1.

Convex subsets
First recall the following characterisation of relatively sequentially compact sets.
Theorem 1.1 (Hausdorff). Let (X , ρ) be a complete metric space. Then, M ⊂ X is
relatively sequentially compact if and only if, for every ϵ > 0, M admits a finite ϵ-net.
Exercise 1.5.2. Let X be a Banach space. Show that the convex hull of a relatively se-
quentially compact set is also relatively sequentially compact.
Proof. Let M be an arbitrary relatively sequentially compact subset of X. Denote by
Conv(M ) the convex hull of M . By Theorem 1.1, it suffices to show that for any ϵ > 0 the
convex hull Conv(M ) admits a finite ϵ-net. For an arbitrary ϵ > 0, since M is relatively
sequentially compact, M has a finite (ϵ/2)-net Nϵ/2 (M ) = {m1 , . . . , mr }.


Step 1. We show that Conv Nϵ/2 (M ) has a finite (ϵ/2)-net in this step. By Theorem 1.1,


it suffices to show that Conv Nϵ/2 (M ) is relatively sequentially compact. To this end, let
r
X r
X
xn = ani mi , with n ∈ N, ani = 1, and ani ≥ 0 for every 1 ≤ i ≤ r,
i=1 i=1


be an arbitrary sequence of points in Conv Nϵ/2 (M ) . As {an1 }n∈N is a sequence of
real numbers in [0, 1], it has a converging subsequence {ank 1 }k∈N . Replacing {xn }n∈N
by {xnk }k∈N , we can assume that the sequence {an1 }n∈N of real numbers converges to some
a1 ∈ [0, 1]. Applying the same argument inductively to every 1 ≤ i ≤ r, we can also assume
that
lim ani = ai ∈ [0, 1]
n∈N
for every 1 ≤ i ≤ r. Since ri=1 ani = 1 for every n ∈ N, we also have that
P
r
X
ai = 1,
i=1
which shows that
r
X

lim xn = ai mi ∈ Conv Nϵ/2 (M ) .
n→∞
i=1


Thus, Conv Nϵ/2 (M ) admits a finite (ϵ/2)-net
 


N := Nϵ/2 Conv Nϵ/2 (M ) .


Note that Conv Nϵ/2 (M ) is also a Banach space as it is a finite-dimensional normed space.

Step 2. Let x = tj=1 bj yj be an element in Conv(M ) (where tj=1 bj = 1 and bj ≥ 0

P P

for each 1 ≤ j ≤ t). As Nϵ/2 (M ) is an (ϵ/2)-net of M , every yj has a corresponding mk(j) ,

where k(j) is one of {1, . . . , r} depending on j, such that
ϵ
yj − mk(j) < .
2
Pt
Set m := j=1 bj mk(j) , then we have
t t t
X X ϵ X ϵ
∥x − m∥ = bj (yj − mk(j) ) ≤ bj yj − mk(j) < · bj = .
2 2
j=1 j=1 j=1
1
2


On the other hand, since m ∈ Conv Nϵ/2 (M ) , there is an n ∈ N satisfying
ϵ
∥m − n∥ <
2
by Step 1, hence we can conclude that
∥x − n∥ ≤ ∥x − m∥ + ∥m − n∥ < ϵ.
That is, N is a finite ϵ-net of Conv(M ). □

2. The fixed point theorems

Recall following result on existence of fixed points.
Theorem 2.1 (Schauder). Let X be a B ∗ -space and C ⊂ X a closed convex subset. Suppose
that T : C → C is a continuous map such that T (C) is relatively sequentially compact. Then,
T has a fixed point in C.
Lemma 2.2. Let f : X → Y be a continuous map of topological spaces. Let M ⊂ X be a
compact subset of X. Then, f (M ) is also a compact subset of Y .
Proof. Let {Vi }i∈I be an arbitrary open covering of f (M ). Then, {f −1 (Vi )}i∈I is an open
covering of M . As M is compact, M admits a finite open covering {f −1 (Vi )}i∈J , where J
is a finite subset of the index set I. Then, {Vi }i∈J is also a finite open covering of f (M ),
so f (M ) is also compact. □
Exercise 1.5.3. Let X be a B ∗ -space, and let C ⊂ X be a compact convex subset. Let
T : C → C be a continuous map. Show that T admits a fixed point in C.
Proof. Since C is compact, it is a closed subset of X (as C is then sequentially compact).
By Lemma 2.2, f (C) is also compact, so sequentially compact. Thus, by Theorem 2.1, T
has a fixed point in C. □
Exercise 1.5.5. Let A be an n × n matrix with aij > 0 for all 1 ≤ i, j ≤ n. Show that
there exist λ > 0 and x = (b1 , . . . , bn )T ∈ Rn , where bi ≥ 0 for every 1 ≤ i ≤ n and x ̸= 0,
such that
Ax = λx.
Proof. Consider the simplex inside Rn :
 n
X 
T n
C := x = (x1 , . . . , xn ) ∈ R xi = 1 with xi ≥ 0 for every i = 1, 2, . . . , n .
i=1
Set f : Rn → Rn by
Ax
f (x) := Pn .
j=1 (Ax)j
It is clear that C is a compact convex subset of Rn and that f is continuous. As aij > 0
for all 1 ≤ i, j ≤ n, for every x ∈ C, the image f (x) also belongs to C. Thus, f induces
P f : C → C. By Lemma 2.2 and Theorem 2.1, f admits a fixed point
a well-defined map
x ∈ C. Set λ = nj=1 (Ax)j . Since Ax ∈ C, λ > 0, and so Ax = λx as desired. □
Exercise 1.5.6. Let K(x, y) be an R>0 -valued continuous function on [0, 1] × [0, 1]. Set
Z 1
T : C[0, 1] → C[0, 1] by u(y) 7→ (T u)(x) := K(x, y)u(y)dy.
0
Show that there exist λ > 0 and nonzero u ≥ 0 in C[0, 1] such that
T u = λu.
 3

Proof. Denote by ∥·∥ the canonical norm via taking the maximums. Set
 Z 1 
C := u ∈ C[0, 1] u ≥ 0 and u=1
0
and
Tu
f : C → C, u 7→ R 1 .
0 (T u)(x)dx
Since K(x, y) is R>0 -valued, f is well-defined.

Step 1. It is clear that C is convex. Let {ui }i∈N be a Cauchy sequence in C. As

(C[0, 1], ∥·∥) is complete, ui → u for some u ∈ C[0, 1]. It is easy to see that u ∈ C. Thus,
C is a closed convex subset of C[0, 1].

Step 2. Fix u, v ∈ C. As K(x, y) is R>0 -valued and continuous on the compact set
[0, 1] × [0, 1], there exist A, B > 0 such that A < K(x, y) < B for any (x, y) ∈ [0, 1] × [0, 1].
Then, we have
R1 R1
0 K(x, y)u(y)dy K(x, y)v(y)dy
|f (u) − f (v)| =   −  0 
R1 R1 R1 R1
0 0 K(x, y)u(y)dy dx 0 0 K(x, y)v(y)dy dx
   
R1 R1 R1 R1
0 (T v)(x)dx · 0 K(x, y)u(y)dy − 0 (T u)(x)dx · 0 K(x, y)v(y)dy
=     .
R1 R1
0 (T u)(x)dx · 0 (T v)(v)dx

The numerator N can be rewritten as

Z 1  Z 1 Z 1  Z 1
(T v)(x)dx · K(x, y)u(y)dy − (T v)(x)dx · K(x, y)v(y)dy
0 0 0 0
Z 1  Z 1 Z 1  Z 1
+ (T v)(x)dx · K(x, y)v(y)dy − (T u)(x)dx · K(x, y)v(y)dy
0 0 0 0


Z 1 Z 1

= T (u − v) (x) · (T v)(x)dx + (T v)(x) · T (v − u) (x)dx.
0 0
Notice that


Z 1
T (u − v) (x) ≤ B · |u − v| (y)dy ≤ B · ∥u − v∥ ,
0
Z 1 Z 1Z 1 
(T v)(x)dx ≤ B · u(y)dy dx = B,
0 0 0
Z 1
(T v)(x) ≤ B · v(y)dy = B, and that
0
Z 1 Z 1Z 1 


T (v − u) (x)dx ≤ B · |u − v| (y)dy dx ≤ B · ∥u − v∥ .
0 0 0
Thus, we can make the estimate
Z 1 Z 1
T (v − u) (x)dx ≤ 2B 2 · ∥u − v∥ .



|N | ≤ T (u − v) (x) · (T v)(x)dx + (T v)(x) ·
0 0
4

On the other hand, the denominator satisfies that

Z 1 Z 1Z 1 
(T u)(x)dx ≥ A · u(y)dy dx = A.
0 0 0
Then, we can conclude the estimate that
|N | 2B 2
|f (u) − f (v)| =
≤ 2 · ∥u − v∥ ,
|M | A
which shows that f : C → C is continuous, indeed, equicontinuous.

Step 3. By the same estimate in Step 2, for any u ∈ C, we have

B
|f (u)| ≤ ,
A
hence the set C is uniformly bounded. On the other hand, since K(x, y) is a continuous
function on the compact set [0, 1] × [0, 1], K(x, y) is equicontinuous, that is, for every ϵ > 0,
there exists a δ > 0 (depending only on ϵ) such that |K(x, y) − K(x′ , y ′ )| < ϵ for all (x, y)
and (x′ , y ′ ) satisfying ∥(x, y) − (x′ , y ′ )∥ < δ. Thus, for all x1 , x2 satisfying |x1 − x2 | < δ, we
have
R1 R1
0 K(x1 , y)u(y)dy − 0 K(x2 , y)u(y)dy
|f (u)(x1 ) − f (u)(x2 )| = R1
0 (T u)(x)dx
R1
|K(x1 , y) − K(x2 , y)| u(y)dy
≤ 0
A
R1
ϵ · 0 u(y)dy ϵ
≤ = .
A A
Hence, f (C) is a uniformly equicontinuous set of functions in C[0, 1]. By Arzelà-Ascoli
Theorem, we see that f (C) is a relatively sequentially compact subset of C[0, 1]. By The-
orem 2.1, f admits a fixed point u ∈ C. As K(x, y) > 0 for any (x, y) ∈ [0, 1] × [0, 1] and
u ≥ 0 is nonzero,
Z 1Z 1 
λ := K(x, y)u(y)dy dx > 0.
0 0
Therefore, u satisfies the equation T u = λu. □

3. Inner products
Exercise. Let X be a vector space equipped with an inner product ⟨·, ·⟩X . Denote by ∥·∥X
the norm induced by the inner product. Denote by (X, ∥·∥X ) the completion of X with
respect to the norm ∥·∥X , and by ι : X → X the inclusion map, that is,
• ι is an isometry of normed spaces, and
• ι(X) is dense in X.
For every x, y ∈ X such that
x = lim ι(xn ) and y = lim ι(yn )
n→∞ n→∞
for sequences {xn }n∈N and {yn }n∈N in X, define
⟨x, y⟩X := lim ⟨xn , yn ⟩X .
n→∞
Show that
(1) ⟨·, ·⟩X is well-defined,
 5

(2) ⟨·, ·⟩X is an inner product, and

(3) the norm ∥·∥X on X is equal to the norm induced by ⟨·, ·⟩X .
Proof. Step 1. Let {xn }n∈N and {yn }n∈N be sequences in X such that
x = lim ι(xn ) and y = lim ι(yn ).
n→∞ n→∞
As ι is an isometry, for any ϵ > 0, there exists an N (ϵ) ∈ N such that
∥xn ∥X < ∥x∥X + 1, ∥yn ∥X < ∥y∥X + 1,
ϵ/2 ϵ/2
∥xn − xm ∥X < , and ∥yn − ym ∥X <
∥y∥X + 1 ∥x∥X + 1
for any n, m ≥ N (ϵ). By Cauchy-Schwarz inequality, we have
|⟨xn , yn ⟩X − ⟨xm , ym ⟩X | = |⟨xn , yn ⟩X − ⟨xn , ym ⟩X + ⟨xn , ym ⟩X − ⟨xm , ym ⟩X |
≤ |⟨xn , yn − ym ⟩X | + |⟨xn − xm , ym ⟩X |
≤ ∥xn ∥ · ∥yn − ym ∥ + ∥xn − xm ∥ · ∥ym ∥
ϵ ϵ
< + = ϵ.
2 2
Thus, {⟨xn , yn ⟩}n∈N is a Cauchy sequence in C, hence the limit limn→∞ ⟨xn , yn ⟩ exists.

Step 2. Let {x′n }n∈N and {yn′ }n∈N be two other sequences in X such that
x = lim ι(x′n ) and y = lim ι(yn′ ).
n→∞ n→∞
Then, Cauchy-Schwarz inequality shows that
⟨xn , yn ⟩X − ⟨x′n , yn′ ⟩X = ⟨xn , yn ⟩X − ⟨xn , yn′ ⟩X + ⟨xn , yn′ ⟩X − ⟨x′n , yn′ ⟩X
≤ ⟨xn , yn − yn′ ⟩X + ⟨xn − x′n , yn′ ⟩X
≤ ∥xn ∥X · yn − yn′ X
+ xn − x′n X
· yn′ .
Since
lim ∥xn ∥X = ∥x∥X , lim yn′ X
= ∥y∥X , lim yn − yn′ X
= 0, and lim xn − x′n = 0,
n→∞ n→∞ n→∞ n→∞
we see that
lim ⟨xn , yn ⟩X = lim ⟨x′n , yn′ ⟩X ,
n→∞ n→∞
which is equivalent to saying that ⟨·, ·⟩X is well-defined. That is, (1) holds.

Step 3. The conjugate-bilinear and conjugate-symmetric properties follow immediately

from the definition of ⟨·, ·⟩X . It suffices to show that ⟨·, ·⟩X is positive-definite. Let {xn }n∈N
be a sequence in X such that x = limn→∞ ι(xn ), then it is clear that ⟨x, x⟩X ≥ 0. Now,
assume ⟨x, x⟩X = 0. Then,
0 = ⟨x, x⟩X = lim ⟨xn , xn ⟩X = lim ∥xn ∥X = ∥x∥X .
n→∞ n→∞

As ∥·∥X defines a norm on X, we can conclude that x = 0. Thus, (2) also holds.

Step 4. Let {xn }n∈N be a sequence in X such that x = limn→∞ ι(xn ). Then,
⟨x, x⟩X = lim ⟨xn , xn ⟩X = lim ∥xn ∥X = ∥x∥X ,
n→∞ n→∞
which shows (3). □
6

Santai Qu
Institute of Geometry and Physics, University of Science and Technology of China,
Hefei, Anhui Province, China
Email: [email protected]