Professional Documents
Culture Documents
Course 221
David Simms
School of Mathematics
Trinity College
Dublin
Typesetting
Eoin Curran.
1
2
Contents
1 Lebesgue Measure 7
1.1 Algebra of Subsets . . . . . . . . . . . . . . . . . . . . . . . . 7
1.2 The Interval Algebra . . . . . . . . . . . . . . . . . . . . . . . 9
1.3 The length measure . . . . . . . . . . . . . . . . . . . . . . . . 11
1.4 The σ-algebra . . . . . . . . . . . . . . . . . . . . . . . . . . . 13
1.5 The outer measure . . . . . . . . . . . . . . . . . . . . . . . . 14
1.6 Extension of measure to σ-algebra, using outer measure . . . . 16
1.7 Increasing Unions, Decreasing Intersections . . . . . . . . . . . 20
1.8 Properties of Lebesgue Measure . . . . . . . . . . . . . . . . . 22
1.9 Borel Sets . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 24
2 Integration 29
2.1 Measure Space, Measurable sets . . . . . . . . . . . . . . . . . 29
2.2 Characteristic Function . . . . . . . . . . . . . . . . . . . . . . 30
2.3 The Integral . . . . . . . . . . . . . . . . . . . . . . . . . . . . 31
2.4 Monotone Convergence Theorem . . . . . . . . . . . . . . . . 34
2.5 Existence of Monotone Increasing Simple Functions converg-
ing to f . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 36
2.6 ’Almost Everywhere’ . . . . . . . . . . . . . . . . . . . . . . . 39
2.7 Integral Notation . . . . . . . . . . . . . . . . . . . . . . . . . 45
2.8 Fundamental Theorem of Calculus . . . . . . . . . . . . . . . 46
2.9 Fatou’s Lemma . . . . . . . . . . . . . . . . . . . . . . . . . . 48
2.10 Dominated Convergence Theorem . . . . . . . . . . . . . . . . 49
2.11 Differentiation under the integral sign . . . . . . . . . . . . . . 50
3 Multiple Integration 53
3.1 Product Measure . . . . . . . . . . . . . . . . . . . . . . . . . 53
3.2 Monotone Class . . . . . . . . . . . . . . . . . . . . . . . . . . 56
3.3 Ring of Subsets . . . . . . . . . . . . . . . . . . . . . . . . . . 57
3.4 Integration using Product Measure . . . . . . . . . . . . . . . 59
3.5 Tonelli’s Theorem . . . . . . . . . . . . . . . . . . . . . . . . . 63
3
3.6 Fubini’s Theorem . . . . . . . . . . . . . . . . . . . . . . . . . 64
4 Differentiation 71
4.1 Differentiation . . . . . . . . . . . . . . . . . . . . . . . . . . . 71
4.2 Normed Space . . . . . . . . . . . . . . . . . . . . . . . . . . . 72
4.3 Metric Space . . . . . . . . . . . . . . . . . . . . . . . . . . . 73
4.4 Topological space . . . . . . . . . . . . . . . . . . . . . . . . . 73
4.5 Continuous map of topological spaces . . . . . . . . . . . . . . 75
4.6 Homeomorphisms . . . . . . . . . . . . . . . . . . . . . . . . . 77
4.7 Operator Norm . . . . . . . . . . . . . . . . . . . . . . . . . . 77
4.8 Differentiation . . . . . . . . . . . . . . . . . . . . . . . . . . . 80
4.9 Notation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 83
4.10 C r Functions . . . . . . . . . . . . . . . . . . . . . . . . . . . 84
4.11 Chain Rule . . . . . . . . . . . . . . . . . . . . . . . . . . . . 87
4
7.13 orientation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 140
5
6
Chapter 1
Lebesgue Measure
1. ∅ ∈ A
2. E ∈ A =⇒ E 0 ∈ A. E 0 = {x ∈ X : x 6∈ E}
1. (a, b] = {x ∈ R : a < x ≤ b}
2. (−∞, b] = {x ∈ R : x ≤ b}
3. (a, ∞) = {x ∈ R : a < x}
4. (−∞, ∞) = R
7
We want to assign a ‘length’ to each element of the interval algebra I, so
we want to allow ‘∞’ as a length. We adjoin to the real no’s the 2 symbols
∞ and −∞ to get the Extended Real Line:
R ∪ {−∞, ∞} = [−∞, ∞]
1. −∞ < x < ∞ ∀x ∈ R
2. Addition
(a) ∞ + ∞ = x + ∞ = ∞ + x = ∞
(b) (−∞) + (−∞) = x + (−∞) = (−∞) + x = −∞
3. Multiplication
(a) ∞∞ = ∞ = (−∞)(−∞)
(b) ∞(−∞) = (−∞)∞ = −∞
∞ x>0
(c) x∞ = ∞x = 0 x=0
−∞ x < 0
−∞ x>0
(d) x(−∞) = (−∞)x = 0 x=0
∞ x<0
1. m(∅) = 0
S∞
2. if
P∞ E = j=1 Ej is countably disjoint with Ej ∈ A ∀j then m(E) =
j=1 m(Ej ). (m is Countably Additive.)
8
E
Note:
E = E1 ∪ · · · ∪ Ek
(disjoint) where each Ei is of the form (a, b], (−∞, b], (a, ∞), or(−∞, ∞). The
length of E is m(E) = m(E1 ) + m(E2 ) + · · · + m(Ek ) where m(a, b] = b − a.
a− a a+
9
Theorem 1.2.1. (Heine-Borel-Lebesgue) The closed interval is compact.
Let {Vi }i∈I be a family of open sets in R which cover the closed interval
[a, b]. Then ∃ a finite number of them: Vi1 , . . . , Vik (say) which cover [a, b]
(i.e. [a, b] ⊂ Vi1 ∪ · · · ∪ Vik .)
Proof. Put
a c b
a ∈ K =⇒ K 6= ∅
Let
c = sup K
Then
a ≤ c ≤ b c ∈ Vj (say)
Vj is open, therefore
∃ > 0 s.t. (c − , c + ) ⊂ Vj
and
∃ k ∈ K s.t. c − < k ≤ c,
Then
[a, min(c + , b)] ⊂ (Vi1 ∪ · · · ∪ Vir ∪ Vj )
2
therefore
min(c + , b) ∈ K
2
But c + 2
6∈ K (as c = sup K) =⇒ b ∈ K as required.
10
1.3 The length measure
Theorem 1.3.1. The length function
m : I −→ [0, ∞]
(a1 , b1 ], . . . , (an , bn ]
so that:
a a1 b1 a2 b2 a3 b3 an bn b
then
Pn
j=1 (bj − aj ) = b1 − a1 +b2 − a2 +b3 − a3 + · · · +bn − an
≤ a2 − a +a3 − a2 +a4 − a3 + · · · +b − an
=b−a
therefore ∞
X
(bj − aj )
j=1
11
By the compactness of [a, b] ∃ a finite number of these open sets which
cover [a, b], and which by renumbering and discarding some intervals if nec-
essary we can take as:
(a − 0 , a + 0 ), (a1 , b1 + 1 ), . . . , (an , bn + n )
so that:
a − 0 a + 0 a2 b2 + 2 an bn + n
Therefore ∞
X
(bj − aj ) = b − a
j=1
as required.
Let m :−→ [0, ∞] be a measure m on an algebra A of subsets of X.
T henm(E) ≤ m(F )
E ⊂ F =⇒ and
m(F ∩ E 0 ) = m(F ) − m(E) if m(F ) 6= ∞
Then
F = E ∪ (F ∩ E 0 )
12
Theorem 1.3.2. (m is subadditive on countable unions)
∞
[ ∞
X
m( )≤ m(Ei )
j=1 i=1
E1 E2 E3 En
∞
[
Ej = E1 ∪ (E2 ∩ F1 0 ) ∪ (E3 ∩ F2 0 ) ∪ · · ·
j=1
S∞
m( j=1 Ej ) = m(E1 ) +m(E2 ∩ F1 0 ) +m(E3 ∩ F2 0 ) + · · ·
≤ m(E1 ) +m(E2 ) +m(E3 ) +···
hence the result: ∞ ∞
[ X
m( Ej ) ≤ m(Ej )
j=1 j=1
13
1.5 The outer measure
Definition We define the outer measure m̂ associated with m to be the
function:
m̂ : {all subsets of X} −→ [0, ∞]
given by:
∞
X
m̂(E) = inf m(Ej )
j=1
where the inf is taken over all sequences Ej of elements of A such that:
∞
[
E⊂ Ej
j=1
Proof. 1. if
E∈A
and ∞
[
E⊂ Ej with Ej ∈ A
j=1
then: ∞
X
m(E) ≤ m(Ej )
j=1
therefore
m(E) ≤ m̂(E)
2. if
E∈A
then
E, ∅, ∅, . . .
is a sequence of elements of A whose union contains E. Therefore:
m̂(E) ≤ m(E) + 0 + 0 + · · ·
therefore
m̂(E) ≤ m(E)
14
Theorem 1.5.2. E ⊂ F =⇒ m̂(E) ≤ m̂(F )
0 , 1 , 2 , . . . > 0
such that
∞
X
j =
j=0
and s.t.
∞
X
m(Bij ) ≤ m̂(Ei ) + i
j=1
Then
∞
[ ∞ [
[ ∞
Ei ⊂ Bij
i=1 i=1 j=1
and
∞ X
X ∞ ∞
X
m(Bij ) ≤ m̂(Ei ) +
i=1 j=1 i=1
therefore
∞
[ ∞
X
m̂( Ej ) ≤ m̂(Ej )
j=1 j=1
as required.
15
Definition We call a subset E ⊂ X measurable w.r.t m if:
m̂(A) = m̂(A ∩ E) + m̂(A ∩ E 0 )
for all A ⊂ X
(E splits every set A into two pieces whose outer measures add up.)
A
2
1
16
Proof. 1. A ⊂ M
Let E ∈ A, let A ⊂ X
Need to show:
m̂(A) = m̂(A ∩ E) + m̂(A ∩ E 0 )
For > 0. Let
F1 , F2 , . . .
be a sequence in A s.t.
∞
[
A⊂ Fj
j=1
and s.t.
∞
X
m(Fj ) ≤ m̂(A) +
j=1
A
Fi
Then
≤ m̂(A) +
17
therefore,
m̂(A) ≤ m̂(A ∩ E) + m̂(A ∩ E 0 ) ≤ m̂(A) +
∀ > 0, and therefore
m̂(A) = m̂(A ∩ E) + m̂(A ∩ E 0 )
as required.
2. M is an algebra
let E, F ∈ M ; Let A ⊂ X
4
3
1 2
Then
m̂(A) = m̂(1 + 2) +m̂(3 + 4) since E ∈ M
18
A
E1 E2 En
Fn
therefore
E∈M
and M is closed under countable disjoint unions. But any countable
union:
E1 ∪ E 2 ∪ E 3 ∪ · · ·
19
can be written as a countable disjoint union
E1 ∪ (E2 ∩ F1 0 ) ∪ (E3 ∩ F2 0 ) ∪ · · ·
where
Fn = E 1 ∪ · · · ∪ E n
therefore M is closed under countable unions as required.
4. m̂ is countably additive on M
Put A = E in (*) to get
∞ ∞
!
X [
m̂(Ek ) = m̂ Ek
k=1 k=1
as required.
E
E2
E1
E1 ⊂ E 2 ⊂ · · · ⊂ E j ⊂ · · ·
is an increasing sequence of sets such that
∞
[
Ej = E
j=1
E1
E2
E
20
E1 ⊃ E 2 ⊃ · · · ⊃ E j ⊃ · · ·
is a decreasing sequence of sets such that
∞
\
Ej = E
j=1
E = E1 ∪ (E2 ∩ E1 0 ) ∪ (E3 ∩ E2 0 ) ∪ · · ·
= limn→∞ m(En )
as required.
Theorem 1.7.2.
Ej ↓ E
=⇒ lim m(Ej ) = m(E)
m(E1 ) 6= ∞ j→∞
Proof.
(E1 ∩ Ej 0 ) ↑ (E1 ∩ E 0 )
therefore:
lim m(E1 ∩ Ej 0 ) = m(E1 ∩ E 0 )
j→∞
and
lim [m(E1 ) − m(Ej )] = m(E1 ) − m(E)
j→∞
so
lim m(Ej ) = m(E)
j→∞
as required.
21
Example
1
(a − , b] ↓ [a, b]
n
a−
1 a b
n
= limn→∞ (b − a + n1 )
=b−a
as expected.
Xi ↑ X
(−n, n] ↑ R
−n
O n
22
Theorem 1.8.1. (Uniqueness of Extension) Let m be a σ-finite measure on
an algebra A of subsets of X.
Let M be the collection of measurable sets w.r.t. m.
Let l be any measure on M which agrees with m on A.
Then l(E) = m̂(E) ∀E ∈ M
we have
P∞ P∞
l(E) ≤ i=1 l(Ai ) = i=1 m(Ai ) since l = m on A
Therefore,
As a consequence we have:
Lemma 1.8.2. Let m be the Lebesgue measure on R. Then for each mea-
surable set E and each c ∈ R we have:
2. m(cE) = |c|m(E)
23
1. define a measure mc on M by
mc (E) = m(E + c)
then
mc (E) = m(cE)
Then
mc (a, b] = m(c(a, b])
m(ca, cb] c > 0
= m{0} c=0
m[cb, ca) c < 0
cb − ca
= 0
ca − cb
= |c|(b − a)
= |c|m(a, b]
Therefore mc agrees with |c|m on the interval algebra. Therefore mc
agrees with |c|m on M .
V ⊂ G(V) ⊂ W
24
Thus G(V) is the smallest σ-algebra of subsets of X which contains V. G(V)
is called the σ-algebra generated by V.
Ia
a V
[
V = Ia
a∈V
V ∈ G(I)
implies
V ⊂ G(I) =⇒ G(V) ⊂ G(I)
25
We have:
The following theorem shows that any Lebesgue measurable set in R can
be obtained from a Borel set by removing a set of measure zero.
Theorem 1.9.2. Let E be a Lebesgue measurable subset of R. Then there
is a Borel set B containing E such that
B − E = B ∩ E0
m(B) = m(E)
B
B-E
E
and s.t. ∞
X 1
m(E) ≤ m(Ii ) ≤ m(E) +
i=1
k
S∞
Put Bk = i=1 Ii then E ⊂ Bk , Bk is Borel, and
1
m(E) ≤ m(Bk ) ≤ m(E) +
k
T∞
Put B = k=1 Bk . Then E ⊂ B, B is Borel, and
1
m(E) ≤ m(B) ≤ m(E) +
k
for all k. So
m(E) = m(B) =⇒ m(B − E) = 0
since m(E) is finite.
26
2. Suppose m(E) = ∞.
−(k + 1) −k k k+1
Put
Ek = {x ∈ E : k ≤ |x| < k + 1}
Then ∞
[
E= Ek
k=0
Ek ⊂ B k
and
m(Bk − Ek ) = 0
S∞
Put B = k=1 Bk . Then E ⊂ B, B is Borel, and
∞
[
B−E ⊂ (Bk − Ek )
k=1
therefore ∞ ∞
X X
m(B − E) ≤ m(Bk − Ek ) = 0=0
k=1 k=1
as required.
B2
B1
E1 E2
27
28
Chapter 2
Integration
f : X −→ [−∞, ∞]
measurable if
f −1 (α, ∞] = {x ∈ X : f (x) > α}
is a measurable ∀α ∈ R.
R
α f
>α >α
{E ⊂ R : f −1 (E) is measurable}
29
is a σ-algebra containing the sets
{(α, β] : α, β ∈ R}
therefore contains the Borel sets. Therefore f −1 is measurable for each Borel
set B ⊂ R.
E E
a1 , a2 , . . . , an
(say). Each simple function φ can be written in a unique way as:
φ = a 1 χE 1 + · · · + a n χE n
where a1 , . . . , an are distinct and
X = E1 ∪ · · · ∪ En
is a disjoint union.
30
2.3 The Integral
Definition If φ is a non-negative measurable simple function with φ =
a1 χE1 + · · · + an χEn ; X = E1 ∪ · · · ∪ En disjoint union we define the integral
of φ w.r.t. the measure m to be
Z
φ dm = a1 m(E1 ) + · · · + an m(En ) ∈ [0, ∞]
a1
a2
E1 E2
31
R P
(φ + ψ) = ck m(Gk )
P P
= k ck {i,j:ai +bj =ck } m(Ei ∩ Fj )
P
= i,j (ai + bj )m(Ei ∩ Fj )
P P
= ai m(Ei ∩ Fj ) + bj m(Ei ∩ Fj )
P P
= i ai m(Ei ) + j bj m(Fj )
R R
= φ dm + ψ dm
which proves 1.
Then λ is a measure on M .
Proof. Let
φ = a 1 χE 1 + · · · a n χE n
for ai ≥ 0. Then
R
λE = E
φ dm
R
= φχE dm
R
= (a1 χE1 ∩E + · · · + an χEn ∩E ) dm
therefore
λ = a 1 mE 1 + · · · + a n mE n
is a linear combination of measures with non-negative coefficients. Therefore
λ is a measure.
32
S∞
Corollary 2.3.2. 1. If E = n=1 En is a countable disjoint union then
Z ∞ Z
X
φ dm = φ dm
E n=1 En
S∞
2. If E = n=1 En is a countable increasing union then
Z Z
φ dm = lim φ dm
E n→∞ En
T∞ R
3. If E = n=1 En is a countable decreasing intersection and E1
φ dm <
∞ then Z Z
φ dm = lim φ dm
E n→∞ En
where the sup is taken over all simple measurable functions φ such that:
0≤φ≤f
φ f
33
f
PSfrag replacements f χE
Ε
We then have:
R R
1. f ≤ g =⇒ f dm ≤ g dm
R R
2. E ⊂ F =⇒ E f dm ≤ F f dm
1.
fn ≤ fn+1 ≤ f
for all n. Therefore:
Z Z Z
fn dm ≤ fn+1 dm ≤ f dm
34
2. Let φ be a simple measurable function s.t.:
0≤φ≤f
φ f
fn
αφ
f
fn
En
Z Z Z
αφ dm ≤ fn dm ≤ fn dm
En En
Let n → ∞: Z Z
α φ dm ≤ lim fn dm
so: Z Z
f dm ≤ lim fn dm
hence Z Z
f dm = lim fn dm
35
Definition In probability theory we have a measure space:
X , M , P
sample events
space
P {x : f (x) ∈ B}
is the probability that the random variable f takes value in the Borel set
B ⊂ R. If
f = a 1 χE 1 + · · · + a n χE n
with a1 , . . . , an ; E1 , . . . , En disjoint, and E1 ∪ · · · ∪ En = X, then the proba-
bility that f takes value ai is
P {x ∈ X : f (x) = ai } = P (Ei )
and Z
f dP = a1 P (E1 ) + · · · + an P (En )
36
Proof. Put each integer n > 0:
k k k+1
2n 2n
≤ f (x) < 2n
k = 0, 1, 2, . . . , n2n − 1
φn (x) =
n f (x) ≥ n
k+1
2n
k
2n
PSfrag replacements
Then
1. 0 ≤ φn (x) ≤ φn+1 (x)
37
Theorem 2.5.3. Let fn be a sequence of non-negative measurable functions
X −→ [0, ∞]. Then:
Z X ∞ ∞ Z
X
( fn ) dm = fn dm
n=1 n=1
Therefore: Z X ∞ ∞ Z
X
( fr ) dm = fr dm
r=1 r=1
M CT P∞ R
= k=1 f χEk dm
P∞ R
= k=1 Ek
f dm
P∞
= k=1 λ(Ek )
therefore λ is countably additive, as required.
S
Corollary 2.5.5. 1. if E = ∞ k=1 Ek countable disjoint union then
Z X∞ Z
f dm = f dm
E k=1 Ek
R R
2. if Ek ↑ E then limk→∞ Ek f dm = E f dm
R R R
3. If Ek ↓ E and E1 f dm < ∞, then limk→∞ Ek f dm = E f dm
38
2.6 ’Almost Everywhere’
Definition Let f, g : X −→ [0, ∞]. Then we say that f = g almost every-
where (a.e.) or f (x) = g(x) almost all x ∈ X (a.a.x) if {x ∈ X : f (x) 6= g(x)}
has measure zero.
1. Put
1
En = {x ∈ X : f (x) > }
n
for each integer n > 0, so
1
f> χE
n n
1
n
PSfrag replacements
En
R
Suppose that f dm = 0. Then
Z
1
0 = f dm ≥ m(En )
n
so
m(En ) = 0
for all n. But En ↑ E, so
therefore
f = 0 a.e.
0 ≤ f ≤ lim nχE
39
nχE
PSfrag replacements
E
therefore R
0 ≤ f dm
R
≤ lim nχE dm
M CT R
= lim nχE dm
= lim nm(E)
= lim 0 = 0
so Z
f dm = 0
Proof.
f χE = 0 a.e.
therefore Z Z
f dm = f χE dm = 0
E
40
Proof. Let f = g on E and m(E 0 ) = 0 then
Z Z Z Z Z Z
f dm = f dm + f dm = g dm + g dm = g dm
E E0 E E0
R
Thus changing f on a set of measure zero makes no difference to f dm.
Also
M CT R
= lim E
fn dm + 0
R R
= lim f dm +
E n E0
fn dm
R
= lim fn dm
X −→ [0, ∞]
41
f+
PSfrag replacements
f
f−
f
PSfrag replacements
and we write Z Z Z
+
f dm = f dm − f − dm
42
Proof. 1. f = f1 − f2 , therefore f + ≤ f1 ; f − ≤ f2 . So
Z Z Z Z
+ −
f dm ≤ f1 dm < ∞; f dm ≤ f2 dm < ∞
therefore f is integrable.
2. f = f1 − f2 = f + − f − . So
f1 + f − = f + + f 2
therefore
Z Z Z Z
− +
f1 dm + f dm = f dm + f2 dm
so Z Z Z Z Z
+ −
f1 dm − f2 dm = f dm − f dm = f dm
as required.
Proof. R R
f + = g + a.e. =⇒ f + dm = g + dm
R R
f − = g − a.e. =⇒ f − dm = g − dm
so
Z Z Z Z Z Z
+ − + −
f dm = f dm − f dm = g dm − g dm = g dm
2. if f is integrable then
Z Z
f dm ≤ |f | dm
43
Proof. 1.
R R
f integrable ⇔ f + dm < ∞ and f − dm < ∞
R
⇔ (f + + f − ) dm < ∞
R
⇔ |f | dm < ∞
⇔ |f | integrable
2. R R R
f dm = f + dm − f − dm
R R
≤ f + dm + f − dm
R
= |f | dm
(say) (f1 , f2 real) is called integrable if f1 and f2 are integrable and we define:
Z Z Z
f dm = f1 dm + i f2 dm
Thus Z Z
< f dm = (<f ) dm
Z Z
= f dm = (=f ) dm
and Z Z
(cf ) dm = c f dm
44
Thus, the set L(X, R, m) of all integrable real valued functions on X is
a real vector space, and the set L(X, C, m) of all integrable complex valued
functions on X is a complex vector space. And on each space:
Z
f −→ f dm
is a linear form.
If f : X −→ C is integrable then
Z Z
f dm = f dm eiθ (say) θ real
therefore
R (real) R
f dm = e−iθ f dm
R (real)
−iθ
= e f dm
R
= <[e−iθ f ] dm
R −iθ
≤ e f dm
R
= |f | dm
therefore Z Z
f dm ≤ |f | dm
if a ≤ b. It follows that
Z b Z c Z c
f (x) dx + f (x) dx = f (x) dx
a b a
and Z b
dx = b − a
a
45
∀a, b, c.
Notice that x is a dummy symbol and that
Z b Z b Z b
f (x) dx = f (y) dy = f (t) dt = · · ·
a a a
just as
n
X n
X
a i bi = a j bj = · · ·
i=1 j=1
|f (t + h) − f (t)| ≤ ∀|h| ≤ δ
by continuity of f . Therefore
F (t+h)−F (t)
h
− f (t)
R R t+h
t+h
= h1 t f (x) dx − 1
h t
f (t) dx
R
1 t+h
= |h| t [f (x) − f (t)] dx
1
≤ |h|
|h| =
F 0 (t) = G0 (t)
46
for all t. Therefore
F (t) = G(t) + c
for all t, where c is constant. Therefore
Z b
G(b) − G(a) = F (b) − F (a) = G0 (x) dx
a
as required.
Proof. Put Z t
G(t) = f (x) dx
t1
= G(g(t4 )) − G(g(t3 ))
= G(t2 ) − G(t1 )
R t2
= t1
f (x) dx
as required.
To deal with sequences which are not monotone we need the concepts of
lim inf and lim sup.
Definition Let
{an } = a1 , a2 , a3 , . . .
47
be a sequence in [−∞, ∞]. Put
b1 ≤ b2 · · · ≤ bn ≤ bn+1 ≤ · · · ≤ cn+1 ≤ cn ≤ · · · ≤ c2 ≤ c1
Define:
lim inf an = lim bn
lim sup an = lim cn
Then
lim inf an ≤ lim sup an
and an converges iff lim inf an = lim sup an (= lim an ).
Proof. Put
gr = inf{fr , fr+1 , . . .}
so that
· · · ≤ gr ≤ gr+1 ≤ · · ·
Now:
fn ≥ g r ∀n ≥ r
R R
=⇒ fn ≥ gr ∀n ≥ r
R R
=⇒ lim inf fn ≥ gr ∀r
R R M CT R R
=⇒ lim inf fn ≥ lim gr = lim gr = lim inf fn
as required.
48
2.10 Dominated Convergence Theorem
Theorem 2.10.1. Lebesgue’s Dominated Convergence Theorem, DCT Let
fn be integrable and fn → f . Let
|fn | ≤ g
R R
for all n where g is integrable. Then f is integrable and f = lim fn
Proof. |f | = lim |fn | ≤ g. Therefore |f | is integrable, and therefore f is
integrable. Now
g ± fn ≥ 0
therefore Z Z
lim inf[g ± fn ] ≤F AT OU lim inf [g ± fn ]
so Z Z Z
[g ± f ] ≤ g + lim inf ± fn
so Z Z
± f ≤ lim inf ± fn
Which gives: R R
: f ≤ lim inf fn
R R
: − f ≤ − lim sup fn
therefore Z Z Z Z
f ≤ lim inf fn ≤ lim sup fn ≤ f
which is equivalent to: Z Z
lim fn = f
as required.
For series this leads to:
Theorem 2.10.2. Dominated convergence theorem for series Let fn be a
sequence of integrable functions such that
X∞ Z
|fn | dm < ∞
n=1
P∞
Then n=1 fn is (equal a.e. to) an integrable function and
Z X ∞ X∞ Z
( fn ) dm = fn dm
n=1 n=1
49
Proof. Put
sn = f 1 + f 2 + · · · + f n
sum to n terms. Then
∞
X
|sn | ≤ |f1 | + · · · + |fn | ≤ |fr | = g
r=1
(say). Then
R R P∞
g dm = r=1 |fr | dm
M CT P∞ R
= r=1 |fr | dm < ∞
Therefore g is integrable (a.e. equal to an integrable fn). Therefore lim sn is
integrable and Z Z
DCT
lim sn dm = lim sn dm
therefore
∞
Z X ∞ Z
X
fr dm = fr dm
r=1 r=1
as required.
Proof. Put Z
F (t) = f (x, t) dx
50
Let a ≤ t ≤ b. Choose a sequence tn in [a, b] s.t. lim tn = t and tn 6= t. Then,
by the Mean Value Theorem:
|f (x, tn ) − f (x, t)| =|tn − t| ∂f
∂t
(x, c(n, x, t))
≤ |tn − t|g(x)
Therefore
f (x, tn ) − f (x, t)
≤ g(x)
tn − t
so R
lim F (tntn)−F
−t
(t)
= lim f (x,tn )−f (x,t)
tn −t
dx
DCT R
= lim f (x,tntn)−f
−t
(x,t)
dx
R ∂f
= ∂t
(x, t) dx
and therefore Z
dF ∂f
= (x, t) dx
dt ∂t
as required.
51
52
Chapter 3
Multiple Integration
B A×B
PSfrag replacements
A
53
S
Lemma 3.1.1. Let A×B = ∞ i=1 Ai ×Bi be a rectangle written as a countable
disjoint union of rectangles. Then
∞
X
l(A)m(B) = l(Ai )m(Bi )
i=1
Proof. We have
∞
X
χA×B = χAi ×Bi
i=1
∞
X
=⇒ χA (x)χB (y) = χAi (x)χBi (y)
i=1
Fix x and integrate w.r.t. m term by term using the Monotone Convergence
Theorem: ∞
X
χA (x)m(B) = χAi (x)m(Bi )
i=1
as required.
54
then ∞
[
Ai × B i = (Ai ∩ Cj ) × (Bi ∩ Dj )
j=1
and hence,
P∞ P∞ P ∞
i=1 l(Ai )m(Bi ) = i=1 j=1 l(Ai ∩ Cj )m(Bi ∩ Dj )
P∞
= j=1 l(Cj )m(Dj )
Dj
PSfrag replacements
Bi
Di
Bj Ai Cj
S∞
2. countably additive Let E = i=1 Ei be a countable disjoint union with
E, Ei ∈ A.
ni
[
Ei = Aij × Bij
j=1
Therefore ∞ X
ni ∞
X X
π(E) = l(Aij )m(Bij ) = π(Ei )
i=1 j=1 i=1
55
Definition The measure π on A extends to a measure (also denoted by π
and called the product of the measures l and m) on the σ-algebra (denoted
L × M) of all subsets of X × Y which are measurable w.r.t. π.
Similarly, if
(X1 , M1 , m1 ), . . . , (Xn , Mn , mn )
is a sequence of measure spaces then we have a product measure on a σ-
algebra of subsets of
X1 × · · · × X n
s.t.
π(E1 × · · · × En ) = m1 (E1 )m2 (E2 ) . . . mn (En )
In particular, starting with 1-dimensional Lebesgue measure on R we get
n-dimensional Lebesgue Measure on
R × · · · × R = Rn
= P (E × Y )P (X × F ) by independence
= P1 (E)P2 (F )
1. En ↑ E, En ∈ M =⇒ E ∈ M
2. En ↓ E, En ∈ M =⇒ E ∈ M
56
Definition If V is a non-empty collection of subsets of X and if M is the
intersection of all the monotone classes of subsets of X which contain V, then
M is a monotone class, called the monotone class generated by V.
M is the smallest monotone class containing V and
V ⊂ monotone class ⊂ σ-algebra
generated by V generated by V
57
1. ME ⊂ M by definition
Fn ↑ F, Fn ∈ ME
=⇒ (E ∩ Fn 0 ) ↓ (E ∩ F 0 ) (E ∪ Fn ) ↑ (E ∪ F ) (E 0 ∩ Fn ) ↑ (E 0 ∩ F )
with E ∩ Fn 0 E ∪ Fn E 0 ∩ Fn all ∈ M
=⇒ E ∩ F 0 E∪F E0 ∩ F all ∈ M
(M i monotone
=⇒ F ∈ ME
3. E ∈ R
=⇒ R ⊂ ME since R is a ring. Therefore ME = M by (i), (ii) since
M is smallest monotone class containing R.
4. F ∈ ME ∀E ∈ R, F ∈ M by (iii). Therefore E ∈ MF ∀E ∈ R, F ∈
M.
Therefore
R ⊂ MF ∀F ∈ M
and therefore
MF = M ∀F ∈ M
as required.
E ∈ M =⇒ E 0 = E 0 ∩ X ∈ M
58
S∞
Also, if En is a sequence in M and E = 1 En put
Fn = E 1 ∪ · · · ∪ E n ∈ M
g : Y −→ [0, ∞]
F : X × Y −→ [0, ∞]
If E ⊂ X × Y write
Ex = {y ∈ Y : (x, y) ∈ E}
E y = {x ∈ X : (x, y) ∈ E}
59
y
Ex
x E
y
Proof. To show
Z
π(E) = m(Ex ) dx (3.1)
X
A ⊂ N ⊂ G(A)
(a) A ∈ N .
If E = A × B is a rectangle then
m(Ex ) = χA (x)m(B)
60
E
then Z
m(Ex ) dx = l(A)m(B) = π(E)
X
A⊂N
(En )x ↑ Ex
for each x ∈ X.
En
61
Then R R
X
m(Ex ) dx = X
lim m((En )x ) dx
M CT R
= lim X
m((En )x ) dx
= lim π(En )
= π(E)
= l(X)m(Y ) − π(E)x
= π(X × Y ) − π(E)
= π(E 0 ).
An ↑ X, Bn ↑ Y
(say) with
l(An ) < ∞, m(Bn ) < ∞
Then
Zn ↑ (X × Y )
(E ∩ Zn )x ↑ Ex
for all x ∈ X.
62
Zn E
So: R R
X
m(Ex ) dx = X
lim m((E ∩ Zn )x ) dx
M CT R
= lim X
m((E ∩ Zn )x ) dx
= lim π(E ∩ Zn )
= π(E)
as required.
(notation as before.)
63
Proof. To show
Z Z Z
F (x, y) dy dx = F (x, y) dx dy (3.2)
X Y X×Y
2. (General Case)
∃ monotone increasing sequence of non-negative measurable functions
with F = lim Fn
R R R R
X Y
F (x, y) dy dx = X Y
lim F n (x, y) dy dx
M CT R R
= X
lim Y
Fn (x, y) dy dx
M CT R R
= lim X Y
Fn (x, y) dy dx
(1.) R
= lim X×Y
Fn (x, y) dx dy
M CT R
= X×Y
lim Fn (x, y) dx dy
R
= X×Y
F (x, y) dx dy
64
Proof.
R R R
X×Y
F (x, y) dx dy = X×Y
F + (x, y) dx dy − X×Y
F − (x, y) dx dy
T onelli R R R R
= X Y
F + (x, y) dy dx − X Y F − (x, y) dy dx
R R
Therefore Y F + (x, y) dy, and Y F − (x, y) dy are each finite a.a.x., and each
has a finite integral w.r.t. x. (*)
So:
F (x, y) = F + (x, y) − F − (x, y)
is integrable w.r.t y a.e., and:
Z Z Z
+
F (x, y) dy = F (x, y) dy − F − (x, y) dy
Y Y Y
1.
Z Z Z
χE (x + c) dx = χE−c (x) dx = m(E − c) = m(E) = χE (x) dx
2.
Z Z Z
1 1 1
χE (cx) dx = χ 1 E (x) dx = m( E) = m(E) = χE (x) dx
c c |c| |c|
Therefore, these are true for f simple, and therefore true for f non-negative
measurable (by MCT), since ∃fn simple, s.t. fn ↑ f . Therefore, these are
true for f = f + − f − integrable.
65
We Rnow see how toR deal with integration on change of variable on Rn .
1
Recall f (cx) dx = |c| f (x) dx.
66
3. To replace row i by c row j, multiply A by:
1
..
.
1
D= c
1
..
.
1
B1 B2 · · · Bk A = I
and therefore,
A = Bk−1 · · · B2−1 B1−1
is a product of matrices of type N, P or D.
Now, if the theorem holds for matrices A, B then it also holds for AB
since:
Z Z Z Z
1 1 1
f (ABx) dx = f (Ax) dx = f (x) dx = f (x) dx
| det B| | det B|| det A| | det AB|
1. Let
1 c 0 ··· 0
0 1 0 ··· 0
0 0 1 0 ···
N = ..
.
..
.
1
(say). det N = 1. Then
R R
f (N x) dx = R f (x1 + cx2 , x2 , . . . , xn ) dx1 dx2 · · · dxn
= f (x1 , x2 , . . . , xn ) dx1 dx2 · · · dxn
(by Fubini, and translation invariance)
1
R
= | det N |
f (x) dx
67
2. Let
0 1 0 ··· 0
1 0 0 ··· 0
0 0 1 0 ···
P = ..
.
..
.
1
(say). det P = −1. Then
R R
f (P x) dx = R f (x2 , x1 , . . . , xn ) dx1 dx2 · · · dxn
= f (x1 , x2 , . . . , xn ) dx1 dx2 · · · dxn
(by Fubini)
1
R
= | det P |
f (x) dx
3. Let
c 0 0 ··· 0
0 1 0 ··· 0
0 0 1 0 ···
D= ..
.
..
.
1
(say). det N = c. Then
R R
f (Dx) dx = fR(cx1 , x2 , . . . , xn ) dx1 dx2 · · · dxn
1
= |c| f (x1 , x2 , . . . , xn ) dx1 dx2 · · · dxn
(by Fubini)
1
R
= | det D|
f (x) dx
A
Corollary 3.6.4. if E ⊂ Rn is measurable and Rn −→ Rn is a linear homo-
morphism then
m(A(E)) = | det A|m(E)
Proof. R
m(E) = R χE (x) dx
= χA(E) R (AX) dx
= | det1 A| χA(E) (x) dx
= | det1 A| m(A(E))
68
as required.
69
70
Chapter 4
Differentiation
4.1 Differentiation
f
If R −→ R is a real valued function od a real variable then the derivative of
f at a is defined to be:
f (a + h) − f (a)
f 0 (a) = lim (4.1)
h−→0 h
We want to define the derivative f 0 (a) when f is a vector-valued function of
a vector variable:
f : M −→ N
where m, N are real (or complex) vector spaces.
We cannot use Equation (4.1) directly since we don’t know how to divide
f (a + h) − f (a), which is a vector in N , by h, which is a vector in M
So, we rewrite Equation (4.1) as:
f (a + h) = f (a) + f 0 (a)h + φ(h)
|{z} | {z } |{z}
(constant) (linear in h) (remainder)
where
φ(h) f (a + h) − f (a)
lim = − f 0 (a)
h−→0 h h
This suggests that we take M, N to be normed spaces and define f 0 (a) to
be a linear operator such that
f (a + h) = f (a) + f 0 (a)h + φ(h)
where
kφ(h)k
lim =0
khk−→0 khk
71
a+h a+h
f φ(h)
h f’(a) h
f(a+h)
a f(a)
This f 0 (a)h is the linear approx to the change in f when variable changes by
h from a to a + h.
72
4.3 Metric Space
Definition a set X is called a metric space if a function
D : X × X −→ R
1. d(x, y) ≥ 0
2. d(x, y) = 0 ⇔ x = y
d(x, y) = kx − yk
1. ∅ and X belong to V
S
2. if {Vi }i∈I is any family of elements of V then i∈I vi belongs to V. i.e.
V is closed under unions
We call the elements of V the open sets of the topological space X, or open
in X.
BX (a, r) ⊂ V
73
V
r
a
Theorem 4.4.1. if X is a metric space then each c ∈ X and s > 0 the ball
BX (c, s) is open in X.
Then
x ∈ BX (a, r)
=⇒ d(x, c) < s
=⇒ x ∈ BX (c, s)
74
BX (c, s)
BX (a, r)
x
s r
c a
PSfrag replacements
V open in Y =⇒ f −1 V open in X
f : X −→ Y
i.e.
f BX (a, δ) ⊂ BY (f (a), )
75
f −1 BY (f (a), ) BY (f (a), )
ε
f(a)
PSfrag replacements δ
a
∃δ > 0 s.t.
BX (a, δ) ⊂ f −1 BY (f (a), )
and
f BX (a, δ) ⊂ BY (f (a), )
a
f −1 V
f(a)
PSfrag replacements
therefore,
BX (a, δ) ⊂ f −1 V
76
4.6 Homeomorphisms
Definition a map f : X −→ Y of topolgical spaces is called homeomorphism
if:
1. f is bijective
kT xk
kT k = sup
x∈M kxk
x6=0
We have:
x
1. If x 6= 0 and y = kxk
then
1
kyk = kxk = 1
kxk
77
x
x
kxk
PSfrag replacements
and
kT k = sup kT yk
y∈M
kyk=1
kT k
y
T y
PSfrag replacements
2. if α is a scalar then
3. if S, T : M −→ N then
kS+T k = sup k(S+T )yk ≤ sup (kSyk+kT yk) ≤ sup kSyk+sup kT yk = kSk+kT k
kyk=1 kyk=1
78
4.
kT xk
kT k = 0 ⇐⇒ supx6=0 kxk
=0
⇐⇒ kT xk = 0 ∀x
⇐⇒ T x = 0 ∀x
⇐⇒ T = 0
(by 2,3,4 the operator norm is a norm).
Theorem 4.7.1. 1. if T : M −→ N and x ∈ M then
kT xk ≤ kT k kxk
T S
2. If L −→ M −→ N then
kST k ≤ kSk kT k
Proof. 1.
kT xk
≤ kT k
kxk
∀x 6= 0, be definition. therefore,
kT xk ≤ kT k kxk
for all x
2.
kST k = supkyk=1 kST yk
= kSk kT k
Note:
1. if M is finite dimensional then every choice of norm on M defines the
same topology on M
2. a sequence xr of points in a topological space X is said to converge to
a ∈ M if, for each open set V containing a ∃N s.t. xr ∈ V ∀ r ≥ N .
For a normed space M this is the same as: for each > 0 ∃N s.t.
kxr − ak < ∀ r ≥ N .
79
4.8 Differentiation
Definition let
f
M ⊃ V −→ N
where M, N are normed spaces and V is open in M . Let a ∈ V . THen f is
differentiable at a if ∃ a linear operator
f 0 (a)
M −→ N
such that
f (a + h) = f (a0) + f 0 (a)h + φ(h)
where
kφ(h)k
−→ 0 as khk −→ 0
khk
d
= dt
f (a + th)t=0
a+h
a+th
a
Proof.
f (a + th) = f (a) + f 0 (a)th + φ(th)
therefore,
f (a + th) − f (a) 0
φ(th)
kφ(th)k
− f (a)h
=
= khk
t t
kthk
tends to 0 as t −→ 0.
80
f’(a) h
f(a+h)
h a+h
f(a+th)
a+th f
a
f(a)
f (a + h) = (a + h)3 = |{z}
a3 + 3a 2
|{z} h + 3ah 2 3
| {z+ h}
f (a) f 0 (a)h φ(h)
so f 0 (a) = 3a2 .
f (A + H) = (A + H)3
A3 + A
= |{z} 2
| H + AHA
2
{z + HA}
f (A) f 0 (A)H
2
+ AH
| + HAH{z+ H 2 A + H}3
φ(H)
3kAk kHk2
≤ kHk
= 3kAk kHk −→ 0
as kHk −→ 0
therefore, f is differentiable at A, and
is given by
f 0 (A)H = A2 H + AHA + HA2
81
f
Theorem 4.8.2. Let Rn ⊃ V −→ Rm be a differentiable function with
f = (f 1 , . . . , f m ) = (f i )
Then
0 ∂f i
f =
∂xj
Proof. For each a ∈ V we have
f 0 (a) : Rn −→ Rm
∂f
= ∂xj
(a)
∂f 1 m
= ∂xj
(a), . . . , ∂f
∂xj
(a)
therefore,
0 ∂f i
f (a) = (a)
∂xj
for all a ∈ V . Therefore,
0 ∂f i
f =
∂xj
which is the Jacobian matrix.
f
Theorem 4.8.3. Let M ⊃ V −→ N1 × · · · × Nk where
f (x) = f 1 (x), . . . , f k (x)
Then f 1 ,
. . . , f k differentiable at a ∈ V =⇒ f is differentiable at a, and
0 0
f 0 (a)h = f 1 (a)h, . . . , f k (a)h .
Proof. (k = 2)
f
M ⊃ V −→ N1 × N2
f (x) = f 1 (x), f 2 (x)
Take any norms on M, N1 , N2 and define a norm on N! × N2 by
82
Then
f (a + h) = (f 1 (a + h), f 2 (a + h))
0 0
= f 1 (a) + f 1 (a)h + φ1 (h), f 2 (a) + f 2 (a)h + φ2 (h)
0 0
= (f 1 (a), f 2 (a)) + f 1 (a)h, f 2 (a)h + φ1 (h), φ2 (h)
| {z } | {z }
linear in h remainder
Now:
k (φ1 (h), φ2 (h)) k kφ1 (h)k kφ2 (h)k
= +
khk khk khk
tends to 0 as h −→ 0. Therefore, f is differentiable at am and
0 0
f 0 (a)h = f 1 (a)h, f 2 (a)h
as required.
4.9 Notation
1. Given a function of n variables
f
Rn ⊃ V −→ R
x1 , x2 , . . . , x n
the usual co-ordinate functions on Rn and shall often denote the partial
derivative of f at a w.r.t. j th variable by:
∂f f (a1 ,a2 ,...,aj +t,...,an )−f (a1 ,a2 ,...,aj ,...,an )
∂xj
(a) = limt−→0 t
d
= dt
f (a + tej )t=0
2. given a function
f
R2 ⊃ V −→ R
83
V open, we often denote by x, y the usual co-ordinate functions instead
of x1 , x2 and write
∂f ∂f
for
∂x ∂x1
∂f ∂f
for
∂y ∂x2
etc.
3. given
f
M ⊃ V −→ N
V open, if f is differentiable at a, ∀a ∈ V , we say that f is differentiable
on V and call the function on V :
f 0 : a −→ f 0 (a)
f (2) = (f 0 )0
f (3) = ((f 0 )0 )0
4.10 C r Functions
Theorem 4.10.1. Let
f
Rn ⊃ V −→ R
∂f
V open. Then f is C 1 ⇐⇒ ∂xi
exists and is continuous for i = 1, . . . , n.
Proof. (case n = 2)
f
R2 ⊃ V −→ R, f (x, y)
∂f ∂f
1. if f is C 1 , then ,
∂x ∂y
exist and
0 ∂f ∂f
f = ,
∂x ∂y
∂f ∂f
therefore, f 0 is continuous, and therefore, ∂x
and ∂y
are continuous
84
∂f ∂f
2. Let ,
∂x ∂y
exists and be continuous on V . Then
∂f ∂f
f (a + h) = f (a) + (a)h1 + (a)h2 + φ(h)
∂x ∂y
φ(h) = f (a + h) − f (a + h2 e2 ) − ∂f
∂x
(a)h1
∂f
2
+f (a + h e2 ) − f (a) − ∂y (a)h2
∂f
= ∂x
(m1 )h1 − ∂f
∂x
(a)h1
+ ∂f
∂y
(m2 )h2 − ∂f
∂x
(a)h2
a + h 2 e2
m2 a+h
m1
PSfrag replacements a
Therefore,
kφ(h)k ∂f ∂f |h1 | ∂f ∂f |h2 |
≤ (m1 ) − (a) + (m2 ) − (a) −→ 0
khk ∂x ∂x khk ∂y ∂y khk
∂f ∂f
as h −→ 0 since ,
∂x ∂y
are continuous at a.
∂f
Therefore, f is differentiable at a and f 0 (a) = ∂x
(a), ∂f
∂y
(a) . Also,
0 ∂f ∂f
f = ∂x , ∂y is continuous.
85
∂r f
Corollary 4.10.2. f is C r ⇐⇒ ∂xi1 ···∂xir
exists and is continuous for each
i1 , . . . , ir .
Theorem 4.10.3. if
f
Rn ⊃ V −→ R V open
∂2f
is C 2 then ∂xi ∂xj
is a symmetric matrix.
Proof. Let
f
Rn ⊃ V −→ R
be C 2 . Need to show:
∂2f ∂2f
=
∂x∂y ∂y∂x
b+h - +
b + -
a x a+h
is contained in V . Put
86
Then
f (a + h, b + k) − f (a + h, b)
−f (a, b + k) + f (a, b) = g(a + h) − g(a)
∂ f 2
= hk ∂y∂x (c, d) some b ≤ d ≤ b + k by MVT
and, similarly:
f (a + h, b + k) − f (a + h, b)
∂2f
−f (a, b + k) + f (a, b) = kh ∂x∂y (c0 , di0 )
∂2f ∂2f 0 0
(c, d) = (c , d )
∂y∂x ∂x∂y
Now, let (h, k) −→ (0, 0). Then (c, d) −→ (a, b), and (c0 , d0 ) −→ (a, b).
Therefore,
∂2f ∂2f
(a, b) = (a, b)
∂y∂x ∂x∂y
∂2f ∂2f
by continuity of ∂y∂x
and ∂x∂y
87
Proof. we have
f [g(ah )] = f [g(a) + T h + φ(h)] where T = g 0 (a)
and kφ(h)k
khk
−→ 0
as khk −→ 0
Example
d
f (g 1 (t), . . . , g n (t)) = (f · g)0 (t) = f 0 (g(t)) g 0 (t)
dt | {z } |{z}
1×n n×1
where
g f
R ⊃ U −→ V −→ R
V open in Rn , and f, g differentiable. Then
d 1
dt
g (t)
f 0 (g(t))g 0(t) = ∂f ∂f
(g(t)), . . . ∂x
..
∂x1 n (g(t)) .
d n
dt
g (t)
∂f ∂f
= ∂x1
(g(t)) dtd g 1 (t) +···+ ∂xn
(g(t)) dtd g n (t)
is the usual chain rule.
88
Chapter 5
s.t.
f (a + h) = f (a) + f 0 (a)h + φ(h)
where |φ(h)|
|h|
−→ 0 as h −→ 0. f 0 (a) is a 1 × 1 complex matrix, i.e. a complex
number, and
f (a + h) − f (a) 0
|φ(h)|
− f (a) =
h |h|
which tends to 0 as h −→ 0. Therefore
f (a + h) − f (a)
f 0 (a) = lim
h−→0 h
We call f 0 (a) the derivative of f at a. f 0 is called holomorphic on V if
0 df
f (a) exists ∀ a ∈ V . We write f = dy .
Example
f : C −→ C, f (z) = z n
89
then,
f (z+h)−f (z) (z+h)n −z n
limh−→0 h
= limh−→0 h
n(n−1) n−2
= limh−→0 [nz n−1 + 2
z h + higher powers h]
= nz n−1
Example
f (z) = z
then,
f (z + h) − f (z) h 1 h real
= =
h h −1 h pure imaginary
which does not converge as h −→ 0.
d dg df
2. dy
fg = f dy + g dy
df
d f g dy −f dg
3. dy g
= g2
dy
if g 6= 0
d
4. dz
f (g(z)) = f 0 (g(z)) g 0 (z) chain rule
By definition C = R2
z = x + iy = (x, y)
√ i i
and the operation
by i = −1
of mult C −→ C is the operator R2 −→ R2
0 −1
with matrix , because
1 0
90
preserves addition and commutes with mult by real scalars, for each a ∈ V ,
and f is holomorphic in V iff f 0 (a) also commutes with multiplication by
complex scalars. ⇐⇒ f 0 (a) also commutes with multiplication by i.
∂u ∂u ∂u ∂u
∂x ∂y 0 −1 0 −1 ∂x ∂y
⇐⇒ ∂v ∂v = ∂v ∂v
∂x ∂y
1 0 1 0 ∂x ∂y
∂u ∂v ∂v
∂y
− ∂u
∂x
− ∂x − ∂y
⇐⇒ ∂v ∂v = ∂u ∂u
∂y
− ∂x ∂x ∂y
∂u ∂v
∂x
= ∂y
⇐⇒ Cauchy-Riemann Equations
∂v ∂u
∂x
= − ∂y
∂v
v = 2xy ∂x
= 2y = − ∂u
∂x
∂ 1 ∂
= ∂x
[u + iv] = i ∂y
[u + iv]
therefore,
∂f ∂u ∂v ∂v ∂u
= +i = −i
∂z ∂x ∂x ∂y ∂y
this also gives the Cauchy-Riemann Equations.
t −→ α(t)
91
is called a (parametrical) path in R fom α(t1 ) to α(t2 ) and if f, g are complex-
valued continuous on V we write
Z Z t2
d d
(f dx + g dy) = [f (α(t)) x(α(t)) + g(α(t)) y(α(t))] dt
α t1 dt dt
and call it the integral of the differential form f dx + g dy over the path α. If
σ
[s1 , s2 ] −→ [t1 , t2 ]
We have
R R s2 d d
β
(f dx + g dy) = s1
[f (β(s)) ds x(β(s)) + g(β(s)) ds y(β(s))] ds
R s2 d d
= s1
[f (α(σ(s))) ds x(α(σ(s))) + g(α(σ(s))) ds y(α(σ(s)))] ds
R s2
= s1
[f (α(σ(s)))(x · α)0 (σ(s)) + g(α(σ(s)))(y · α)0 (σ(s))]σ 0 (s) ds
R t2
= t1
[f (α(t)) dtd x(α(t)) + g(α(t)) dt
d
y(α(t))] dt
R
= α
(f dx + g dy)
Definition If f is C 1 on V we write
∂f ∂f
df = dx + dy
∂x ∂y
A differential form of this type is called exact, and is called the differential
of f .
92
If α is a path from a to b, α(t1 ) = a, α(t2 ) = b, then
R R t2
α
df = t1
[f (α(t)) dtd x(α(t)) + g(α(t)) dtd y(α(t))] dt
R t2
= [ d f (α(t))] dt
t1 dt
= f (α(t2 )) − f (α(t1 ))
∂u ∂v ∂v
= ∂x
dx − ∂x
dx + i ∂x dx + i ∂u
∂x
dy
∂u ∂v
= ∂x
+ i ∂x ( dx + i dy)
= f 0 (z) dz
therefore, if f is holomorphic on V open then
1. for any path α in V from a to b:
Z
f 0 (z) dz = f (b) − f (a)
α
path independence
93
2. for any closed path α in V :
Z
f 0 (z) dz = 0
α
in particular,
2. if α is a closed path,
Z
z n dz = 0 n 6= 1
α
But:
1
therefore, z
cannot be the derivative of a holomorphic function on C.
α
Definition if [t1 , t2 ] −→ C is a path in C then we write
Z t2
L(α) = |α0 (t)| dt
t1
= L(α)
94
Theorem 5.2.1. (Estimating a complex integral)
Let f be bounded on α:
|f (α(t))| ≤ M
(say), t1 ≤ t ≤ t2 . then
Z
f (z) dz ≤ M L(α)
α
Proof. R
R
f (z) dz = t2 f (α(t))α0 (t) dt
α t1
R t2
≤ t1
|f (α(t))| |α0(t)| dt
R t2
≤ M t1
|α0 (t)| dt
= M L(α)
∂ ∂ ∂ ∂
and ∂y u = ∂x (−v), ∂y v = ∂x u (by Cauchy-Riemann).
Therefore, the neccessary conditions for path-independence are satisfied,
but not always sufficient, e.g.f (z) = z1 . However, we have
Proof. write Z
i(T ) = f (z) dz
δt
95
and join the mid-points of the sides to get 4 triangles
S1 , S2 , S3 , S4
then
4
X
i(T ) = i(Sj )
j=1
and therefore
4
X
|i(T )| ≤ |i(Sj )| ≤ 4|i(T1 )|
j=1
T1
T1 , T2 , . . . , T n , . . .
with
|i(T )| ≤ 4n |i(Tn )|
T∞
Let j=1 Tj = {c}. Then
R
i(Tn ) = δTn
f (z) dz
R
= δTn
[f (c) + f 0 (c)(z − c) + |z − c|φ(z − c)] dz
R
= δTn
|z − c| φ(z − c) dz where |φ(z − c)| −→ 0 as |z − c| −→ 0
96
Tn
c
z
Let > 0. Choose δ > 0 s.t. |φ(z − c)| < ∀ |z − c| < δ. Let L =length of T .
Choose n s.t. length of Tn = ”ln < δ.
Then
l l l2
|i(Tn )| ≤ n n = n
2 2 4
therefore |i(T )| ≤ l2 , and therefore i(T ) = 0.
[a, z] ⊂ V ∀z ∈ V
f
Theorem 5.3.2. Let C ⊃ V −→ C be a holomorphic function on an open
star-shaped set. Then f = F 0 for some complex-differentable function F on
V and hence: Z
f (z) dz = 0
α
97
w+h
R w+h
F (w + h) = a
f (z) dz
Rw R w+h
= a
f (z) dz + w
f (z) dz by Cauchy
Z w+h
= F (w) + f (w)h + [f (z) − f (w)] dy
w
| {z }
φ(h)
Let > 0. Choose δ > 0 w.t. |f (z) − f (w)| < ∀|z − w| < δ. Therefore
|φ(h)| ≤ |h|
|φ(h)|
lim =0
h−→0 |h|
98
Theorem 5.4.1. Let a ∈ C ad α : [t1 , t2 ] −→ C − {a} be a closed path. The
Z
1 dz
2πi α z − a
d
dt
[α(t) − a]e−β(t) = [α0 (t) − β 0 (t){α(t) − a}] e−β(t)
= 0
α(t) − a
eβ(t) =
α(t1 ) − a
therefore,
α(t2 ) − a
eβ(t2 ) = =1
α(t1 ) − a
and therefore, β(t2 ) = 2nπi, with n an integer.
2nπi
a
eβ(t)
β(t)
frag replacements
99
Therefore,
Z Z t2
dz α0 (s)
= ds = β(t2 ) = 2nπi
α z−a t1 α(s) − a
as required.
Theorem 5.4.2. Let C be a circle and a ∈ C. Then
1. if a is inside C, then C has winding number 1 about a:
Z
dz
= 2πi
C z −a
100
C
C1
α
a
R dz
R dz
R dz
R dz
R dz
α z−a
+ β z−a
+ γ z−a
= C z−a
− C1 z−a
R R
0 + 0 = 0 = C
− C1
therefore,
Z Z Z 2π
dz dz ireiθ
= = dθ = 2πi
C z−a C1 z−a 0 reiθ
(put z = a + reiθ ).
101
V
C
w
Proof. Let > 0. Choose a circle C1 centre w, radius r (say) inside C such
that |f (z) − f (w)| ≤ ∀z ∈ C1 by continuity of f .
C1
α
w
f (z)
Let α, β, γ be as indicated. Then, by integrating z−w :
Z Z Z Z Z
− = + + =0+0+0=0
C C1 α β γ
f (z)
since each of α, β, γ are contained in a star-shaped set on which z−w
is a
holomorphic function of z. Therefore,
R f (z) R f (z)
C z−w
dz = C1 z−w dz
R f (w) R f (z)−f (w)
= C1 z−w
dz + C1 z−w
dz
= 2πi f (w) + 0
because: Z
f (z) − f (w)
dz ≤ 2πr = 2π
C1 z−w r
for all > 0. Hence result.
102
f
Let C ⊃ V −→ C be holomorphic on V open. Let w ∈ V . Pick a circle
C around w such that C and it’s interior ⊃ V . We have:
Z
1 f (z)
f (w) = dz
2πi C z − w
differentiating w.r.t. w under the integral sign:
Z
0 1 f (z)
f (w) = dz
2πi C (z − w)2
r z
w1
PSfrag replacements C
103
5.6 Term-by-term differentiation, analytic func-
tions, Taylor series
Theorem 5.6.1. Let {fn (z)} be a sequence of functions which is uniformly
convergent to the function f (z) on a path α. Then
Z Z
lim fn (z) dz = f (z) dz
n−→∞ α α
≤ l(α) ∀n≥N
hence the result.
TheoremP 5.6.2. (term by term differentiation)
Let ∞ i=1 fn (z) be a series of holomorphic functions on an open set V ,
which converges uniformly on a circle C which, together with it’s interior, is
contained in V .
Then the series converges inside C to a holomorphic function F (say)
and ∞
X
(k)
f = fn(k)
n=1
inside C
Proof. Let w be inside C, then λ = inf z∈C |z − w| > 0.
1 1
=⇒ ≤ k+1 ∀z ∈ C
|z − w| k+1 λ
P∞ fn (z) f (z)
Therefore, n=1 (z−w)k+1 converges uniformly to (z−w)k+1
z ∈ C. Therefore,
∞ Z Z
X fn (z) f (z)
dz = dz
n=1 C (z − w)k+1 C (z − w)k+1
and therefore,
∞
X
fn(k) (w) = f (n) (w)
n=1
as required.
104
f
Definition Let C ⊃ V −→ C with V opn. Then f is called analytic on V
if, for each a ∈ V ∃ r > 0 and constants c0 , c1 , c2 , . . . ∈ C, such that:
f (z) = c0 + c1 (z − a) = c2 (z − a)2 + · · · ∀ z s.t. |z − a| < r
The R.H.S. is called the Taylor series of f about a.
R
r
a
PSfrag replacements
P
Proof. LetP0 < ρ < R < r. cn (z − a)n converges for z − a = R, and
therefore, cn Rn converges. Therefore, ∃K s.t. |cn Rn | ≤ K ∀n. Therefore
K ρ n
|cn (z − a)n | ≤ |cn | |z − a|n ≤ n ρn = K
R R
for all |z − a| ≤ ρ.
Therefore, we can differentiate term by term n times:
f (n) (z) = n!cn + (n + 1)!cn+1 |z − a| + · · · higher powers of z − a
1 (n)
Therefore, f (n) (a) = n!cn , and therefore, cn = n! f (a).
The Taylor coefficents are uniquely determined, f is C ∞ , and
∞
X 1 (n)
f (z) = f (z − a)n on |z − a| < r
n=1
n!
105
f
Theorem 5.6.4. Let C ⊃ V −→ C be holomorphic on open V . Let C be
any circle, centre a s.t. C and it’s interior are contained in V . Then f has
a Taylor series about a convergent inside C. Hence f is analytic on V .
Proof. Let w be inside C. Then
1
R f (z)
f (w) = 2πi C z−w
dz
1
R f (z)
= 2πi C (z−a)−(w−a)
dz
1
R f (z)
= 2πi C (z−a)[1− w−a ]
dz
z−a
1
R f (z) P∞
w−a n
= 2πi C z−a n=1 z−a
dz
Z
P∞ 1 n f (z)
= n=1 (w − a) dz
2πi C (z − a)n=1
| {z }
cn
P∞ 1 (n)
= n=0 n! f (c)(w − a)n
as required.
106
Chapter 6
Further Calculus
kf (x − f (y)k ≤ k kx − yk
Therefore,
R1
kf (y) − f (x)k ≤ 0
kf 0 [ty + (1 − t)x]k ky − xk dt
R1
≤ 0
k ky − xk dt
= k ky − xk
as required.
107
6.2 Contracting Map
Theorem 6.2.1. Let M be a finite dimensional real vector space with norm.
Let
f
M ⊃ V −→ M
be a C 1 function, V open in M , f (0) = 0, f 0 (0) = 1.
Let 0 < < 1, and let B be a closed ball centre O s.t.
k1 − f 0 (x)k ≤ ∀x ∈ B
Then
1.
kf (x) − f (y)k ≥ (1 − ) kx − yk ∀ x, y ∈ B (6.1)
Thus fB is injective.
2. (1 − )B ⊂ f (B) ⊂ (1 + )B
f (B)
r f (1 − )r
PSfrag replacements B
(1 + )r
108
2.
MV T
k(1 − f )(x) − (1 − f )(y)k ≤ kx − yk ∀ x, y ∈ B
therefore, kx − y| − kf (x) − f (y)k ≤ kx − yk, and so
≤ kxk + kak
≤ r + (1 − )r ∀x ∈ B
= r
therefore, g(x) ∈ B ∀ x ∈ B. So g maps B into B and is contracting.
Therefore, by the contraction mapping theorem ∃x ∈ B s.t. g(x) = x.
i.e. x − f (x) + a = x
i.e. f (x) = a. Therefore, a ∈ f (B), and (1 − )B ⊂ f (B) as required.
109
3. f and f −1 are C r
V is C r diffeomorphic to W if ∃ a C r -diffeomorphism V −→ W .
Example
f
R −→ (0, ∞)
y = ex
PSfrag replacements
x = ln y
f 0 (a) : M −→ N
fW : W −→ f (W )
is a C r diffeomorphism.
110
N
f(W)
W
a
M
F (x) = T f (x + a) − T f (a)
We have:
F (0) = T f (a) − T f (a) = 0
U+a f(U+a)
a f
f(a)
T
+a
T f(a)
0
F 0 - T f(a)
U
F(U)
111
f (U + a) by a C r diffeomorphism. Now
F 0 (x) = T · f 0 (x + a)
=⇒ F 0 (0) = T · f 0 (a) = 1M
Choose a closed ball B centre 0 of positive radius s.s.
1
kF 0 (x) − 1M k ≤ ∀x ∈ B
2
and also s.t. det F 0 (x) 6= 0. Then by the previous theorem (with = 12 ) we
have:
FB is injective
and
1
kF (x) − F (y)k ≥ kx − yk ∀ x, y, ∈ B
2
and
1
B ⊂ F (B)
2
F F(B)
F −1
U 1
2
B
PSfrag replacements
112
F
y+l x+h
G
y
x
F(U)
F 0 , G, [·]−1
G C s =⇒ G C s+1 ∀0 ≤ s < r
therefore G is C r as required.
113
Example Let f, g be C r on an open set containing (a, b)
t
W
g(a,b)
b
W’
y
x a f(a,b) z
Let
∂f ∂f
∂(f, g) ∂x ∂y
= 6= 0
∂(x, y) ∂g ∂g
∂x ∂y
at (a, b). Then (f, g) maps an open neighbourhood W of (a, b) onto an open
neighbourhood W 0 of (f (a, b), g(a, b) by a C r diffeomorphism. Therefore, for
each z, t ∈ W 0 ∃ unique (x, y) ∈ W such that
z = f (x, y)
t = g(x, y)
and x = h(z, t), y = k(z, t) where h and k are C r .
114
Chapter 7
y(V)
y n (x) y(x)
x V
rag replacements
X
y 1 (x) Rn
115
Example On the set V = {y 6= 0 or x > 0} in R2 the functions r, θ given
by
x = r cos θ
y = r sin θ
−π < θ < π
θ
(x,y)
π
r
(r, θ)
θ
PSfrag replacements v
θ
−π
f = F (y 1 , . . . , y n )
∂f ∂F 1
= (y , . . . , y n )
∂y i ∂xi
116
Note:
∂f ∂F
∂y i
(a) = ∂xi
(y 1 (a), . . . , y n (a))
d
= dt
F (y(a) + tei )t=0
d
= dt
f (αi (t))t=0
y(a) + tei
αi (t)
a y(a)
rag replacements
therefore, along curve αi all coords y 1 , . . . , y n are constant except ith coord
y i and αi (t) is parameterised by change t in ith coord y i .
αi is called the ith coordinate curve at a.
7.2 C r -manifold
Definition Let y 1 , . . . , y n with domain V , and z 1 , . . . , z n with domain W
be two coordinate systems on X. Then these two systems are called C r -
compatible if each z i is a C r function of y 1 , . . . , y n, and each y i is a C r
function of z 1 , . . . , z n on V ∩ W .
We call X an n-dimensional C r -manifold if a collection of n-dimensional
coordinate systems is given whose domains cover X and which are C r -
compatible.
117
Example the 2-sphere S 2 given by:
x1 + y 1 + z 2 = 1 in R3
x=x x=x
p p
y= 1 − x2 − y 2 z= 1 − x2 − y 2
118
act on the differentiable functions f at a (i.e. functions f which are real-
valued differentiable functions of y 1 , . . . , y n on an open neighbourhood of a)
and are defined by:
∂ ∂f
j
f = j (a)
∂y a ∂y
The operators (7.1) are linearly independent since
i
∂ y i = αj ∂y (a) = αj δ i = αi
αj j (7.2)
∂y j a ∂y j
j ∂
=⇒ α j
= 0 =⇒ αi = 0 ∀i (7.3)
∂y a
The real vector space with basis (7.1) is denoted Ta X and is called the tangent
space of X at a. If v ∈ Ta X then (7.2) shows that v has components vy i
w.r.t. basis (7.1)
α̇(t)
α(t)
PSfrag replacements
i.e.
d
α̇(t)f = dt
f (α(t))
d
= dt
F (y 1 (α(t)), . . . , y n (α(t)) if f = F (y 1 , . . . , y n )
∂F
= ∂xj
(y 1 (α(t)), . . . , y n (α(t))) dtd y j (α(t))
∂f
= ∂y j
(α(t)) dtd y j (α(t))
d j
= dt
y (α(t)) ∂y∂ j f
α(t)
119
˙ is the tangent vector with components
therefore, α(t)
d j
y (α(t)) = ẏ j (t)
dt
i.e.
j ∂
α̇(t) = ẏ (t) j
∂y α(t)
The tangent space Ta X does not depend on the choice of (compatible) coor-
dinates at a since if z 1 , . . . , z n is another coordinate system at a, then
∂
∂z j a
= velocity vector of j th coordinate curve of z 1 , . . . , z n at a
∂y i ∂
= ∂z j ∂y i
∈ Ta X
a
therefore,
∂ ∂
,··· , n
∂z 1 a ∂z a
∂y i
is also a basis for Ta X, and ∂z j
(a) is the transition matrix from z 1 , . . . , z n to
y 1, . . . , y n.
Therefore, every tangent vector (at a) is the velocity vector of some curve,
and vice versa.
120
∂
therefore, dya1 , . . . , dyan is the basis of Ta X ∗ which is dual to the basis ∂y∂ 1 , · · · , ∂y n
a a
of Ta X.
Thus the linear form dyai gives the ith component of tangent vectors at a:
therefore,
∂f
dfa = (a) dyaj
∂y j
(which is called the chain rule for differentials.)
x 7−→ Sx
which assigns to each x ∈ W a tensor of fixed type over the tangent space
Ta X at x. e.g.
Sx : Tx X × (Tx X)∗ × Tx X −→ R
We can add tensor fields, contract them, form tensor products and wedge
products by carrying out these operations at each point x ∈ W . e.g.
(R + S)x = Rx + Sx
(R ⊗ S)x = Rx ⊗ Sx
Definition A tensor field with no indices is called a scalar field (i.e. a real-
valued function); a tensor field with one upper index is called a vector field ;
a tensor field with r skew-symmetric lower indices is called a differential r-
form; a tensor field with two lower indices is called a metric tensor if it is
symmetric and non-singular at each point.
121
If y 1 , . . . , y n is a coordinate system with domain W and S is a tensor field
on W with (say) indices of type down-up-down, then
∂
S = αi j k dy i ⊗ ⊗ dy k
∂y k
where the scalar fields αi j k are the components of S w.r.t. the coordinates
y i.
If f is a scalar field, then df is a differential 1-form. If v is a vector field,
then vf is the scalar field defined by:
if v is a vector field then we can lower it’s index to get a differential 1-form
ω such that
hω, ui = (v|u)
Conversely, raising the index of a 1-form gives a vector field. Raising the
index of df gives a vector field gradf called the gradient of f :
hω, ui = αi ωi
122
∂f
if f is a scalar field df = ∂y i
dy i
∂f
hdf, ui = uf = αi
∂y i
if
∂ ∂
= gij
∂y i ∂y j
then
(·|·) = gij dy i ⊗ dy j
and
ds2 = gij dy i dy j
gradf has components
∂f
g ij
∂y j
therefore,
∂f ∂
gradf = g ij
∂y j ∂y i
so
∂f ∂f
kgradf k = (gradf |gradf ) = hdf, gradf i = g ij
∂y i ∂y j
Example On R3 with the usual coordinate functions x, y, z the usual metric
tensor is
(·|·) = dx ⊗ dx + dy ⊗ dy + dz ⊗ dz
with components
1 0 0
= g ij
0 1 0
gij =
0 0 1
∂
, ∂, ∂
∂x ∂y ∂z
are orthonormal vector fields
∂f ∂f ∂f
df = dx + dy + dy
∂x ∂y ∂y
and
∂f ∂ ∂f ∂ ∂f ∂
gradf = + +
∂x ∂x ∂y ∂y ∂z ∂z
The line element is
ds2 = (dx)2 + (dy)2 + (dz)2
123
If r, θ, φ are spherical polar cordinates on R3 then
x = r sin θ cos φ
y = r sin θ sin φ
z = r cos θ
therefore,
therefore,
1 0 0
gij = 0 r 2 0
0 0 r 2 sin2 θ
and
1 0 0
ij 1
g = 0 r2
0
1
0 0 r 2 sin2 θ
so
∂f ∂f ∂f
df = dr + dθ + dφ
∂r ∂θ ∂φ
and df has components:
∂f ∂f ∂f
, ,
∂r ∂θ ∂φ
therefore, gradf has components
∂f 1 ∂f 1 ∂f
, ,
∂r r 2 ∂θ r 2 sin2 θ ∂φ
therefore,
∂f ∂ 1 ∂f ∂ 1 ∂f ∂
gradf = + 2 + 2 2
∂r ∂r r ∂θ ∂θ r sin θ ∂φ ∂φ
124
7.5 Pull-back, Push-forward
Definition Let X and Y be manifolds and
φ
X −→ Y
a continuous map. Then for each scalar field f on Y we have a scalr field
φ∗ f = f · φ
on X (we assume that φ∗ f is C ∞ for each C ∞ f ).
φ∗ f is called the pull-back of f to X under φ.
φ∗ f f
φ φ(a)
a
rag replacements
X Y
rag replacements
φ φ∗ α̇(t)
a
α̇(t)
α(t) φ(a)φ(α(t))
125
Thus the velocity vector of α(t) pushes forward to the velocity vector of
φ(α(t))
φ
Theorem 7.5.1. Let X −→ Y and let y 1 , . . . , y n be coordinates on X at a
and let
φ1 (x), . . . , φn (x)
be the coordinates of φ(x) w.r.t. a coordinate system z 1 , . . . , z n on Y at φ(a).
Then the push-forward
φ∗
Ta X −→ Tφ(a) Y
has matrix
∂φi
(x)
∂y j
Proof. h i
i th
component of φ∗ ∂y∂ j = φ∗ ∂y∂ j zi
a a
∂
= ∂y j a
[φ∗ z i ]
∂φi
= ∂y j
(a)
since φi (x) = z i (φ(x)), so φi = φ∗ z i .
Thus the matrix of φ∗ is the same as the matrix of the derivative in the
case Rn −→ Rn . The push-forward φ∗ is often called the derivative of φ at
a.
The chain rule for vector spaces
(ψ · φ)0 (x) = ψ 0 (φ(x)) φ0 (x)
corresponds, for manifolds, to:
Theorem 7.5.2. (Chain rule for maps of manifolds; functorial property of
the push-forward)
φ ψ
Let X −→ Y −→ Z be maps of manifolds. Then
(ψ · φ)∗ = ψ∗ · φ∗
Proof. Let α̇(t) be a tangent vector on X, then:
(ψ · φ)∗ α̇(t) = velocity vector of ψ(φ(α(t)))
= ψ∗ [φ∗ α̇(t)]
as required.
126
φ
Definition Let X −→ Y be a map of manifolds and ω a tensor field on Y
with r lower indices. Then we define the pull-back of ω to X under φ to be
the tensor field φ∗ ω on X having r lower indices and defined by:
all v1 , . . . , vr ∈ Tx X, ∀x ∈ X. i.e.
We note the φ∗ on tensor fields preserves all the algebraic operations such
as additions, tensor products, wedge products.
We have:
φ
Theorem 7.5.3. if X −→ Y and f is a scalar field on Y , then
φ∗ df = dφ∗ f
= [φ∗ v]f
= v[φ∗ f ]
= h(dφ∗ f )x , vi
127
(say) so that on an open neighbourhood of a in X they form a coordinate
system on X. Any two such coordinate systems are C r -compatible. Thus X
is an (n − l)-dimensional C r manifold.
Proof.
∂f 1 ∂f 1 ∂f 1
∂x1
··· ∂xl
··· ∂xn
f 0 = ... .. ..
. .
∂f l ∂f l ∂f l
∂x1
··· ∂xl
··· ∂xn l×n
0 0
Let a ∈ X. THen f (a) has rank l, so the matrix f (a) has l linearly inde-
pendent columns: the first l columns (say). Put
F = (f 1 , . . . , f l , xl + 1, . . . , xn )
then
∂f 1 ∂f 1
∂x1
··· ∂xl
0 ···
.. ..
. .
∂f l ∂f l
∂x1
··· ∂xl
0 ···
0
F =
0 ··· 0 1 0
.. .. ..
. . .
0 ··· 0 0 1 l×n
therefore,
∂f 1 ∂f 1
∂x1
(a) ···
∂xl
(a)
det F 0 (a) =
.. ..
. . 6= 0
∂f l ∂f l
∂x 1 (a) · · · ∂x l (a)
PSfrag replacements
n−l
Rl R X Rn−l
Rn−l
W
F F (W )
U a
G
Rl c Rl
128
By the inverse function theorem, F maps an open neighbourhood W of a onto
an open neighbourhood F (W ) by a C r -diffeomorphism with inverse G (say).
Note that F , and hence also G leave the coordinates xl+1 , . . . , xn unchanged.
F maps W ∩ X homeomorphically onto {c} × U where U is open in Rn−l .
Thus (xl+1 , . . . , xn ) maps W ∩ X homeomorphically onto U and is there-
fore an (n − l)-dimensional coordinate system on X with domain W ∩ X.
Also, if
G = (G1 , . . . , Gl , xl+1 , . . . , xn )
on F (W ) then
x1 = G1 (c1 , . . . , cl , xl+1 , . . . , xn )
.. ..
. .
x = Gl (c1 , . . . , cl , xl+1 , . . . , xn )
l
Note: the proof shows that if (say) the first l columns of the matrix
∂f i
∂xj
(a)
are linearly independent, then the l equations in n unknowns
f 1 (x1 , . . . , xn ) = c1
.. ..
. .
f l (x1 , . . . , xn ) = cl
Example 1.
f
Rn ⊃ V −→ R
a C r function.
0 ∂f ∂f ∂f
f = = 1
,..., n = ∇f
∂xi 1×n ∂x ∂x
129
∂f
If (say) ∂x1
(a) 6= 0 at a ∈ X then
x2 , x3 , . . . , x n
!
∂f 1 ∂f 1
∂f i ∂x1
··· ∂xn
= ∂f 2 ∂f 2
∂xj ∂x1
··· ∂xn
f 1 (x1 , . . . , xn ) = c1
f 2 (x1 , . . . , xn ) = c2
(two equations in n unknowns) then X is an (n − 2)-dimensional C r
manifold. If (say)
∂(f 1 , f 2 )
6= 0 at a ∈ X
∂(x1 , x2 )
7.7 Constraints
If f 1 , . . . , f l are C r functions on an open set V in Rn and c = (c1 , . . . , cn ) we
consider the equations
f 1 = c1 , . . . , f l = cl
as a system of constraints which are satisfied by points on the constraint
manifold :
1 1 l l ∂f i
X = x ∈ V : f (x) = c , . . . , f (x) = c , rank j (x) = l
∂x
i
Let X −→ Rn be the inclusion map i(x) = x ∀x ∈ X. Then for each scalar
field f on V :
(i∗ f )(x) = f (i(x)) = f (x) ∀x ∈ X
130
the pull-back i∗ f is the restriction of f to X. We call i∗ f the constrained
function F . If ω is a differential r-form on V the i∗ ω is called the constrained
r-form ω. Similarly if (·|·) is a metric tensor on V and ds2 the line element
the i∗ (·|·) is the constrained matrix and i∗ ds2 is the constrained line element.
For each a ∈ X the push-forward
i
∗
Ta X −→ T a Rn
f (x, y, z) = c f Cr
We have:
∂f ∂f ∂f
df = dx + dy + dz unconstrained
∂x ∂y ∂z
Pulling back to X, where f is constant, we get:
∂f ∂f ∂f
0= dx + dy + dz constrained
∂x ∂y ∂z
If (say) ∂f
∂z
6= 0 at a ∈ X then, by the implicit function theorem, the con-
strained functions x, y are coordinates on X in a neighbourhood W of a and
constrained z is a C r function of x, y on W .
z = F (x, y)
say. Now
∂f . ∂f ∂f . ∂f
dz = − dx − dy
∂x ∂z ∂y ∂z
Therefore,
∂z ∂F ∂f . ∂f
= =−
∂x
y ∂x ∂x ∂z
∂z ∂F ∂f . ∂f
= =−
∂y x ∂y ∂y ∂z
The usual line element on R3 is
131
pulling back to X gives:
h . . i2
ds2 = (dx)2 + (dy)2 + − ∂f
∂x
∂f
∂z
dx − ∂f
∂y
∂f
∂z
dy
. 2 ∂f ∂f
. 2
∂f ∂f
= 1+ ∂x ∂z
2
(dx) + 2 ∂x ∂y
∂f 2
dx dy + 1 + ∂f
∂y
∂f
∂z
(dy)2
( )
∂y
r = const
we have:
ds2 = r 2 (dθ)2 + r 2 sin2 (dφ)2
Thus the constrained line element has components:
2
r 0
0 r 2 sin2 θ
f 1 = c1 , . . . , f l = cl
132
Thus n
X ∂F i
dF = 0 + dx constrained
i=l+1
∂xi
∂F
This is zero at a critical point a, so ∂x i (a) = 0, i = l + 1, . . . , n, and
hence
l
X ∂F
dFa = i
(a)df i
i=1
∂f
at a critical point a of constrained F . If we put
∂F
λi = (a) = rate of change of F at a with respect to the ith constraint
∂f i
we have:
dF = λ1 df 1 + · · · + λl df l
at a where the scalars λ1 , . . . , λl are called Lagrange multipliers.
Since dF is a linear combination of df 1 , . . . , df l at a we can take the
wedge product to get the equations:
dF ∧ df 1 ∧ · · · ∧ df l = 0
f 1 = c1 , . . . , f l = cl
which must hold at any critical point of constrained F .
df = 0 constrained
therefore,
hdf, α̇(t)i = 0
for all curves in X. Hence the system of l linear equations
df 1 = 0, . . . , df l = 0
give the tangent space to X at each point. Also if (·|·) is a metric tensor
with domain V then
133
for all curves in X. Therefore gradf is orthogonal to the tangent space to X
at each point. Hence
gradf 1 , . . . , gradf l
is a basis for the normed space to X at each point.
At a critical point of constrained F we have:
dF = λ1 df 1 + · · · + λl df l
and raising the index gives:
gradF = λ1 gradf 1 + · · · + λl gradf l
thus gradF is normal to the constraint manifold X at each critical point of
constrained F .
134
7.11 Integral of Pull-back
φ
Theorem 7.11.1. Let V −→ φ(V ) be a C 1 diffeomorphism of open V in Rn
onto open φ(V ) in Rn , with det φ0 > 0. Let ω be an n-form on φ(V ). Then
Z Z
∗
φω= ω
V φ(V )
φ
ω
φ∗ ω
Sfrag replacements
V φV
Let
ω = f (x1 , . . . , xn )dx1 ∧ · · · ∧ dxn
φ = (φ1 , . . . , φn )
i.e. φi (x) = xi (φ(x)) i.e. φi = xi · φ = φ∗ xi .
Then
φ∗ ω = f (φ1 , . . . , φn )dφ1 ∧ · · · ∧ dφn
∂φ 1 ∂φn
i1
= f (φ(x)) ∂x i1 dx ∧ ···∧ ∂xin
dxin
∂φi
= f (φ(x)) det ∂xj
(x) dx1 ∧ · · · ∧ dxn
therefore,
Z Z Z
∗ 0
φω= f (φ(x)) det φ (x)dx1 dx2 . . . dxn = f (x) dx
V V φ(V )
135
Corollary 7.11.2. Let y 1 , . . . , y n be positively oriented coordinates with do-
main V open in Rn . Then
Z Z
1 n 1 n
f (y , . . . , y )dy ∧ · · · ∧ dy = f (y1 , . . . , yn )dy1 dy2 . . . dyn
V y(V )
| {z } | {z }
non-dummy y i with wedge Lebesgue integral over y(V ). y1 , . . . , yn dummy. No wedge
y = (y 1 , . . . , y n )
y ∗ xi xi
PSfrag replacements
V y(V )
Proof.
y ∗ xi = x i · y = y i
therefore,
R
LHS = V
y ∗ [f (x1 , . . . , xn ) dx1 ∧ cdots ∧ dxn ]
R
= y(V )
f (x1 , . . . , xn ) dx1 ∧ · · · ∧ dxn
R
= y(V )
f (x1 , . . . , xn ) dx1 dx2 . . . dxn x1 , . . . , xn dummy
= RHS
136
PSfrag replacements W
V
a b
A0 = {x ∈ X : x 6∈ Z}
is open in X.
A0
PSfrag replacements A
{x ∈ X : Sx 6= 0}
supp S
PSfrag replacements S 6= 0
137
1. Fi ≥ 0
2. F1 + · · · + Fk = 1 on A ( partition of unity)
3. each supp Fi is contained in the domain Vi of a coordinate system on
X.
Proof. (sketch) Let
1
e− t t>0
g(t) =
0 t≤0
g(t)
PSfrag replacements
let
g[1 − t2 ]
b(t) =
g(1)
b(t)
PSfrag replacements
1
−1 1
138
then h is C ∞ , supp h ⊂ V , 0 ≤ h ≤ 1, h(a) = 1; ’bump’ at a.
Let W = {x ∈ X : h(x) > 0}. then W is an open neighbourhood of a
and h > 0 on W .
Thus, for each a ∈ A we have an open neighbourhood Wa of a and a
scalar field fa s.t. ha > 0 on Wa , ha is C ∞ , supp ha ⊂ a coord domain Va ,
0 ≤ ha ≤ 1.
Va
Wa
2r
a
rag replacements r
y(a)
Rn
( )
ha i
ha1 +···+hak
6 0
ifhai =
Fi =
0 ifhai = 0
139
h a2
h a1
PSfrag replacements F1 + F 2 = 1
F1 F2
7.13 orientation
Definition Let y 1 , . . . , y n with domain V , z 1 , . . . , z n with domain W be two
coordinate systems on a manifold X. Then they have the same orientation
if
∂(y 1 , . . . , y n) ∂y i
= det >0
∂(z 1 , . . . , z n ) ∂z j
on V ∩ W .
Since
∂ ∂y i ∂
=
∂z j ∂z j ∂y i
this means that ∂z∂ 1 a , · · · , ∂z∂n a has the same orientation in Ta X as ∂y∂ 1 , · · · , ∂y∂n
a a
each a ∈ V ∩ W .
We call X oriented if a family of mutually compatible coordinate systems
is given on X, whose domains cover X and any two of which have the same
orientation.
∂(x,y)
Example x = r cos θ, y = r sin θ. Then ∂(r,θ)
= r > 0. Therefore x, y and
r, θ have the same orientation.
140
Definition Let ω be a differential n-form with compact support on an ori-
ented n-dimensional Hausdorff manifold X. We define
Z
ω
X
W
V
PSfrag replacements X
y z
y(V ) z(W )
φ
ω2
ω1
ω = f (y 1 , . . . , y n) dy 1 ∧ · · · ∧ dy n = g(z 1 , . . . , z n ) dz 1 ∧ · · · ∧ dz n (say) on V ∩ W
= y ∗ ω1 = z ∗ ω2
141
(say). So ω1 = φ∗ ω2 where φ = z · y −1 : y(V ∩ W ) −→ z(V ∩ W ).
Therefore,
R R R R
y(V )
f (x 1 , . . . , x n ) dx 1 . . . dx n = y(V )
ω 1 = y(V ∩W )
ω 1 = ω
z(V ∩W ) 2
R R
= ω =
z(W ) 2 z(W )
g(x1 , . . . , xn ) dx1 . . . dxn
as required.
F1 + · · · + F k = 1
ω1 + · · · + ω k = ω
similarly Pk R
= i=1 X
Fi ω
x2 + y 2 + z = 2, z>0
We have:
2x dx + 2y dy + dz = 0
142
constrained to X. Therefore,
h 3
i√2
2 1 2
= 2π 3 8
(1 + 4r ) 2
0
π
= 6
[27 − 1]
13
= 3
π
143
144
Chapter 8
Complex Analysis
The RHS is called the Laurent series of f about a. The coefficient cn are
uniquely determined by:
Z
1 f (z)
cn = dz
2πi C (z − a)n+1
w
PSfrag replacements C2
a
C1 C
145
Then
1
R f (z)
f (w) = 2πi C2 z−w
dz
1
R f (z) 1
R f (z)
= 2πi C z−a
d− 2πi C1 z−w
dz
1
R f (z) 1
R f (z)
= 2πi C (z−a)−(w−a)
dz + 2πi C1 (w−a)−(z−a)
dz
1
R f (z) 1
R f (z)
= 2πi C (z−a)[1− w−a ]
dz + 2πi C1 (w−a)[1− z−a ]
dz
z−a w−a
1
R f (z) P∞
w−a n 1
R f (z) P∞
z−a n
= 2πi C (z−a) n=0 z−a
dz + 2πi C1 (w−a) n=0 w−a
dz
Z Z
P∞ 1 f (z) P −n−1 1 f (z)
= n=0 (w − a)
n
dz + ∞n=0 (w − a) dz
2πi C (z − a) n+1 2πi C! (z − a)−n
| {z } | {z }
+ve powers (w − a) -ve powers (w − a)
as required.
∞
X c−2 c−1
f (z) = cn (z − a)n = · · · + 2
+ +c0 + c1 (z − a) + · · ·
n=−∞
(z − a) z − a
| {z }
p(z)
P
is the Laurent series of f at a. p(z) = −1 n
n=−∞ cn (z−a) is called the principal
part of f at a. p(z) is holomorphic on C − {a}.
f (z) − p(z) is holomorphic in V (defining it’s value at 0 to be c0 ).
For any closed curve α in C − {a}:
R R h c−2 c−1
i
α
p(z) dz = α
···+ (z−a)2
+ z−a
dz
R dz
= · · · + 0 + c1 α z−a
1
R dz
where W (α, a) = 2πi α z−a
is the winding number of α about a. and
Res(f, a) = c−1 is the residue of f at a
146
8.2 Residue Theorem
Theorem 8.2.1 (Residue Theorem). Let f be holomorphic on V −{a1 , . . . , an }
where a1 , . . . , an are distinct points in a star-shaped (or contractible) open set
V . Then for any closed curve α in V − {a1 , . . . , an } we have:
Z n
X
f (z) dz = 2πi Res(f, aj ) W (α, aj )
α j=1
as required.
order -3:
c−2 c−2 c1
f (z) = 3
+ 2
+ + c0 + c1 (z − a) + · · · c−3 6= 0
(z − a) (z − a) (z − a)
147
n = −1: simple pole; n = −2: double pole.
If f has order n and g has order m:
f (z) = (z − a)n f1 (z) f1 (a) 6= 0
f (z) (z)
g(z)
= (z − a)n−m gf11 (z) f
g
has order n − m
The residue theorem is very useful for evaluating integrals:
R∞
Example to evaluate 0 xx2sin
+a2
x
, a > 0 we put
z eiz
f (z) = z 2 +a2
[eiz is easier to handle than sin z]
z eiz
= (z−ia)(z+ia)
ia
−R R
PSfrag replacements
−ia
Choose a closed contour that goes along x-axis −R to R (say) then loops
around a pole, upper semi-circle α (say).
Z R Z
x eix z eiz
2 2
dx + 2 2
dz = 2πi Res(f, ia)
−R x + a α z +a
148
c
1. if f (z) = z−ia + c0 + c1 (z − ia) + · · · on neighbourhood of ia then
(z − ia)f (z) = c + c0 (z − ia) + c1 (z − ia)2 + · · · on neighbourhood of
ia. Then limz−→ia (z − ia)f (z) = c = Res(f, ia). Therefore
f (z)
Res(f, ia) = c = limz−→ia z−ia
z eiz ia −a
= limz−→ia z+ia
= 2ia
e
1 −a
= 2
e
R2
Rπ
≤ R2 −a2 0
e−R sin t dt
which −→ 0 as R −→ ∞ since
Z π Z π
−R sin t DCT
lim e dt = lim e−R sin t dt = 0
R→∞ 0 0 R→∞
Therefore Z ∞
x(cos x + i sin x) e−a
dx + 0 = 2πi
−∞ x2 + a 2 2
so Z ∞
x sin x
dx = πe−a
−∞ x2 + a 2
R∞ sin x eiz
Example to calculate −∞ x
dx put f (z) = z
holomorphic except for
simple pole at z = 0.
α
PSfrag replacements
−R −r r R
149
Choose a closed contour along x-axis from −R to −r, loops around 0 by
β(t) = reit then r to R then back along α(t) = Reit .
Z −r Z Z R Z
eix eiz eix eiz
dx + dz + dx + dz = 0
−R x β z r x α z
2. R R
eiz π eiR(cos t+i sin t) iReit
α z
dz = 0 Reit
dt
Rπ
≤ 0
e−R sin t dt
which −→ 0 as R −→ ∞.
eiz 1
3. z
= z
+ g(z), g holomorphic in neighbourhood of 0. Therefore
Z Z Z
eiz dz
dz = + g(z) dz
β z β z β
Let sup |g(z)| = M (say on a closed ball centre 0 redius δ > 0 (say).
Z
g(z) dz ≤ M πr −→ 0 as r −→ 0
β
if r ≤ δ Z Z Z
π π
dz ireit
=− dt = − i dt = −iπ ∀r
β z 0 reit 0
therefore Z
eiz
lim dz = −iπ
r−→0 β z
therefore Z Z
0 ∞
eix eix
dx − iπ + dx + 0 = 0
−∞ x 0 x
so Z ∞
cos x + i sin x
dx = iπ
−∞ x
and Z ∞
sin x
dx = π
−∞ x
150
8.3 Uniqueness of analytic continuation
Theorem 8.3.1 (Uniqueness of analytic continuation). Let f, g be holo-
morphic on a connected open set V . Let a ∈ V and let {zk } be a sequence in
V (6= a) converging to a s.t.
f (zk ) = g(zk ) ∀k
Then f = g on V .
Proof. Put F = f − g, so F (zk ) = 0. To show F = 0 on V .
= (z − a)m Fm (z)
151
152
Chapter 9
≤ k k
=
153
therefore (1 − )B1 ⊂ T (c)φB ⊂ (1 + )B1 , where B1 is a translate of B to
new centre T (c)φ(x). Therefore
m(φB)
=⇒ (1 − )n ≤ | det T (c)| ≤ (1 + )n
m(B)
m(φB)
=⇒ lim | det T (c)| =1
m(B)
m(φB)
=⇒ | det T (a)| lim =1
m(B)
Therefore:
m(φB) 1
lim = = | det φ0 (a)|
m(B) | det T (a)|
as required
Proof. Let > 0. Choose δ > 0 s.t. |f (x) − f (a)| < ∀ kx − ak < δ. Then
each cube B containing a of radius ≤ 2δ we have:
R R
B f (x) dx B
[f (x) − f (a)] dx m(B)
m(B) − f (a) = ≤ =
m(B) m(B)
hence result.
{V × W : W open in Rs } (9.1)
154
are all open in Rr+s . Therefore the σ-algebra generated by Eqn(9.1):
{V × B : B Borel in Rs }
consists of Borel sets in Rr+s . Hence for fixed B Borel in Rs , the set:
{V × B : V open in Rr (9.2)
are all Borel sets in Rr+s . Therefore the σ-algebra generated by Eqn(9.2):
{A × B : A Borel in Rr }
B − E = B ∩ E0
E ⊂ Rn−1 × R
A1 × B − 1, A2 × B2 , . . .
155
S
2. Let m(E) = ∞. Put Ek = {x ∈ E : k ≤ |x| < k + 1}. E = ∞ k=0 Ek is
a countable disjoint union, and m(Ek ) < ∞. For each integer k choose
by 1. Borel Bk s.t. Ek ⊂ Bk and m(Bk ∩ Ek0 ) = 0
S
Put B = Bk . Then E ⊂ B and m(B ∩ E 0 ) = m(B) − m(E) as
required. ???????????
156
Therefore Eqn(9.4) holds for F ∈ monotone class generated by R.
Therefore Eqn(9.4) holds for F ∈ σ-algebra generated by R (monotone
class lemma). So Eqn(9.4) holds for any Borel set F ⊂ V
But R
λ(Bk ) m(φBk )− Bk | det φ0 (x)| dx
limh→∞ m(B k)
= limh→∞ m(Bk )
= 0
a contradiction.
R Therefore λ(B) = 0 for all cubes B. Therefore,
m(φB) = B | det φ0 (x)| dx for all cubes B as required.
7. Now suppose Z has measure zero, and choose a Borel set B s.t.
Z⊂B⊂V
157
Then Z
m(φZ) ≤ m(φB) = | det φ0 (x)| dx = 0
B
since B has measure zero. Therefore φZ has measure zero.
Therefore under a C 1 diffeomorphism we have:
158