Notes on Mathematical Analysis
Notes on Mathematical Analysis
FIN
Contents
1 Real Numbers 3
1.1 Cuts . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
1.2 Limits of a sequence in R . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5
2 Metric Spaces 11
2.1 Definitions and Examples . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11
2.2 Sequences and Completeness . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13
2.3 Continuity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16
2.4 Compactness . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18
2.5 Connectedness . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 20
2.6 Baire Category Theorem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22
2.7 Odds and Ends . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 23
4 Continuity 42
4.1 Limit of Functions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 42
4.2 Continuous Functions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 43
4.3 Discontinuity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 44
4.4 Monotonic Functions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 44
4.5 Semi-continuity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 45
4.6 Odds and Ends . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 46
1
6 Riemann-Steljtes Integral 62
6.1 Existence of the Riemann-Steljtes Integral . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 62
6.2 Extend to bounded functions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 71
6.3 Integration and Differentiation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 72
6.4 Odds and Ends . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 73
2
Chapter 1
Real Numbers
1.1 Cuts
p
Proposition 1.1.1. There is no number r = q in Q such that r2 = 2.
2
Proof. Suppose there is such r = pq with gcd(p, q) = 1. Then 2 = r2 = pq2 ⇐⇒ p2 = 2q 2 . Then 2|p2 ,
which implies that 2|p. Hence, by assuming that p = 2k, we have q 2 = 2k 2 . Similarly, 2|q and therefore
gcd(p, q) ≥ 2, a contradiction.
Definition 1.1.2. A cut in Q is a pair of subsets A, B of Q such that
1. A ∪ B = Q, A, B ̸= ∅, A ∩ B = ∅.
2. For any a ∈ A, b ∈ B, we have a < b.
3. A contains no largest elements, i.e., for any a ∈ A, there exists a′ ∈ A such that a < a′ .
In this case, we use the notation A|B to denote a cut.
Definition 1.1.3. A real number is a cut in Q. The class of real numbers is denoted by R.
Definition 1.1.4. If x = A|B and y = C|D, then we define x ≤ y if A ⊆ C. It is clear that x ≤ y if and
only if D ⊆ B. Also, x < y if x ≤ y but A ̸= C.
Definition 1.1.5. A relation ≤ on a set X is called a linear ordering if for any a, b, c ∈ X:
1. a ≤ a.
2. If a ≤ b, b ≤ c, then a ≤ c.
3. If a ≤ b and b ≤ a, then a = b.
4. a ≤ b or b ≤ a.
Proposition 1.1.6. ≤ is a linear ordering on R.
Proof. It is clear that 1,2 in the definition 1.1.5 are satisfied. For 3, suppose x = X1 |X2 , y = Y1 |Y2 ∈ R with
x ≤ y, y ≤ x. Then X1 = Y1 and therefore X2 = Y2 ; hence, x = y.
As for 4, we claim that for any cut A|B, if a ∈ A, then a′ ∈ A for any a′ ≤Q a. Indeed, if a′ ∈ / A, then
a ∈ B. However, this implies that a′ > a, which is a contradiction. Hence, given cuts x = X1 |X2 , y = Y1 |Y2
′
with x ̸≤ y, there exists a ∈ X1 \Y1 . Then a ∈ Y2 and therefore y1 <Q a for any y1 ∈ Y1 . Since a ∈ X1 ,
y ∈ Y1 and y ≤ x, as required.
Corollary 1.1.6.1. A cut A|B contain from below, that, if a ∈ A, then a′ ∈ A for any a′ ≤Q a.
Definition 1.1.7. M ∈ R is an upper bound of a set S ⊆ R if for any s ∈ S, we have s ≤ M . If an upper
bound exists, then S is called bounded above. Moreover, if M is the smallest such upper bounded,i.e., for
any upper bound M ′ of S, M ≤ M ′ , then we say M is the least upper bound or the supremum of S,
denoted by sup(S).
3
Theorem 1.1.8. The set R satisfies the supremum property, that is, for any nonempty set S ⊆ R which
is bounded above, then sup(S) exists.
Proof. Let C ⊆ R be any non-empty collection of cuts which is bounded above, say be the cut X|Y . Define
C = {a ∈ Q : a ∈ A for some cut A|B ∈ C}
and D = Q\C. We shall verify that (C, D) is indeed a cut:
Since C is bounded above, there is a cut M = M1 |M2 such that A ⊆ M1 for any cut A|B ∈ C. Since
M2 ̸= ∅, there exists a t ∈ Q such that t ∈
/ A for any cut A|B ∈ C. Hence, C ̸= Q and therefore D ̸= ∅.
Also, since S is not empty set, C ̸= ∅. It is clear that C ∪ D = Q and C ∩ D = ∅.
Given c ∈ C, d ∈ D, we want to prove that c < d. Since d ∈ D, for any cut A|B ∈ C, d ∈ B. Then
c < d by definition.
Suppose a ∈ C, then there exists a cut A|B ∈ C such that a ∈ A. Then there exists a′ ∈ A such that
a′ > a. Note that a′ ∈ C, then we are done.
Let z = C|D. Clearly, it is an upper bound for C since A ⊆ C for any cut A|B ∈ C. Let z ′ = C ′ |D′ be
any upper bound for C. By assumption, we have A|B ≤ C ′ |D′ for all A|B ∈ C, we see that A for member C
is contained in C ′ . Hence C ⊆ C ′ , so z ≤ z ′ . Then z = sup(S).
Definition 1.1.9. For a rational r ∈ Q, we define r∗ = (A, B), where A = {a ∈ Q : a < r} and B = Q\A.
It is clear that r∗ is a cut.
Definition 1.1.10. For x = A|B, y = C|D ∈ R, we define x + y = E|F by
E = {r ∈ Q : r = a + c for some a ∈ A, c ∈ C.}
And −x = G|H by G = {r ∈ Q : r = −b for some b ∈ B\ max(B)} and H = Q\G. It is clear that (R, +) is
an abelian group with identity elements 0∗ .
Definition 1.1.11. We say that a cut is positive if 0∗ < x and negative if x < 0∗ .
Definition 1.1.12. If x = A|B and y = C|D are postive cuts then the product x · y = E|F is given by
E = {r ∈ Q : r ≤Q 0 or r = ac for some a, c >Q 0, r = ac}
and F = Q\E. If x > 0 and y < 0∗ , we define x · y by x · y = −(x · (−y)). When x < 0∗ is defined similarly.
∗
Thus, x ≤ |x|.
Proposition 1.1.15. For any x, y ∈ R, we have |x + y| ≤ |x| + |y|.
Proof. Since x + y, −x − y ≤ |x| + |y|, we are done.
Proposition 1.1.16. For any x ∈ R, |x| ≥ 0 and |x| = 0 if and only if x = 0.
Proposition 1.1.17. For any x, y ∈ R, |x − y| = |y − x|.
4
1.2 Limits of a sequence in R
Definition 1.2.1. Let (an ) be a sequence in R. We say that an converges to the limit b ∈ R, denoted by
an → b or limn an = b, if for any ϵ > 0, there exists N ∈ N such that
|an − b| < ϵ
for any n ≥ N . Also, we say a double indexed sequence (anm ) in R converges to the limit b ∈ R, denoted
by an,m → b or limn,m an,m = b, if for any ϵ > 0, there exists N ∈ N such that for any n, m ≥ N , we have
|an,m − b| < ϵ.
|an − am | < ϵ.
5
Hence, take T = max{nN , M }, then for any n ≥ T , we have
|an − b| ≤ |an − anN | + |anN − b| < ϵ,
as required.
Definition 1.2.7. For a sequence (an ) in R, we say that an is increasing if an ≤ an+1 for any n ∈ N;
decreasing if an ≥ an+1 for any n ∈ N. If (an ) is increasing or decreasing, we say that (an ) is monotone.
Lemma 1.2.8. Let (an ) be a sequence in R. Then (an ) has a monotone subsequence.
Proof. For (an ), we say that ak is a peak if for any n ≥ k, ak ≥ an . Set an1 = a1 and i = 1, then we will
construct a monotone subsequence by the below algorithm:
If there is a m > nj such that anj < am , then set anj+1 = am and nj ← nj+1 .
Otherwise, anj is a peak, then set mi = nj and n1 ← nj+1 , i ← i + 1.
If there is infinitely many nj ’s, then there is an increasing subsequence (anj ). If not, then there are infinitely
many peaks and (ami ) is a decreasing subsequence.
Proposition 1.2.9. Let (an ), (bn ), (cn ) be sequences in R.
1. If an → a for some a ∈ R, then can → ca for any c ∈ R.
2. If an → a, bn → b for some a, b ∈ R, then an + bn → a + b.
3. If an → a, bn → b for some a, b ∈ R, then an bn → ab.
1
4. If an → a for some a ∈ R and an , a ̸= 0, then an → a1 .
5. If a2n , a2n+1 → a for some a ∈ R, then an → a.
6. an → a for some a ∈ R if and only if ani → a for any subsequence (anj ).
7. If for any subsequence (ani ), there is a subsequence (anij ) converges to a ∈ R, then an → a.
8. If (an ), (bn ) and (cn ) are two sequences such that an ≤ bn ≤ cn for sufficient large n, both (an ) and
(cn ) converges and lim an = lim cn , then (bn ) converges and lim an = lim bn = lim cn .
9. If (an ), (bn ) are two convergent sequences and an ≤ bn for all sufficiently large n, then lim an ≤ lim bn .
10. If (an ), (bn ) are two sequences, an ≤ bn for all sufficiently large n, then:
(a) if an → ∞ as n → ∞, then bn → ∞;
(b) if bn → −∞ as n → ∞, then an → −∞.
Proof. Let (an ), (bn ) and (cn ) be sequences in R.
1. Let an → a for some a ∈ R. If c = 0, then can = 0 for all n ∈ N and therefore can → 0 = ca. If
c ̸= 0, then for any ϵ > 0, there exists N ∈ N such that for any n ≥ N , we have |an − a| < cϵ ⇐⇒
|can − ca| < ϵ. Then can → ca.
2. Let an → a, bn → b for some a, b ∈ R, then there exists N, M ∈ N such that for any n ≥ N, m ≥ M ,
|an − a| , |bm − b| < 2ϵ . Therefore, for n ≥ max{N, M },
|(an + bn ) − (a + b)| ≤ |an − a| + |bn − b| < ϵ.
Hence, an + bn → a + b.
ϵ
3. By 1.2.4, there exists M > 0 such that |an | < M . Then there exists N1 ∈ N such that |bn − b| < 2M
whenever n ≥ N1 . Hence,
ϵ
|an bn − ab| = |an bn − an b + an b − ab| ≤ |an | |bn − b| + |b| |an − a| < + |b| |an − a| .
2
If b = 0, then we are done. Otherwise, there exists N ′ ∈ N such that |an − a| < ϵ
2|b| whenever n ≥ N2 ,
then take N = max{N1 , N2 } and for any n ≥ N , we have |an bn − ab| < ϵ.
6
|a|
4. Since a ̸= 0, |a| > 0. Therefore, there exist N ∈ N such that |an − a| < 2 whenever n ≥ N . Hence,
|a| 1 2 ϵ|a|2
|an | > 2 ⇐⇒ < whenever n ≥ N . Also, there exists M ∈ N such that |an − a| <
|an | |a| 2
′
whenever n ≥ M . Take N = max{N, M }, then
1 1 |an − a| 2 |an − a|
− < < 2 <ϵ
an a |an a| |a|
whenever n ≥ N ′ .
5. Let a2n , a2n+1 → a for some a ∈ R. Then there exists N1 , N2 ∈ N such that whenever 2n + 1 ≥ N1 and
2n ≥ N2 , we have |a2n+1 − a| < ϵ and |a2n − a| < ϵ. Take N = max{N1 , N2 }, then for any n ≥ N ;
therefore, if n = 2k + 1 for some k ∈ N, then 2k + 1 ≥ N1 and |a2k+1 − a| < ϵ; otherwise, n = 2k for
some k ∈ N, then 2k ≥ N2 and |a2k − a| < ϵ. In either case, we have |an − a| < ϵ. Thus, an → a.
6. Let an → a for some a ∈ R and (ani ) is a subsequence. Then there exists N ∈ N such that for any
n ≥ N , we have |an − a| < ϵ. Since ni ≥ i for all i ∈ N, |ani − a| < ϵ for any i ≥ N . Conversely, since
(an ) itself is a subsequence of (an ), we are done.
7. If an ̸→ a, then exists ϵ0 > 0 such that for any k ∈ N, there exists nk ≥ k such that |ank − a| ≥ ϵ0 .
Without loss of generality, suppose (nk ) is increasing, then (ank ) is a subsequence of (ak ). Then for
any subsequence (anki ), anki − a ≥ ϵ0 , then we are done by contrapositive.
8. Suppose for any n ≥ N1 , we have an ≤ bn ≤ cn and let lim an = lim cn = L ∈ R. Now since
lim an = lim cn = L, there exists N2 ∈ N such that L − 2ϵ < an ≤ cn < L + 2ϵ for any n ≥ N2 . Hence,
take N = max{N1 , N2 }, for any n ≥ N , we have |bn − L| < ϵ and therefore lim bn = L, as required.
9. Let lim an = L and lim bn = M and suppose, to get a contradiction, L > M . Then L − M > 0. Since
lim an = L, there exists N1 ∈ N such that |an − L| < L−M
2 =⇒ an > L+M 2 > M for any n ≥ N1 .
Then an − M > 0 for all n ≥ N1 . Fix n ≥ N1 , since lim bn = M , there exists N2 ∈ N such that for
any m ≥ N2 , we have |bm − M | < an −M
2 =⇒ bm < an +M2 < an . Hence, take k > max{N1 , N2 }, then
bk < ak , a contradiction.
10. Suppose an ≤ bn and an → ∞. Then for amy M > 0, there exists N ∈ N such that bn ≥ an > M for
any n ≥ N , and therefore bn → ∞. For the condition bn → −∞ is similar.
1. s is an upper bound of S.
2. for any ϵ > 0, there exists x ∈ S such that s − ϵ < x ≤ s.
Then sup(S) exists and is equal to s.
Theorem 1.2.11 (Monotone convergent theorem). Let (an ) be a bounded monotone sequence in R. Then
(an ) converges.
Proof. First, suppose an is increasing and bounded above. Hence the supremum sup({an : n ∈ N}) exists
by 1.1.8, say a ∈ R. Given ϵ > 0, by 1.2.10, there exists N ∈ N such that |aN − a| < ϵ. Then since an is
increasing, for any n ≥ N , |an − a| < ϵ.
Now for decreasing subsequence an bound below, notice that −an is increasing and bound above, then
we have −an → a for some a ∈ R by the above discussion. By 1.2.9, an → −a, as required.
7
Corollary 1.2.11.1. Let (xn ) be an increasing (resp. decreasing) sequence in R. Then:
1. If (xn ) is bounded above (resp. below), then (xn ) converges.
2. If (xn ) is not bounded above (resp. below), then (xn ) diverges to ∞ (resp. −∞).
Proof. Suppose (xn ) is bounded above and increasing. Then by 1.2.11, (xn ) converges.
Now suppose (xn ) is not bounded above and increasing, then for any M > 0, there exists m ∈ N such
that am > M . Hence, for any n ≥ m, xn ≥ xm > M .
Theorem 1.2.12 (Bolzano-Weierstrass). Suppose (an ) is a bounded sequence in R, then (an ) has a conver-
gent subsequence.
Theorem 1.2.14 (Archimedean property). For any x, y ∈ R with x > 0, there exists n ∈ N such that
nx > y.
Proof. Suppose not. Then the set S = {nx : n ∈ N} is bounded above by y. Hence, by 1.1.8, sup(S) exists,
say s ∈ R. Therefore, there exists m ∈ N such that s − x < mx ≤ s since x > 0 and 1.2.10. However, we
have (m + 1)x > s and (m + 1)x ∈ S, a contradiction.
Corollary 1.2.14.1 (denseness of Q). Suppose x, y ∈ R with y > x, then the interval (x, y) contains a
rational number.
Proof. Since y > x, by 1.2.14, there exists n ∈ N such that n(y − x) > 1 ⇐⇒ ny > nx + 1. By 1.2.14
again, there exists m′ ∈ N such that m′ > nx, −m′′ < nx and therefore nx ∈ (−m′′ , m′ ). Then there exists
m ∈ N such that m − 1 ≤ nx < m. Hence, x < m m
n . Moreover, since ny > nx + 1 ≥ m, we have y > n , as
required.
1
Corollary 1.2.14.2. Given ϵ > 0, there exists n ∈ N such that n < ϵ.
1
Proof. Since ϵ > 0, by 1.2.14, there exists nϵ > 1 ⇐⇒ ϵ > n.
Theorem 1.2.15 (ϵ-principle). Let a, b ∈ R. If a ≤ b + ϵ for any ϵ > 0, then a ≤ b. In particular, if for any
ϵ > 0, |x − y| ≤ ϵ, then a = b.
a−b a−b a+b a+b
Proof. Suppose not. Then 2 > 0. Hence, a ≤ b + 2 = 2 . However, 2 < a, a contradiction.
Definition 1.2.16. For a subset S ⊆ R, we say that the infimum of S is the largest lower bound of S,
denoted by inf(S).
Proposition 1.2.17. If a subset S ⊆ R is nonempty and bounded below, then inf(S) exists.
Proof. Since S ⊆ R is bounded below and nonempty, −S = {−s : s ∈ S} is bounded above and nonempty.
Then by 1.1.8, sup(−S) exists. Now we clam that − sup(−S) = inf(S). It is clear that − sup(−S) is a
lower bound of S. Now suppose l is a lower bound of S, then −l is an upper bound of −S and therefore
sup(−S) ≤ −l ⇐⇒ l ≤ − sup(−S). Hence, inf(S) = − sup(−S).
Proof. By 1.2.17, inf(S) exists. Suppose s is an element that satisfies the properties above and u is a lower
bound of S. If s < u, then by 2, there exists x ∈ S such that s ≤ x < s + u−s u+s u+s
2 = 2 . But since 2 < u,
u+s
2 is a lower bound of S, a contradiction.
8
Proposition 1.2.18. Let A and B ⊆ R be two nonempty sets.
1. If sup(A) and sup(B) exist (as finite numbers) and for any a ∈ A, there exists b ∈ B such that a ≤ b,
then sup(A) ≤ sup(B).
2. If inf(A) and inf(B) exist (as finite numbers) and for any b ∈ B, there exists a ∈ A such that a ≤ b,
then inf(A) ≤ inf(B).
Proof.
1. Suppose sup(A) > sup(B), then by 1.2.10, there exists a ∈ A such that sup(B) < a ≤ sup(A), a
contradiction.
2. Suppose inf(A) > inf(B), then by 1.2.17.1, there exists b ∈ B such that inf(A) > b ≥ inf(B), a
contradiction.
Proposition 1.2.19. Let X be a nonempty set and f, g : X → R be two functions. If f (x) ≤ g(x) for any
x ∈ X, then supx∈X f (x) ≤ supx∈X g(x) and inf x∈X f (x) ≤ inf x∈X g(x).
Proof. It is equivalent to prove that sup f (X) ≤ sup g(X) and inf f (X) ≤ inf g(X).
1. For the case supx∈X g(x) = ∞ is clear. Now suppose supx∈X g(x) ∈ R (notice that since X ̸= ∅, g(X) ̸=
∅ and therefore sup g(X) ̸= −∞). Since f (x) ≤ g(x) ≤ sup g(X) for all x ∈ X, sup g(X) is an upper
bound of f (X). Hence, sup f (X) exists and sup f (X) ≤ sup g(X).
2. For the case inf f (X) = −∞ is clear. Now suppose inf f (X) ∈ R. Since inf f (X) ≤ f (x) ≤ g(x) for
any x ∈ X, inf f (X) is a lower bound of g(X). Hence, inf g(X) exists and inf f (X) ≤ inf g(X).
1. if c ≥ 0, then sup(cX) = c sup(X) and inf(cX) = c inf(X), where cX = {cx : x ∈ X} (we define that
0 · ∞ = 0 = 0 · (−∞) here);
2. if c ≤ 0 and X is bounded, then sup(cX) = c inf(X) and inf(cX) = c sup(X);
3. inf(X)+inf(Y ) ≤ inf(X +Y ) ≤ sup(X +Y ) ≤ sup(X)+sup(Y ), where X +Y = {x+y : x ∈ X, y ∈ Y }.
Proof.
1. If c = 0, then cX = {0} and therefore sup(cX) = inf(cX) = 0 = 0 sup(X) = 0 inf(X). Now let c > 0:
If X is not bounded above, then cX is not bounded above, hence sup(cX) = c sup(X) = ∞.
If X is bounded above, then sup(cX), sup(X) ∈ R. Notice that since x ≤ sup(X) for all x ∈ X,
cx ≤ c sup(X) for all x ∈ X, then c sup(X) is an upper bound of cX. Now given ϵ > 0, by 1.2.10,
there exists x ∈ X such that sup(X) − cϵ < x ≤ sup(X) ⇐⇒ c sup(X) − ϵ < cx ≤ c sup(X). By
1.2.10 again, c sup(X) = sup(cX).
The cases when X is (not) bounded below are similar.
2. If c = 0 and X is bounded, then sup(cX) = c inf(X) = inf(cX) = c sup(X) = 0. Now let c < 0:
Since X is bounded, sup(cX), inf(X) ∈ R. Notice that since inf(X) ≤ x for all x ∈ X, c inf(X) ≥
cx for all x ∈ X and therefore c inf(X) is an upper bound of X. Given ϵ > 0, by 1.2.17.1, there
ϵ
exists x ∈ X such that inf(X) ≤ x < inf(X) + −c =⇒ c inf(X) − ϵ < cx ≤ c inf(X). By 1.2.10,
c inf(X) = sup(cX).
The proof of inf(cX) = c sup(X) is similar.
9
Definition 1.2.21. For a nonempty S ⊆ R that is not bounded above, we define sup(S) = ∞; not bounded
below, we define inf(S) = −∞. By convention, we define sup(∅) = −∞ and inf(∅) = ∞.
n
Proposition 1.2.22. For every
√ real x > 0 and every integer n > 0 there is a unique y > 0 such that y = x.
This number y is denoted by n x.
Proof. The uniqueness is clear since 0 < y1 < y2 implies y1n < y2n . Let
E = {t ∈ R : t > 0, tn < x}.
x n x
Since ( 1+x ) < 1+x < t, E is nonempty. Since (1 + x)n > x, hence (1 + x) is an upper bound of E. Then
by 1.1.8, sup(E) exists, say y ∈ R.
To prove that y n = x, we will show that each y n < x and y n > x leads to a contradiction.
n x−y n
Assume that y < x. Then δ := n(y+1)n+1 > 0. Now take h such that 0 < h < min{δ, 1}. Then
10
Chapter 2
Metric Spaces
11
S
Om the other hand, since G ∩ Y = p∈S (BX (r; rp ) ∩ Y ) ⊆ S, we have G ∩ Y = S. Conversely, suppose there
is a open set G ⊆ X such that S = G ∩ Y , then for p ∈ S, there exists r > 0 such that B(p; rp ) ⊆ G =⇒
BY (p; rp ) ⊆ S. Then S is open in Y , as required.
Definition 2.1.6. Let (X, d) be a metric space. Then the interior of A, denoted by int(A), is the set
defined by [
int(A) = {G : G is open}.
G⊆A
∂A = A ∩ X\A.
12
Proof. Since int(A) is open, if A = int(A), then A is clearly open. Conversely, suppose A is open, then for
any x ∈ A, there exists r > 0 such that B(x; r) ⊆ A. Hence, by 2.1.7, x ∈ int(A). Hence, A = int(A).
Since A is closed, if A = A, A is closed. Conversely, suppose A is closed and x ∈ A. By 2.1.7, for any
r > 0, B(x; r) ∩ A ̸= ∅. Then x ∈ A as otherwise there exists r > 0 such that B(x; r) ⊆ (X\A) since (X\A)
is open. Therefore, A ⊆ A =⇒ A = A.
Suppose x ∈ A. If x ∈ int(X\A), then by 2.1.7, there exist r > 0 such that B(x; r) ⊆ X\A, a contradic-
tion. Conversely, if x ∈ X\ int(X\A), then for any r > 0, B(x; r) ∩ A ̸= ∅, then x ∈ A.
Let B = X\A, then B = X\ int(X\B) ⇐⇒ X\A = X\ int(A) ⇐⇒ int(A) = X\X\A.
Suppose x ∈ ∂A, then x ∈ A. Moreover, if x ∈ int(A), there exists r > 0 such that B(x; r) ⊆ A, by
2.1.7, x ∈/ ∂A. Therefore, x ∈ X\ int(A). Hence, x ∈ A ∩ (X\ int(A)) = A\ int(A). Conversely, suppose
x ∈ A\ int(A), then for any r > 0 such that B(x; r) ∩ A ̸= ∅, B(x; r) ∩ (X\A) ̸= ∅, then x ∈ ∂A.
Suppose A ⊆ B and x ∈ A, then for any r > 0 such that B(x; r) ∩ A ̸= ∅ =⇒ B(x; r) ∩ B ̸= ∅ ⇐⇒
x ∈ B. S S
Suppose x ∈ α∈Γ Gα , then there exists α ∈ Γ such that x ∈ Gα ⊆ α∈Γ Gα . Suppose Γ is finite, say
Sn Sn
Γ = {γ1 , · · · , γn } and x ∈ j=1 Gγj . If x ∈
/ j=1 Gγj , then there exists rj > 0 such that B(x; rj ) ∩ Gγj = ∅.
Take r = min{r1 , · · · , rn } > 0, thus B(x; r) ∩ Gγj = ∅ for any j, a contradiction.
Proof. Suppose D is dense in X and x ∈ X. Then since D = X, for any x ∈ X, r > 0, B(x; r) ∩ D ̸= ∅.
Conversely, suppose for any x ∈ D, r > 0, B(x; r) ∩ D ̸= ∅, by 2.1.7, x ∈ D and therefore X = D.
T
Proposition
T 2.1.12. Let (X, d) be a metric space. T If Aα (α ∈ T
Γ) are subsets of X, then int α∈Γ Aα ⊆
α∈Γ int(Aα ). If , in addition, Γ is finite, then int( α∈Γ Aα ) = α∈Γ int(Aα ).
T T
Proof. Suppose x ∈ int α∈Γ Aα . Then T there exists r > 0 such that B(x; r) ⊆ α∈Γ Aα ; thus, for any
α ∈ Γ, B(x; r) ⊆ Aα and therefore x ∈ α∈Γ int(Aα ). T
n
Suppose Γ is finite, say Γ = {γ1 , · · · , γn } and x ∈ j=1 int(Aγj ). Then for any j, there exists rj > 0,
Tn Tn
B(x; rj ) ⊆ Aγj . Take r = {r1 , · · · , rn } > 0, and B(x; r) ⊆ j=1 Aγj , then x ∈ int( j=1 Aγj ).
13
Proof. Let (an ) be sequences in metric space (X, d).
1. Let a2n , a2n+1 → a for some a ∈ X. Then there exists N1 , N2 ∈ N such that whenever 2n + 1 ≥ N1
and 2n ≥ N2 , we have d(a2n+1 , a) < ϵ and |a2n − a| < ϵ. Take N = max{N1 , N2 }, then for any n ≥ N ;
therefore, if n = 2k + 1 for some k ∈ N, then 2k + 1 ≥ N1 and d(a2k+1 , a) < ϵ; otherwise, n = 2k for
some k ∈ N, then 2k ≥ N2 and d(a2k , a) < ϵ. In either case, we have d(an , a) < ϵ. Thus, an → a.
2. Let an → a for some a ∈ X and (ani ) is a subsequence. Then there exists N ∈ N such that for any
n ≥ N , we have d(an , a) < ϵ. Since ni ≥ i for all i ∈ N, d(ani , a) < ϵ for any i ≥ N . Conversely, since
(an ) itself is a subsequence of (an ), we are done.
3. If an ̸→ a, then exists ϵ0 > 0 such that for any k ∈ N, there exists nk ≥ k such that d(ank , a) ≥ ϵ0 .
Without loss of generality, suppose (nk ) is increasing, then (ank ) is a subsequence of (ak ). Then for
any subsequence (anki ), d(anki , a) ≥ ϵ0 , then we are done by contrapositive.
2. A = A ∪ A′ .
3. A is a closed set if and only if it contains all its limit points.
Proof. Suppose x is a limit point of A. Then for any n ∈ N, there exists xn ∈ A such that xn ̸= x with
d(xn , x) < n1 . Then by 1.2.14, xn → x with xn ̸= x for all n ∈ N. Conversely, suppose xn → x with xn ̸= x
for all n ∈ N, then for all ϵ > 0, there exists n ∈ N such that d(xn , x) < ϵ and xn ̸= x.
Suppose x ∈ A\A, then for n ∈ N, there exists xn ∈ A such that d(xn , x) < n1 . Hence, x ∈ A′ . Conversely,
let x ∈ A∪A′ and r > 0. If x ∈ A, then B(x; r)∩A ̸= ∅. If x ∈ / A, then x ∈ A′ and therefore B(x; r)∩A ̸= ∅.
′
Thus, A = A ∪ A .
2.1.9 2
A is a closed set ⇐⇒ A = A ⇐⇒ A ∪ A′ = A ⇐⇒ A′ ⊆ A.
Definition 2.2.6. Let (X, d) be a metric space. If A ⊆ X and x ∈ X, then the distance from x to A is
defined by
d(x, A) = inf{d(x, a) : a ∈ A}.
Proposition 2.2.10. Let (an ) be a convergent sequence in metric space (X, d). Then (an ) is Cauchy.
Proof. Since an → a for some a ∈ X, if ϵ > 0, then there exist N ∈ N such that for any n, m ≥ N , we have
ϵ
d(a, an ), d(a, am ) < .
2
Then
d(an , am ) ≤ d(an , a) + d(a, am ) < ϵ
whenever n, m ≥ N , as required.
Proposition 2.2.11. Let (an ) be a Cauchy sequence in metric space (X, d).
14
1. (an ) is bounded.
2. If there exists a subsequence (ani ) such that ani → b for some b ∈ R, then an → b.
Proof. Since (an ) is Cauchy, there exists N ∈ N such that d(aN , am ) < 1 for any m ≥ N . Given x ∈ X, let
M = max{d(x, a1 ) + 1, d(x, a2 ) + 1, · · · , d(x, aN ) + 1}. Then for any n < N , it is clear that d(x, an ) < M .
For n ≥ N , since
d(x, an ) ≤ d(an , aN ) + d(aN , x) < 1 + d(x, aN ) ≤ M,
we also have d(x, an ) < M . Hence, (an ) is bounded.
Now suppose there exists a subsequence (ani ) such that ani → b for some b ∈ R. Then there exists N ∈ R
such that for any j ≥ N ,
ϵ
d(anj , b) < .
2
Also, there exists M such that for any n, m ≥ M ,
ϵ
d(an , am ) < .
2
Hence, take T = max{nN , M }, then for any n ≥ T , we have
as required.
Definition 2.2.12. Let (X, d) be a metric space and A ⊆ X. Then the diameter of A, denoted by diam(A),
is defined by
diam(A) = sup{d(x, y) : x, y ∈ A}.
If diam(A) < ∞, we say that A is bounded.
Theorem 2.2.13 (Cantor). A metric space (X, d) is complete if and only if whenever (Fn ) is a sequence of
nonempty subsets satisfying
1. each Fn is closed;
2. F1 ⊇ F2 ⊇ · · · ;
3. diam Fn → 0,
T∞
then n=1 Fn is a single point.
Proof. Suppose X is complete and (Fn ) is as in the statement of the theorem. For each n ∈ N let xn ∈ Fn .
Given ϵ > 0, then let N be such that diam(Fn ) < ϵ for n ≥ N . Thus, if m, n ≥ N , then 2 implies
xn , xm ∈ FN and so d(xn , xm ) ≤ diam FN < ϵ. Thus (xn ) is a CauchyTsequence; since (X, d) is complete,
∞
thereT is an x ∈ X such that xn → x. Since each Fn is closed, x ∈ n=1 Fn . If there is another point
∞
y ∈ n=1 Fn , then d(x, y) ≤ diam Fn for each n ≥ 1. By 3, y = x.
Assume that X satisfies the stated conditions and (xn ) is a Cauchy sequence. Let Fn = {xn , xn+1 , · · · }.
Clearly, 1 and 2 are satisfied. If ϵ > 0, then let N be such that d(xn , xm ) < ϵ for m, n ≥ N .TBut for k ≥ N ,
∞
diam Fk = sup{d(xn , xm ) : n, m ≥ k} ≤ ϵ Thus (Fn ) satisfies the three conditions, so that n=1 Fn = x for
some points x. But for any n ≥ 1, d(x, xn ) ≤ diam Fn → 0. There fore xn → x and X is complete.
Proposition 2.2.14. If (X, d) is a complete metric space and Y ⊆ X, then (Y, d) is complete if and only if
Y is closed in X.
Proof. Suppose Y is closed and (yn ) is a Cauchy sequence in Y . Then (yn ) is a Cauchy sequence in X. Since
X is complete, (yn ) converges to some y ∈ X. Since Y is closed, by 2.2.2, y ∈ Y and therefore Y is complete.
Suppose (Y, d) is complete and (yn ) is a converge sequence such that yn → y ∈ X. By 2.2.2 again, it is
sufficient to prove that y ∈ Y . Since Y is complete and 2.2.10, y ∈ Y , as required.
Proposition 2.2.15. Let (X, d) be a metric space. A subset A ⊆ X is bounded if and only if for any x ∈ X,
there exists r > 0 such that A ⊆ B(x; r).
15
Proof. Suppose A is bounded and x ∈ X. Let diam(A) = M < ∞, then for any a, b ∈ A, d(a, b) ≤ M . Fix
a ∈ A and let d(x, a) = r′ , r = r′ + M + 1. Then for any b ∈ A, d(x, b) ≤ d(x, a) + d(a, b) ≤ r′ + M < r,
hence b ∈ B(x; r).
Suppose for any x ∈ X, there exists r > 0 such that A ⊆ B(x; r). Then for any a, b ∈ A, d(a, b) ≤
d(a, x) + d(x, b) = 2r =⇒ diam(A) ≤ 2r < ∞, as required.
Corollary 2.2.15.1. Let (X, d) be a metric space and B1 , · · · , Bn are bounded subsets in (X, d). Then
S n
j=1 Bj is bounded.
Sn
Proof. Given x ∈ X, by 2.2.15, it is sufficient to prove that there exists r > 0 such that j=1 Bj ⊆ B(x; r).
By 2.2.15 again, there exists r1 , · · · , rn > 0 such that Bj ⊆ B(x; rj ). Therefore, we take r = maxj {rj }, then
we are done.
This proposition relates the definitions between bounded sets and bounded sequences.
2.3 Continuity
Definition 2.3.1. If (X, d) and (Y, ρ) are two metric spaces, a function f : X → Y is continuous at a point
x0 ∈ X if for every ϵ > 0, there exists δ > 0 such that f (B(x0 ; δ)) ⊆ B(f (x0 ); ϵ). If f is continuous at every
point in X, then f is a continuous function.
Proposition 2.3.2. Let (X, d), (Y, ρ) be two metric spaces and f : X → Y be a function. Then the following
are equivalent:
1. f is continuous at x ∈ X;
2. for any sequence (xn ) in X converging to x, we have f (xn ) → f (x).
Proof. Suppose f is continuous at x ∈ X and (xn ) is a sequence in X which converges to x. Then given
any ϵ > 0, there exists δ > 0 such that f (B(x; δ)) ⊆ B(f (x); ϵ). With such δ, there exists N ∈ N such
that xn ∈ B(x; δ) whenever n ≥ N . Hence, d(f (xn ), f (x)) < ϵ whenever n ≥ N . Since ϵ is arbitrary,
f (xn ) → f (x).
Suppose f is not continuous at x ∈ X, then there exists ϵ0 > 0 such that for any δ > 0, there exists y ∈ X
such that d(x, y) < δ and d(f (x), f (y)) ≥ ϵ0 . For any n ∈ N, take δ = n1 , and pick xn such that d(x, xn ) < n1
and d(f (x), f (xn )) ≥ ϵ0 . Hence, xn → x but f (xn ) ̸→ f (x), as required.
Theorem 2.3.3. If (X, d) and (Y, ρ) are metric spaces and f : X → Y is a function, then the following
statements are equivalent:
1. f is a continuous function.
2. If U is an open subset of Y , then f −1 (U ) is an open subset of X.
3. If D is an closed subset of Y , then f −1 (D) is an closed subset of X.
Proof. (1 ⇒ 2) : Suppose f is a continuous function and U is open in Y . Let x ∈ f −1 (U ), then f (x) ∈ U .
Therefore, there exists ϵ > 0 such that B(f (x); ϵ) ⊆ U . Since f is continuous, there exists δ > 0 such that
f (B(x; δ)) ⊆ B(f (x); ϵ) ⊆ U =⇒ B(x; δ) ⊆ f −1 (U ). Then f −1 (U ) is open.
(2 ⇒ 3) : Let D be a closed subset of Y . Then Y \D is open, hence f −1 (Y \D) = X\f −1 (D) is open and
therefore f −1 (D) is closed.
(3 ⇒ 2) : Let U be an open subset of Y . Then Y \U is closed, hence f −1 (Y \U ) = X\f −1 (U ) is closed
and therefore f −1 (U ) is open.
(2 ⇒ 1) : Suppose x ∈ X and let ϵ > 0. By 2.1.3, for any ϵ > 0, B(f (x); ϵ) is open. Hence f −1 (B(f (x); ϵ))
is open. Since x ∈ f −1 (B(f (x); ϵ)), there exists δ > 0 such that B(x; δ) ⊆ f −1 (B(f (x); ϵ)) =⇒ f (B(x; δ)) ⊆
B(f (x); ϵ).
Proposition 2.3.4. If (X, d) is a metric space and A ⊆ X is a nonempty set, then
for any x, y ∈ X.
16
Proof. For any a ∈ A, we have d(x, a) ≤ d(x, y) + d(y, a). Take infimum over a ∈ A, we have d(x, A) ≤
d(x, y) + d(y, A). Since A ̸= ∅, d(x, A) − d(y, A) ≤ d(x, y). Reverse the role of x and y, thus we get the
inequality.
We always assume that R is with metric d defined by d(x, y) := |x − y|.
Corollary 2.3.4.1. If (X, d) is a metric space and A ⊆ X is a nonempty set, then f : X → R defined by
f (x) = d(x, A) is a continuous function.
Proof. Given ϵ > 0, take δ = ϵ, then for any x, y ∈ X, d(f (x), f (y)) ≤ d(x, y) < ϵ whenever d(x, y) < δ = ϵ.
Since x, y are arbitrary, then f is continuous.
Proposition 2.3.5. If (X, d) is a metric space and f, g : X → R are continuous functions, then f + g and
f g are continuous. If f ̸= 0, then f1 is continuous.
Proof. It is clear by 2.3.2 and 1.2.9.
Corollary 2.3.5.1. Let (X, d) be a metric space. Then C(X) := {f : X → R : f is continuous.} is a vector
space over R.
Proposition 2.3.6. If (X, d1 ), (Y, d2 ), (Z, d3 ) are metric spaces and f : X → Y, g : Y → Z are continuous,
then g ◦ f is continuous.
Proof. Let U ⊆ Z be an open set. Then (g ◦ f )−1 (U ) = f −1 (g −1 (U )) is open since g −1 (U ) is an open subset
in Y and f pulls open sets to open sets.
Theorem 2.3.7 (Urysohn’s Lemma). Let (X, d) be a metric space. If A and B are two disjoint closed
subsets of X, then there is a continuous function f : X → R having the following properties:
1. 0 ≤ f (x) ≤ 1 for all x ∈ X;
2. f (x) = 0 for all x ∈ A;
Definition 2.3.8. If (X, d) and (Y, ρ) are metric spaces, then a map f : X → Y is uniformly continuous
if for every ϵ > 0, there is a δ > 0 such that ρ(f (x), f (y)) < ϵ whenever d(x, y) < δ.
Proposition 2.3.9. Let (X, d) and (Y, ρ) are metric spaces.
1. If f : X → Y is uniformly continuous and (xn ) is a Cauchy sequence in X, then (f (xn )) is a Cauchy
sequence in Y .
2. If A ⊆ X and Y is complete, and f : A → Y is uniformly continuous, then f can be extended to a
uniformly continuous function f : A → Y .
Proof. Given ϵ > 0, then there exists δ > 0 such that d(x, y) < δ implies ρ(f (x), f (y)) < ϵ. Since (xn ) is
Cauchy, there exists N ∈ N such that d(xn , xm ) < δ. Hence, ρ(f (xn ), f (xm )) < ϵ.
We first claim that whenever (an ) and (bn ) is a sequence in A with an , bn → x ∈ X, then limn f (xn ) =
limn f (yn ) (notice that f (x) may not be defined since x may not be in A).
17
Proof of claim. To prove that limn f (an ) = limn f (bn ), we consider the sequence (cn ) = (a1 , b1 , a2 , b2 , · · · ).
Since limn an = limn bn = x, by 2.2.4, cn → x. Then f (cn ) is convergent sequence by 2.2.10, 1 and the
completeness of Y . Therefore, by 2.2.4, limn f (an ) = limn f (bn ).
Now we define f˜ : A → Y by f˜(x) = limn f (xn ), where (xn ) is a sequence in A which converges to x ∈ X.
By claim, f˜ is well-defined and for any x ∈ A, x, x, x, · · · → x, f˜ is an extension of f .
It remains to prove that f˜ is uniformly continuous. Let ϵ > 0, take δ > 0 such that ρ(f (a), f (b)) < 3ϵ
whenever a, b ∈ A and d(a, b) < δ. Assume x, y ∈ A and d(x, y) < 3δ . So there exist sequences (an ) and (bn )
such that an → x and bn → y. Let N1 ∈ N such that ρ(f (an ), f (x)), ρ(f (bn ), f (y)) < 3ϵ whenever n ≥ N1 .
This for n ≥ N1 ,
2ϵ
ρ(f (x), f (y)) ≤ ρ(f (x), f (an )) + ρ(f (an ), f (bn )) + ρ(f (bn ), f (y)) < + ρ(f (an ), f (bn )).
3
Now we can choose N2 ∈ N such that when n ≥ N2 , we have d(an , x), d(bn , y) < 3δ . Thus d(an , bn ) ≤
d(an , x) + d(x, y) + d(y, bn ) < δ. But since an , bn ∈ A, this implies ρ(f (an ), f (bn )) < 3ϵ whenever n ≥ N2 .
Then take N = max{N1 , N2 }, we have ρ(f (x), f (y)) < 3ϵ whenever x, y ∈ A whenever d(x, y) < 3δ . Therefore,
f˜ is uniformly continuous.
2.4 Compactness
S d) be a metric space and E ⊆ X. Then a collection of open sets G is called an
Definition 2.4.1. Let (X,
open cover of E if E ⊆ G∈G G and G is open for any G ∈ G.
Definition 2.4.2. Let (X, d) be a metric space. A subset K ⊆ X is compact if every open cover of K has
a finite subcover.
Proposition 2.4.3. Let (X, d) be a metric space.
1. If K ⊆ X is compact, then K is closed and bounded.
2. If K ⊆ X is compact and F ⊆ K is closed, then F is compact.
3. Let f : X → (Z, ρ) be a continuous function mapping from X to metric space (Z, ρ).
Proof.
S Let x ∈ X\K. For each p ∈ K, let rp > 0 be the radius such that B(x; Snrp ) ∩ B(p; rp ) = ∅. Then
p∈X B(p; rp ) ⊇ K. Therefore, by definition, there exist p1 , · · · , pn such that j=1 B(pj ; rpj ) ⊇ K. Then
take r = min1≤j≤n {rj } and S therefore B(x; r) ⊆ X\K; hence, X\K is open and Sn K is closed.
thus
Given r > 0, since p∈K B(p; r) ⊇ K, there exist p1 , · · · , pn such that j=1 B(pj ; r) ⊇ K. Hence,
diam(K) ≤ 2nr < ∞, as required.
Suppose {Gα : α ∈ ∆} is an open covering of F (we can assume that Gα ∩ F ̸= ∅ for any α ∈ Γ), then
{Gα : α ∈ ∆} ∪ S {X\F } is an open covering of K. Therefore, there exist T1 , · · · , Tn (probably containing
n
X\F ) such that j=1 Tj ⊇ K. Then discard X\F in Tj ’s, which results in a finite subcover of F .
f (X). Then α∈∆ f −1 (Gα ) ⊇ X. By 2.3.3, there exists
S
Suppose {Gα : α ∈S∆} is an open covering of S
n n
G1 , · · · , Gn such that j=1 f −1 (Gj ) ⊇ X. Hence, j=1 Gj ⊇ f (X).
Corollary 2.4.3.1 (EVT). If (X, d) is a compact metric space and f : X → R is a continuous function,
then there are points a and b in X such that f (a) ≤ f (x) ≤ f (b) for all x in X.
Proof. By 2.4.3, f (X) is closed and bounded. By 1.1.8 and 1.2.17, inf f (X) and sup f (X) exist. Then by
1.2.17.1 and 1.2.10, there exist two sequences converge to sup f (X) and inf f (X), respectively. Hence, by
2.2.2, sup f (X) and inf f (X) are both in f (X).
Definition 2.4.4. Let (X, d) be a metric space. A subset Sn K ⊆ X is totally bounded if for any radius
r > 0 there are points x1 , · · · , xn in K such that K ⊆ k=1 B(xk ; r).
TnA collection F of subsets of K has the
finite intersection property (FIP) if whenever F1 , · · · , Fn ∈ F, j=1 Fj ̸= ∅.
Theorem 2.4.5. The following statements are equivalent for a closed subset K of a metric space (X, d).
1. K is compact.
18
T
2. If F is a collection of closed subsets of K having the FIP, then it holds that F ∈F F ̸= ∅.
3. Every sequence in K has a convergent subsequence (which converges in K). (Here, K is called se-
quentially compact.)
4. Every infinite subset of K has a limit point.
Lemma 2.4.6 (Lebesgue). Let (X, d) be a metric space and K be a sequentially compact subset of X.
Then for any open covering G of K, there exists r > 0 such that for any x ∈ K, there is a G ∈ G such that
B(x; r) ⊆ G.
Proof. Suppose not, then for any r > 0, there exists a x ∈ K such that B(x; r) ∩ (X\G) ̸= ∅ for any G ∈ G.
Then for each n ∈ N, pick r = n1 and the corresponding points xn ∈ K, which constructs a sequence (xn ).
Since K is sequentially compact, there exists a convergent subsequence (xnk ) which converges to some p ∈ K.
Since G is an open covering, there exists G ∈ G such that p ∈ G. Since G is open, there exists r > 0 such
that B(p; r) ⊆ G. Since xnk → G, there exists nk ∈ N such that xnk ∈ B(p; r/2) with n1k < 2r . Then for any
y ∈ B(xnk ; n1k ), we have d(y, x) ≤ d(y, xnk ) + d(xnk , x) < r =⇒ B(xnk ; n1k ) ⊆ G, a contradiction.
Theorem 2.4.7 (Heine-Cantor). If (X, d) is a compact metric space and f : X → (Z, ρ) is a continuous
function, then f is uniformly continuous.
Proof. Since f is continuous,
S for every x ∈ X and ϵ > 0, there exists δ = δx > 0 such that f (B(x; δx )) ⊆
B(f (x); ϵ/2). Since x∈X B(x; δx /2) ⊇ X and X is compact, there exists x1 , · · · , xn such that
n
[ δj
B(xj ; ) ⊇ X.
j=1
2
19
Let δ ′ = min1≤j≤n (δj /2), then for any x, y ∈ X with d(x, y) < δ ′ , there exists xj ∈ X such that
d(x, xj ) < δj /2 and therefore d(y, xj ) ≤ d(y, x) + d(x, xj ) < δj . Hence, x, y ∈ B(xj ; δj ) and therefore
d(f (x), f (y)) ≤ d(f (x), f (xj )) + d(f (xj ), f (y)) < ϵ, as required.
Proposition 2.4.8. A compact metric space (X, d) is separable.
S∞
Proof. Since for each n ∈ N, there exists a finite set Fn such that X ⊆ x∈Fn B(x; n1 ), by setting F = j=1 Fj ,
S
for any p ∈ X and r > 0, there exists n1 < r and hence x ∈ B(p; r) =⇒ B(p; r) ∩ F ̸= ∅. By 2.1.11, we are
done.
2.5 Connectedness
Definition 2.5.1. A metric space (X, d) is connected if there are no subsets A, B ⊆ X such that:
1. A, B are open;
2. A, B ̸= ∅;
3. A ∩ B = ∅;
4. A ∪ B = X.
Proposition 2.5.2. Let (X, d) be a metric space. Then the followings are equivalent:
1. X is connected;
2. The only subsets of X which are open and closed are X and ∅;
3. The only subsets of X with empty boundary are X and the empty set.
4. X cannot be written as the union of two nonempty sets A, B such that A ∩ B, A ∩ B = ∅ (in this case,
A, B are called the separation of X, denoted by A|B).
5. All continuous
( functions from X to {0, 1} are constant, where {0, 1} is the metric space with metric
1 i ̸= j
d(i, j) = , where i, j ∈ {0, 1}.
0 i=j
1. A, B are open;
2. A, B ̸= ∅;
3. A ∩ B = ∅;
4. A ∪ B = X,
a contradiction.
(5 ⇒ 1) : Suppose there exists A, B ⊆ X satisfying the definition. Then we defined f : X → {0, 1} by
f (A) = {0} and f (B) = {1}. Since the open sets in {0, 1} are ∅, {0}, {1}, {0, 1}, f is continuous by 2.3.3, a
contradiction.
20
Proposition 2.5.3. A subset of R, say E, is connected if and only if for any a, b ∈ E, [a, b] ⊆ E.
Proof. Suppose E is connected and a, b ∈ E. Suppose there is a c ∈ / E with a ≤ c ≤ b, then E can be wriiten
as E ∩ (−∞, c) ∪ E ∩ (c, ∞). Let A = E ∩ (−∞, c) and B = E ∩ (c, ∞). Since 2.1.5, A and B are open in E
and A ∩ B = ∅, A ∪ B = E and A, B are both nonempty, E is not connected, a contradiction.
Suppose E is not connected, then by 2.5.2, there is a separation A|B of E. Since A, B ̸= ∅, there exists
a ∈ A, b ∈ B. Without loss of generality, we assume that a < b. Define c = sup(A ∩ [a, b]) ⊆ A ∩ [a, b] ⊆ A.
Hence, c ∈/ B. In particular, a ≤ c < b. If c ∈
/ A, then a < c < b and c ∈
/ E, a contradiction.
/ B, then there exists c1 with c < c1 < y and c1 ∈
If c ∈ A, then c ∈ / B. Then a < c1 < b and c1 ∈ / E
(c1 ∈
/ A as otherwise c1 ∈ A ∩ [a, b] with c1 > c, a contradiction).
Proposition 2.5.4. Let (X, d), (Z, ρ) be metric spaces and f : X → Z be a continuous function. If X is
connected, then f (X) is connected.
Proof. Suppose not, then there exists two nonempty, disjoint open (in f (X)) subsets A, B such that A ∪ B =
f (X). Then f −1 (A) ∪ f −1 (B) = X, by 2.3.3, a contradiction.
Corollary 2.5.4.1. If f : (X, d) → R is continuous, then for any a, b ∈ f (X), c ∈ R with a < c < b, c ∈ f (X).
Proposition 2.5.5. Let (X, d) be a metric space. Suppose U, V ⊆ X are two disjoint open sets. If A ⊆ U ∪V
is a connected set, then A ⊆ U or A ⊆ V .
Proof. Suppose A ∩ U, A ∩ V ̸= ∅, then A ∩ U, A ∩ V are two nonempty disjoint open sets (open in A by
2.1.5) whose union is A, a contradiction.
Corollary 2.5.5.1. Let (X, d) be a metric space. If A ⊆ X is a dense and connected subset, then X is
connected.
Proof. Suppose X is not connected. Then there exist nonempty disjoint open sets U, V whose union is X.
Hence, (A ∩ U ), (A ∩ V ) are nonempty (since A is dense) disjoint open sets (open in A) whose union is A, a
contradiction.
Corollary 2.5.5.2. Let (X, d) be a metric space. If A ⊆ X is a connected subset and A ⊆ B ⊆ A, then B
is connected.
Proof. Take A as a subspace of X, then A is dense and connected in A. Let UB ⊆ B be a nonempty open
set, then since UB ⊆ A, by 2.1.5, there exists an open set U ⊆ A such that UB = B ∩ U . Since A is dense in
A, U ∩ A ̸= ∅, therefore B ∩ U ∩ A = A ∩ U = ∅, which shows that A is dense in B.
Hence, A as a dense and connected subset of B, B is connected.
Proposition 2.5.6. Let (X, d) be a metric space.
S∞
1. If (En ) is a sequence of connected subsets of X such that En ∩ En+1 ̸= ∅, then E = n=1 En is
connected.
2. If {ES
α : α ∈ ∆} is a collection of connected subsets of X such that Ei ∩ Ej ̸= ∅ for any i, j ∈ ∆, then
E = α∈∆ Eα is connected.
Proof. Let f : E → {0, 1} be a continuous function. By 2.5.2, f |Ej is a constant for any j. Since En ∩En+1 ̸=
∅, f is a constant. By 2.5.2 again, E is connected.
2 is similar to 1.
Definition 2.5.7. If (X, d) is a metric space, then a component of X is a maximal connected subset of X.
Proposition 2.5.8. For any metric space (X, d), every connected set is contained in a component. distinct
components are disjoint, and the union of all the components is the entire space.
S subset D of X, let CD denote the collection of all connected subsets of X that contain
Proof. Fix a connected
D. By 2.5.6, C = A∈CD A is connected. Clearly, C is a component and contains D.
Given x ∈ X, since {x} is connected (by 2.5.2), x is contained in a component. Therefore, the union of
all components is X.
Finally, if C and D are two distinct components such that C ∩ D ̸= ∅, then C ⊂ C ∪ D and C ∪ D is
connected by 2.5.6, which contradicts to the definition of components.
21
Corollary 2.5.8.1. Every open set in R can be written as an at most countable union of open intervals.
Proof. Let G be an open set in R. Consider the family of components CG . By 2.5.8, we only need to prove
that each component is an open interval. Let g ∈ CG and p ∈ g. Since G is open, there exists r > 0 such
that B(p; r) ⊆ G. Then since B(p; r) is connected and B(p; r) ∩ g ̸= ∅, by 2.5.6, B(p; r) ∪ g is connected.
Hence, B(p; r) ⊆ g and therefore g is open. Now by 2.5.3, g is an interval, then we are done.
Definition 2.5.9. Let (X, d) be a metric space and E ⊆ X. If x, y ∈ E and ϵ > 0, say that there is an
ϵ-chain from x to y when there is a finite number of points x1 , · · · , xn ∈ E such that
1. for 1 ≤ k ≤ n, B(xk ; ϵ) ⊆ E;
2. for 2 ≤ k ≤ n, xk−1 ∈ B(xk ; ϵ);
3. x1 = x and xn = y.
Proposition 2.5.10. Let (X, d) be metric space.
1. If G ⊆ X is a open subset of X, then every component of G is open. If, in addition, X is separable,
then there are countably many components (of G).
2. X is connected if and only if there is an ϵ > 0 such that for any x, y ∈ X, there is an ϵ-chain in X from
x to y (in this case, we say that X is ϵ-chainable) .
Proof.
1. If H is a component of G and x ∈ H, choose r > 0 such that B(x; r) ⊆ G. Then by 2.5.6, B(x; r) ∪ H
is connected and therefore B(x; r) ⊆ H. Now suppose X is separable, there exists a dense subset D of
X. Now since each component is open, each component contains a point of D. By 2.5.8, each point of
D is assigned to a component, hence there are countably many components of G.
2. Suppose X is ϵ-chainable and x ∈ X. Let H be a component containing x. Let y ∈SX. Since X is
n
ϵ-chainable, there exists a ϵ-chain x1 , · · · , xn ∈ X from x to y. Then by 2.5.6, B = j=1 B(xj ; ϵ) is
connected. Therefore, as a connected subset containing x, B ⊆ H. In particular, y ∈ H and since y is
arbitrary, H = X.
Conversely, suppose X is connected. Fix a ∈ X and ϵ > 0, let
22
T∞
Proof. To show that U = n=1 Un is dense, it suffices to show that for any nonempty open set G, G ∩ U ̸= ∅
(by 2.1.11). Since U1 is dense and open, there is an open ball B(x1 ; r1 ) with r1 < 1 such that B(x1 ; r1 ) ⊆
G ∩ U1 . This is the first step in an introduction argument that for each n ≥ 2 there is an open ball
B(xn ; rn ) < n−1 such that
B(xn ; rn ) ⊆ B(xn−1 ; rn−1 ) ∩ Un ⊆ G.
(since U2 is dense and open, B(x1 ; r1 ) ∩ U2 ̸= ∅, therefore there exists an open ball B(x2 ; r2 ) ⊆ B(x1 ; r1 ) ∩ U2
with r2 < 2−1 . Also, B(x1 ; r1 ) ∩ U2 ⊆ B(x1 ; r1 ) ∩ U2 ⊆ B(x1 ; r1 ) ⊆ G ∩ U1 ⊆ G and continue the process
inductively.) Therefore, for any n > N ,
B(xn ; rn ) ⊆ B(xN ; rN ) ∩ UN ⊆ G ∩ UN .
Therefore, for n, m ≥ N , d(xn , xm ) < 2N −1 , and so (xn ) is a Cauchy sequence. Since X is complete, there
exists x ∈ X such that xn → x. Hence, when n > m > N , x ∈ B(xn ; rn ) ⊆ B(xm ; rm ) ⊆ G ∩ UN . Since N
is arbitrary, x ∈ G ∩ U . Thus, G ∩ U ̸= ∅, then we are done.
Corollary
S∞ 2.6.1.1. If (X, d) is a complete metric space and (Fn ) is a sequence of closed subsets such that
X = n=1 Fn , then there is an n such that we have int(Fn ) ̸= ∅.
Proof.
T∞ Suppose int(Fn ) T
= ∅ for each n and set Un = X\Fn . Then each Un is open and dense in X. Therefore,
∞
n=1 Un is dense. But n=1 Un = ∅, a contradiction.
Corollary 2.6.1.2. Let (X, d) be a complete metric spaceSand (En ) be a sequence such that int(En ) = ∅
∞
for each n (in this case, En is nowhere dense). Then X\ n=1 En is dense in X.
S∞ S∞ T∞
Proof. It suffices to show that X\ n=1 En is dense in X. Notice
T∞ that X\ n=1 En = n=1 (X\En ). Let
Un = X\En , then each Un is dense and open in X. Therefore, n=1 Un is dense in X, as required.
23
Corollary 2.7.6.1. Rk has a countable base.
Proof. We only need to show that Rk is separable and it is clear that Qk is a countable dense subset of Rk .
(For more detail, given x ∈ Rk and r > 0, since there is an open cube centered at x contained in the open
ball, by 1.2.14.1, there exists a q ∈ Qk ∩ B(x; r).)
Definition 2.7.7. Let (X, d) be a metric space. Then subset P ⊆ X is said to be perfect if P is closed and
every point of P is a limit point of P .
Definition 2.7.8. Let (X, d) be a metric space and E ⊆ X be a subset. Then we say that x ∈ X is a
condensation point of S if for any r > 0, B(x; r) ∩ S is uncountable. It is clear that if x is a condensation
point of S, x is a limit point of S.
Proposition 2.7.9. Let (X, d) be a separable metric space. Suppose E ⊆ X is an uncountable subset and
let P be the set of all condensation points of E. Then P is perfect and (X\P ) ∩ E is at most countable.
Proof. By 2.7.6, there exists a countable base {Vn }. Let W be the union of those Vn for which E ∩ Vn is at
most countable. We shall prove that P = X\W .
Given p ∈ P , in order to get a contradiction, suppose p ∈ W . Then there exists Vn such that p ∈ Vn and
Vn ∩ S are countable. However, since Vn is open, there exists r > 0 such that B(p; r) ⊆ Vn . Since p ∈ P ,
B(p; r) ∩ S is uncountable, a contradiction. Hence, p ∈ X\W .
Conversely, let p ∈ X\W . Suppose, to get a contradiction, there exists r > 0 such that B(p; r) ∩ S is
at most countable. Since B(p; r) is open, B(p; r) is an union of subcollection of {Vn }, say {Vni }. Since
B(p; r) ∩ S is uncountable, Vni ∩ S is at most countable for each i. Then the union of Vni ’s is a subset of W ,
which implies that p ∈ W , which is a contradiction.
Therefore, P = X\W . As a complement of an open set (W is open by 2.1.4), P is closed. It then remains
to show that every point of P is a limit point of P .
Let p ∈ P . Suppose, to get a contradiction, there exists r > 0 such that B(p; r) ∩ P = {p}. Then for
any y ∈ B(p; r)\{p}, there exists ry > 0 such that B(y; ry ) ∩ S is at most countable (since y ∈ / P ) and
B(y; ry ) ⊆ B(p; r). Since B(y; ry ) is open, there exists Vny ∈ {Vn } such that y ∈ Vny ⊆ B(y; ry ) ⊆ B(p; r)
and Vny ∩ S is at most countable. Hence,
[
(B(p; r)\{p}) ∩ S = (Vny ∩ S),
y∈B(p;r)\{p}
which is countable since it is a countable union of at most countable sets. Hence, B(p; r) ∩ S is countable, a
contradiction.
Note that (X\P ) ∩ S = W ∩ S is countable by definition.
Corollary 2.7.9.1. Let (X, d) be a separable metric space and F be an uncountable closed set F ⊆ X.
Then there exists disjoint A, B ⊆ X such that A is perfect, B is countable and F = A ∪ B.
Proof. By 2.7.9, the set of condensation points of F , P , is perfect and (X\P ) ∩ F = F \P is countable. Take
A = P and B = F \P , then we are done.
Proposition 2.7.10. Let P be a nonempty perfect set in Rk . Then P is uncountable.
Proof. Since P has limit points, P must be infinite. Suppose, to get a contradiction, P is uncountable, and
denote the points of P by a sequence (xn ). We shall construct a sequence (Vn ) of open sets, as follows.
Let V1 be an open set containing x1 . Now we assume that Vn has been constructed such that
1. Vn ⊆ Vn−1 .
2. xn−1 ∈
/ Vn .
3. Vn ∩ P ̸= ∅.
For Vn+1 , since Vn ∩ P is not empty, there exists an open set Vn+1 which satisfies 1,2 and 3.
Let Kn =TVn ∩ P . Since Vn is closed and bounded, by 2.4.5.1, Vn is compact. Since xn ∈ / Kn+1 , no point
∞ T∞
of P lies in n=1 Kn . Since Kn ⊆ P , this implies
T∞ that n=1 Kn = ∅. But each K n is compact, nested
(Kn+1 ⊆ Kn ) and nonempty (by 3). By 2.4.5, n=1 Kn is nonempty, a contradiction.
24
Corollary 2.7.10.1. Every interval [a, b] with a < b is uncountable.
Proposition 2.7.11. Let (X, d) be a metric space and (xn ) be a sequence in X. Then the set L = {p ∈ X :
there exists a subsequence (xnk ) such that xnk → p} is closed in X.
Proof. By 2.1.8, it suffices to show that ∂L ⊆ L. Let p ∈ ∂L. Then for any δ > 0, there exists q = qδ
such that d(p, q) < δ (by 2.1.7) and therefore there exists xδ ∈ {xn : n ∈ N} such that d(p, xδ ) < δ. Take
xn1 ∈ {xn } such that xn1 ̸= q. Notice that such xn1 must exist as otherwise, L has only one element, and
there is nothing to prove. Take δ = d(p, xn1 ). Suppose n1 , · · · , nk are chosen such that d(p, xnj ) < δj . Then
δ
there exists nk+1 > nk such that d(p, xnk+1 ) < k+1 . Then xnk → p and therefore p ∈ L.
25
Chapter 3
26
2. If ϵ > 0, there are infinitely many numbers of n ∈ N such that x∗ + ϵ < xn , but there are only finite
numbers such that x∗ − ϵ < xn .
3. If wm = inf{xn : n ≥ m}, then x∗ = sup{vm : n ∈ N}.
4. If wm = inf{xn : n ≥ m}, then x∗ = lim(wm ).
5. If L is the set of b ∈ R such that there exists a subsequence of xn which converges to b, then x∗ = inf(L).
Theorem 3.1.4. Let (xn ) and (yn ) be bounded sequences of real numbers. Then:
1. lim inf(xn ) ≤ lim sup(xn ).
2. If c ≥ 0, then lim inf(cxn ) = c lim inf(xn ) and lim sup(cxn ) = c lim sup(xn ).
3. If c ≤ 0, then lim inf(cxn ) = c lim sup(xn ) and lim sup(cxn ) = c lim inf(xn ).
4. lim inf(xn ) + lim inf(yn ) ≤ lim inf(xn + yn ) ≤ lim sup(xn + yn ) ≤ lim sup(xn ) + lim sup(yn ).
5. If xn ≤ yn for all n, then lim inf(xn ) ≤ lim inf(yn ) and also lim sup(xn ) ≤ lim sup(yn ).
6. lim sup(xn ) = lim inf(xn ) if and only if (xn ) converges and lim xn = lim sup xn = lim inf xn .
Proof.
1. By 3.1.2 and 3.1.3, lim inf(xn ) = inf(L) ≤ sup(L) = lim sup(xn ).
2. If c = 0, then lim inf(cxn ) = lim sup(cxn ) = c lim sup(xn ) = c lim inf(xn ) = 0. Now let c > 0, then by
1.2.9, for any subsequence (xnk ), xnk converges if and only if cxnk converges and therefore cL is the
1.2.20
set of b ∈ R such that there exists a subsequence of cxnk → b. By 3.1.3, lim inf(cxn ) = inf(cL) =
1.2.20
c inf(L) = c lim inf(xn ) and lim sup(cxn ) = sup(cL) = c sup(L) = c lim sup(xn ).
3. Similar to 2.
4. Apply 3.1.2, 3.1.3 and 1.2.20, we are done.
5. Apply 1.2.19 and 1.2.9, we are done.
6. Notice that of lim inf(xn ) = lim inf(xn ), then there exist b ∈ R such that every subsequence of (xn ) has
a convergent subsequence converges to b. Now apply 1.2.9, and we are done.
27
If x∗ = −∞, then L = {−∞}. Hence, for any real M , there exists m such that for any n ≥ m,
sn ≤ M as for otherwise there exists a subsequence (xnk ) which is bounded below by some K ∈ R and
therefore, by 1.2.8, there exists a monotone further subsequence (xnkj ) which is bounded below. Then
by 1.2.11.1, it is either that xnkj → ∞ or xnkj → x ∈ R, a contradiction. Therefore, sn → −∞.
Given x > x∗ , suppose, to get a contradiction, sn ≥ x for infinitely many n. In that case, there is a number
y ∈ E such that y ≥ x > s∗ , a contradiction.
Now it remains to show the uniqueness. Suppose there are two numbers, p and q ∈ R, which satisfy 2
and suppose p < q. choose x such that p < x < q. Since p satisfies 2, we have sn < x for n ≥ N for some
N ∈ N. But then q cannot satisfy 2 since there is no subsequence converges to q, a contradiction.
Theorem 3.2.3. Let (xn ) be a sequence of real numbers. Let L be the set in the definition 3.2.1. Then the
following are equivalent x∗ ∈ R:
1. x∗ = lim inf xn
2. x∗ ∈ L and for any x < x∗ , there exists an integer N ∈ N such that n ≥ N implies x < xn .
Proposition 3.2.4. Let (sn ), (tn ) be two sequences in R.
1. lim sup sn = lim inf sn = s if and only if lim sn = s.
2. If sn ≤ tn for n ≥ N , where N is fixed, then lim inf sn ≤ lim inf tn and lim sup sn ≤ lim sup tn .
Proof. Suppose lim sn = s. Then for all the subsequences of sn , say snk , lim snk = s. Then by 3.1.2 and
3.2.3, lim sup sn = lim inf sn = s. Conversely, suppose lim sup sn = lim inf sn = s, then for any subsequence
of sn , say snk , lim snk = s. By 1.2.9, lim sn = s.
Suppose sn ≤ tn for n ≥ N , where N is fixed. Suppose, to get a contradiction, lim inf sn > lim inf tn .
Then by 3.2.3, there exists M such that sn > lim inf tn for any n ≥ M . By 3.2.3, there exists a subsequence
tnk such that tnk → lim inf tn . Hence, there exists K ∈ N such that sn > tnk for all k ≥ K. Take
N ′ = max{nK , N, M }, then sN ′ > tN ′ , a contradiction.
Theorem 3.2.5.
1
1. If p > 0, then lim = 0.
np
√
2. If p > 0, then lim n p = 1.
√
3. lim n n = 1.
nα
4. If p > 0 and α ∈ R, then lim = 0.
(1 + p)n
5. If |x| < 1, then lim xn = 0.
Proof.
1
1. Given ϵ > 0, then by 1.2.14, take n > ϵ−1/p ⇐⇒ < ϵ.
np
√
2. If p > 1, take xn = n p − 1. Then xn > 0, and, by the binomial theorem, 1 + nxn ≤ (1 + xn )n = p, so
that
p−1
0 < xn ≤
.
n
Hence xn → 0. If p = 1, then the result is trivial. If 0 < p < 1, the result is obtained by taking
reciprocals.
√
3. Let xn = n n − 1. Then xn ≥ 0, and, by binomial theorem,
s
n n(n − 1) 2 2
n = (1 + xn ) ≥ xn =⇒ ≥ xn ≥ 0.
2 (n − 1)
√
By 1.2.9, xn → 0 ⇐⇒ n
n → 1.
28
4. Let k be an integer such that k > max{α, 0}. For n > 2k,
(np)k
n n k n(n − 1) · · · (n − k + 1) k
(1 + p) > p = p > k .
k k! 2 k!
Hence
nα 2k k!
0< n
< k nα−k
(1 + p) p
for any n > 2k. Since α − k < 0, nα−k → 0, by 1.
5. Take α = 0 in 4.
3.3 Series
Definition 3.3.1. For a sequence (an ) in R, we define
n
X
sn ≡ an ,
k=1
4. If an ≤ 0
and if (sn ) is bounded, then
P
an converges (and the converse also holds).
and if (sn ) is not bounded, then sn → −∞.
P P
5. If there exists N ∈ N such that |an | ≤ cn for any n ≥ N0 , and if cn converges, then an converges.
6. (Comparison test) Let (dn ) be a nonnegative sequence. If there exists N ∈ N such that an ≥ dn for
n ≥ N , then:
if
P P
dn diverges, then an diverges.
if
P P
an converges, then dn converges.
P P P P
7. If an converges, then kan converges and kan = k an for any k ∈ R.
P P
8. If |an | converges, then an converges.
Proposition 3.3.3.
29
1. Let x ∈ R. If |x| < 1, then
∞
X 1
xn = .
n=0
1−x
If |x| ≥ 1, the serie diverges.
P 1
2. n! converges.
Proof.
1. It is well-known that for x ̸= 1,
1 − xn+1
sn = .
1−x
1
Then by 1.2.9, sn → 1−x if |x| < 1. For |x| ≥ 1, since lim xn =
̸ 0, by 3.3.2, we are done.
2. Since
n
X 1 1 1 1 1 1
sn = =1+1+ + ··· + < 1 + 1 + + 2 + · · · + n−1 .
k! 1·2 1 · 2···n 2 2 2
k=0
Then by 1 and 3.3.2, sn converges.
∞
X 1
Remark. We define e = , called the natural number(constant). Also, log n denotes the logarithm
n=0
n!
of n to the base e.
n
1
Proposition 3.3.4. lim 1 + = e.
n
n n
X 1 1
Proof. Let sn = , tn = 1 + . By the binomial theorem,
k! n
k=0
1 1 1 1 2 1 1 n−1
tn = 1 + 1 + 1− + 1− 1− + ··· + 1− ··· 1 − ≤ sn .
2! n 3! n n n! n n
By 3.2.4, lim sup tn ≤ e. Next, fixed m ∈ N,
1 1 1 1 m−1
tn ≥ 1 + 1 + 1− + ··· + 1− ··· 1 − .
2! n m! n n
Then by 3.2.4,
lim inf tn ≥ sm
for any m. Hence, lim inf tn ≥ e by 3.2.4 again.
n
X k
X
Proof. Let sn = ak , tk = 2j a2j . Then for n ≤ 2k ,
k=1 j=0
30
1
P
Corollary 3.4.1.1 (p-test). converges if p > 1 and diverges if p ≤ 1.
np
P k 1
2 2kp = 2(1−p)k converges. By 3.3.3,
P P k(1−p)
Proof. By 3.4.1, it is equivalent to determine when will 2
converges if and only if 21−p < 1 ⇐⇒ 1 < p, as required.
Corollary 3.4.1.2 (advanced p-test). If p > 1,
∞
X 1
n=2
n(log n)p
∆an ≡ an+1 − an .
Theorem 3.4.3 (generalized Cauchy condensation test, Schlomilch test). Let (an ) be a nonnegative decreas-
ing sequence and un be a positive and strictly increasing sequence of integers such that there exists M > 0
such that
∆un
<M
∆un−1
∞
X ∞
X
for any n ≥ 1. Then an converges if and only if ∆un aun converges.
n=1 n=0
n
X k
X
Proof. Let sn = ak , tk = uj auj . Then for n < uk ,
k=1 j=0
0 −1
uX
sn ≤ (a1 + · · · au0 −1 ) + (au0 + au0 +1 + · · · + au0 +∆u0 −1 ) + · · · + (auk + · · · + auk +∆uk −1 ) ≤ aj + tk .
j=1
(If u0 = 1, then the sum at the RHS is 0.) On the other hand, if n > uk , then
1 −1
uX
tk
sn > (a1 + · · · + au1 −1 ) + (au1 + · · · + au1 +∆u0 −1 ) + · · · + (auk + · · · + auk +∆uk−1 −1 ) > aj + .
j=1
M
Then (sn ) and (tk ) are either both bounded or both unbounded. This completes the proof by 3.3.2.
P 1
Example. Investigate the convergence of √ .
2 n
∆bk+1 1
Solution. Consider the sequence bn = n2 . Since
P
∆bk ≤ 3 for all k ≥ 1, by 3.4.3, √
2 n converges if and
P ∆bn P 2k+1
only if 2n = 2k
converges. Then, since
2k+3
2k+1 1
lim2k+1
= < 1,
n→∞
2k
2
31
P P
1. If bn converges, then an converges also.
P P
2. If an diverges, then bn diverges also.
Proof. By multiplying the inequalities given, we have
aN +k bN +k aN
≤ ⇐⇒ aN +k ≤ bN +k .
aN bN bN
Then the rest follows by comparison test.
1
Example. Let (an ) be a positive sequence such that a1 = 1, a2 = 4 and
n−1
Y k−1
an =
k+2
k=2
P
for n ≥ 3. Show that an converges.
1
P
Solution. Consider the sequence bn = n(n+1) . It is well-known that bn converges (by telescoping). Now
notice that
an+1 n n bn+1
= ≤ = .
an n+3 n+2 bn
P
Then by 3.4.4, an converges.
Proposition 3.4.5 (limit comparison test). Let (an ), (bn ) be two nonnegative
P sequences.PSuppose that there
exists N ∈ N and M ≥ 0 such that for any n ≥ N, an ≤ M bn . Then if bn converges, an converges.
P P
Proof. Since bn converges, an is bounded. Then the rest follows by 3.3.2.
Corollary 3.4.5.1 (limit comparison test for calculus students). Let (an ), (bn ) be two nonnegative sequences.
an
Suppose that there exists N ∈ N such that an , bn > 0 for all n ≥ N . Now assume that lim = L ∈ [0, ∞].
bn
P P
1. If 0 < L < ∞, then an and bn either both diverge or converge.
P P
2. If L = ∞ and an converges, then bn converges.
P P
3. If L = 0 and bn converges, then an converges.
Proof.
1. If 0 < L < ∞, then there exists N ′ > N , M > L and m ∈ (0, L) such that for any n ≥ N ′ ,
mbn ≤ an ≤ M bn . Then by 3.4.5, we are done.
′ 1
∞, then there exists
2. If L =P P M > 0 and N > N such that for any n ≥ N , an ≥ M bn ⇐⇒ bn ≤ M an .
Since an converges, bn converges.
3. If L = 0, then there exists N ′ > N such that an ≤ bn for n ≥ N ′ . Since
P P
bn converges, an
converges.
Corollary 3.4.5.2 (limit comparison test second version). Let (an ) and (bn ) be strictly positive sequences
and suppose that
an an
lim sup = L↑ , lim inf = L↓ .
bn bn
P P
1. If L↑ < ∞ and bn converges, then an converges also.
P P
2. If L↓ > 0 and an converges, then an converges also.
Proof.
1. Suppose L↑ < ∞, then there exists PM ∈ R such that M > L P↑ . By 3.2.2, there exists N ∈ N such that
for any n ≥ N , an < M bn . Since bn converges, by 3.4.5, an converges.
32
2. Suppose L↓ > 0, then there exists m ∈ (0, L↓ ). By 3.2.3, there exists N ∈ N such that N bn < an ⇐⇒
bn < N1 an . Hence, by 3.4.5,
P
bn converges.
P
Proposition 3.4.6 (Abel-Olivier-Prinsgheim). Let (an ) be a decreasing, nonnegative sequence. If an
converges, then lim nan = 0.
Proof. Given ϵ > 0, by 3.3.2, there exists N ∈ N such that for all n ≥ N ,
2n
X
ak < ϵ.
n+1
Then lim na2n = 0 and therefore lim 2na2n = 0 by 1.2.9. Thus also
2n + 1
0 ≤ (2n + 1)a2n+1 ≤ (2na2n ).
2n
Proof.
1. Suppose there exists N ∈ N and c > 0 such that for any n ≥ N ,
1
ρn ≥ c ⇐⇒ (ηn an − ηn+1 an+1 ) ≥ an+1 .
c
Now taking sum at both sides from N to m, we get
m+1
X 1 1
ak ≤ (η0 a0 − ηm+1 am+1 ) ≤ η0 a0 .
c c
k=N +1
P
Hence, the partial sum of (an ) is bounded, and therefore, by 3.3.2, an converges.
an ηn+1 an+1 ηn
2. Suppose there exists N ∈ N such that ρn ≤ 0 ⇐⇒ an+1 ≤ ηn ⇐⇒ an ≥ ηn+1 for all n ≥ N .
P 1 P
Since ηn diverges, by 3.4.4, an diverges.
33
Diverge if n ηn
ηn+1 − 1 ≤ 1 for sufficiently large n.
for anyPn ≥ N by 3.2.2. Then for n ≥ N , we have |an | < β n . Since β ∈ (0, 1), β n converges. Then by
P
3.3.2, |an | converges.
If α > 1, then by 3.2.2 again, there exists a subsequences (ank ) such that
p
nk
|ank | → α.
an+1 P
2. if lim inf an > 1, then an diverges.
Proof.
an+1
1. Since lim sup = α < 1, there exists β ∈ (α, 1). Then by 3.2.2, there exists M ∈ N such that
an
an+1
< β ⇐⇒ |an+1 | < β |an |. Hence, fixed k ∈ N, |aN +k | < β k |aN |. Since β < 1,
P k
an β converges.
P
Now by 3.4.5, |an | converges absolutely.
an+1 ank+1
2. Since lim inf an = γ > 1, by 3.2.3, there exists a subsequence (ank ) such that ank → γ and
ank+1 > |ank | for sufficiently large k. Since an ̸= 0 for all n ≥ N . lim inf |ank | ≥ |anN | > 0. Then by
3.3.2, (an ) converges.
34
Proposition 3.5.4. For any sequence (cn ) of positive numbers, then
cn+1 √
lim inf ≤ lim inf n cn ;
cn
√ cn+1
lim sup n cn ≤ lim sup .
cn
Proof. We shall prove the second inequality; the proof of the first is similar. Let α = lim sup cn+1cn . Then
if α = ∞, there is nothing to prove. Now suppose that α ∈ R, then for any β > α, by 3.2.2, there exists
N ∈ N such that for any n ≥ N , cn+1
cn < β ⇐⇒ cn+1 < βcn . Then for n ≥ N , we have cn < β
n−N
cN =
n −N √ √
β β cN =⇒ lim sup cn ≤ β. Since β is arbitrary, we have lim sup cn ≤ α.
n n
P
Theorem 3.5.7 (Riemann rearrangement theorem). Let (an ) be a sequence in R such that an converges
conditionally. Suppose
−∞ ≤ α ≤ β ≤ ∞.
P ′
Then there exists a rearrangement an with partial sums s′n such that
lim inf s′n = α, lim sup s′n = β.
Proof. Put
|an | + an |an | − an
pn = , qn = .
2 2
P P
Then
P pn −P qn = an , pn + qn = |an | , pn , qn ≥ 0.PThe series pn , qn must both diverge (as otherwise, both
pn and qn converge, which implies that |an | converges, a contradiction).
Now let P1 , P2 , P3 , · · · denote the nonnegative terms of (an ), in the order in which they occur, and let
Q1 , Q2 , Q3 , · · ·P
be thePabsolute values of Pthe negative
P terms of (an ), also in their original order.
The series Pn , Qn differ from pn , qn only by zero terms, and are therefore divergent.
Choose real sequences (αn ), (βn ) such that αn → α, βn → β, αn < βn , β1 > 0.
Let m1 , k1 be the smallest integer such that
P1 + · · · + Pm1 > β1 ,
P1 + · · · + Pm1 − Q1 − · · · − Qk1 < α1 .
Let m2 , k2 be the smallest integers such that
P1 + · · · + Pm1 − Q1 − · · · − Qk1 + Pm1 +1 + · · · + Pm2 > β2 ,
P1 + · · · + Pm1 − Q1 − · · · − Qk1 + Pm1 +1 + · · · + Pm2 − Qk1 +1 − · · · − Qk2 < α2
P P
and continue this way. This is possible since Pn and Qn diverge.
If xn , yn denote the partial sum of this rearrangement whose last terms are Pmn and −Qkn , then
|xn − βn | ≤ Pmn , |yn − αn | ≤ Qkn .
Since Pn → 0 and Qn → 0 as n → ∞, we see that xn → β and yn → α.
Finally, it is clear that no number less than α for greater than β can be a subsequential limit of the partial
sums of this rearrangement by its construction.
35
3.6 Summation by Parts
Lemma 3.6.1 (summation by parts, Abel’s lemma). Suppose (an ) and (bn ) are two sequences in R. Then
for 0 ≤ m ≤ n,
Xn n−1
X
ak ∆bk = an bn+1 − am bm − bk+1 ∆ak
k=m k=m
Corollary 3.6.1.1. Suppose (an ) and (bn ) are two sequences in R. Put
n
X
An = ak
k=0
by 3.6.1.
Proposition 3.6.2 (Dirichlet’s test). Let (an ), (bn ) be two sequences in R. Suppose
1. the partial sum An of (an ) form a bounded sequence;
2. (bn ) is decreasing;
3. lim bn = 0.
P
Then an bn converges.
ϵ
Proof. Take M such that |An | ≤ M for all n ∈ N. Given ϵ > 0, there is an integer N such that bN ≤ 2M .
Then for N ≤ m ≤ n, we have
n n−1 n−1
!
3.6.1.1
X X X
ak bk = An bn − Am−1 bm − Ak ∆bk ≤ M bn + bm + bk − bk+1 ≤ 2M bm ≤ ϵ.
k=m k=m k=m
P
Then by 3.3.2, ak bk converges.
Corollary 3.6.2.1 (alternating test). Suppose (an ) is a sequence in R such that
1. (|an |) is decreasing;
2. a2m−1 ≥ 0, a2m ≤ 0 (m ∈ N);
3. lim an = 0.
P
Then an converges.
P (−1)n
Example. n converges.
36
3.7 Power Series
Definition 3.7.1. Given a sequence (an ) of real numbers, the series
∞
X
cn xn
n=0
cn xn .
P
Remark. R is called the radius of convergence of
∆an
there exists N ∈ N such that n ≥ N implies ∆bn < r and bn > 0. Hence, for n ≥ N , it follows that
an+1 bN aN
<r 1− + := cn .
bn+1 bn+1 an+1
an an
Since cn → r as n → ∞, bn < q for sufficiently large n. Similarly, if −∞ < L ≤ ∞ and p > L, bn > p.
Hence,
If L = ∞, then for any p ∈ R, an
bn > p for sufficiently large n and therefore an
bn → ∞.
If L = −∞, then for any q ∈ R, abnn < q for sufficiently large n and therefore abnn → −∞.
37
If L ∈ R, then for any p, q ∈ R with p < L < q, we have p < an
bn < q for sufficiently large n and
therefore abnn → L.
2. We shall prove that bn > 0 for all n ∈ N. Suppose that for some N ∈ N, bN ≤ 0. Then since (bn ) is
strictly decreasing, bN +1 < 0. Furthermore, for any n ≥ N + 1, |bn | > |bN +1 |, a contradiction. Suppose
−∞ ≤ L < ∞. Then there exists r, q such that L < r < q. Since ∆a ∆an
∆bn → L, ∆bn < r for sufficiently
n
large n. Notice that an = (an − an+1 ) + · · · + (am − am+1 ) + am+1 for all m ≥ n. Hence, for sufficiently
large n and m ≥ n, we have
an bm+1 am+1
an < r(bn − bm+1 ) + am+1 ⇐⇒ <r 1− + := cm .
bn bn bn
Since cm → r as m → ∞, abnn < q for all sufficiently large n. Similarly, if −∞ < L ≤ ∞ and L < p,
an
bn < p for sufficiently large p. Then the rests are similar.
Definition 3.8.2. For a sequence (an ) of real numbers. We define the Cesaro sum σn by
a1 + · · · + an
σn = .
n
Proposition 3.8.3. Suppose (an ) is a sequence in R that converges to a ∈ R. Then its Cesaro sum (σn )
also converges to a.
Proof. Since σn = a1 +···+an
n , by 3.8.1, lim σn = lim an+1
1 = a since an → a.
Put
γn = a0 βn + a1 βn−1 + · · · + an β0 .
38
P
Hence, it suffices to show that γn → 0 since An B → AB. Let α = |an |. Given ϵ > 0, since βn → 0, there
exists N ∈ N such that for any n ≥ N , |βn | < ϵ. Therefore, for n ≥ N ,
By 3.2.4, we have
lim sup |γn | ≤ ϵα.
Since ϵ is arbitrary, γn → 0, as required.
Remark. The product of two convergent series may actually diverge. The series
∞
X (−1)n 1 1
√ = 1 − √ + √ − ···
n=0
n+1 2 3
converges by 3.6.2.1. However, if we product of the series with itself and obtain
n
X 1
cn = (−1)n p .
k=0
(n − k + 1)(k + 1)
Since
n 2 n 2 n 2 n
X 2 2(n + 1)
(n − k + 1)(k + 1) = +1 − −k ≤ +1 =⇒ |cn | ≥ = →1
2 2 2 n+2 n+2
k=0
P
as n → ∞. Hence, by 3.3.2, cn diverges.
Proposition 3.9.3. Let (an ) and (bn ) be two real sequences. Then
lim inf an + lim inf bn ≤ lim inf(an + bn ) ≤ lim sup(an + bn ) ≤ lim sup an + lim sup bn
3. Let α, β ∈ R. Then for any α′ > α, β ′ > β, there exists N ∈ N such that an < α′ , bn < β ′ for any
n ≥ N . Hence an + bn < α′ + β ′ . By 3.2.4 again, we have lim sup(an + bn ) ≤ α′ + β ′ . But since α′ , β ′
are arbitrary, lim sup(an + bn ) ≤ α + β = lim sup an + lim sup bn , as required.
P
Exercise 3.9.4. Let (an ) be a nonnegative sequence. Prove that the convergence of an implies the
convergence of
X √an
.
n
Proof. By Cauchy inequality, we have
m √ m √
! ! !2 v v
m m um um qX rX
X X 1 X ak X ak uX uX 1 1
ak ≥ ⇐⇒ ≤ t ak t ≤ ak .
k2 k k k2 k2
k=1 k=1 k=1 k=1 k=1 k=1
√ √
P an P an
Hence, the partial sum of n is bounded, by 3.3.2, n converges.
39
P
Proposition 3.9.5 (Abel’s test).
P Let (an ) and (bn ) be two sequences in R. If an converges and if (bn ) is
monotonic and bounded, then an bn converges.
Proof.
1. Suppose (bn ) is decreasing to b ((bn ) converges
P P βn = bn − b. Hence, βn is
by 1.2.11). Then define
decreasing
P to
P 0 andPthe partial sum of an is bounded. By 3.6.2, an βn converges and therefore
an bn = an b + an βn converges.
2. Suppose (bn ) is increasing to b. Define βn = b − bn , then the rest is similar.
Lemma 3.9.6. Let (an ) be a nonzero sequence in R. Then lim |an | = 0 if and only if lim |a1n | = ∞.
1
Proof. Suppose lim |an | = 0. Then for any M > 0, since M > 0, there exists N ∈ N such that for any n ≥ N ,
1 1 1
|an | < M ⇐⇒ |an | > M . Hence, lim |an | = ∞.
Conversely, suppose lim |a1n | = ∞. Then for any ϵ > 0, since 1ϵ > 0, there exists N ∈ N such that
1 1
|an | > ϵ ⇐⇒ |an | < ϵ. Hence, lim |an | = 0.
Corollary 3.9.6.1. Let (an ) be a nonzero sequence in R. Then lim |a1n | = 0 if and only if lim |an | = ∞.
P
Exercise 3.9.7. Let (an ) be a sequence such that an > 0, sn = a1 + · · · + an and an diverges.
P an
1. Prove that 1+an diverges.
2. Prove that
an+1 an+1 am sn
+ ··· + ≥1−
sn+1 sn+1 sm sm
P an
for any m > n ≥ 1 and deduce that sn diverges.
3. Prove that
an 1 1
2
≤ −
sn sn−1 sn
P an
for n ≥ 2 and deduce that s2n converges.
N +k
X an 1
<
sn 2
n=N +1
whenever k ≥ 1. Hence,
N +k
1 X an sN sN 1
> ≥1− =⇒ ≥ ⇐⇒ sN +k ≤ 2sN ,
2 sn sN +k sN +k 2
n=N +1
a contradiction.
40
3.
an sn − sn−1 sn − sn−1 1 1
2
= 2
< = −
sn sn sn sn−1 sn−1 sn
for all n ≥ 2. Now since
N N
X an X 1 1 1 1 1
2
< − = − < ,
s
n=2 n n=2
sn−1 sn s1 sN s1
P an
by 3.3.2, s2n converges.
s0 +···+sn
Proposition 3.9.8. Let (sn ) be a sequence in R, a0 = s0 , an = sn − sn−1 for n ≥ 1 and σn = n+1 . If
(nan ) is bounded and σn converges, then sn converges to the same number.
Proof. Let M > 0 such that |nan | < M for all n ∈ N, and lim σn = σ. Given n ∈ N, for m < n,
n
m+1 1 X m+1 1
(σn − σm ) + (sn − sj ) = sn + (σn − σm ) − ((n + 1)σn − (m + 1)σm )
n−m n − m j=m+1 n−m n−m
= sn − σn .
By above, we shall estimate |sn − sj | :
n n
X X 1 (n − j) (n − m − 1)M
|sn − sj | = ak ≤M ≤M ≤ .
k j+1 m+2
k=j+1 k=j+1
41
Chapter 4
Continuity
called the limit of f at p, if for any ϵ > 0, there exists δ > 0 such that
lim f (x) = q
x→p
if and only if
lim f (pn ) = q
for every sequence (pn ) in E such that
pn ̸= p, lim pn = p.
Proof. Suppose lim f (x) = q and (pn ) is a sequence converges to p with pn ̸= p. Given ϵ > 0, since
x→p
lim f (x) = p, there exists δ > 0 such that f (B∗ (p; δ)) ⊆ B(q; ϵ). For this δ, since pn → p, there exists N ∈ N
x→p
such that for any n ≥ N , pn ∈ B∗ (p; δ) =⇒ f (pn ) ∈ B(q; ϵ). Hence f (pn ) → q.
Conversely, suppose f (x) ̸→ q as x → p. Then there exists ϵ0 > 0 such that for any δ > 0 there exists a
point x = xδ ∈ E such that d(f (x), q) ≥ ϵ0 but 0 < d(x; p) < δ. Take δn = n1 , we have construct a sequence
(pn ) such that pn → p, pn ̸= p but d(f (pn ), q) ≥ ϵ0 for all n ∈ N.
Proposition 4.1.3. Suppose E ⊆ X, a metric space, p is a limit point of E, f and g are real-valued functions
on E, and
lim f (x) = A, lim g(x) = B.
x→p x→p
Then:
1. lim (f + g)(x) = A + B;
x→p
f A
3. lim (x) = , if B ̸= 0.
x→p g B
Proof. Follows by 4.1.2 and 1.2.9.
42
Proposition 4.1.4 (Squeeze theorem). Suppose g, f, h : E ⊆ X → R are three functions mapping from
a subset E of metric space X to R. Let p be a limit point of E. If there exists δ0 > 0 such that for any
x ∈ B∗ (p; δ0 ) such that g(x) ≤ f (x) ≤ h(x) and lim g(x) = lim h(x) = q, then lim f (x) = q.
x→p x→p x→p
Proof. Given ϵ > 0, lim g(x) = lim h(x) = q, there exists δ1 , δ2 > 0 such that g(B∗ (p; δ1 )) ⊆ B(q; ϵ) and
x→p x→p
h(B∗ (p; δ2 )) ⊆ B(q; ϵ). Hence, take δ = min{δ1 , δ2 , δ3 } > 0, for x ∈ B∗ (p; δ), we have
h(B(p; δ)) = g(f (B(p; δ))) ⊆ g(B(f (p); η)) ⊆ B(h(p); ϵ).
Proposition 4.2.2. Let f1 , f2 , · · · , fk be real functions on a metric space X, and let f be the mapping of
X into Rk defined by
f (x) = (f1 (x), · · · , fk (x)),
then f is continuous if and only if each of the functions f1 , · · · , fk is continuous.
Proof. It follows from the inequalities
v
u k
uX 2
|fj (x) − fj (y)| ≤ ∥f (x) − f (y)∥ = t |fi (x) − fi (y)| ,
j=1
for j = 1, · · · , k
Corollary 4.2.2.1. If f and g are continuous functions mapping from a metric space X to Rk , then f + g
and f · g are continuous on X.
Definition 4.2.3. We define the coordinate functions φj : Rk → R, j = 1, · · · , k by φj (x1 , · · · , xk ) = xj .
(We usually abuse the notation: writing φj as xj ) A finite product of coordinate functions and a constant is
mk
called a monomial, i.e., cxm 1 m2 0
1 x2 · · · xk , where c ∈ R, m1 , · · · , mk ∈ N ∪ {0} (we define xj ≡ 1). A finite
sum of monomials is called a polynomial.
Proposition 4.2.4. All polynomials are continuous.
Proof. By 2.3.5, we only need to show that the coordinate functions φj : Rk → R are all continuous, which
is clear by
|φj (x) − φj (y)| ≤ ∥x − y∥ .
Proof. Since
|∥x∥ − ∥y∥| ≤ ∥x − y∥ ,
we are done.
43
Definition 4.2.6. We say that a function f : X → Y , where X and Y are metric spaces, is a homeomor-
phism if f is a continuous bijection and f −1 is also continuous.
Proposition 4.2.7. Suppose f : X → Y , where X and Y are metric spaces, is a continuous bijection. Then
if X is compact, f is a homeomorphism.
Proof. We want to show that f −1 is also continuous, which, by 2.3.3, is equivalent to prove that f (V ) is
open for any open subset V ⊆ X. Since V is open, R\V is closed and therefore compact by 2.4.3. Hence
f (R\V ) = Y \f (V ) is compact by 2.4.3 again, where the equation holds since f is a bijection. Hence, f (V )
is open since the compact set is closed by 2.4.3.
4.3 Discontinuity
Definition 4.3.1 (One sided limit). Let f be a function maps (a, b), where a, b ∈ R with a < b, into some
metric space X and let a ≤ x < b. We write
f (x+) = q ∈ X
if f (tn ) → q as n → ∞, for all sequences (tn ) in (x, b) such that tn → x. Similarly, for a < x ≤ b, we define
f (x−) = q ∈ X if f (tn ) → q as n → ∞, for all sequences (tn ) in (a, x) such that tn → x.
Proposition 4.3.2. Let f be a function maps (a, b), where a, b ∈ R with a < b, into some metric space X.
1. Let a ≤ x < b. Then f (x+) = q ∈ X if and only if for any ϵ > 0, there exists δ > 0 such that
(x, x + δ) ⊆ (a, b) and f ((x, x + δ)) ⊆ B(q; ϵ).
2. Let a < x ≤ b. Then f (x−) = q ∈ X if and only if for any ϵ > 0, there exists δ > 0 such that
(x − δ, x) ⊆ (a, b) and f ((x − δ, x)) ⊆ B(q; ϵ).
Proof. We shall only prove 1 since 1 and 2 are similar. Suppose the RHS holds. Then for any ϵ > 0, there
exist δ > 0 such that (x.x + δ) ⊆ (a, b) and f ((x, x + δ)) ⊆ B(q; ϵ). Let (tn ) be a sequence in (x, b) such that
tn → x. Then there exist N ∈ N such that for any n ≥ N , f (tn ) ∈ B(q; ϵ). Hence, f (tn ) → q as n → ∞.
Conversely, suppose the RHS does not hold. Then there exist ϵ0 ≥ 0 such that for any δ > 0 with
(x, x + δ) ⊆ (a, b), there exist t ∈ (x, x + δ) such that d(f (t), q) ≥ ϵ0 . Take N ∈ N large enough such that
(x, x + N1 ) ⊆ (a, b) (this can be done by 1.2.14.2). Let δn = N +n1
, then the corresponding (tn ) is a sequence
lies in (x, b) such that tn → x and d(f (tn ), q) ≥ ϵ9 for all n ∈ N.
Corollary 4.3.2.1. Let f be a function maps (a, b), where a, b ∈ R with a < b, into some metric space X.
Then for any x ∈ (a, b), lim f (t) = q ∈ X if and only if
t→x
f (x+) = f (x−) = q.
Definition 4.3.3. Let f be defined on (a, b). If f is not continuous at a point x, and if f (x+) and f (x−)
both exist, then f is said to have a discontinuity of the first kind, or simple discontinuity, at x.
Otherwise, the discontinuity is said to be of the second kind.
44
Proof. Since f is increasing on (a, b), for any t ∈ (a, x), f (t) ≤ f (x). Hence, sup f (t) exists and sup f (t) ≤
a<t<x a<t<x
f (x). Let sup f (x) = A, we shall prove that f (x−) exists and A = f (x−). Given ϵ > 0, by 1.2.10, there
a<t<x
exists t ∈ (a, x) such that A − ϵ < f (t) ≤ A. Since f is increasing, for any t′ ∈ (t′ , x), A − ϵ < f (t′ ) ≤ A.
Hence, take δ = x − t, we are done by 4.3.2.
The other side is similar.
Proposition 4.4.2. Let f be decreasing on (a, b). Then f (x+) and f (x−) exist at every point of x of (a, b).
More precisely,
inf f (t) = f (x−) ≥ f (x) ≥ f (x−) = sup f (t).
a<t<x x<t<b
Since x1 < x2 implies f (x1 +) ≤ f (x2 −), we see that r(x1 ) ̸= r(x2 ) if x1 ̸= x2 . Therefore, r is an injection
which sends the discontinuities of f to Q, then we are done.
4.5 Semi-continuity
Definition 4.5.1. For a function f : E ⊆ X → R = R ∪ {±∞}, where X is a metric spaces, and a limit
point p of E, we define
lim sup = lim sup{f (x) : x ∈ E ∩ B∗ (p; ϵ)},
x→p ϵ→0
45
2. If f (x0 ) > −∞, then f is lsc at x0 if and only if for any m < f (x0 ), there exists δ > 0 such that
f (x) > m for all x ∈ E ∩ B(x0 ; δ).
Proof.
1. If f is usc at x0 , then given any M > f (x0 ) ≥ lim sup f (x) = inf sup{f (x) : x ∈ E ∩ B∗ (x0 ; ϵ)}, by
x→x0 ϵ>0
1.2.17.1, there exists δ > 0 such that sup{f (x) : x ∈ E ∩ B∗ (x0 ; δ)} < M . Hence, for any x ∈ B(x0 ; δ),
f (x) < M since f (x0 ) < M .
Conversely, suppose for any M > f (x0 ), there exists δ > 0 such that f (x) < M for any x ∈ E ∩B(x0 ; δ).
Then sup{f (x) : x ∈ B(x0 ; δ)} ≤ M . Hence, for any ϵ ∈ (0, δ), sup{f (x) : x ∈ B(x0 ; ϵ)} ≤ M . Hence,
lim sup{f (x) : x ∈ B(x0 ; ϵ)} ≤ M .
ϵ→0
2. Similar as 1.
Corollary 4.5.3.1. f is continuous at x0 if and only if |f (x0 )| < ∞ and f is both usc and lsc at x0 .
Proposition 4.5.4. Let f : E ⊆ X → R be a function, where X is a metric space.
2. If f (x) > −∞ for all x ∈ E, E is compact, and f is lsc, then f attains minimum in E.
Proof. 2 is similar to 1, so we would only prove 1. Put Ek = {x ∈ E : f (x) < k}, where k ∈ N. Then by
4.5.4, each Ek is open in E. Since f (x) < ∞ for all x S ∈ E, {Ek : k ∈ N} is an open cover of E. Since E
n
is compact, there exists k1 , · · · , kn ∈ N such that E = j=1 Ekj . Take N = max{k1 , · · · , kn }, then f (x) is
bounded by N . Therefore, sup f (X) exists.
Let sup f (X) = M , by 1.2.10, for any k ∈ N, Ck = {f (x) ≥ M T − k1 } is not empty and closed in E (by
∞
4.5.4). Notice that {Ck } has FIP property (recall 2.4.5), hence C = k=1 Ck is not empty. Let x0 ∈ C, then
1
f (x0 ) ≥ M − n for all n ∈ N. Therefore, by 1.2.9, f (x) ≥ M and hence f (x) = M since M = sup f (X), as
required.
f (E) ⊆ f (E)
for every set E ⊆ X. Show, by the example, that f (E) can be a proper subset of f (E).
46
Proof. Let x ∈ E. Then by 2.2.2, there exists a sequence (xn ) in E such that xn → x. By 2.3.2, f (xn ) → f (x).
Since (f (xn )) is a sequence in f (E), by 2.2.2 again, f (x) ∈ f (E).
Set f : (0, ∞) → R by f (x) = x1 . Then f in continuous in (0, ∞). Let E = N, then E = N. Therefore,
1 1
f (E) = :n∈N ⊊ : n ∈ N ∪ {0} = f (E).
n n
Proposition 4.6.2. Let f be a continuous real function on a metric space X. Let Zf be the set of all p ∈ X
at which f (p) = 0. Then Z(f ) is closed.
Proof. Let p be a point in Z(f ). Then there exists a sequence (pn ) in Z(f ) such that pn → p. Then since
f (pn ) = 0 for every n ∈ N, and by 2.3.2, f (pn ) → f (p), f (p) = 0 ⇐⇒ p ∈ Z(f ). Then we are done by
2.1.9.
Proposition 4.6.3. Let f and g be continuous mappings of a metric space X into a metric space Y , and
let E be a dense subset of X. Then f (E) is dense in f (X) and if f (p) = g(p) for all x ∈ E, f ≡ g.
Proof. Since E is dense in X, 4.6.1, f (E) = f (X) ⊆ f (E) =⇒ f (E) = f (X) and therefore f (E) is dense in
f (X). Now suppose f (p) = g(p) for all x ∈ E. Then a continuous function h ≡ f − g is zero on E. Hence,
h(X) = h(E) ⊆ h(E) = {0} = {0} and h ≡ 0 ⇐⇒ f ≡ g.
Proposition 4.6.4. If f is a real continuous function defined on a closed set E ⊆ R, prove that there exists
continuous real functions g on R such that g|E ≡ f .
S
Proof. By 2.5.8.1, X\E is an at most countable union of open intervals, say R\E = i∈C (ai , bi ), where C is
at most countable then we define a function g : R → R by
f (x), x ∈ E;
f (bi )−f (ai )
f (ai ) + bi −ai (x − ai ), x ∈ (ai , bi ) : finite interval;
g(x) = f (ai ), x ∈ (ai , bi ) : ai ∈ R, bi = ∞;
f (bi ), x ∈ (ai , bi ) : ai = −∞, bi ∈ R;
0, x ∈ (ai , bi ) : ai = −∞, bi = ∞.
1. Let p is a limit point of E. We shall prove that g(p+) = g(p−) = g(p)(= f (p)). Since f is continuous
at p, for any ϵ > 0, there exists δ ′ > 0 such that
by construction of g. Hence, |f (ai ) − f (p)| , |f (bi ) − f (p)| < ϵ and thus |f (x) − f (p)| < ϵ.
Consequently, g is continuous at p.
2. If p is not a limit point of E, then there exists δ > 0 such that B∗ (p; δ) ∩ E = ∅. Hence, there exists
two open intervals I = (ai , bi ), J = (aj , bj ) such that bi = aj = p. So g(p+) = g(p−) = g(p) by the
construction of g, which says that g is continuous at p.
47
Corollary 4.6.4.1. For a function f : E ⊆ R → Rk , where E is closed. There exists a continuous extension
g : R → Rk of f .
Proposition 4.6.5. If f : E ⊆ Rk → Rn , we define the graph of f , Gf , by Gf = {(x, f (x)) : x ∈ E} ⊆
Rk+m . Suppose that E is compact, prove that f is continuous if and only if Gf is compact.
Proof. Since x, f are both continuous function, the function f : E → Gf defined by f (x) = (x, f (x)) is
continuous and f (E) = Gf . Therefore, Gf is compact by 2.4.3.
Conversely, let π : Gf → E be a projection map defined by
π(x, f (x)) = x,
then f ◦ π and π ◦ f are both identity functions, which implies that π is a bijection. Since π is continuous
(∥π(x, f (x)) − π(y, f (y))∥ = ∥x − y∥) on a compact set Gf , by 4.2.7, f is continuous. By 4.2.2, f is continuous.
Proof. By 2.3.9, f can be extended into f : E → Y , which is also uniformly continuous. Since E is closed
and bounded, by 2.4.5.1, E is compact. Hence, by 2.4.3, f (E) is compact and hence bounded. Since
f (E) = f (E) ⊆ f (E), f (E) is bounded.
Proposition 4.6.7. Suppose K and F are disjoint sets in a metric space X, K is compact and F is closed.
Then there exists δ > 0 such that d(p, q) > δ if p ∈ K and q ∈ F .
Proof. Since K and F are disjoint, the function d(x, F ) : X → R is a continuous function (by 2.3.4.1) which is
positive on K. By 2.4.3.1, there exists m ∈ K such that d(m, F ) ≤ d(x, F ) for all x ∈ K. Let δ = d(m,F
2
)
> 0.
Then for any p ∈ K, q ∈ F , d(p, q) ≥ d(p, F ) ≥ d(m, F ) > δ.
Exercise 4.6.8. Suppose f : R → R such that for any a, b, c ∈ R with f (a) < c < f (b), there exists
x ∈ R such that f (x) = c for some x between a and b. Suppose also, for every rational r, that the set
Er = {x ∈ R : f (x) = r} is closed. Show that f is continuous.
Proof. Suppose, for a contradiction, f is not continuous at some x0 ∈ R. Then there exists a sequence (xn )
converges to x0 but f (xn ) ̸→ f (x0 ) by 2.3.2. Hence, there exists a subsequence (x′n ) and ϵ0 > 0 such that
|f (x′n ) − f (x0 )| ≥ ϵ0 for all n ∈ N.
WLOG, suppose that there exists a subsequence (x′′n ) of (x′n ) such that f (x′′n ) ≥ f (x0 ) + ϵ0 > f (x0 ) for
all n ∈ N. By 1.2.14.1, there exists r ∈ Q such that f (x′′n ) ≥ f (x0 ) + ϵ0 > r > f (x0 ). Then by assumption,
there exists x′′′ ′′ ′′′ ′′
n ∈ (x0 , xn ) such that f (xn ) = r for all n ∈ N. However, since (xn ) is a subsequence of (xn ),
x′′n → x0 and therefore x′′′ n → x 0 by 1.2.9, but f (x 0 ) ̸
= r, a contradiction.
Exercise 4.6.9. We say that a function f : X → Y , where X and Y are metric spaces, is open, if f (V )
is open for any open subset V ⊆ X. Prove that every continuous open mapping of R into R is strictly
monotonic.
Proof. Suppose not. Then there exist a, b, c ∈ R such that a < c < b and
Now suppose f (a) ≤ f (c) ≥ f (b). Since f is continuous on [a, b], by 2.4.3.1, f can attain the maximum value
M ∈ R on [a, b].
If f (a) = M or f (b) = M , then f (c) = M .
If f (a) < M and f (b) < M , then there exists d ∈ (a, b) such that f (d) = M .
In either case, there exists d ∈ (a, b) such that f (d) = M . Since (a, b) is open, f ((a, b)) is open and
M ∈ f ((a, b)). Hence, there exists e ∈ (a, b) such that f (e) > M , a contradiction.
The case that f (a) ≥ f (c) ≤ f (b) is similar (consider the minimum value m).
48
Lemma 4.6.10. Let α be an irrational number. Then for any N ∈ N, there exists k, h ∈ Z such that
0 < k ≤ N and
1
|kα − h| < .
N
Proof. For any x ∈ R, define [x] = max{n ∈ Z : n ≤ x} and {x} = x − [x] ∈ [0, 1). It is clear by 1.2.10 that
[x] exists uniquely for every x ∈ R.
Given N ∈ N, divide (0, 1] into N intervals:
1 1 2 N −1
0, , , ,··· , ,1
N N N N
∥(z − c) − k∥ < δ,
a contradiction.
2. We first prove that Z≥0 α + Z is dense in (0, 1]. Given x ∈ [0, 1), it suffices to show that for any ϵ > 0,
there exists a nα + m ∈ Z≥0 α + Z such that |x − (nα − m)| < ϵ. Given ϵ > 0, by 1.2.14.2, there exists
N ∈ N such that N1 < ϵ. By 4.6.10, there exists k, h ∈ N such that 0 < k ≤ N and
1
|kα − h| < < ϵ.
N
Let δ = |kα − h| > 0 (since α is irrational). Then there exists a ≥ 0 (since x ≥ 0) such that
aδ ≤ x < (a + 1)δ.
Hence,
0 ≤ x − aδ < ϵ,
then we are done.
Now for any x ∈ R and ϵ > 0, since there exists nα + m ∈ Z≥0 α + Z such that
49
Proposition 4.6.13. Let f be a convex function defined on (a, b). If a < s < t < u < b, then
f (t) − f (s) f (u) − f (s) f (u) − f (t)
≤ ≤ .
t−s u−s u−t
Therefore, every convex function is continuous.
u−t t−s u−t u−t t−s t−s
Proof. Since t = u−s s + u−s u = u−s s + 1 − u−s u = 1 − u−s s + u−s u, we have
u−t u−t f (u) − f (s) f (u) − f (t)
f (t) ≤ f (s) + 1 − f (u) ⇐⇒ ≤ ,
u−s u−s u−s u−t
u−t u−t f (t) − f (s) f (u) − f (s)
f (t) ≤ f (s) + 1 − f (u) ⇐⇒ ≤ .
u−s u−s t−s u−s
Then for any x′ , y ′ , x, y ∈ (a, b) with x ≤ x′ < y ′ , x < y ≤ y ′ , we have
f (y) − f (x) f (y ′ ) − f (x′ )
≤ .
y−x y ′ − x′
Given x ∈ (a, b), there exists c1 , c2 , d1 , d2 ∈ R such that c1 < c2 < x < d2 < d1 , take y ∈ (c2 , d2 ) but y ̸= x,
we have
f (c2 ) − f (c1 ) f (y) − f (x) f (d2 ) − f (d1 )
≤ ≤ .
c2 − c1 y−x d2 − d1
n o
Thus |f (y) − f (x)| ≤ M |y − x|, where M = max f (dd22)−f (d1 )
−d1 , f (cc22)−f
−c1
(c1 )
+ 1 > 0. Then we are done by
taking y → x and applying 4.1.4.
Proposition 4.6.14. Assume that f is a continuous function defined in (a, b) such that
x+y f (x) + f (y)
f ≤
2 2
for any x, y ∈ (a, b). Prove that f is convex.
Proof. We first prove that for any x1 , · · · , xn ∈ (a, b), we have
Pn Pn
k=1 xk f (xk )
f ≤ k=1
n n
for all n ∈ N. It clearly holds when n = 1, 2. Now suppose that when n = 2p , the hypothesis holds, then for
n = 2p+1 , we obtain
P2p+1 ! P2p ! P2p+1 !!
k=1 xk 1 k=1 xk k=2p +1 xk
f ≤ f +f
2p+1 2 2p 2p
p+1
2X
1
≤ f (xk ).
2p+1
k=1
Now, suppose that Pwhen n = 2p (p > 0), the hypothesis holds, then for n = 2p −1, take arbitrary x1 , · · · , xn ∈
n
xk
(a, b) and set α = k=1n . We, by hypothesis, deduce that
Pn Pn n
! n
α + k=1 xk k=1 x k 1 X 1X
f = f = f (α) ≤ f (α) + f (x k ) =⇒ f (α) ≤ f (xk ).
2p n 2p n
k=1 k=1
m
Now given λ = n ∈ Q ∩ (0, 1) and x, y ∈ (a, b), we have
m m−n 1
f (λx + (1 − λ)y) = f x+ y ≤ (mf (x) + (m − n)f (y)) = λf (x) + (1 − λ)f (y).
n n n
Finally, suppose λ ∈ R ∩ (0, 1), by 1.2.14.1, there exists a sequence (qn ) in (0, 1) ∩ Q such that qn → λ. Hence,
for any x, y ∈ (a, b),
2.3.2,1.2.9
f (qn x + (1 − qn )y) ≤ qn f (x) + (1 − qn )f (y) =⇒ , f (λx + (1 − λ)y) ≤ λf (x) + (1 − λ)f (y),
which proves that f is convex, as required.
50
Chapter 5
f (x + h) − f (x)
φ(h) = ,
h
where h > 0 and x + h ∈ (a, b) and say that f is differentiable at x if lim φ(t) exists. In this case, we
h→0
define f ′ (x) = lim φ(h), called the derivative of f at x. Sometimes we write f ′ (x0 ) as fx (x0 ) or d
dx f (x0 )
h→0
to emphasize that the derivative exists in x-axis.
Proposition 5.1.2. Let f : (a, b) → R be a real-valued function. Then f is differentiable at x ∈ (a, b) if and
only if there exists A ∈ R and δ > 0 such that
f (x + h) − f (x) = Ah + o(h),
o(h)
for any 0 < |h| < δ, where o(h) is a real-valued function such that lim = 0 (In fact, A = f ′ (x)).
h→0 h
Proof. Suppose f is differentiable at x ∈ (a, b), then f ′ (x) exists. Let f ′ (x) = A and given ϵ > 0; thus, by
definition, there exists δ > 0 such that
f (x + h) − f (x)
−A <ϵ
h
o(h)
whenever 0 < |h| < δ. Put f (x + h) − f (x) − Ah = o(h), then f (x + h) − f (x) = Ah + o(h) and h <ϵ
whenever 0 < |h| < δ, as required.
Conversely, consider
f (x + h) − f (x) o(h)
−A =
h h
o(h) o(h)
for any h ∈ (0, δ). Since lim = 0, for any ϵ > 0, there exists δ ′ > 0 such that h < ϵ as 0 < |h| < δ ′ .
h→0 h
Take δ ′′ = min{δ, δ ′ }, then we are done.
Corollary 5.1.2.1. Let f : (a, b) → R be a real-valued function. If f is differentiable at x ∈ (a, b), then f is
continuous at x.
Proof. Since f is differentiable at x, there exists δ > 0 such that
o(h)
where lim = 0, whenever 0 < |h| < δ. Then o(h) → 0 as h → 0, then we are done.
h→0 h
51
Corollary 5.1.2.2. Let f, g : (a, b) → R be two real functions ane let x ∈ (a, b). If f, g are differentiable at
x, then f ± g, f g, fg (if g(x) ̸= 0) are differentiable at x, and
Proof. Since f and g are differentiable at x, by 5.1.2, there exists δ > 0 such that
and
g(x + h) = g(x) + g ′ (x)h + o2 (h),
o1 (h) o2 (h)
where h , h → 0 as h → 0, whenever 0 < |h| < δ. Then:
1. (f ± g)(x + h) = (f ± g)(x) + (f ′ ± g ′ )(x)h + (o1 ± o2 )(h). By 5.1.2 again, (f ± g) is differentiable at x
and (f ± g)′ (x) = (f ′ ± g ′ )(x).
2.
1 1 −g ′ (x)h − o2 (h)
− = .
g(x + h) g(x) g(x + h)g(x)
Hence,
1 1
g(x+h) − −g ′ (x)
g(x) g(x + h) − g(x) o2 (h)
−
≤ + .
h g 2 (x) g(x + h)g 2 (x) hg(x + h)g(x)
′ ′
Then by 4.2.5, 4.1.4, 4.1.3 and 5.1.2.1, we have g1 (x) = −g g2 (x). Now apply 3, we are done.
Proposition 5.1.3 (chain rule). Let f : (a, b) → R be a real-valued function and g : (c, d) → R such that
(c, d) ⊆ f ((a, b)). If f is differentiable at x and g is differentiable at f (x), then (g ◦ f ) : (a, b) → R is
differentiable at x and (g ◦ f )′ (x) = g ′ (f (x))f ′ (x).
Proof. Let y = f (x).
1. If f ′ (x) ̸= 0, then there exists δ1 > 0 such that for any 0 < |h| < δ1 , we have
Since g is differentiable at y, there exists η1 > 0 such that for all 0 < |k| < η1 ,
Put k ′ = f (x + h) − f (x), by 5.1.2.1, k ′ → 0 as h → 0. Hence, given ϵ > 0, there exists 0 < η2 < η1
such that (∗) holds and |o2 (k)/k| < ϵ for all 0 < |k| < η2 . Since k ′ → 0 as h → 0, there exists δ2 > 0
such that |k ′ | < η2 whenever 0 < |h| < δ2 . Since |o1 (h)/h| → 0, there exists δ3 > 0 such that for any
0 < |h| < δ3 , |o1 (h)/h| < ϵ. Since f ′ (x) exists, there exists δ4 > 0 such that for any 0 < |h| < δ4 ,
|(f (x + h) − f (x))/h| < 2 |f ′ (x)|.
52
Finally, put δ = min{δ1 , δ2 , δ3 , δ4 } > 0, then 0 < |k ′ | < η2 and therefore
g(f (x + h)) − g(f (x)) = g(y + k ′ ) − g(y) = k ′ g(y) + o2 (k ′ ) = g ′ (y)f ′ (x)h + o1 (h)g ′ (y) + o2 (k ′ )
g(y + k) − g(y)
<M
k
whenever 0 < |k| < δ1 . Given ϵ > 0, since f ′ (x) = 0, there exists δ > 0 such that
f (x + h) − f (x) ϵ
|f (x + h) − f (x)| < δ1 , <
h M
(g ◦ f )(x + h) − (g ◦ f )(x)
=0<ϵ
h
f (p + h) − f (p)
≥ 0 =⇒ f ′ (p) ≥ 0.
h
Therefore, f ′ (p) = 0, as required.
Theorem 5.2.3 (generalized MVT). If f and g are continuous real valued functions on [a, b] which are
differentiable in (a, b), then there exists a point c ∈ (a, b) at which
53
Proof. Define h(x) = (f (b) − f (a))g(x) − (g(b) − g(a))f (x). Then h is continuous at [a, b]. If h is constant,
this holds for every x ∈ (a, b). If h(t) > h(a) for some t ∈ (a, b), let c be a point on [a, b] at which h attains
its maximum by 2.4.3.1. Since h(a) = h(b), c ∈ (a, b). And therefore by 5.2.2, h′ (c) = 0. For the case
h(t) < h(a) for some t ∈ (a, b) is similar.
Corollary 5.2.3.1 (MVT). If f is a continuous real-valued function defined on [a, b] and differentiable on
(a, b). Then there exists a point c ∈ (a, b) such that
Corollary 5.2.3.3 (Rolle). If f is a continuous real-valued function defined on [a, b] and differentiable on
(a, b). If f (a) = f (b), then there exists a point c ∈ (a, b) such that f ′ (c) = 0
Corollary 5.3.1.3. Suppose f is a real differentiable function on (a, b). If f ′ is monotonic, then f ′ is
continuous.
Proof. By 4.4.2.1 and 5.3.1.2, we are done.
54
5.4 L’Hospital Rules
Theorem 5.4.1 (L’Hospital Rules). Suppose f and g are real and differentiable in (a, b) and g ′ (x) ̸= 0 for
all x ∈ (a, b), where −∞ ≤ a < b ≤ ∞. Suppose
f ′ (x)
→A
g ′ (x)
as x → a. If
f (x), g(x) → 0
as x → a or if
g(x) → ∞
as x → a, then
f (x)
→A
g(x)
as x → a.
Proof. We first consider the case when −∞ ≤ A < ∞. Choose a real number q such that A < q, and then
choose r such that A < r < q. Since fg(x)
(x)
→ A as x → a and g ′ (x) ̸= 0 for all x ∈ (a, b), there exists a point
c ∈ (a, b) such that whenever x, y ∈ (a, c), we have
f ′ (x)
< r, g(x) − g(y) ̸= 0.
g ′ (x)
If a < x < y < c, then by 5.2.3, there exists t ∈ (x, y) such that
f (y)
≤ r < q,
g(y)
55
5.5 Taylor’s Theorem
Lemma 5.5.1. Given n ∈ N, (xn )′ = nxn−1 .
Proof. We first assume x > c, for the case x < c is similar. Given t ∈ [c, x], define
n−1
X f (k) (t)
F (t) = f (t) + (x − t)k
k!
k=1
and
n−1
X g (k) (t)
G(t) = g(t) + (x − t)k .
k!
k=1
Then F and G are continuous on [c, x] and differentiable on (c, x). Therefore, by 5.2.3 there exists x1 ∈ (c, x)
such that
(F (c) − F (x))G′ (x1 ) = (G(c) − G(x))F ′ (x1 ).
(x−t)n−1 (n) (x−t)n−1 (n)
Since F (x) = f (x), G(x) = g(x), F ′ (t) = (n−1)! f (t) and G′ (t) = (n−1)! g (t), by putting t = x1 , we
obtain the formula of the theorem.
Corollary 5.5.3.1. Let f : (a, b) → R be a function such that f (n) (x) exists on (a, b) and f (n−1) (x) is
continuous on (a, b) and c ∈ (a, b). Then, for every x ∈ (a, b), x ̸= c, there exists a point x1 which lies
between x and c such that
n−1
X f (k) (c) f (n) (x1 )
f (x) = f (c) + (x − c)k + (x − c)n .
k! n!
k=1
f (x + h) − f (x) = hv + o(h),
o(h)
where o(h) is a vector-valued function such that lim = 0. In this case v = f ′ (x).
h→0 h
56
Proof. Suppose f is differentiable at x, then given there exists v ∈ Rn such that
f (x + h) − f (x)
lim − v = 0.
h→0 h
Put o(h) = f (x + h) − f (x) − hv, then we are done.
Conversely, suppose there exists v and δ > 0 such that for any h ∈ R with 0 < |h| < δ, we have
f (x + h) − f (x) = hv + o(h),
o(h)
where lim = 0, then
h→0 h
f (x + h) − f (x) o(h)
−v = →0
h h
as h → 0.
Corollary 5.6.2.1. Let f = (f1 , · · · , fn ) : (a, b) → Rn be a function. Then f is differentiable at x ∈ (a, b) if
and only if fj is differentiable at x ∈ (a, b) for all j = 1, · · · , n. In this case, f ′ (x) = (f1′ (x), · · · , fn′ (x)).
Proof. Suppose f is differentiable at x ∈ (a, b), then there exists δ > 0 such that for any h ∈ R with
o(h)
0 < |h| < δ, we have f (x + h) − f (x) = hf ′ (x) + o(h) with lim = 0. Now suppose f ′ (x) = (v1 , · · · , vn )
h→0 h
and o(h) = (o1 (h), · · · , on (h)), then
fj (x + h) − f (x) = hvj + oj (h)
oj (h)
with lim = 0 for each j = 1, · · · , n. Then by 5.1.2, fj is differentiable at x and vj = fj′ (x).
h→0 h
Conversely, suppose f1 , · · · , fn is differentiable at x. Then there exists 0 < δ1 , · · · , δn such that whenever
0 < |h| < δj , we have
fj (x + h) − fj (x) = hfj′ (x) + oj (h)
oj (h)
with lim = 0, where j = 1, · · · , n. Then take δ = min{δ1 , · · · , δn } > 0; thus, we have
h→0 h
f (x + h) − f (x) = h(f1′ (x), · · · , fn′ (x)) + (o1 (h), · · · , on (h)).
By 5.6.2, we are done.
Corollary 5.6.2.2. Suppose f , g : (a, b) → Rn are two functions differentiable at x, then:
1. f ± g is differentiable at x, and (f ± g)′ (x) = (f ′ ± g′ )(x).
2. f · g is differentiable at x, and (f · g)′ (x) = (f ′ · g + f · g′ )(x).
3. If α : (a, b) → R is a function differentiable at x, then (αf ) is differentiable and (αf )′ (x) = (α′ f +αf ′ )(x).
4. If n = 3, then (f × g)′ (x) = (f ′ × g + f × g′ )(x).
Proof. By 5.6.2.1 and 5.1.2.2, we are done.
Proposition 5.6.3 (MVT on vector-valued function). Suppose f : [a, b] → Rn is a continuous mapping on
[a, b] and differentiable on (a, b), then there exists c ∈ (a, b) such that
∥f (b) − f (a)∥ ≤ (b − a) ∥f ′ (c)∥ .
Proof. Let z = f (b) − f (a) and define
φ(t) = z · f (t)
for a ≤ t ≤ b. Then φ is real-valued continuous function on [a, b] which is differentiable on (a, b). Then by
5.2.3.1, there exists c ∈ (a, b) such that
φ(b) − φ(a) = (b − a)φ′ (c).
Hence,
2
∥z∥ = (b − a) |z · f ′ (c)| ≤ (b − a) ∥z∥ ∥f ′ (c)∥ =⇒ ∥z∥ ≤ (b − a) ∥f ′ (c)∥ .
57
5.7 Uniqueness of Inivital Value Problems
Lemma 5.7.1. Suppose that f : [a, b] → Rn is continuous on [a, b] and differentiable on (a, b) with f (a) = 0,
and there is a real number A such that ∥f ′ (x)∥ ≤ A ∥f (x)∥ on (a, b). Then f (x) = 0 for all x ∈ [a, b].
Proof. It clearly holds when A = 0, now suppose A > 0.
Given x0 ∈ (a, b), let M0 = sup ∥f (x)∥ and M1 = sup ∥f ′ (x)∥. Hence, M1 ≤ AM0 since f (a) = 0.
a≤x≤x0 a<x≤x0
Then for any x ∈ (a, x0 ], by 5.2.3.1, there exists η ∈ (a, x) such that
∥f (x) − f (x)∥ = ∥f (x)∥ ≤ (x − a) ∥f ′ (η)∥ ≤ M1 (x0 − a) ≤ AM0 (x0 − a) =⇒ M0 ≤ AM0 (x0 − a).
1
If M0 = 0, then f (x) = 0 for all x ∈ [a, x0 ]. Now suppose M0 > 0, then take x0 such that A(x0 − a) ≤ 2
and thus
M0
0 ≤ M0 ≤ =⇒ M0 = 0
2
and therefore f (x) = 0 for all x ∈ [a, x0 ].
Given ϵ > 0, take a partition P = {x−1 = a < x0 < x1 < · · · < xn = b − ϵ} such that
1
A max (xj − xj−1 ) < .
0≤j≤n 2
We may assume f (x0 ) = 0 by above, then proceed the argument above inducitvely, we obtain that f (x) = 0
for x ∈ [a, b − ϵ]. Since ϵ > 0, f (x) = 0 for all a ≤ x < b. Finally, since f is continuous at b, f (b) = 0 and we
are done.
Proposition 5.7.2. Let φ be a real function defined on a rectangle R in the plane, given a ≤ x ≤ b, α ≤
y ≤ β. A solution of the initial-valued problem
is a continuous function f : [a, b] → R which is differentiable on (a, b) such that f (a) = c, α ≤ f (x) ≤ β, and
Then such a problem has at most one solution if there is a constant A such that
Proof. Suppose f1 and f2 are two solutions. Let f = f1 − f2 . Then f (a) = 0 and
58
Exercise 5.8.2. Suppose f ′ (x) > 0 in (a, b). Prove that f is strictly increasing in (a, b), and let g be its
inverse function. Prove that g is differentiable, and that
1
g ′ (f (x)) =
f ′ (x)
for all x ∈ (a, b).
Proof. By 5.2.3.2, f is strictly increasing. Then since g(f (x)) = x for all x ∈ (a, b), g(f (x)) is differentiable
and (g ◦ f )′ (x) = 1 for all x ∈ I = (a, b). Notice that this implies
(g ◦ f )(t) − (g ◦ f )(x)
lim = 1.
t→x t−x
Hence, given f (x), f (t) ∈ f (I), we have
(g ◦ f )(t) − (g ◦ f )(x) f (t) − f (x)
→1
f (t) − f (x) t−x
as t → x. Then since f ′ (x) > 0, we are done.
Proposition 5.8.3. Let f be a continuous real function on (a, b) and f ′ (x) exists for all (a, b)\{c}, where
c ∈ (a, b). If f ′ (x) → d as x → c, then f ′ (c) exists and f ′ (c) = d.
Proof.
f (c + h) − f (c) 5.4.1 f ′ (c + h)
lim = lim = lim f ′ (c + h) = d.
h→0 h h→0 1 h→0
Proposition 5.8.4. Let f be a differentiable real function defined in (a, b). Prove that f is convex if and
only if f ′ is increasing. Assume that f ′′ (x) exists for every x ∈ (a, b), and prove that f is convex if and only
if f ′′ (x) ≥ 0 for all x ∈ (a, b).
Proof. Suppose x, y ∈ (a, b) with x ̸= y and 0 < λ < 1. Since
λf (x) + (1 − λ)f (y) ≥ f (λx + (1 − λ)y) = f (y + λ(x − y)).
Hence,
f (y + λ(x − y)) − f (y)
f (x) − f (y) ≥ f (y + λ(x − y)) − f (y) = (x − y).
λ(x − y)
Let λ → 0, we have
f (x) − f (y) ≥ f ′ (y)(x − y).
If x > y, we have
f (x) − f (y) y→x
≥ f ′ (y) =⇒ f ′ (x) ≥ f ′ (y).
x−y
Conversely, suppose f ′ is increasing. Given x, y ∈ (a, b) with x > y, by 5.2.3.1, there exists η ∈ (y, x) such
that f (x) − f (y) = f ′ (η)(x − y). Since f ′ is increasing, we have
f ′ (x)(x − y) ≥ f (x) − f (y) = f ′ (η)(x − y) ≥ f ′ (y)(x − y).
Let z = λx + (1 − λ)y, then
f (x) − f (z) ≥ f ′ (z)(x − z) ⇐⇒ f (x) ≥ f (z) + f ′ (z)(x − z)
and
f (z) − f (y) ≤ f ′ (z)(z − y) ⇐⇒ f (y) ≥ f (z) + f ′ (z)(y − z).
Hence,
λf (x) + (1 − λ)f (y) ≥ f (z) + f ′ (z)(λx − λz + (1 − λ)y − (1 − λ)z) = f (z) = f (λx + (1 − λ)y).
Suppose f ′′ exists. Then
f is convex ⇐⇒ f ′ is monotonicity increasing ⇐⇒ f ′′ ≥ 0.
59
Proposition 5.8.5 (Landau-Kolmogorov inequality on half-line). Suppose a ∈ R, f is twice-differentiable
real function on (a, ∞), and M0 , M1 and M2 are least upper bounds of |f (x)| , |f ′ (x)| , |f ′′ (x)|, respectively,
on (a, ∞). Then
M12 ≤ 4M0 M2 .
Proof. Given h > 0 and x ∈ (a, ∞), by 5.5.3.1, there exists η ∈ (x, x + 2h) such that
1
f (x + 2h) = f (x) + 2hf ′ (x) + 2h2 f ′′ (η) ⇐⇒ f ′ (x) = (f (x + 2h) − f (x)) − hf ′′ (η).
2h
Hence,
1 1
|f ′ (x)| ≤ |f (x + 2h) − f (x)| + h |f ′′ (η)| ≤ M0 + hM2 .
2h h
for all x ∈ (a, ∞). Then,
M0
M1 ≤ + hM2 .
h
If M0 = 0, then f (x) = 0 for all x ∈ (a, ∞) and therefore M1 = 0.
Suppose M2 = 0, then f ′′ (x) = 0 for all x ∈ (a, ∞) and therefore f ′ (x) = α for some α > 0. By 5.2.3.2
apply on g(x) = f (x) − αx, we have g(x) = β for some β ∈ R ⇐⇒ f (x) = αx + β. Hence, M0 = ∞ if
α ̸= 0 and M0 = |β| if α = 0.
q
Hence, we assume that M0 , M2 > 0 and put h = M M2 . Therefore,
0
p
M1 ≤ 2 M0 M2 ⇐⇒ M02 ≤ 4M0 M2 ,
as required.
Exercise 5.8.6. Suppose f is twice-differentiable on (0, ∞), f ′′ is bounded on (0, ∞) and f (x) → 0 as x → 0.
Then f ′ (x) → 0 as x → ∞.
Proof. Given ϵ > 0, there exists M > 0 such that whenever x > M , |f (x)| < ϵ2 . Let M0 , M1 and M2 be
the least upper bound of f, f ′ and f ′′ on (M, ∞) respectively. Then M0 ≤ ϵ Since f ′′ is bounded on (0, ∞),
M2 ∈ R. Then by 5.8.5,
M12 ≤ 4M0 M2 ≤ 4M2 ϵ2 .
√ √
Therefore, M1 ≤ 2 M2 ϵ. Hence, |f ′ (x)| ≤ 2 M2 ϵ. Since ϵ is arbitrary, f ′ (x) → 0 as x → ∞.
Exercise 5.8.7. If
C1 Cn
+ ··· +
C0 + = 0,
2 n+1
where C0 , · · · , Cn ∈ R, prove that the equation
C0 + C1 x + · · · + Cn−1 xn−1 + Cn xn = 0
60
Proof. By 5.2.3.2, it is sufficient to prove that
xf ′ (x) − f (x)
g ′ (x) = ≥ 0 ⇐⇒ xf ′ (x) − f (x) ≥ 0
x2
for x > 0. Given x > 0, by 5.2.3.1,
f (x) − f (0) = f (x) = xf ′ (η)
for some η ∈ (0, x). Since f ′ is increasing,
as required.
61
Chapter 6
Riemann-Steljtes Integral
respectively.
It is clear that for any refinement P ′ , we have L(g, f, P ) ≤ L(g, f, P ′ ) ≤ U (g, f, P ′ ) ≤ U (g, f, P ).
We now define the lower R-S integral and upper R-S integral by
Z b Z b
gdf = sup L(g, f, P ) and f dg = inf U (g, f, P ).
a P a P
Z b Z b
If f dg, then we say that g is R-S integrable w.r.t f over I, and we write g ∈ R(f, I). This
gdf =
a a
Z b
common value is denoted by gdf . We say that g is the integrand and f is the integrator.
a
If f (x) = x, then we say that g is Riemann integrable over I, and we write g ∈ R(I). (In this case,
“f ” in the notations of upper sum and lower sum are omitted.)
Since if a = b, any upper sum and any lower sum equal to 0 for any bounded function g, we will always
assume that the interval discussed below is not degenerated, that is, only containing a single point.
Z a Z b
By convention, we define gdf = − gdf .
b a
Proposition 6.1.2. Let I = [a, b] and g be a bounded real-valued function defined on I and let α be an
increasing function on I. Then g ∈ R(α, I) if and only if for any ϵ > 0, there exists a partition P of I such
that
U (g, α, P ) − L(g, α, P ) < ϵ.
Proof. Suppose g ∈ R(α, I). Given ϵ > 0, there exist partitions P1 , P2 of I such that
Z b Z b
ϵ ϵ
U (g, α, P ) − gdα < and gdα − L(g, α, P2 ) < .
a 2 a 2
Take P = P1 ∪ P2 , then U (g, α, P ) − L(g, α, P ) < ϵ.
Now suppose the converse holds. Then for any ϵ > 0, we have
Z b Z b
0≤ gdα − gdα < ϵ.
a a
62
Since ϵ is arbitrary, we are done.
Corollary 6.1.2.1. Let I = [a, b] and g is a bounded real-valued function defined on I and let α be an
increasing function on I. If g ∈ R(α, I), then for any ϵ > 0, there exists a partition P = {x0 , · · · , xn } of [a, b]
such that for any pi ∈ [xi , xi+1 ], i = 0, · · · , n − 1,
n−1
X Z b
g(pi )∆i α − gdα < ϵ.
i=0 a
Also, since Z
L(g, α, P ) ≤ gdα ≤ U (g, α, P ),
we are done.
Remark. Notice that we can replace P by any refinement P ′ with the result remaining true.
Corollary 6.1.2.2. Let I = [a, b] and g is a bounded real-valued function defined on I and let α be an
increasing function on I. If g ∈ R(α, I) and [c, d] ⊆ [a, b], then g ∈ R([c, d]). Moreover, if a < c < b, then
Z b Z c Z b
gdα = gdα + gdα.
a a c
Proof. Since g ∈ R(α, I), for any ϵ > 0, by 6.1.2, there exists a partition P = {x0 , · · · , xn } of [a, b] such that
Now discard the nonnegative terms (Mk − mk )∆k α such that [xk , xk+1 ] ⊆ R\[c, d], by 6.1.2 again, g ∈
R(α, [c, d]).
Let P be a partition of [a, b], then P ′ = P ∪ {c, d} is a refinement of [a, b] and therefore
Z c Z b
′
U (g, α, P ) ≥ U (g, α, P ) ≥ gdα + gdα
a c
Z c
since U (g, α, P ′ ) can be partitioned into the upper sums of the intervals [a, c] and [c, b]. Hence, gdα +
Z b a
gdα is a lower bound of all upper sums of [a, b]. Now given ϵ > 0, there exist partitions P1 and P2 of
c
[a, c] and [c, b], respectively, such that
Z c Z b
ϵ
0 ≤ U (g, α, P1 ) − gdα, U (g, α, P2 ) − gdα < .
a c 2
Since P = P1 ∪ P2 is a partition of [a, b] and
63
Lemma 6.1.3. Let g : X → R be a function, where X is a nonempty set and let sup g(X) = M, inf g(X) = m.
Then
M − m = sup |g(x) − g(y)| .
x,y∈X
Proof. Since X is nonempty, we have M ≥ m, m < ∞, M > −∞ and therefore the LHS is well defined. Given
x, y ∈ X, since m ≤ g(x), g(y) ≤ M , we have |g(x) − g(y)| ≤ M −m and therefore sup |g(x) − g(y)| ≤ M −m.
If M − m = 0, the equality clearly holds. Now suppose M > m.
Take M ′ < M, m′ > m such that M ′ > m′ . By 1.2.10, there exists x, y ∈ X such that g(x) > M ′ and
g(y) < m′ . Hence, g(x) − g(y) = |g(x) − g(y)| > M ′ − m′ . Since M ′ , m′ is arbitrary, we have M − m =
sup |g(x) − g(y)|, as required.
Proposition 6.1.4. Let I = [a, b] and g be a bounded real-valued function defined on I and let α be an
increasing function on I. If there is a number L such that for any ϵ > 0, there exists a partition Pϵ of I such
that for any refinement P = {x0 , · · · , xn } and any pi ∈ [xi , xi+1 ],
n−1
X
g(pi )∆i α − L < ϵ,
i=0
Z b
then g ∈ R(α, I) and L = gdα.
a
Proof. Given ϵ, η > 0, there exists a partition P = {x0 , · · · , xn } of I such that for any pk ∈ [xk , xk+1 ],
n−1
X ϵ
g(pk )∆k α − L < .
i=0
4
Now, by 6.1.3, there exists xk , yk ∈ Ik such that Mk − mk < g(xk ) − g(yk ) + η. Hence,
n−1 n−1
! n−1
!
X X X
U (g, α, P )−L(g, α, P ) = (Mk −mk )∆k α ≤ g(xk )∆k α − L + L − g(yk )∆k α +η(α(b)−α(a)).
k=0 k=0 k=0
Hence,
ϵ
U (g, α, P ) − L(g, α, P ) ≤ + η(α(b) − α(a)).
2
Since η is arbitrary, by 6.1.2, g ∈ R(α, I).
For the equality, since pk are arbitrary,
ϵ ϵ
U (g, α, P ) ≤ L + , L(g, α, P ) ≥ L − .
4 4
Notice that Z b Z b
ϵ ϵ
L − ≤ L(g, α, P ) ≤ gdα ≤ U (g, α, P ) ≤ L + =⇒ gdα − L < ϵ,
4 a 4 a
Z b
and therefore gdα = L.
a
Proposition 6.1.5. Let I = [a, b] and g be a bounded real-valued function defined on I and let α be an
increasing function on I. If g is continuous on [a, b], then g ∈ R(α, I).
Proof. Since g is continuous on [a, b], by 2.4.5.1 and 2.4.7, g is uniformly continuous on [a, b].
Let ϵ > 0 be given, take η > 0 such that (α(b) − α(a))η < ϵ. Since g is uniformly continuous, there exists
δ > 0 such that whenever x, y ∈ [a.b] with |x − y| < δ, we have |g(x) − g(y)| < η.
If P = {x0 , · · · , xn } is any partition of [a, b] such that ∥P ∥ < δ, then Mi − mi ≤ η (by 6.1.3) and therefore
n−1
X
U (g, α, P ) − L(g, α, P ) = (Mi − mi )∆i α ≤ η(α(b) − α(a)) < ϵ.
i=0
64
Proposition 6.1.6. If g is monotonic on I = [a, b], and if α is increasing and continuous on [a, b], then
g ∈ R(α, I).
Proof. Let ϵ > 0 be given. For any positive integer n, choose a partition P = {x0 , · · · , xn } such that
jα(b) + (n − j)α(a)
α(xj ) = , j = 0, · · · , n.
n
This is possible since α is continuous (2.5.4.1).
We suppose that g is increasing (for the case g is decreasing is similar). Then
Mj = g(xj ), mj = g(xj−1 )
for j = 1, · · · , n. Hence,
n
α(b) − α(a) X α(b) − α(a)
U (g, α, P ) − L(g, α, P ) = (g(xi ) − g(xi−1 )) = (g(b) − g(a)) < ϵ
n j=1
n
Let h = φ◦g, Ij = [Ij , Ij+1 ] for j = 0, · · · , n−1, Mj = supIj g(x), mj = inf Ij g(x), Mj∗ = supIj h, m∗j = inf Ij h.
Divide the number 0, · · · , n into two classes: i ∈ A if Mi − mi < δ, i ∈ B if Mi − mi ≥ δ.
For i ∈ A, our choice of δ shows that Mi∗ − m∗i ≤ ϵ (by 6.1.3).
For i ∈ B, let K = sup |φ(t)|, then Mi∗ − m∗i ≤ 2K. Hence,
X X X
δ ∆i α ≤ (Mi − mi )∆i α < δ 2 =⇒ ∆i α < δ.
i∈B i∈B i∈B
65
Therefore,
X X
U (h, α, P ) − L(h, α, P ) = (Mi∗ − m∗i )∆i α + (Mi∗ − m∗i )∆i α ≤ ϵ(α(b) − α(a)) + 2Kδ.
i∈A i∈B
Corollary 6.1.8.1. Let I = [a, b] and g be a bounded real-valued function defined on I and let α be an
increasing function on I. If g ∈ R(α, I), then |g| ∈ R(α, I), g 2 ∈ R(α, I).
Lemma 6.1.9. Let c ̸= 0 and X ⊆ R. Define cX = {cx : x ∈ X}. Then:
1. if c > 0, sup(cX) = c sup(X) and inf(cX) = c inf(X).
Z b Z b Z b
5. If g1 ∈ R(β, I), then g1 ∈ R(α + β, I) and g1 d(α + β) = g1 dα + g1 dβ.
a a a
Z b Z b
6. For any c ≥ 0, g1 ∈ R(cα, I) and g1 d(cα) = c g1 dα.
a a
7. g1 g2 ∈ R(α, I).
Proof.
1. Given ϵ > 0, by 6.1.2, there exists a partition P1 and P2 of [a, b] such that
ϵ
U (g1 , α, P1 ) − L(g1 , α, P1 ), U (g2 , α, P2 ) − L(g2 , α, P2 ) < .
2
66
Take P = P1 ∪ P2 , then we have
if c < 0, then
Z b
U (cg1 , α, P ) = cL(g1 , α, P ) ≥ c g1 dα.
a
Z b Z b
Since P is arbitrary, (cg1 )dα ≥ c g1 dα. Apply similar argument on the lower sum, we deduce
Z b a a
Z b
that (cg1 )dα ≤ c g1 dα.
a a
3. Let g = g2 − g1 , then by 1 and 2, g ∈ R(α, I). Since g ≥ 0 every lower sum of g is not less than 0.
Z b Z b Z b
Hence, gdα ≥ 0 ⇐⇒ g2 dα ≥ g1 dα.
a a a
67
Hence g ∈ R(α + β, I). Let Π be a partition of [a, b], since
Z b Z b
U (g1 , α + β, Π) = U (g1 , α, Π) + U (g1 , β, Π) ≥ g1 dα + g1 dβ,
a a
Z b Z b
g1 dα + g1 dβ is a lower bound of upper sum. With same ϵ, there exist partitions P3 , P4 of [a, b]
a a
such that Z b Z b
ϵ
U (g1 , α, P3 ) − g1 dα, U (g1 , β, P4 ) − g1 dβ < .
a a 2
′
Take P = P3 ∪ P4 , then
! !
Z b Z b Z b Z b
′
U (g1 , α + β, P ) − g1 dα + g1 dβ = U (g1 , α, P ) + U (g1 , β, P ) − g1 dα + g1 dβ < ϵ.
a a a a
Z b Z b Z b
Then by 1.2.10, g1 d(α + β) = g1 dα + g1 dβ.
a a a
Lemma 6.1.12. If a < s < b, g is bounded on [a, b], continuous at s and α(x) = I(x − s), then g ∈ R(α, I)
and Z b
gdα = g(s).
a
Proof. Consider the partitions P = {x0 , x1 , x2 , x3 }, where x0 = a, and x1 = s < x2 < x3 = b. Then
U (P, g, α) = M2 , L(P, g, α) = m2 .
Put
N
X ∞
X
α1 (x) = cn I(x − sn ), α2 (x) = cn I(x − sn ).
n=1 n=N +1
68
Z b N
X
Then by 6.1.10 and 6.1.12, gdα1 = cn g(sn ). Since α2 (b) − α2 (a) < ϵ,
a j=1
Z b
gdα2 ≤ M ϵ.
a
Hence,
Z b N
X
gdα − cn g(sn ) ≤ M ϵ.
a j=1
Define g2 by
xi − t t − xi−1
g2 (t) = g(xi−1 ) + g(xi )
∆i−1 x ∆i−1 x
if t ∈ [xi−1 , xi ], then g2 is continuous.
Notice that
xi − t t − xi−1 xi − t t − xi−1
|g(t) − g2 (t)| = + g(t) − g(xi−1 ) − g(xi )
∆i−1 x ∆i−1 x ∆i−1 x ∆i−1 x
xi − t xi − t t − xi−1 t − xi−1
≤ g(t) − g(xi−1 ) + g(t) − g(xi )
∆i−1 x ∆i−1 x ∆i−1 x ∆i−1 x
≤ Mi−1 − mi−1
2
Remark. We can replace |g − g2 | by |g − g2 | by replacing ϵ/(M − m + 1) by ϵ.
Theorem 6.1.15. Let I = [a, b] and g be a bounded real-valued function defined on I and let α be an
increasing function on I. If α′ exists on [a, b] and α′ ∈ R(I), then f ∈ R(α, I) if and only if f α′ ∈ R(I). In
this case,
Z b Z b
f dα = f (x)α′ (x)dx.
a a
69
Proof. Let ϵ > 0 be given, there exists a partition P = {x0 , · · · , xn } of [a, b] such that
∆i α = α′ (ti )∆i x
we have
n−1
X n−1
X
f (si )∆i α − f (si )α′ (si )∆i x ≤ M ϵ.
i=0 i=0
Therefore,
n−1
X
f (si )∆i α ≤ U (f α′ , P ) + M ϵ
i=0
U (f, α, P ) ≤ U (f α′ , P ) + M ϵ.
Similarly,
U (f α′ , P ) ≤ U (f, α, P ) + M ϵ.
Thus
|U (f, P, α) − U (f α′ , P )| ≤ M ϵ.
Notice that this remains true when P is replaced by any refinement. Hence,
Z b Z b Z b Z b
f dα − f α′ dx ≤ M ϵ =⇒ f dα = f α′ dx.
a a a a
The equality of the lower integral follows in the exactly same way. The theorem follows.
Theorem 6.1.16 (change of variables). Let I = [a, b]. Suppose φ is a strictly increasing continuous function
that maps on an interval [A, B] onto [a, b]. Suppose α is increasing on [a, b] and f ∈ R(α, I). Define β and g
on [A, B] by
β(y) = α(φ(y)), g(y) = f (φ(y)).
Then g ∈ R(β, I) and
Z B Z b
gdβ = f dα.
A a
Proof. To each partition P = {x0 , · · · , xn } of [a, b] corresponds a partition Q = {y0 , · · · , yn } of [A, B], so
that xi = φ(yi ). All partitions of [A, B] are obtained in this way. Since the values taken by f on [xi−1 , xi ]
are exactly the same as those taken by g on [yi−1 , yi ], we see that
70
Remark. If α(x) = x and φ′ ∈ R on [A, B], we have
Z b Z B Z B
f dx = f (φ(y))dφ = f (φ(y))φ′ (y)dy
a A A
by 6.1.15.
Proposition 6.1.17. Let I = [a, b] and g ∈ R(α) on [a, b], where α is an strictly increasing function. If
g > 0 on [a, b], then
Z b
gdα > 0.
a
Z b
Proof. Suppose gdα = 0. Then there exists a partition P1 = {x0 < · · · < xn } such that
a
1
U (g, α, P ) <
α(b) − α(a)
. Hence, there exists a subinterval I1 = [xj , xj+1 ] such that supI1 < 1. Put xj = x1 , xj+1 = y 1 and P = P 1 .
Now, we will construct a sequence of subintervals as below:
There exists a subinterval Ik = [xk , y k ] such that supIk g < k1 .
Z b
Since gdα = 0, there exists a partition P k = [xk1 , · · · , xkm ] such that
a
1
U (g, α, P k ) < .
(k + 1)(α(y k ) − α(xk ))
Hence, there exists a subinterval Ik+1 = [xkj , xkj+1 ] ⊆ Ik such that supIk+1 g < 1
k+1 (by taking union,
we can assume that xk , y k ∈ P k ). Let xkj = xk+1 , xk+1 = xkj+1 .
Hence, we have construct a sequence of nested interval (Ik ) such that supIk g < k1 . By 2.4.5, = J
T
k∈N Ik ̸= ∅.
However, for any x ∈ J, g(x) ≤ 0, a contradiction.
Proof. Let ϵ > 0 be given, and let P = {x0 , · · · , xn } be the partition of I such that for any tk ∈ [xk , xk+1 ],
we have
n−1 Z b
X ϵ
g(tk )∆k f − gdf < .
i=0 a 2
71
Hence,
n−1
X n
X n
X
A− f (tk )∆k g = g(xk )(f (xk ) − f (tk )) + g(xk−1 )(f (tk ) − f (xk−1 )).
k=0 k=1 k=1
Now take P ′ = P ∪ {t1 , · · · , tn }, then since P ′ is a refinement of P ,
Z b X
A− gdf − f (tk )∆k g < ϵ.
a
Z b Z b
Since P and the tk are arbitrary it follows that f ∈ R(g, I) and A − gdf = f dg.
a a
Then F is continuous on [a, b]; furthermore, if f is continuous at x0 ∈ [a, b], then F is differentiable at x0 ,
and
F ′ (x0 ) = f (x0 ).
Proof. Since f ∈ R(I), f is bounded and there exists M > 0 such that |f (x)| ≤ M for all x ∈ [a, b]. Then
Z y
|F (x) − F (y)| = f (t)dt ≤ M |x − y|
x
Proof. Let ϵ > 0 be given. Choose a partition P = {x0 , · · · , xn } of [a, b] so that for any pi ∈ [xi , xi+1 ] we
have
n−1
X Z b
f (pi )∆i x − f (x)dx < ϵ
i=0 a
72
Lemma 6.3.3 (mean value theorem for definite integrals). Let I = [a, b] and f : [a, b] → R be a continuous
function, then there exists ξ ∈ [a, b] such that
Z b
f dx = f (ξ)(b − a).
a
Proof. By 2.4.3.1, there exists x1 , x2 ∈ [a, b] such that f (x1 ) = M := max[a,b] f (x) and f (x2 ) = m :=
min[a,b] f (x). Then by 6.1.10,
Z b
f (x2 )(b − a) ≤ f dx ≤ f (x1 )(b − a).
a
If M = m, then we can pick any x ∈ [a, b] as ξ. Otherwise, by 2.5.4.1, there exists ξ lies in the interior of
interval with endpoints x1 , x2 such that
Z b Z b
1
f dx = f (ξ) ⇐⇒ f (ξ)(b − a) = f dx
b−a a a
as required.
Proof. Let P = {t0 , · · · , tn } be a partition of [a, b]. Then ti ≤ x ≤ ti+1 for some i = 0, · · · , n − 1. Hence,
Z b Z b
U (f, α, P ) = α(ti+1 ) − α(ti ) ≥ 0 and L(f, α, P ) = 0. Hence f dα ≥ 0 and f α = 0.
a a
Now given ϵ > 0, since α is continuous at x0 , there exists δ > 0 such that whenever t ∈ [a, b] with
|x − t| < δ, we have |α(x) − α(t)| < ϵ. Then we take a partition P with ∥P ∥ < δ. Hence, we have
Z b
U (f, α, P ) < ϵ and therefore f dα < ϵ.
a
Corollary 6.4.1.1. Let I = [a, b]. Suppose α increases on [a, b], α is continuous on [a, b]. If f ∈ R(α, I) and
g : [a, b] → R is a bounded function such that f ≡ g for all but a finite points, then g ∈ R(α, I) and
Z Z
f dα = gdα.
Proof. Suppose Z = {x ∈ [a, b] : f (x) ̸= g(x)} = {x0 , · · · , xn }, and yj = g(xj ) − f (xj ) for j = 0, · · · , n. Let
h(x) = g(x) − f (x) and (
1 x = xj
Ij (x) =
0 x ̸= xj
P 6.1.10 6.1.10
on [a, b]. Then h ≡ yj Ij ∈ R(α, I) ⇐⇒ g − f ∈ R(α, I) =⇒ g ∈ R(α, I) and
Z b n
X Z b Z b Z b
hdα = yj Ij dα = 0 ⇐⇒ f dα = gdα.
a j=0 a a a
Z b
Proposition 6.4.2. Suppose f ≥ 0, f is continuous on [a, b], and f (x)dx = 0. Then f ≡ 0 on [a, b].
a
y
Proof. Suppose f (x) = y > 0 for some x ∈ [a, b]. Then there exists c, d ∈ [a, b] with c < d such that f (x) > 2
for all x ∈ [c, d]. Then
Z b Z c Z d Z b
6.1.10 y
f (x)d(x) = f (x)dx + f (x)dx + f (x)dx ≥ (d − c) > 0,
a a c d 2
a contradiction.
73
Proposition 6.4.3. Suppose f ∈ R([a, b]) for every b > a where a is fixed. Define
Z ∞ Z b
f (x)dx = lim f (x)dx
a b→∞ a
Theorem 6.4.4 (Holder’s inequality). Let p and q be positive real numbers such that
1 1
+ = 1.
p q
1. If u ≥ 0 and v ≥ 0, then
up vq
uv ≤ + .
p q
Proof.
1. Fix u ≥ 0, let
vq up
f (v) = + − uv.
q p
Then
f ′ (v) = v q−1 − u.
Then f ′ (v) ≤ 0 on [0, u1/(q−1) ) and f ′ (v) ≥ 0 on [u1/(q−1) , ∞) and f ′ (u1/(q−1) ) = 0. Hence, by 5.2.3.2
and 5.2.2, f (v) attains minimum at v = u1/(q−1) ⇐⇒ v q = up . Also,
f (u1/(q−1) ) = up − uup/q = 0.
up vq
uv ≤ + = 1 =⇒ ∥f g∥1 ≤ ∥f ∥p ∥g∥q .
p q
74
Z b Z b
p
Now suppose ∥f ∥p = 0 ⇐⇒ |f | dx = 0. We shall prove that |f g| dx = 0.
a a
Z b
First suppose |f | dx = A > 0. Then there exists a partition P = {x0 , · · · , xn } of [a, b] such that
a
n−1
X A
m I ∆i α > ,
i=0
2
where inf |f | = mi . Hence, by piegeon-hole principle, there exists k ∈ {0, · · · , n − 1} such that
[xi ,xi+1 ]
A
mk ∆ k α ≥ .
2n
Now we consider
n−1
X
p
L(|f | , α, P ) = mpi ∆i α ≥ mpk ∆k α > 0
i=0
Z b
p
and therefore |f | ≥ mpk ∆k α > 0, a contradiction.
a
Hence ∥f ∥1 = 0. Since |g| ≤ M for some M > 0, we have
Z b Z b
|f g| dα ≤ |f | M dα = 0.
a a
Corollary 6.4.4.1. Let α be an increasing function on I = [a, b]. Suppose f, g ∈ R(α, I), then
∥f + g∥2 ≤ ∥f ∥2 + ∥g∥2 .
Proof.
Z b Z b 6.4.4
2 2 2 2 2 2
∥f + g∥2 = |f + g| dα = ∥f ∥2 + ∥g∥2 +2 |f g| dα ≤ ∥f ∥2 + ∥g∥2 + 2 ∥f ∥2 ∥g∥2 = (∥f ∥2 + ∥g∥2 )2 .
a a
75
Chapter 7
76
2. Conversely, if Y is complete, then for any x ∈ E, (fn (x)) converges. Let f (x) = lim fn (x) for x ∈ E,
we shall prove that fn ⇒ f . Let ϵ > 0 be given, there exists N ∈ N such that for any m, n ≥ N , we
have
d(fn (x), fm (x)) < ϵ
for any x ∈ E. Since fm (x) → f (x) for any x ∈ E, we have
d(fn (x), f (x)) ≤ ϵ
and therefore fn ⇒ f .
Definition 7.1.4. Suppose (fn ) is a sequence of real-valued functions defined on a set E. We define
∥fn ∥∞ = sup |fn (x)| .
x∈E
Proof. Let ϵ > 0 be given. By 7.1.3, there exists N ∈ N such that for any n, m ≥ N , we have
d(fm (t), fm (t)) ≤ ϵ.
Let t → x,we have
d(An , Am ) ≤ ϵ
for n, m ≥ N . Hence, (An ) is Cauchy and therefore converges, say to A.
Next,
d(f (t), A) ≤ d(f (t), fn (t)) + d(fn (t), An ) + d(An , A).
We first choose n such that
ϵ
d(f (t), fn (t)) ≤
3
for all t ∈ E and such that
ϵ
d(An , A) ≤ .
3
This, for this n, we take t so close to x such that
ϵ
d(fn (t), An ) ≤
3
if t ∈ V ∩ E, t ̸= x. Hence,
d(f (t), A) ≤ ϵ
for all such t.
77
Corollary 7.1.6.1. Suppose fn ⇒ f on a set E in a metric space to a complete metric space Y . If fn are
all continuous, then f is continuous.
Proposition 7.1.7. Let fk , f : X → Y , where X, Y are metric space. Suppose (mk ) is a nonnegative
sequence with mk → 0 as k → ∞. If d(fk (x), f (x)) ≤ mk for all x ∈ X, then fk ⇒ f on X.
Proof. Given ϵ > 0, since mk → 0, there exists N ∈ N such that k ≥ N implies mk < ϵ. Hence,
d(fk (x), f (x)) < ϵ for all x ∈ X and k ≥ N . Hence, fk ⇒ f on X.
Theorem 7.1.8 (MCT for functions). Suppose K is compact, and
1. (fn ) is a sequence of continuous functions on K to R,
2. fn converge to a continuous function f pointwisely on K,
3. fn ≥ fn+1 for all n ∈ N.
Then fn ⇒ f uniformly on K.
Proof. Let gn = fn − f . Then gn is continuous, gn → 0 and gn ≥ gn+1 . We shall prove that gn ⇒ 0 on K.
Let ϵ > 0 be given. Let Kn be the set for all x ∈ K with gn (x) ≥ ϵ. Then Kn is closed by 2.3.3 and hence
compact by 2.4.3. Since gn ≥ gb+1 ,Twe have Kn ⊇ Kn+1 . Fix
T x ∈ K. Since gn (x) → 0, we see that x ∈/ Kn
if n is sufficiently large. Thus x ∈
/ Kn . In other words, Kn = ∅. Hence, KN is empty for some N . It
follows that
0 ≤ gn < ϵ
for all x ∈ K and for all n ≥ N . This proves the theorem.
7.1.8 remains true if we replace the condition fn ≥ fn+1 by fn ≤ fn+1 . The proof is similar (replace
gn = fn − f by gn = f − fn ).
Definition 7.1.9. If X, Y are metric spaces, Cb (X, Y ) will denote the set of all continuous and bounded
functions from X to Y . Notice that if Y is complete, then Cb (X, Y ) is a complete metric space with metric
ρ(f, g) = sup{d(f (x), g(x)) : x ∈ X} by 7.1.3.
fn − ϵn ≤ f ≤ fn + ϵn .
Hence,
Z b Z b Z b Z b Z b Z b
(fn − ϵn )dα ≤ f dα ≤ f dα ≤ (fn + ϵn )dα =⇒ 0 ≤ f dα − f dα ≤ 2ϵn V (α; [a, b]) → 0
a a a a a a
Z b Z b
as n → ∞. Therefore, f ∈ R(α, I) and f dα = lim fn dα.
a a
78
7.3 Uniform Convergence and Differentiation
Proposition 7.3.1. Suppose (fn ) is a sequence of real-valued functions, differentiable on [a, b] and such
that (fn (x0 )) converge for some point x0 on [a, b]. If (fn′ ) converges uniformly on [a, b], then (fn ) converges
uniformly on [a, b], to a function f , and
f ′ (x) = lim fn′ (x).
|x − t| ϵ ϵ
|fn (x) − fm (x) − fn (t) + fm (t)| ≤ ≤
2(b − a) 2
|fn (x) − fm (x)| ≤ |fn (x) − fm (x) − fn (x0 ) + fm (x0 )| + |fn (x0 ) − fm (x0 )| ≤ ϵ,
Also, we have
ϵ
|φn (t) − φm (t)| ≤ ,
2(b − a)
so (φn ) converges uniformly, for t ̸= x. Since fn ⇒ f , we have
φn (t) ⇒ φ(t)
lim lim φn (t) = lim φ(t) = f ′ (x) = lim fn′ (x) = lim lim φn (t).
t→x n→∞ t→x n→∞ n→∞ t→x
Proposition 7.3.2. There exists a real continuous function on the real line which is nowhere differentiable.
Proof. Define
φ(x) = |x|
for x ∈ [−1, 1] and extend the definition of φ(x) to all real x by requiring that
φ(x + 2) = φ(x).
79
Hence, φ is continuous on R1 . Define
∞ n
X 3
f (x) = φ(4n x).
n=0
4
Now, let F = {choose a function g ∈ Ur1 ,··· ,rℓ if Ur1 ,··· ,rℓ ̸= ∅ : 1 ≤ i ≤ ℓ, 1 ≤ ri ≤ ki }, then F is finite.
Let f ∈ B, there exist g0 ∈ F such that
d(f (xi ), g(xi )) ≤ d(f (xi ), yi,ri ) + d(yi,ri , g0 (xi )) < 2ϵ
Sℓ
for 1 ≤ i ≤ ℓ. For x ∈ A ⊆ k=1 B(xk , δ), there exists 1 ≤ i ≤ ℓ such that d(x, xi ) < δ. Then
d(f (x), g0 (x)) ≤ d(f (x), f (xi )) + d(f (xi ), g0 (xi )) + d(g0 (xi ), g(x)) < 4ϵ.
S
Hence, f ∈ B(g) , 4ϵ) and then B ⊂ g∈F B(g, ϵ).
80
Lemma 7.4.3. Suppose (xn ) is a sequence in a totally bounded space X. Then (xn ) has a Cauchy subse-
quence.
Proof. WLOG, suppose A = {x n : n ∈ N} is an infinite set. Since X is totally bounded, there exist
Sm
p1 , · · · , pm ∈ X such that X = i=1 B(pi , ϵ). Then there exists pj such that A ∩ B(pj , ϵ) is infinite. Hence,
there exists a Cauchy subsequence.
Theorem 7.4.4 (Arzela-Ascoli theorem). Let M, N be two metric spaces and A ⊆ M be compact. Suppose
B ⊆ C(A, N ) is compact if and only if B is closed, equicontinuous and pointwise compact.
Proof. Suppose that B ⊆ C(A, N ) is compact.
B is closed.
Sm
Given ϵ > 0, since B is compact, there exist f1 , · · · , fm ∈ B such that B ⊆ k=1 B(fk , ϵ). Since A is
compact, f1 , · · · , fm are uniformly continuous. Hence, there exists δS> 0 such that for any x, y ∈ A
m
with d(x, y) < δ, d(fi (x), fi (y)) < ϵ for all 1 ≤ i ≤ m. Let f ∈ B ⊆ k=1 B(fk , ϵ). Then for x, y ∈ A
with d(x, y) < δ, d(f (x), f (y)) ≤ d(f (x), fi (x)) + d(fi (x), fi (y)) + d(fi (y), f (y)) < 3ϵ. This implies that
B is equicontinuous.
Let y0 ∈ A and {Ui : i ∈ I} be an open cover of By0 . Given f ∈ B, we have f (y0 ) ∈ i∈I Ui and Ui is
S
open. There exists ϵf > 0 such that B(f (y0S ), ϵf ) ⊆ Ui for some i ∈ I. Notice that since B is compact,
m
there exists f1 , · · · , fm ∈ B such that B ⊆ k=1 B(fk , ϵfk ). Thus, for all f ∈ B, since f ∈ B(fk0 , ϵfk0 )
Sm
for some 1 ≤ k0 ≤ m, f (y0 ) ∈ B(fk0 (y0 ), ϵfk0 ) ⊆ Uik0 for some ik0 ∈ I. So, By0 ⊆ k=1 Uik .
Assume that B ⊆ C(A, N ) is closed, equicontinuous and pointwise compact. By 7.4.2, B is totally bounded.
Hence, we only need to claim that B is complete (by 2.4.5). Assume that (fk ) is a Cauchy sequence in B.
Then for x ∈ A, (fk (x)) is a Cauchy sequence in Bx . Since Bx is compact, fk (x) converges to a limit, say
f (x). Since fk is a Cauchy sequence in B, there exists N ∈ N such that m, n ≥ N implies d(fn (x), fm (x)) < ϵ.
Taking m → ∞, d(fn (x), f (x)) ≤ ϵ for all x ∈ A, n ≥ N . Therefore, fk ⇒ f on A and then fk → f in
C(A, N ). Since B is closed, f ∈ B and we are done.
Proposition 7.4.5. Let M, N be two metric spaces and A ⊆ M be compact. Suppose B ⊆ C(A, N ). If B
is equicontinuous and pointwise compact, then for any sequence (fk ) in B, (fk ) has a uniformly convergent
subsequence (which converges in C(A, N )).
Proof. Since A and Bx are compact for all x ∈ A, by 7.4.2, B is totally bounded. Put (fk ) be a sequence
in B, by 7.4.3, there exists a Cauchy subsequence (fkj ). Put x ∈ A, then (fkj (x)) is Cauchy and therefore
converges (since Bx is complete), say converges to f (x). Given ϵ > 0, since (fkj ) is Cauchy, there exists
N ∈ N such that for any j1 , j2 ≥ N and x ∈ A,
Let j2 → ∞, we have
d(fkj1 (x), f (x)) ≤ ϵ,
which implies that fkj1 ⇒ f on A, as required.
Lemma 7.4.6. Let A be a countable set. If (fn ) is a pointwise bounded sequence of Rn -valued functions on
a countable set E, then (fn ) has a subsequence (fnk ) such that (fnk (x)) converges for every x ∈ E.
Proof. We first enumerate E = {pn : n ∈ N}. Given p1 ∈ E, since (fn (p1 )) is bounded, there exists a
convergent subsequence (fn1 (p1 )) by 1.2.12. We shall construct (fnk ) inductively as follows:
Given (fnk ), since f2k (pn+1 ), f3k (pn+1 ) is bounded, by 1.2.12 again, there exists a convergent subsequence
(fnk+1 (pn+1 )).
Hence, for each k ∈ N, (fnk+1 ) is a subsequence of (fnk ). Define a sequence (gn ) by gn = fnn . Then (gn ) is a
subsequence of (fn ) and (gn ) converges at pn for all n ∈ N, as required.
Proposition 7.4.7. Let M be a metric space and A ⊆ M be a compact subset. Suppose B ⊆ C(A, Rn ) is
equicontinuous and pointwise bounded. Then for any sequence (fn ) in B, fk has a uniformly convergence
subsequence (fnk ) with fnk ⇒ f ∈ C(A, Rn ).
81
Proof. Since A is compact, by 2.4.8, there exists a countable dense subset D of A. Then by 7.4.6, there exists
a subsequence (fnk ) such that (fnk ) converge at every point x ∈ D.
Given ϵ > 0, then
since B is equicontinuous, there exists δ > 0 such that for any k ∈ N, x, y ∈ A with d(x, y) < δ, we
have ∥fnk (x) − fnk (y)∥ < ϵ.
since D is dense in A, A ⊆ d∈D B(d, δ).
S
since (fnk (d)) converges for any d ∈ D, there exists Nd ∈ N such that for any k, m ≥ Nd , ∥fnk (d) − fnm (d)∥ <
ϵ.
Hence, for any x ∈ X, pick d ∈ D such that d(x, d) < δ. Therefore, for any k, m ≥ Nd ,
∥fnk (x) − fnm (x)∥ ≤ ∥fnk (x) − fnk (d)∥ + ∥fnk (d) − fnm (d)∥ + ∥fnm (d) − fnm (x)∥ < 3ϵ,
as required.
Proposition 7.4.8. If K, M are two metric spaces with K compact and (fn ) is a uniformly convergent in
C(K, M ), then {fn : n ≥ 1} is equicontinuous.
Proof. Since K is compact, fn is uniformly continuous for each n ∈ N. Given ϵ > 0, since (fn ) is uniformly
convergent, there exists N ∈ N such that for any n ≥ N, x ∈ K, we have d(fn (x), fN (x)) < ϵ/3. Since
fn is uniformly continuous, there exists δn > 0 such that for any x, y ∈ K with d(x, y) < δn , we have
d(fn (x), fn (y)) < ϵ. Take δ = min1≤n≤N {δn } > 0. Hence, for n = 1, · · · , N with x, y ∈ K and d(x, y) < δ,
we have d(fn (x), fn (y)) < ϵ. For n ≥ N and d(x, y) < δ, it follows that
d(fn (x), fn (y)) ≤ d(fn (x), fN (x)) + d(fN (x), fN (y)) + d(fN (y), fn (y)) < ϵ,
as required.
82
then by 6.1.16, Z 1 Z 1
Pn (x) = f (x + t)Qn (t)dt = f (t)Qn (t − x)dt,
−1 0
and the last integral is clearly a polynomial in x. Thus (Pn ) is a sequence of polynomials.
Given ϵ > 0, take δ > 0 such that whenever x, y ∈ R with |x − y| < δ, we have
ϵ
|f (y) − f (x)| < .
2
Let M = sup |f (x)|. It follows that
Z 1
|Pn (x) − f (x)| ≤ |f (x + t) − f (x)| Qn (t)dt
−1
−δ
ϵ δ
Z Z Z 1
≤ 2M Qn (t)dt +
Qn (t)dt + 2M Qn (t)dt
−1 2 −δ δ
√ ϵ
≤ 4M n(1 − δ 2 )n + < ϵ
2
for sufficiently large n, which proves the theorem.
Lemma 7.5.2. For every interval [−a, a], there is a sequence of real polynomials (Pn ) such that Pn (0) = 0
and such that Pn ⇒ |x| on [−a, a].
Proof. By 7.5.1, there exists a sequence (Pn∗ ) of real polynomials which converges to |x| uniformly on [−a, a].
In particular, Pn∗ (0) → 0 as n → ∞. The polynomials Pn (x) = Pn∗ (x) − Pn∗ (0) have desired properties.
Definition 7.5.3. A family of real functions A defined on a set E is said to be an algebra if
1. f + g ∈ A
2. f g ∈ A
3. cf ∈ A
for all f, g ∈ A and c ∈ R. If A has the property that f ∈ A whenever fn ∈ A and fn ⇒ f on E, then A
is said to be uniformly closed. Let B be the set of functions which are the limit of uniformly convergent
sequences of members of A. Then B is called the uniform closure of A.
Proposition 7.5.4. Let B be the uniform closure of an algebra A of bounded functions. Then B is uniformly
closed algebra.
Proof. It is clear that B is closed, hence it suffices to prove that B is an algebra.
Given f, g ∈ B, it is also a routine to prove that fn + gn ⇒ f + g, fn gn →⇒ f g and cfn ⇒ cf .
Definition 7.5.5. Let A be a family of functions on a set E. Then A is said to separate points on E if
every pair of distinct x1 , x2 ∈ E there corresponds a function f ∈ A such that f (x1 ) ̸= f (x2 ). If to each
x ∈ E, there corresponds a function g ∈ A such that g(x) ̸= 0, we say that A vanishes nowhere on E.
Proposition 7.5.6. Suppose A is an algebra of functions on a set E, A separates points on E, and A
vanishes nowhere on E. Suppose x1 , x2 are distinct points of E, and c1 , c2 ∈ R. Then A contains a function
f such that
f (x1 ) = c1 , f (x2 ) = c2 .
Proof. The assumption show that A contains functions g, h and k such that
g(x1 ) ̸= g(x2 ), h(x1 ) ̸= 0, k(x2 ) ̸= 0.
Put
u = gk − g(x1 )k, v = gh − g(x2 )h.
Then u, v ∈ A, u(x1 ) = v(x2 ) = 0, u(x2 ) ̸= 0 and v(x1 ) ̸= 0. Therefore
c1 v c2 u
f= +
v(x1 ) u(x2 )
has the desired properties.
83
Theorem 7.5.7 (Stone-Weierstrass). Let A be an algebra on real continuous functions on a compact set K.
If A separates points on K and if A vanishes nowhere on K, then the uniform closure B of A consists of all
real continuous functions on K.
We shall divide the proof into four steps.
1. If f ∈ B, then |f | ∈ B.
Proof. Let a = sup |f (x)| and ϵ > 0 be given. By 7.5.2, there exist real numbers c1 , · · · , cn such that
n
X
cj y j − |y| < ϵ
j=1
Proof. Since
(f + g) + |f − g|
max(f, g) =
2
and
f + g − |f − g|
min(f, g) = ,
2
then by 1 we are done.
3. Given a real function f , continuous on K, a point x ∈ K, and ϵ > 0, there exists a function gx ∈ B
such that gx (x) = f (x) and
gx (t) > f (t) − ϵ
for all t ∈ K.
Proof. Since A ⊆ B and A separates points and vanishes nowhere on K, by 7.5.6, for every y ∈ K, we
can find a function hy ∈ B such that
for all t ∈ Jy . Since K is compact, there exists a finite set of points y1 , · · · , yn such that
n
[
K⊆ Jyj .
j=1
Put
gx = max(hy1 , · · · , hyn ) ∈ B,
gx (x) = f (x) and gx (t) > f (t) − ϵ for all t ∈ K.
84
4. Given a real function f , continuous on K and ϵ > 0, there exists a function h ∈ B such that
Proof. By 3, there exists gx ∈ B such that gx (t) > f (t) − ϵ for all t ∈ K. By the continuity of gx , there
exists open sets Vx containing x, such that
for all t ∈ Vx . Since K is compact, there exists a finite set of points x1 , · · · , xm such that
m
[
K⊆ Vxj .
j=1
Put
h = min(gx1 , · · · , gxm ).
By 2, h ∈ B and
|h(t) − f (t)| < ϵ
for all t ∈ K.
It is easy to see that s can be tranformed into a trapezoidal function g, such that 0 ≤ g ≤ s and
Z b Z b
ϵ
sdx ≤ gdx + . Hence, there exists a continuous function g such that 0 ≤ g ≤ f and
a a 2
Z b Z b
f dx ≤ gdx + ϵ
a a
Lemma 7.6.3. Let (fn ) be a decreasing sequence of bounded integral functions on [a, b]. If fn → 0 on [a, b],
then Z b
lim fn dx = 0.
a
85
Proof. By 7.6.2, for any ϵ > 0 and n ∈ N, there exists a continuous function gn on [a, b] such that 0 ≤ gn ≤ fn
and Z b Z b
ϵ
fn dx ≤ gn dx + n .
a a 2
For each n, we set hn = (g1 , · · · , gn ). Then 0 ≤ hn ≤ gn ≤ fn , hn is continuous on [a, b] and (hn ) is decreasing
Z b
to 0 everywhere on [a, b]. Hence, by 7.1.8, hn ⇒ 0 and lim hn dx = 0 by 7.2.1.
a
Now notice that for each j = 1, · · · , n,
n
X
0 ≤ gn = gj + (gn − gj ) ≤ gj + max(gj , · · · , gn ) − gj ≤ gj + (max(gj , · · · , gn ) − gj );
j=1
hence
n
X
0 ≤ gn ≤ hn + (max(gj , · · · , gn ) − gj ).
j=1
Therefore,
Z b Z b
lim fn dx ≤ ϵ =⇒ lim fn dx = 0.
a a
Theorem 7.6.4 (bounded convergence theorem). Let (fn ) be a sequence of Riemann-integrable functions
defined on a bounded and closed interval [a, b], which converges on [a, b] to a Riemann-integrable function f .
If there exists a constant M > 0 satisfying |f | ≤ M on [a, b] for all n, then
Z b
lim |fn − f | dx = 0.
a
In particular,
Z b Z b
lim fn (x)dx = f dx.
a a
Proof. WLOG, we assume that 0 ≤ fn ≤ M for all n and fn → f on [a, b]. For each n, we set pn (x) =
supk≥0 (fn+k (x)). Then 0 ≤ fn ≤ pn and the sequence (pn ) decreases everywhere to zero on [a, b]. Indeed,
lim fn (x) = lim sup fn (x) = lim pn (x) = 0 for all x ∈ [a, b]. Hence, by 7.6.3,
Z b
lim pn dx = 0,
a
and so,
Z b Z b
0 ≤ lim fn dx ≤ lim pn dx = 0,
a a
86
Proof. If max(lim sup an , lim sup bn ) = ∞, the result is trivial. Now suppose M ∈ R with
then there exists N ∈ N such that for any n ≥ N , an , bn < M by 3.2.2. Hence, for any n ≥ N , max(an , bn ) <
M and therefore
lim sup max(an , bn ) ≤ M
by 3.2.4. Since M is arbitrary, lim sup max(an , bn ) ≤ max(lim sup an , lim sup bn ).
Theorem 7.6.6 (dominated convergence theorem). Let f ∈ R([a, b]) be the limit of a convergence sequence
(fn ) in R([a, b]). If there exist a g ≥ 0 such that g ∈ R([a, b]) and |fn | ≤ g on [a, b] for all n ∈ N, then
Z b
lim |fn − f | = 0.
a
Z b Z b
In particular, lim fn dx = f dx.
a a
Proof. By considering the sequence (|fn − f |), which satisfies |fn − f | ≤ |f | + g on [a, b] and for all n ∈ N,
which |f | + g ∈ R([a, b]), WLOG, we may assume fn ≥ 0 for all n and that f = 0 on [a, b]. Put
for each pair of indices m ≥ n. Then gm,n ≥ 0, gm,n ∈ R([a, b]) and 0 ≤ gm,n ≤ g for all m ≥ n. Hence, for
each n, the sequence (gm,n ) and consequently, the sequence
Z b
gm,n dx
a
is increasing and bounded in m > n. Hence, for each ϵ > 0 and for each n, there exists an index mn > n
such that mn < mn+1 and
Z b Z b
ϵ
0≤ gn,k dx − gn,mn dx ≤ n ,
a a 2
for all k ≥ mn . Let un = gn,mn . Then by 7.6.5,
Since max(uj , · · · , un ) = max(f1 , · · · , fmn ) − uj = gj,mn − gj,mj and mn > mj for all n > j, we conclude
that Z b
ϵ
max(uj , · · · , un ) − uj dx < j
a 2
for all 1 ≤ j ≤ n, and so by the above inequality
Z b Z b
1
0≤ fn dx ≤ min(u1 , · · · , un ) + ϵ 1 − n .
a a 2
From lim un (x) = 0 for all x ∈ [a, b], it follows that the sequence (min(u1 , · · · , un )) decreases to 0 pointwisely.
Hence, by 7.6.2,
Z b
lim min(u1 , · · · , un )dx = 0,
a
and finally we obtain that
Z b
lim fn dx = 0
a
and the proof is finished.
87
Lemma 7.6.7. Let f ∈ R([a, b]) be a nonnegative function and the limit function of sequence (fn ) of
nonnegative Riemann integrable functions on [a, b]. Then
Z b
lim (f − fn )+ dx = 0,
n→∞ a
Proof. Since fn , f ≥ 0, we have f − fn ≤ f on [a, b]. Hence, (f − fn )+ ≤ f , then the result follows from
7.6.6.
Theorem 7.6.8 (Fatou’s lemma). Let f ∈ R([a, b]) be a nonnegative function and the limit function of
sequence (fn ) of nonnegative Riemann integrable functions on [a, b]. We have
Z b Z b
0≤ f dx ≤ lim sup fn dx.
a a
88
Chapter 8
on (a, b).
Theorem 8.1.2. Suppose the series
∞
X
cn xn
n=0
Then f (x) converges uniformly on [−R + ϵ, R − ϵ] for all ϵ > 0. The function f is continuous an differentiable
in (−R, R), and
X∞
′
f (x) = ncn xn−1 .
n=1
cn (R − ϵ)n converges absolutely (by 3.7.2), by 7.1.5, f (x) uniformly converges on [−R + ϵ, R − ϵ].
P
and since
√
Since n n → 1 as n → ∞, we have
p p
lim sup n n |cn | = lim sup n |cn |,
89
Then
∞
X
f (1− ) = cn .
n=0
for 0 ≤ x < 1. For x < 1, these series converges absolutely by 3.7.2; hence, by be multiplied according to
3.9.2; hence,
f (x)g(x) = h(x)
on 0 ≤ x < 1. Hence, by 8.1.3,
f (x) → A, g(x) → B, h(x) → C
as x → 1. Hence, AB = C.
Theorem 8.1.4 (Fubini theorem on sequences). Given a double sequence (aij ) and suppose that
∞ X
X ∞
|aij | < ∞.
i=1 j=1
Then
∞ X
X ∞ ∞ X
X ∞
aij = aij .
i=1 j=1 j=1 i=1
Proof. Let E be a countable set, consisting of the points x0 , x2 , x2 , · · · , and suppose xn → x0 as n → inf ty.
Define
∞
X n
X X∞
fi (x0 ) = aij , fi (xn ) = aij , g(x) = fi (x).
j=1 j=1 i=1
Hence,
∞
X
|fi (xn ) − fi (x0 )| ≤ |aij | → 0
j=n+1
90
then |fi (x)| ≤ bi for any x ∈ E, by 7.1.5, g converges uniformly, so g is continuous at x0 (by 7.1.6.1). It
follows that
∞ X
X ∞ ∞
X
aij = fi (x0 )
i=1 j=1 i=1
= g(x0 )
= lim g(xn )
n→∞
∞
X
= lim fi (xn )
n→∞
i=1
X∞ X n
= lim aij
n→∞
i=1 j=1
Xn X ∞ ∞ X
X ∞
= lim aij = aij .
n→∞
j=1 i=1 j=1 i=1
the series converging in |x| < R. If −R < a < R, then f can be expanded in a power series about a point
x = a which converges in |x − a| < R − |a|, and
∞
X f (n) (a)
f (x) = (x − a)n .
n=0
n!
Proof. We have
∞ ∞ X
n
X
n n n−m
X
f (x) = cn (x − a + a) = cn a (x − a)m .
n=0 n=0 m=0
m
Since
∞ X
n ∞
X n n−m m
X
cn a (x − a) = |cn | (|x − a| + |a|)n < ∞
n=0 m=0
m n=0
we have
∞ X
X ∞
f (x) = sn,m
n=0 m=0
∞ X ∞
8.1.4
X
= sn,m
m=0 n=0
∞ X∞
X n n−m
= a (x − a)m
cn
m=0 n=m
m
∞ ∞ !
X X n n−m
= cn a (x − a)m .
m=0 n=m
m
91
an xn and bn xn converges in the segment S = (−R, R). Let E be
P P
Theorem 8.1.6. Suppose the series
the set of all x ∈ S at which
∞
X ∞
X
an xn = bn xn .
n=0 n=0
If E has a limit point in S, then the series are equal for all x ∈ S.
Proof. Let A be the limit points of E in S and B = S\A. It is clear that B is open. Notice that if A is
also open, then since S is connected, by 2.5.2, either A or B is empty. However, by hypothesis, A ̸= ∅ and
therefore A = S. Let
X∞
f (x) = (an − bn )xn ,
n=0
then f is continuous by 7.1.6.1 and therefore S = A ⊆ E =⇒ S = E.
Hence, it suffices to show that A is open. If x0 ∈ A, by 8.1.5,
∞
X
f (x) = dn (x − x0 )n ,
n=0
where |x − x0 | < R − |x0 |. We claim that dn = 0 for all n. Otherwise, let k be the smallest nonnegative
integer such that dk = 0. The
f (x) = (x − x0 )k g(x)
where
∞
X
g(x) = dk+m (x − x0 )m .
m=0
Since g is continuous at x0 and g(x0 ) = dk ̸= 0, there exists a δ > 0 such that g(x) ̸= 0 if |x − x0 | < δ.
Hence, f (x) ̸= 0 if 0 < |x − x0 | < δ, but since f (x0 ) = 0, it is a contradiction.
Thus dn = 0 for all n so that f (x) = 0 for all x such that |x − x0 | < R − |x0 |. This shows that A is open,
then we are done.
for all x ∈ R. Also, since E(x) converges absolutely at all x ∈ R, given x, y ∈ R, by 3.9.2,
∞ Xn ∞
X xk y n−k X 1
E(x)E(y) = = (x + y)n = E(x + y).
n=0
k!(n − k)! n=0
n!
k=0
Moreover, by the definition of e, we have E(1) = e and therefore E(n) = en for all n ∈ N. Hence, for
p= m
n , we have
(E(p))n = E(pn ) = E(m) = em =⇒ E(p) = em/n = ep
and therefore E(r) = er for all rational r (since E(−p)E(p) = 1).
Recall 1.2.22, we define
ex = sup{er : r ∈ Q, r ≤ x} = sup{E(r) : r ∈ Q, r ≤ x} = E(x−) = E(x)
since E(x) is a strictly increasing function (for 0 ≤ x < y, we have E(x) < E(y) and therefore E(−x) <
E(−y), which implies that E(x) is strictly increasing) and continuous.
It is clear that for any x ≥ 0, E(x) ≥ E(0) = 1 > 0. Hence, by E(−x)E(x) = 1, we have E(x) > 0 for all
x ∈ R.
92
Proposition 8.2.2. Let E(x) be defined on R as above. Then:
1. E(x) is continuous and differentiable for all x ∈ R;
2. E ′ (x) = E(x);
3. E(x) is strictly increasing function of x, and ex > 0;
4. E(x + y) = E(x)E(y);
5. E(x) → ∞ as x → ∞, E(x) → 0 as x → −∞;
6. E(xy) = (E(x))y ;
5. Notice that E(x) > x whenever x > 0. Then as x → ∞, E(x) → ∞. Hence, given ϵ > 0, there exists
M ∈ R such that whenever x > M ,
1
E(x) > =⇒ E(−x) < ϵ.
ϵ
Therefore, E(x) → 0 as x → −∞.
6. Since E(x + y) = E(x)E(y), for any fixed x ∈ R, we have E(nx) = (E(x))n . Then for any rational
n
p= m ∈ Q, we have
Now, fix x > 0. For any y ∈ R and ϵ > 0, since E(t) is continuous at xy, there exists δ > 0 such that
whenever t ∈ R with |t − xy| < δ, we have |E(t) − E(xy)| < 2ϵ . By 1.2.14.1, take p, q ∈ Q such that
q < y < p |q − y| , |p − y| < xδ =⇒ |qx − xy| , |px − xy| < δ. Hence,
Since xq < xy < xp, we have E(qx) < E(xy) < E(yp). On the other hand, since E(qx) =
(E(x))q , E(px) = (E(x))p with E(x) > E(0) = 1. By 1.2.22, we have
Therefore
1.2.15
|E(x)y − E(xy)| < ϵ =⇒ E(x)y = E(xy).
Hence, for the case x < 0 and y ∈ R, we have
1
E(xy)−1 = E(−xy) = E(−x)y = =⇒ E(xy) = E(x)y ,
E(x)y
as required.
7. Notice that when x > 0,
xn+1 (n + 1)!
E(x) > ⇐⇒ xn E(−x) < →0
(n + 1)! x
as x → ∞.
93
Definition 8.2.3. Since E ′ (x) > 0 for all x ∈ R, we can define a inverse function L(x) : (0, ∞) → R by
(L ◦ E)(x) = x. Also, since E ′ (x) > 0, L(x) is differentiable on (0, ∞) and
1 1
L′ (E(x)) = = .
E ′ (x) E(x)
Hence,
1
L′ (y) =
y
for all y > 0. Moreover, notice that L(1) = L(E(0)) = 0 and therefore, by 6.3.2,
Z y Z y
1
L(y) = L′ (x)dx = dx,
1 1 x
which defines the logarithm function. Sometimes we use the notation log(x) or ln(x) instead of L(x).
Remark. For x > 0, α ∈ R, since xα = E(L(x))α = E(αL(x)), we have
d α 5.1.3 α
(x ) = E(αL(x)) = αxα−1 .
dx x
provided the limit exists. Similarly, for a real function g on (c, a] and g ∈ R([t, a]) for any t ∈ (c, a], we define
Z a Z a
gdx = lim gdx
c x→c t
provided the limit exists. Such integrals are called indefinite integral.
We shall show that such an integral agrees with the old one. Suppose g ∈ R([c, a]), then g ∈ R([t, a]) for
any t ∈ (c, a] and
Z a Z a Z t
gdx − gdx = gdx ≤ M (t − c) → 0
c t c
as t → c, where M = sup[c,a] |f |.
Definition 8.3.2. For 0 < x < ∞, we define
Z ∞
Γ(x) = tx−1 e−t dt,
0
tn+2 (n + 2)!
et > ⇐⇒ e−t tn < .
(n + 2)! t2
Moreover, since tx−1 = E(L(t)(x − 1)) < E(L(t)(n + 2)) = tn+2 for all t > 0, by 6.1.10,
b b
b−1
Z Z
(n + 2)!
γ1 (b) ≤ tn e−t dt ≤ dt = (n + 2)! ≤ (n + 2)!
1 1 t2 b
94
for any b > 1. Since γ1 (b) is increasing (since tx−1 e−t ≥ 0 for all t ∈ R), we have shown that lim γ1 (b)
b→∞
converges. Therefore, since tx−1 e−t is continuous at [0, 1], Γ(x) converges for any x ≥ 1.
For x < 1, put Z 1
γ2 (a) = tx−1 e−t dt,
a
−t
where a ∈ (0, 1] Notice that e < 1 for all t ∈ [0, 1]; hence,
Z 1
1 1
γ2 (a) ≤ tx−1 dt = (1 − ax ) ≤ .
a x x
Since γ2 (a) increases as a → 0, γ2 (a) converges as a → 0. The rest is analogous to the previous case.
Proposition 8.3.3.
1. For x > 0, Γ(x + 1) = xΓ(x).
2. Γ(n + 1) = n! for n ∈ Z≥0 .
3. log Γ is convex on (0, ∞).
Proof.
1. Using 6.2.2, we have
1
Γ(x) = Γ(x + 1) ⇐⇒ xΓ(x) = Γ(x + 1).
x
2. Since Γ(0 + 1) = 1 = 1!, it then follows by 1.
3. Given p, q ∈ (0, ∞) with (1/p) + (1/q) = 1, it follows that
tx/p+y/q−1 e−t = t(x/p+y/q)−(1/p+1/q) e−t(1/p+1/q) = t(x−1)/p e−t/p t(y−1)/q e−t/q .
Thus, by 6.4.4 (it is an easy exercise that it applies to indifinite integral), we have
1/p 1/q
x y x y
Γ + ≤Γ Γ .
p q p q
This is equivalent to 3.
95
Corollary 8.3.3.2.
n!nx
Γ(x) = lim .
n→∞ x(x + 1) · · · (x + n)
The special case x = y = 1/2 gives Γ(1/2) = π. Now, by using a new substitution t = s2 , we have
Z ∞
2
Γ(x) = 2 s2x−1 e−s ds.
0
96
3. S(0) = C(p) = 0.
4. C(−x) = C(x).
5. C(p − x) = S(x), S(p − x) = C(x).
6. S(x + y) = S(x)C(y) + C(x)S(y).
√
7. C(p/2) = S(p/2) = 1/ 2.
√
8. S(−p/2) = −S(p/2) = −1/ 2.
9. S(−p) = −1.
1. Notice that C(0) = C(p − p) = (C(p))2 + (S(p))2 ≥ 1 > 0 and C(0) = C(0 − 0) = (C(0))2 + (S(0))2 ≥
(C(0))2 , then we are done.
2. 1 = C(0) = C(x − x) = (C(x))2 + (S(x))2 .
3. 1 = (C(0))2 + (S(0))2 = 1 + (S(0))2 and therefore S(0) = 0. 1 = (C(p))2 + (S(p))2 = (C(p))2 + 1 and
therefore C(p) = 0.
7. Notice that S(p/2) = C(p − p/2) = C(p/2) and 1√ = S(p) = S(p/2 + p/2) = 2S(p/2)C(p/2) =
2(S(p/2))2 . Since S(p/2) ≥ 0, C(p/2) = S(p/2) = 1/ 2.
8. 0 = S(p/2 + (−p/2)) = S(p/2)C(p/2) + C(p/2)S(−p/2) = C(p/2)(S(p/2) + S(−p/2)); hence, S(p/2) +
S(−p/2) = 0, as required.
9. S(−p) = 2S(−p/2)C(−p/2) = −1.
97
Remark. By 4, 10 and 14, we know that S and C are uniquely determined once the values of S and C on
[0, p] are determined.
Proposition 8.4.3.
1. C is decreasing on [0, p]; S is increasing on [0, p].
3. Notice that C(p) = 0 = 1−1, hence C(p/2) ≥ 1−1/2. Apply 2 inductively, we have C(p/2n ) ≥ 1−1/2n .
Hence,
4.1.4
1 ≥ C(p/2n ) ≥ 1 − 1/2n =⇒ lim C(p/2n ) = 1.
4. For any ϵ > 0, by 3, there exist n ∈ N such that 1 − C(p/2n ) < ϵ. Take δ = p/2n , since C is decreasing
on [0, p], C(x) ≥ C(p/2n ) for all x ∈ [0, δ]. Hence, 1 − C(x) < ϵ for all x ∈ [0, δ]. Moreover, since
C(−x) = C(x), for any x ∈ [−δ, 0], 1 − C(x) < ϵ and therefore C(x) is continuous at 0.
For S, since C(p/2n ) ≥ 1 − 1/2n , (S(p/2n ))2 ≤ 1/2n−1 − 1/2n and therefore
√
0 ≤ S(p/2n ) ≤ 2/ 2n+1
as h → 0.
6. It suffices to prove that C is uniquely defined on [0, p] since S(x) = C(p − x) for all x ∈ [0, p]. If C(x)
is known, then C(kx) is known by 8.4.2 if kx ∈ [0, p]. Then by 8.4.2 again, C(kx/2n ) is known. Notice
that C(p) = 1, then we cen get C(x) for any x = kp/2n ∈ [0, p]. Since {kp/2n : k, n ∈ N} ∩ [0, p] is
dense in [0, p], by 4.6.3, C(x) is uniquely determined on [0, p].
For p = 1, we define F (x) = S(x)/x on (0, 1], which is a continuous function on (0, 1].
Lemma 8.4.4. For any x an positive integer n ∈ N such that 0 < x < nx ≤ 1, the sequence F (x), F (2x), · · · , F (nx)
strictly decreases.
98
Proof. Let n = 2. Note there exists m ∈ N such that x ≥ 21n and therefore C(x) ≤ C(1/2n ) < 1. It follows
that
S(2x) 2S(x)C(x)
F (2x) = = = F (x)C(x) < F (x).
2x 2x
Assume: F (kx) < F ((k − 1)x). To show: F ((k + 1)x) < F (kx). From the hypothesis, after simplifying, we
get kS(x) − S(kx) < kS(kx)C(x) − kC(kx)S(x). But
Hence, kS(kx)C(x) + kC(kx)S(x) < kS(kx) + S(kx),i.e., F ((k + 1)x) < F (kx), as required.
F (y) − F (x2 ) ≤ |F (y) − F (x2 )| < ϵ = F (x0 ) − F (x1 ) < F (x0 ) − F (x2 ).
Proof. Given ϵ > 0, by 8.4.6, there exists N ∈ N such that π/2 − F (2−n ) < ϵ for any n ≥ N . Let δ = 2−N .
Then for any x ∈ (0, δ), there exists m ≥ N such that 2−m < x. Hence, by 8.4.5, F (2−m ) > F (x) > F (2−N )
and therefore 0 < π2 − F (x) < ϵ, as required.
99
Proof. For x ∈ R and h ̸= 0, we have
100
Chapter 9
Proposition 9.1.2.
1. If A ∈ L(Rn , Rm ), then ∥A∥ < ∞ and A is uniformly continuous mapping from Rn to Rm .
2. If A, B ∈ L(Rn , Rm ) and c ∈ R, then
Proof.
cj ej ∈ Rn with ∥x∥ = 1, then |cj | ≤ 1 for all j. Therefore,
P
1. Given x =
X X X
∥Ax∥ = cj Aej ≤ |cj | ∥Aej ∥ ≤ ∥Aej ∥ .
Hence, X
∥A∥ ≤ ∥Aej ∥ < ∞.
Since
∥Ax − Ay∥ ≤ ∥A∥ ∥x − y∥ ,
then A is Lipschitz and therefore uniform continuous.
and
6.1.9
∥cAx∥ = |c| ∥Ax∥ =⇒ ∥cA∥ = |c| ∥A∥ .
101
3. Let x ∈ Rn , we have
α ∥x∥ = α A−1 Ax ≤ ∥Ax∥ ≤ ∥(A − B)x∥ + ∥Bx∥ ≤ β ∥x∥ + ∥Bx∥ =⇒ (α − β) ∥x∥ ≤ ∥Bx∥ .
(α − β) B −1 y ≤ ∥y∥ =⇒ B −1 ≤ (α − β)−1 .
Hence,
9.1.2 β
B −1 − A−1 = B −1 (A − B)A−1 ≤ B −1 ∥A − B∥ A−1 ≤ →0
α(α − β)
as B → A.
9.2 Differentiation
Definition 9.2.1. Suppose G is an open set in Rn , f maps G into Rm and x ∈ G. If there exists a linear
transformation A of Rn into Rm such that
∥f (x + h) − f (x) − Ah∥
lim = 0,
h→0 ∥h∥
f ′ (x) = A.
102
Proof. Suppose A1 , A2 ∈ L(Rn , Rm ) are two linear transformations such that
oj (h)
where j = 1, 2 and → 0 as h → 0. Then
h
∥h(A1 − A2 )∥ = ∥o2 (h) − o1 (h)∥ .
fi (x + tej ) − fi (x)
Dj fi (x) = lim ,
t→0 t
called a partial derivative, provided the limit exists. If for all i and j, Dj fi (x) exists, then we define the
Jacobian of f at x, denoted by J(f , x) ∈ L(Rn , Rm ), by
∂fi
We often write Dj fi (x) and J(f , x) as and J(x), respectively (provided there is no ambiguity for the
∂xj
latter one).
Proposition 9.2.5. Suppose f = (f1 , · · · , fm ) maps an open set G ⊆ Rn into Rm , and f is differentiable at
a point x ∈ G. Then the partial derivative (Dj fi )(x) exists, and
m
X
f ′ (x)ej = (Dj fi )(x)ui (1 ≤ j ≤ n).
i=1
f (x + tej ) − f (x)
lim = f ′ (x)ej .
t→0 t
Notice that
m
X
f (x + tej ) − f (x) = (fi (x + tej ) − fi (x))ui ,
i=1
103
Corollary 9.2.5.1. If, in addition, m = 1 and f attains local extreme at x, then f ′ (x)v = 0 for all v ∈ G.
Proof. Similar to 5.2.2, this time, we consider each partial derivative of f .
Proposition 9.2.6. Suppose f maps a convex open set G ⊆ Rn into Rm , f is differentiable in G, and there
is a real number M such that
∥f ′ (x)∥ ≤ M
for every x ∈ G. Then
∥f (b) − f (a)∥ ≤ M ∥b − a∥
for all a, b ∈ G.
Proof. Given a, b ∈ G, define g : [0, 1] → G by
g(t) = ta + (1 − t)b.
Hence, by 5.6.3,
∥f ◦ g(1) − f ◦ g(0)∥ = ∥f (b) − f (a)∥ ≤ M ∥b − a∥ .
Hence,
|(Dj fi )(y) − (Dj fi )(x)| ≤ ∥f ′ (y) − f ′ (x)∥
and therefore Dj fi is continuous.
Conversely, suppose the partial derivatives Dj fi exist and is continuous on G for 1 ≤ i ≤ m, 1 ≤ j ≤ n.
Then given x ∈ G and ϵ > 0, there exists r > 0 such that for any y ∈ S := B(x; r) ⊆ G (since G is open),
ϵ
|Dj fi (x) − Dj fi (y)| ≤
nm
P
for all 1 ≤ j ≤ n, 1 ≤ i ≤ m. Put h = hj ej with ∥h∥ < r, v0 = 0 and vk = h1 e1 + · · · + hk ek for
1 ≤ k ≤ n. Then for any i,
n
X
fi (x + h) − fi (x) = (fi (x + vj ) − f (x + vj−1 )).
j=1
Since ∥vk ∥ < r for all 1 ≤ k ≤ n and since S is convex, the segment with end points x + vj−1 and x + vj lie
in S. Since vj = vj−1 + hj ej , by 5.2.3.1, there exists θj ∈ (0, 1) such that
104
and therefore
n n
X ϵ X ϵ
fi (x + h) − fi (x) − hj (Dj fi )(x) ≤ 2
|hj | ≤ ∥h∥ .
j=1
n j=1 n
n m
Then define D(x) ∈ L(R , R ) by
Xn n X
X m
D(x) hj ej = hj Dj fi (x)ui .
j=1 j=1 i=1
Therefore,
n
X
∥f (x + h) − f (x) − D(x)h∥ ≤ m max fi (x + h) − fi (x) − hj (Dj fi )(x) ≤ ϵ ∥h∥ ,
1≤i≤m
j=1
which implies that f ′ (x) = D(x). Moreover, there exists r′ > 0 such that for any 1 ≤ j ≤ n, 1 ≤ i ≤ m and
y ∈ S ′ = B(x, r′ ) ⊆ G,
ϵ
∥Dj fi (x) − Dj fi (y)∥ < .
2m2 n2
P
Hence, for any v = vj ej with ∥v∥ = 1, it follows that
n X
m n X
m
X X ϵ
∥(D(x) − D(y))v∥ = vj (Dj fi (x) − Dj fi (y))ui ≤ |vj | · |(Dj fi (x) − Dj fi (y))| <
j=1 i=1 j=1 i=1
2
and therefore ∥D(x) − D(y)∥ < ϵ, which implies that f ∈ C 1 (G), as required.
when n, m are large enough by 3.3.2 and 3.3.3. Hence, (xn ) is a Cauchy sequence. Since X is complete,
xn → x for some x ∈ X. Notice that since φ is a contraction, φ is continuous and therefore
as required.
105
9.4 The Inverse Function Theorem
Theorem 9.4.1 (Inverse Function Theorem). Suppose f is a C 1 -mapping of an open set G ⊆ Rn into Rn ,
f ′ (a) is invertible for some a ∈ G, and b = f (a). Then
1. there exist open sets U and V in Rn such that a ∈ U, b ∈ V , f is a injection on U and f (U ) = V .
Therefore f is invertible on U .
2. if g is the inverse of f , defined in V by
g(f (x)) = x
and g ∈ C (V ).
1
Proof.
1. Put f ′ (a) = A, choose λ so that
2λ A−1 = 1
(notice that if A−1 = 0, then A is not invertible). Since f ′ is continuous at a, there is an open ball
U := B(a; r) ⊆ G, such that
∥f ′ (x) − A∥ < λ
for all x ∈ U . Fixed y ∈ Rn , we define a function φy on G by
φy (x) = x + A−1 (y − f (x)).
Then f (x) = y if and only if φy (x) = x.
Since φ′y (x) = I − A−1 f ′ (x) = A−1 (A − f ′ (x)), then for all x ∈ U ,
9.1.2 1
φ′y (x) < A−1 λ = .
2
Hence, by 9.2.6,
1
∥φy (x1 ) − φy (x2 )∥ ≤ ∥x1 − x2 ∥
2
for all x1 , x2 ∈ U . Thus, φy has at most one fixed point in U , so that f (x) = y for at most one x ∈ U .
Next, put V = f (U ), and pick y0 ∈ V . Then y0 = f (x0 ) for some (and thus unique) x0 ∈ U . Let
B := B(x0 ; r0 ) be an open ball such that B lies in U . We will show that B(y0 ; λr0 ) ⊆ V .
Fix y with ∥y − y0 ∥ < λr0 . If x ∈ B,
1 r0
∥φy (x) − x0 ∥ ≤ ∥φy (x) − φy (x0 )∥ + ∥φy (x0 ) − x0 ∥ ≤ ∥x − x0 ∥ + ≤ r0 ;
2 2
hence φy (x) ∈ B. Hence, φy is a contraction of B into B. Note that B is complete and therefore φy
has only one fixed point x ∈ B. For this x, f (x) = y. Thus y ∈ f (B) ⊆ f (U ) = V . Hence, V is open
and we are done.
2. Pick y ∈ V, y + k ∈ V with k ̸= 0. Then there exist x ∈ U, x + h ∈ U (h ̸= 0 since f is a bijection from
U to V ), so that y = f (x), y + k = f (x + h). Then
φ0 (x + h) − φ0 (x) = h − A−1 k.
Hence,
1 1
h − A−1 k ≤ ∥h∥ =⇒ A−1 k ≥ ∥h∥ =⇒ ∥h∥ ≤ 2 A−1 ∥k∥ = λ−1 ∥k∥ .
2 2
Note that by 4.2.7, f ′ (x) is invertible. Since
g(y + k) − g(y) − T k = h − T k = −T (f (x + h) − f (x) − f ′ (x)h),
∥g(y + k) − g(y) − T k∥ ∥T ∥ ∥f (x + h) − f (x) − f ′ (x)h∥
≤ · .
∥k∥ λ ∥h∥
As k → 0, h → 0 and therefore g′ (y) = T = (f ′ (g(y)))−1 . Note that g, f and inverse function is
continuous by 4.2.7 and 5.1.2.1; thus, g′ is continuous.
106
Corollary 9.4.1.1. If f is a C 1 -mapping of an open set G ⊆ Rn into Rn and if f ′ (x) is invertible for every
x ∈ G, then f is open.
Theorem 9.5.3 (Implicit Function Theorem). Let f be a C 1 -mapping of an open set G ⊆ Rn+m into Rn ,
such that f (a, b) = 0 for some point (a, b) ∈ G. Put A = f ′ (a, b) and assume Ax is invertible. Then there
exist open sets U ⊆ Rn+m and W ⊆ Rm , with (a, b) ∈ U and b ∈ W , having the following property:
1. To every y ∈ W corresponds a unique x such that
(x, y) ∈ U and f (x, y) = 0.
107
9.6 Derivatives of Higher Order
Definition 9.6.1. Suppose f is a real function defined in an open set E ⊆ Rn , with partial derivatives
D1 f, · · · , Dn f . If the functions Dj f are themselves differentiable, then the second order partial deriva-
tives of f are defined by
Dij f = Di Dj f,
where i, j = 1, · · · , n. If all these functions Dij f are continuous in E, we say that f is of class C 2 in E or
that f ∈ C 2 (E).
Lemma 9.6.2. Suppose f is defined in an open set E ⊆ R2 , and D1 f and D21 f exist at every point of E.
Suppose Q ⊆ E is a closed rectangle with sides parallel to the coordinate axes, having (a, b) and (a + h, b + k)
as opposite vertices (h, k ̸= 0). Put
∆(f, Q) = f (a + h, b + k) − f (a + h, b) − f (a, b + k) + f (a, b).
Then there is a point (x, y) in the interior of Q such that
∆(f, Q) = hk(D21 f )(x, y).
Proof. Put u(t) = f (t, b + k) − f (t, b). Then by 5.2.3.1, there exists x ∈ (a, a + h), y ∈ (b, b + k), such that
∆(f, Q) = u(a + h) − u(a) = hu′ (x) = h(D1 f (x, b + k) − D1 f (x, b)) = hkD21 f (x, y).
Proposition 9.6.3 (Peano). Suppose f is defined in an open set E ⊆ R2 , suppose that D1 f, D21 f, D2 f
exist at every point of E, and D21 f is continuous at some point (a, b) ∈ E. Then D12 f exists at (a, b) and
D12 f (a, b) = D21 f (a, b).
Proof. Put A = D21 f (a, b). Given ϵ > 0, let Q be a rectangle as in 9.6.2, and if h and k are sufficiently small,
we have
|A − D21 f (x, y)| < ϵ
for all (x, y) ∈ Q. Thus
∆(f, Q)
−A <ϵ
hk
. Fix h and let k → 0. Since D2 f exists in E, the last inequality implies that
D2 f (a + h, b) − D2 f (a, b)
− A ≤ ϵ.
h
Since ϵ is arbitrary, then it follows that D12 f (a, b) = A.
Corollary 9.6.3.1. If f ∈ C 2 (E), then D21 f = D12 f .
108
Proof. Fixed x ∈ [a, b] and s ∈ (c, d), define
φ(x, t) − φ(x, s)
ψ(x, t) =
t−s
, whenever 0 < |t − s| < δ. Then by 5.2.3.1, there exist ξ between s and t such that ψ(x, t) = D2 φ(x, ξ)
and therefore |ψ(x, t) − D2 φ(x, s)| < ϵ for all x ∈ [a, b]. Hence, ψ t ⇒ D2 φ(x, s) as t → s and therefore
(D2 φ)(x, s) ∈ R(α). Note that
b b
f (t) − f (s)
Z Z
− (D2 φ)(x, s)dα(x) ≤ |ψ(x, t) − (D2 φ)(x, s)| dα ≤ ϵ(α(b) − α(a)),
t−s a a
as required.
109