(x − a) + f (a),

f : U0 + a → T f˜(V0 ) + f (a), bijection.
| {z }
U
Let’s compute f −1 :
Let y = T f˜(x − a) + f (a)
T f˜(x − a) = y − f (a)
f˜(x − a) = T −1 (y − f (a))
x − a = f˜−1 (T −1 (y − f (a))) = g̃(T −1 (y − f (a)))
f −1 (y) = x = g̃(T −1 (y − f (a))) + a
(f −1 )0 (y) = (g̃)0 (T −1 (y − f (a))) · T −1
(f −1 )0 (b) = (g̃)0 (0) · T −1 = T −1 = f 0 (a)−1 .
Real Analysis and Multivariable Calculus Igor Yanovsky, 2005 44

Problem (S’02, #7; W’02, #7; F’03, #6). Suppose F : R2 → R2 is C 1 and that
the Jacobian matrix of F is everywhere nonsingular. Suppose that F (0) = 0 and that
||F (x, y)|| ≥ 1 for all (x, y) with ||(x, y)|| = 1. Denote U = {(x, y) : ||(x, y)|| < 1}.
Prove that F (U ) ⊃ U . T
Hint: Show that F (U ) U is both open and closed in U .
T T
Proof. Since TU is connected, clopenness of F (U ) U in U implies that either F (U ) U =
U or F (U ) U =Tφ. Since there exists a point, namely 0 such that it T is inside both U
and F (U ), F (U ) T U cannot be empty, and thus clopenness of F (U ) U in U would
imply that F (UT ) U = U (which would mean U ⊆ F (U )).
1) Show F (U ) U is open in U .
F (U ) is open in R2 . Say y0 ∈ F (U ), y0 = F (x0 ), F 0 (x0 ) invertible. By inverse func-
tion thm, F maps open set T U0 onto open set V0 ; x0 ∈ U0 ⇒ y0 = F (x0 ) ∈ V0 .
y0 ∈ V0 ⊆ F (U )T⇒ F (U ) U is open in U .
2) Show F (U ) T U is closed in U .
Say xn ∈ F (U ) U , xn → x∗ ∈ U .
xn = F (yn ), yn ∈ U ⊂ U .
There is a subsequence ynk → y ∈ U .
Since F is continuous, F (ynk ) → F (y) = x∗ .
||y|| = 1 ⇒ ||F (y)|| ≥ 1 ⇒ F (y) = x∗ ∈ / U . Contradiction.

13.3 Implicit Function Theorem

(x, y) = (x1 , . . . , xm , y1 , . . . , yn )
Theorem. Suppose G : Rm+n → Rn is C 1 . G(a, b) = 0 for some point (a, b). Partial
derivative matrix ∂G
∂y (a, b) is invertible.
Then there exist open sets U of a in Rm and W of (a, b) in Rm+n and a C 1 mapping
h : U → Rn , such that y = h(x) solves the equation G(x, y) = 0 in W .
Example. f (x, y, z) = 0, ∂f
∂z 6= 0 ⇒ can solve for z = h(x, y), f (x, y, h(x, y)) ≡ 0.
d
Example. m = n = 1. G(x, y) = 0, y = h(x) ⇒ G(x, h(x)) = 0 ⇒ dx G(x, h(x)) = 0
∂G ∂G 0 0 ∂G ∂G ∂G
⇒ ∂x + ∂h h (x) = 0 ⇒ h (x) = − ∂x / ∂h , ∂h =
6 0.
2 2 2 2
√
Say G(x, y) = x + y − 1, x + y − 1 = 0 ⇒ y = ± 1 − x2 , problem at (1, 0).
∂G
∂h cannot be equal to 0.
Problem (S’02, #6). Suppose f : R3 → R is C 1 with ∇f 6= 0 at 0. Show that there
are two other C 1 functions g, h : R3 → R, such that the function
(x, y, z) → (f (x, y, z), g(x, y, z), h(x, y, z))
from R3 to R3 is one-to-one on some neighborhood of 0.
Proof. ∇f 6= 0 ⇒ one of ∂f ∂f ∂f
∂x , ∂y , ∂z is not 0. F = (f, g, h).
 ∂f ∂f ∂f 
∂x ∂y ∂z
 ∂g ∂g ∂g 
∇F =  ∂x ∂y ∂z 
∂h ∂h ∂h
∂x ∂y ∂z

Need to produce functions g, h such that the matrix above is invertible.

If ∂f
∂x (0) 6= 0, let g(x, y, z) = z, h(x, y, z) = y. Then
 ∂f 
∂x 0 0
 0 0 1  and F = F (f (x, y, z), z, y).
0 1 0
Real Analysis and Multivariable Calculus Igor Yanovsky, 2005 45

Similarly, we can find a set of functions g, h by choosing a matrix in each of the other
two cases, i.e. when ∂f ∂f
∂y (0) 6= 0 and ∂z (0) 6= 0.

Problem (F’02, #6). Suppose F : R3 → R2 is C 1 . Suppose for some v0 ∈ R3 and

x0 ∈ R2 that F (v0 ) = x0 and F 0 (v0 ) : R3 → R2 is onto. Show that there is a C 1
function γ : (−², ²) → R3 for some ² > 0, such that

(i) γ 0 (0) 6= 0 ∈ R3 , and

(ii) F (γ(t)) = x0 for all t ∈ (−², ²).

Proof. Since F 0 (v0 ) is onto, the matrix F 0 (v0 ) has rank 2. So, 2 of the 3 columns of
F 0 (v0 ) are linearly independent.
" #
∂f1 ∂f1 ∂f1
F 0 (v0 ) = ∂x1
∂f2
∂x2
∂f2
∂x3
∂f2 .
∂x1 ∂x2 ∂x3 v0

Assume the last two columns are linearly independent.

Consider the function G(x1 , x2 , x3 ) = F (x1 , x2 , x3 ) − x0 . Write (x1 , x2 , x3 ) = (s1 , s2 )
∂G
where s1 = x1 , s2 = (x2 , x3 ). Write v0 = (u1 , u2 ). Then G(u1 , u2 ) = 0 and ∂s 2
(v0 )
is invertible. By Implicit Function Theorem, ∃ ² > 0 and h ∈ C , such that h :1

(u1 − ², u1 + ²) → R2 and G(s1 , h(s1 )) = 0, ∀s1 ∈ (u1 − ², u1 + ²).

⇒ F (s1 , h(s1 )) = x0 , ∀s1 ∈ (u1 − ², u1 + ²).
Define γ(t) : (−², ²) → R3 by γ(t) = (u1 + t, h(u1 + t)). Then γ(t) is a differentiable
curve satisfying
(i) γ 0 (t) = (1, h0 (u1 + t)) 6= 0,
(ii) F (γ(t)) = x0 for all t ∈ (−², ²).
Real Analysis and Multivariable Calculus Igor Yanovsky, 2005 46

13.4 Differentiation Under Integral Sign

∂f
Problem (W’02, #1). f : [a, b] × [c, d] → R ((x, y) → f (x, y)). Suppose ∂y exists on
[a, b] × (c, d) and extends to a continuous function on [a, b] × [c, d]. Let
Z b
F (y) = f (x, y)dx.
a

Then F is differentiable in [a, b] and

Z b
d ∂f
F (y) = (x, y)dx.
dy a ∂y
Z a Z b
d ∂f
⇒ f (x, y)dx = (x, y)dx.
dy b a ∂y
Proof.
¯ Z b ¯ ¯Z bh i ¯¯
¯ F (y + h) − F (y) ∂f ¯ ¯ f (x, y + h) − f (x, y) ∂f
¯ − ¯
(x, y)dx¯ = ¯¯ − (x, y) dx¯¯
¯ h a ∂y a h ∂y
Z b¯ ¯
¯ f (x, y + h) − f (x, y) ∂f ¯
≤ ¯ − (x, y)¯dx ≤ ⇒ ~
¯ h ∂y ¯
a

By MVT, ∃c, 0 < c < 1, such that

f (x, y + h) − f (x, y) ∂f
= (x, y + ch).
h ∂y

||(x, y + ch) − (x, y)|| = c|h| ≤ h.

Since ∂f∂y is continuous on [a, b] × [c, d] ⇒
∂f
∂y is uniformaly continuous on [a, b] × [c, d].
Choose δ such that ||(x, y) − (x0 , y 0 )|| ≤δ ⇒ | ∂f 0 0
∂y (x , y ) −
∂f
∂y (x, y)| ≤ ²
b−a
Z b¯ ¯ Z b
¯ ∂f ∂f ¯ ² ²
⇒ ~ = ¯ ¯
¯ ∂y (x, y + ch) − ∂y (x, y)¯dx ≤ dx = (b − a) = ².
a a b−a b−a