Metric Topology

Metric, Norms, and Inner Product Spaces

Definition.

A distance is a map d:X×X→[0,∞)d: X \times X \to [0, \infty) such that

  • d(x,y)=0d(x, y)=0 if and only if x=yx=y.
  • d(x,y)=d(y,x)d(x, y) = d(y, x).
  • d(x,y)≤d(x,z)+d(y,z)d(x, y) \leq d(x, z) + d(y, z).

For a set XX, if the distance map dd is defined on XX, we call (X,d)(X, d) a metric space. We also say that XX is metric.

Example.

Show that the followings are a metric on R\R.

  • d(x,y)=∣x−y∣1+∣x−y∣d(x, y) = \frac{|x-y|}{1+|x-y|}
  • d(x,y)=∣x−y∣pp≥1d(x, y) = |x - y|^p \quad p \geq 1

Definition.

Let (X,d)(X, d) be a metric space and A⊂XA \subset X. We define the distance between the set and the point as

dist⁡(x,A)=sup⁡y∈Ad(x,y).\dist(x, A) = \sup_{y \in A} d(x, y).

Definition.

Let XX be a vector space. A norm is a map ∥⋅∥:X→[0,∞)\| \cdot \| : X \to [0, \infty) satisfying the followings.

  • ∥x∥=0\left \| x \right \| = 0 if and only if x=0x=0.
  • ∥λx∥=∣λ∣∥x∥\left \| \lambda x \right \| = | \lambda | \left \| x \right \| for all x∈Xx \in X and scalar λ\lambda.
  • ∥x+y∥≤∥x∥+∥y∥\left \|x+y \right \| \leq \left \| x \right \| + \left \| y \right \| for all x,y∈Xx, y \in X.

We call (X,∥⋅∥)(X, \| \cdot \|) A normed space.

Proposition.

A normed space is a metric space.

From now, we define a metric d(x,y)=∥x−y∥d(x, y) = \|x-y\| from a norm. Indeed the converse of \cref{thm:2_norm2metric} is not true in general. Our question is that what additional condition is required so that a normed space has a metric.

Example.

Let (X,∥⋅∥)(X, \|\cdot\|) be a normed space and define d(x,y)=∥x−y∥d(x, y) = \|x - y\|.

  • Prove that dd is a metric.
  • Show that

    d(x+z,y+z)=d(x,y)d(x+z, y+z) = d(x, y)

    for all x,y∈Xx, y \in X.

  • Show that

    d(λx,λy)=∣λ∣d(x,y)d(\lambda x, \lambda y) = |\lambda| d(x, y)

    for all x,y∈Xx, y \in X.

Proposition.

A metric space (X,d)(X, d) is a normed space with its norm ∥x∥=d(x,0)\| x \| = d(x, 0) if and only if

  • d(x+z,y+z)=d(x,y)d(x+z, y+z) = d(x, y) for all x,y∈Xx, y \in X.
  • d(λx,λy)=∣λ∣d(x,y)d(\lambda x, \lambda y) = |\lambda| d(x, y) for all x,y∈Xx, y \in X.

Definition.

An inner product is a map <⋅,⋅>:X×X→R\left < \cdot, \cdot \right > : X \times X\to \R satisfying the followings.

  • <x,x>≥0\left <x, x\right > \geq 0 for all x∈Xx \in X
  • <x,x>=0\left <x, x\right > = 0 if and only if x=0x=0.
  • <λx,y>=λ<x,y>\left < \lambda x , y \right > = \lambda \left <x, y\right > for all x,y∈Xx, y \in X and scalar λ\lambda.
  • <x,y+z>=<x,y>+<x,z>\left <x, y+z\right > = \left <x, y\right > + \left <x, z\right > for all x,y,z∈Xx, y, z \in X
  • <x,y>=<y,x>\left <x, y\right > = \left <y, x\right > for all x,y∈Xx, y \in X

We call (X,<⋅,⋅>)(X, \left <\cdot, \cdot\right >) An inner product space.

Theorem.

{Cauchy-Schwarz Inequality} For an inner product space (X,<⋅,⋅>)(X, \left <\cdot, \cdot\right >),

∣<x,y>∣≤(<x,x>)12(<y,y>)12\left | \left <x, y\right > \right | \leq \left (\left <x,x\right > \right )^{\frac{1}{2}} \left (\left <y,y\right > \right )^{\frac{1}{2}}

Proof. Assume that x,y≠0x, y \neq 0. For all α∈R\alpha \in \R, we have

0≤<αx+y,αx+y>=α2<x,x>+2α<x,y>+<y,y>0 \leq \left <\alpha x + y, \alpha x + y\right > = \alpha^2 \left <x, x\right > + 2 \alpha \left <x, y\right > + \left <y, y\right >

Therefore we derive that (<x,y>)2−<x,x><y,y>≤0\left ( \left <x, y\right > \right )^2 - \left <x,x\right >\left <y,y\right > \leq 0.

Proposition.

An inner product space is a normed space.

From now we define norm from inner product as ∥v∥=<v,v>\|v\| = \sqrt{\left<v, v\right>}. The thing is the triangle inequality could be derived from Cauchy-Schwarz inequality so that a norm is implied by an inner product.

Example.

Let (X,<⋅,⋅>)(X, \left<\cdot, \cdot\right>) be an inner product space and define ∥x∥=<x,x>\| x \| = \sqrt{\left<x, x\right>}.

  • Prove that ∥⋅∥\|\cdot\| is a norm.
  • Show that

    2∥x∥2+2∥y∥2=∥x+y∥2+∥x−y∥22\|x\|^2 + 2\|y\|^2 = \|x+y\|^2 + \|x-y\|^2

    for all x,y∈Xx, y \in X.

  • Show that

    <x,y>=14∥x+y∥2−14∥x−y∥2\left<x, y\right> = \frac{1}{4}\| x+y \|^2 - \frac{1}{4}\| x-y \|^2

    for all x,y∈Xx, y \in X.

Proposition.

A normed space (X,∥⋅∥)(X, \|\cdot\|) is an inner product space with its inner product

<x,y>=14∥x+y∥2−14∥x−y∥2\left<x, y\right> = \frac{1}{4}\| x+y \|^2 - \frac{1}{4}\| x-y \|^2

if and only if

2∥x∥2+2∥y∥2=∥x+y∥2+∥x−y∥22\|x\|^2 + 2\|y\|^2 = \|x+y\|^2 + \|x-y\|^2

for all x,y∈Xx, y \in X. (Parallelogram Law)

Example.

Define

d(x,y)=∣x−y∣,d(x, y) = |x-y|,

then (R,d)(\R, d) is a metric space. Without mentioning, we use this absolute difference as metric on R\R.

Example (Euclidean inner product).

  • Define

    <x,y>2=∑i=1dxiyi,\left< x, y \right>_2 = \sum_{i=1}^d x_i y_i,

    then (Rd,<⋅,⋅>2)(\R^d, \left<\cdot, \cdot\right>_2) is an inner product space.

  • Define an induced norm

    ∥x∥2=(∑i=1d∣xi∣2)12=<x,x>2,\| x\|_2 = \left(\sum_{i=1}^d |x_i|^2\right)^{\frac{1}{2}} = \sqrt{\left<x, x\right>_2},

    then (Rd,∥⋅∥2)(\R^d, \|\cdot\|_2) is a normed space.

  • Define an induced metric

    d2(x,y)=(∑i=1d∣xi−yi∣2)12=∥x−y∥2,d_2(x, y) = \left( \sum_{i=1}^d |x_i - y_i|^2 \right)^\frac{1}{2} = \|x - y\|_2,

    then (Rd,d2)(\R^d, d_2) is a metric space.

Example (operator norm).

Define

∥A∥=sup⁡{∣Ax∣∣∣x∣≤1}\|A\| = \sup \left\{ |Ax| \mid |x| \leq 1 \right\}

for a matrix AA, then (A,∥⋅∥)(A, \|\cdot\|) is a normed space.

Example (lpl^p norm).

  • Define

    ∥x∥p={(∑i=1d∣xi∣p)1p(1≤p<∞)max⁡1≤i≤d∣xi∣(p=∞),\|x\|_p = \begin{cases} \left( \sum_{i=1}^d |x_i|^p \right)^\frac{1}{p} \quad & (1 \leq p < \infty) \\ \max_{1 \leq i \leq d} |x_i| \quad & (p = \infty) \end{cases} ,

    then (Rd,∥⋅∥p)(\R^d, \|\cdot\|_p) is a normed space.

  • Define

    dp(x,y)=∥x−y∥p(1≤p≤∞)d_p(x, y) = \|x - y\|_p \quad (1 \leq p \leq \infty)

    then (Rd,dp)(\R^d, d_p) is a metric space.

Example (LpL^p norm*).

Let

Lp(A)={{f:A→R ∣ (∫A∣f(x)∣pdx)1p<∞}(1≤p<∞){f:A→R∣ess⁡sup⁡x∈A∣f(x)∣<∞}(p=∞).L^p(A) = \begin{cases} \left\{f : A \to \R \,\bigg|\, \left(\int_A |f(x)|^p dx \right)^{\frac{1}{p}} < \infty \right\} \quad & (1 \leq p < \infty)\\ \{ f: A \to \R \mid {\ess\sup}_{x \in A} |f(x)| < \infty \} \quad & (p = \infty). \end{cases}

Define

∥f∥Lp={(∫A∣f(x)∣pdx)1p(1≤p<∞)ess⁡sup⁡x∈A∣f(x)∣(p=∞),\|f\|_{L^p} = \begin{cases} \left(\int_A |f(x)|^p dx \right)^{\frac{1}{p}} \quad & (1 \leq p < \infty) \\ {\ess\sup}_{x \in A} |f(x)| \quad & (p = \infty) \end{cases} ,

then (Lp(A),∥⋅∥Lp)(L^p(A), \|\cdot\|_{L^p}) is a normed space.

Topology from a Metric

Definition.

Let (X,d)(X, d) be a metric space and A⊂XA \subset X.

  • We denote

    B(x,r)={y∈X ∣ d(x,y)<r}B(x, r) = \left \{ y \in X \,|\, d(x, y) < r \right \}

    which is called an open ball of xx with radius rr.

  • AA is open if for every x∈Ax \in A, there exists r>0r > 0 such that B(x,r)⊂AB(x, r) \subset A.
  • AA is closed if X∖AX \setminus A is open, that is, for every x∈X∖Ax \in X \setminus A, there exists r>0r > 0 such that B(x,r)⊂X∖AB(x, r) \subset X \setminus A.
  • x∈Ax \in A is an interior point of AA if there exists r>0r > 0 such that B(x,r)⊂AB(x, r) \subset A.
  • An interior of AA, denoted by A∘A^\circ is a set of all interior points of AA.
  • B⊂AB \subset A is open relative to AA if there exists an open set U⊂XU \subset X such that B=U∩AB = U \cap A.
  • B⊂AB \subset A is closed relative to AA if there exists a closed set V⊂XV \subset X such that B=V∩AB = V \cap A.

The metric gives a nice topological properties to itself. A first step in constructing a topology is nothing but to define an open set. The open ball could be thought of as a circle in R2\R^2, a sphere in R3\R^3, or an open interval in R\R.

It is helpful to sketch an open ball and observe how it varies depending on the space and the metric.

Example.

Draw a unit open ball

B(0,1)={x∈R2∣dp(x,0)<1}B(0, 1) = \{ x \in \R^2 \mid d_p(x, 0) < 1 \}

in R2\R^2 for p=1,2,3,∞p = 1, 2, 3, \infty where

dp(x,y)=(∣x∣p+∣y∣p)1p.d_p(x, y) = \left(|x|^p + |y|^p\right)^\frac{1}{p}.

Proposition.

Every open ball B(x,r)B(x, r) is open.

Proof. For all y∈B(x,r)y \in B(x, r), take r1=r−d(x,y)>0r_1 = r - d(x, y) >0, then

B(y,r1)⊂B(x,r)B(y, r_1) \subset B(x, r)

since for all z∈B(y,r1)z \in B(y, r_1),

d(y,z)<r−d(x,y)  ⟹  d(x,z)≤d(x,y)+d(y,z)<r.d(y, z) < r - d(x, y) \implies d(x, z) \leq d(x,y)+d(y,z) < r.

Theorem.

Let U={Uα}α∈IU = \{U_\alpha\}_{\alpha \in I} be a collection of open sets, where II is an uncountable index set. Then the following sets are open.

  • ⋃α∈IUα\bigcup_{\alpha \in I} U_\alpha
  • ⋂i=1nUi\bigcap_{i=1}^n U_i

Proof. - For every x∈⋃α∈IUαx \in \bigcup_{\alpha \in I} U_\alpha, there exists Uβ∈UU_\beta \in U such that x∈Uβx \in U_\beta.
Hence there is r>0r>0 such that

$$ B(x, r) \subset U_\beta \subset \bigcup_{\alpha \in I} U_\alpha. $$ - For every $$ x \in \bigcap_{i=1}^n U_i $$, note that for $$ 1 \leq i \leq n $$    there exists $$ r_i>0 $$ such that $$ B(x, r_i) \subset U_i. $$    Take $$ r = \min \left \{ r_i \mid 1 \leq i \leq n\right \} $$, then 


$$ x \in B(x, r) \subset \bigcap_{i=1}^n U_i. $$

Theorem.

Let V={Vα}α∈IV = \{ V_\alpha \}_{\alpha \in I} be a collection of closed sets, where II is an uncountable index set. Then the following sets are closed.

  • ⋂α∈IVα\bigcap_{\alpha \in I} V_\alpha
  • ⋃i=1nVi\bigcup_{i=1}^n V_i

Proof. Trivial by De Morgan’s Law.

Example.

Let (X,d)(X, d) be a metric space and A,B⊂XA, B \subset X.

  • Show that A∘=AA^\circ = A if and only if AA is open.
  • Show that A∘=(A∘)∘A^\circ = (A^\circ)^\circ.
  • Show that if A⊂BA \subset B then A∘⊂B∘A^\circ \subset B^\circ
  • Give an example such that A⊄BA \not\subset B even though A∘⊂B∘A^\circ \subset B^\circ.
  • Show that A∘∩B∘=(A∩B)∘A^\circ \cap B^\circ = (A \cap B)^\circ.
  • Show that A∘∪B∘⊂(A∪B)∘A^\circ \cup B^\circ \subset (A \cup B)^\circ.
  • Give an example such that A∘∪B∘≠(A∪B)∘A^\circ \cup B^\circ \neq (A \cup B)^\circ

% \subsection{Limit Points}

Definition.

Let (X,d)(X, d) be a metric space and A⊂XA \subset X.

  • xx is a limit point of AA whenever any open ball of xx has a point in AA other than xx, that is,

    B(x,r)∩(A∖{x})≠∅B(x, r) \cap (A \setminus \{x\}) \neq \emptyset

    for all r>0r>0.

  • We denote A′A' a set of all limit points of AA.
  • xx is an isolated point of AA if it is not a limit point of AA.
  • A⊂XA \subset X is closed if AA contains all of its limit point, that is, A′⊂A.A' \subset A.
  • AA is perfect if all of the point in AA is its limit point, that is, A=A′.A = A'.

Proposition.

Let (X,d)(X, d) be a metric space and A⊂XA \subset X. Then X∖AX \setminus A is open if and only if A′⊂AA' \subset A.

Proof. \begin{steplist} \item (  ⟹  \implies) Suppose not and let x∈X∖Ax \in X\setminus A be a limit point of AA. Since X∖AX \setminus A is open, there exists r>0r>0 such that

B(x,r)⊂X∖A  ⟹  B(x,r)∩(A∖{x})=∅,B(x, r) \subset X \setminus A \implies B(x, r) \cap (A \setminus \{x\}) = \emptyset,

which implies that xx is an isolated point of AA. (contradiction) \item (  ⟸  \impliedby) Suppose not, then there exists x∈X∖Ax \in X\setminus A such that

B(x,r)⊄X∖A  ⟹  B(x,r)∩(A∖{x})≠∅B(x, r) \not\subset X \setminus A \implies B(x, r) \cap (A \setminus \{x\}) \neq \emptyset

for all r>0r > 0, which implies that xx is a limit point of AA. (contradiction) \end{steplist}

Proposition.

XX and ∅\emptyset are both open and closed whenever XX is a metric space.

Our next question is that whether there exists a metric space so that all of its subsets are both open and closed. The following is one example of such a space.

Example.

Let (X,d)(X, d) be a metric space with discrete metric

d(x,y)={1(x≠y)0(x=y).d(x, y) = \begin{cases} 1 \quad& (x \neq y)\\ 0 & (x = y) \end{cases}.

Prove the followings.

  • A⊂XA \subset X is open.
  • A⊂XA \subset X is closed.

Example.

Let (X,d)(X, d) be a metric space and A,B⊂XA, B \subset X.

  • Show that if A⊂BA \subset B then A′⊂B′A' \subset B'.
  • Give an example such that A′⊂B′A' \subset B' but A⊄BA \not\subset B.

% \subsection{Closure and Boundary}

Definition.

Let (X,d)(X, d) be a metric space and A⊂XA \subset X.

  • The closure of AA denoted by A‾\overline{A} is defined as A‾=A∪A′.\overline{A} = A \cup A'.
  • The boundary of AA, denoted by ∂A\partial A is defined as ∂A=A‾∩X∖A‾.\partial A = \overline{A} \cap \overline{X \setminus A}.
  • AA is dense in XX if A‾=X\overline{A} = X.
  • AA is nowhere dense in XX if (A‾)∘=∅\left(\overline{A} \right)^\circ = \emptyset.
  • A collection {Uα}α∈I\{ U_\alpha \}_{\alpha \in I} of open subsets of XX is A base for XX if for all x∈Xx \in X and every open set VV containing xx, there exists UαU_\alpha such that x∈Uα⊂V.x \in U_\alpha \subset V.
  • XX is separable if there exists a countable dense subset of XX.

Proposition.

A‾\overline{A} is the smallest closed set containing A.A.

Proof. Consider a closed set VV such that A⊂V⊂A‾A \subset V \subset \overline{A}. Since A′⊂VA' \subset V, V=A‾V = \overline{A}.

Corollary.

A=A‾A = \overline{A} if and only if AA is closed.

Theorem.

Let (X,d)(X, d) be a metric space and A⊂XA \subset X. x∈∂Ax \in \partial A if and only if

B(x,r)∩A≠∅andB(x,r)∩(X∖A)≠∅B(x, r) \cap A \neq \emptyset \quad \text{and} \quad B(x, r) \cap (X \setminus A) \neq \emptyset

for all r>0r > 0.

Proof. \begin{steplist} \item (  ⟹  \implies) Without loss of generality, it suffices to assume that x∈Ax \in A. We observe

x∈A  ⟹  B(x,r)∩A≠∅x \in A \implies B(x, r) \cap A \neq \emptyset

and

x∈X∖A‾  ⟹  x∈(X∖A)′  ⟹  B(x,r)∩(X∖A)≠∅x \in \overline{X \setminus A} \implies x \in (X \setminus A)' \implies B(x, r) \cap (X \setminus A) \neq \emptyset.

\item (  ⟸  \impliedby) For all r>0r > 0,

B(x,r)∩A≠∅  ⟹  x∈A‾B(x, r) \cap A \neq \emptyset \implies x \in \overline{A}

and

B(x,r)∩(X∖A)≠∅  ⟹  x∈X∖A‾B(x, r) \cap (X \setminus A) \neq \emptyset \implies x \in \overline{X \setminus A}

therefore x∈∂Ax \in \partial A. \end{steplist}

Example.

Let (X,d)(X, d) be a metric space. Prove the followings.

  • (A∘)c=Ac‾(A^\circ)^c = \overline{A^c}
  • (Ac)∘=(A‾)c(A^c)^\circ=\left(\overline{A}\right)^c
  • ∂A=A‾∖A∘\partial A = \overline{A} \setminus A^\circ

As a result of previous example, we observe that A∘A^\circ, ∂A\partial A, and (Ac)∘(A^c)^\circ are a partition of XX, that is, these three sets are disjoint and A∘∪∂A∪(Ac)∘=XA^\circ \cup \partial A \cup (A^c)^\circ = X.

Example.

Let (X,d)(X, d) be a metric space. Prove the followings.

  • ∂A⊂Ac\partial A \subset A^c if and only if AA is open.
  • ∂A⊂A\partial A \subset A if and only if AA is closed.
  • ∂A=∅\partial A = \emptyset if and only if AA is both open and closed.

% \subsection{Bounded Sets}

Definition.

Let (X,d)(X, d) be a metric space.

  • The diameter of AA denoted by diam⁡A\diam A, is defined as

    diam⁡A={sup⁡{d(x,y)∣x,y∈A}(A≠∅)0(A=∅).\diam A = \begin{cases} \sup \left\{ d(x, y) \mid x, y \in A \right\} \quad& (A \neq \emptyset)\\ 0 & (A = \emptyset) \end{cases}.
  • AA is bounded if and only if diam⁡A<∞\diam A < \infty.

Example.

Let (X,d)(X, d) be a metric space and A,B⊂XA, B \subset X.

  • Show that if AA is bounded then so be A‾\overline{A}.
  • Give an example such that AA is not bounded even though A∘A^\circ is.

Sequences and Convergence

% \subsection{Convergence}

Definition.

Let (X,d)(X, d) be a metric space. For a sequence (xn)(x_n) in XX, we say xnx_n converges to x∈Xx \in X, denoted by

lim⁡n→∞xn=x\lim_{n \to \infty}x_n = x

or

xn→xx_n \to x

if for every ε>0\varepsilon >0, there exists n0n_0 such that

n≥n0  ⟹  d(xn,x)<ε.n \geq n_0 \implies d(x_n , x) < \varepsilon.

Theorem.

Let (X,d)(X, d) be a metric space and A⊂XA \subset X. AA is closed if and only if for every sequence (xn)(x_n) in AA that converges in XX, lim⁡n→∞xn∈A\lim_{n \to \infty}x_n \in A.

Proof. (  ⟹  \implies) Suppose lim⁡n→∞xn=x∉A\lim_{n \to \infty} x_n = x \notin A, then for all ε>0\varepsilon >0, there exists n0n_0 such that

n≥n0  ⟹  d(xn,x)<ε  ⟹  B(x,ε)∩(A∖{x})≠∅n \geq n_0 \implies d(x_n , x) < \varepsilon \implies B(x, \varepsilon) \cap (A \setminus \{x\}) \neq \emptyset

therefore x∈A′x \in A'. (contradiction)

(  ⟸  \impliedby) It suffices to show that A′⊂AA' \subset A. Suppose x∈A′x \in A', then for all r>0r > 0,

B(x,r)∩(A∖{x})≠∅B(x, r) \cap (A\setminus\{x\}) \neq \emptyset

therefore we could construct a sequence xn∈Xx_n \in X such that for a given ε>0\varepsilon > 0, there exists n0=min⁡{n∈N ∣ 1n<ε}n_0 = \min\left\{n \in \n \,\bigg|\, \frac{1}{n} < \varepsilon \right\} so that

n≥n0  ⟹  xn∈B(x,1n)∩(A∖{x})n \geq n_0 \implies x_n \in B \left(x, \frac{1}{n}\right) \cap (A\setminus\{x\})

therefore lim⁡n→∞xn=x∈A\lim_{n \to \infty} x_n = x \in A.

Theorem.

For a sequence (xn)(x_n) in a metric space (X,d)(X, d), if

lim⁡n→∞xn=xandlim⁡n→∞xn=y\lim_{n \to \infty} x_n = x \quad \text{and} \quad \lim_{n \to \infty} x_n = y

then x=y.x=y.

Proof. Let ε>0\varepsilon > 0 be given. Then there exists n0n_0 such that

n≥n0  ⟹  d(xn,x)<ε2andd(xn,y)<ε2,n \geq n_0 \implies d(x_n, x) < \frac{\varepsilon}{2} \quad \text{and} \quad d(x_n, y) < \frac{\varepsilon}{2},

which implies

d(x,y)≤d(x,xn)+d(xn,y)<εd(x, y) \leq d(x, x_n) + d(x_n, y) < \varepsilon

hence x=yx = y.

Theorem.

Let (X,d)(X, d) be a metric space and A⊂XA \subset X. x∈A‾x \in \overline{A} if and only if there exists a sequence (xn)(x_n) in AA that converges to xx.

Proof. It suffices to consider the case of x∈A‾∖Ax \in \overline{A} \setminus A.

(  ⟹  \implies) Suppose x∈A‾∖A⊂A′x \in \overline{A} \setminus A \subset A', then for all r>0r > 0, B(x,r)∩(A∖{x})≠∅B(x, r) \cap (A\setminus\{x\}) \neq \emptyset therefore we could construct a sequence xn∈Xx_n \in X so that for all ε>0\varepsilon > 0, there exists n0n_0 such that

n≥n0  ⟹  xn∈B(x,ε)∩(A∖{x}).n \geq n_0 \implies x_n \in B(x, \varepsilon) \cap (A\setminus\{x\}).

(  ⟸  \impliedby) Suppose that lim⁡n→∞xn=x\lim_{n \to \infty} x_n = x, then for all ε>0\varepsilon > 0, there exists n0n_0 such that

n≥n0  ⟹  xn∈B(x,ε)n \geq n_0 \implies x_n \in B(x, \varepsilon)

which implies that

B(x,r)∩A≠∅B(x, r) \cap A \neq \emptyset

for all r>0r > 0.

Definition.

Let (X,d)(X, d) be a metric space and (xn)(x_n) be a sequence in XX.

  • xnx_n is bounded if {xn∣n∈N}\{ x_n \mid n \in \n\} is bounded.
  • For a strictly increasing sequence (nk)(n_k) in N\n, we say (xnk)(x_{n_k}) A subsequence of xnx_n.

Theorem.

If xx is a limit point of {xn∣n∈N}\{x_n \mid n \in \n\}, then there exists a subsequence of xnx_n which converges to xx.

Proof. Since x∈{xn∣n∈N}′x \in \{x_n \mid n \in \n\}', for all r>0r > 0,

B(x,r)∩({xn∣n∈N}∖{x})≠∅B(x, r) \cap (\{x_n \mid n \in \n\} \setminus \{x\}) \neq \emptyset

hence we could construct a subsequence xnkx_{n_k} such that

for a given ε>0\varepsilon >0, take k0=min⁡{k∈N ∣ 1k<ε}k_0 = \min\left\{k \in \n \,\bigg|\, \frac{1}{k} < \varepsilon \right\} so that

k≥k0  ⟹  xnk∈B(x,1k)∩({xn∣n∈N}∖{x})k \geq k_0 \implies x_{n_k} \in B \left(x, \frac{1}{k}\right) \cap (\{x_n \mid n \in \n\} \setminus \{x\})

therefore lim⁡k→∞xnk=x\lim_{k \to \infty} x_{n_k} = x.

% \subsection{Cauchy Sequences}

Completeness

Definition.

Let (X,d)(X, d) be a metric space and (xn)(x_n) be a sequence in XX.

  • xn∈Xx_n \in X is a Cauchy sequence if for all ε>0\varepsilon>0, there exists n0n_0 such that

    m,n≥n0  ⟹  d(xm,xn)<ε.m, n \geq n_0 \implies d(x_m, x_n) < \varepsilon.
  • XX is complete if every Cauchy sequence in XX converges to a point in XX.

Example.

Show that (0,1](0, 1] is not complete.

Proposition.

Let (X,d)(X, d) be a metric space and A⊂XA \subset X. If AA is complete, then AA is closed.

The converse of \cref{prop:2_complete_closed} is indeed not true, as the following example.

Example.

Show that Q\Q is complete but not closed.

Theorem.

Every convergent sequence in a metric space is a Cauchy sequence.

Proof. Suppose (xn)(x_n) in a metric space XX be a sequence such that lim⁡n→∞xn=x∈X\lim_{n \to \infty} x_n = x \in X, then for a given ε>0\varepsilon > 0, there exists n0n_0 such that

n≥n0  ⟹  d(x,xn)<ε2.n \geq n_0 \implies d(x, x_n) < \frac{\varepsilon}{2}.

By the triangle equality,

m,n≥n0  ⟹  d(xn,xm)≤d(xn,x)+d(xm,x)<εm, n \geq n_0 \implies d(x_n, x_m) \leq d(x_n, x) + d(x_m, x) < \varepsilon

therefore (xn)(x_n) is Cauchy.

Theorem.

Every Cauchy sequence in a metric space is bounded.

Proof. Suppose (xn)(x_n) be a Cauchy sequence in a metric space XX. Then there exists n0n_0 such that

m,n≥n0  ⟹  d(xn,xm)<1.m, n \geq n_0 \implies d(x_n, x_m) <1.

Take M=sup⁡{d(xm,xn)∣m,n≤n0}+1M = \sup \{ d(x_m, x_n) \mid m, n \leq n_0 \} + 1 then diam⁡{xn∣x∈N}≤M\diam\{ x_n \mid x \in \n \} \leq M.

Compactness

% \subsection{Compact Sets}

Definition.

Let (X,d)(X, d) be a metric space.

  • A⊂XA \subset X is sequentially compact if every sequence in AA has a subsequence converges to a point in AA.
  • A⊂XA \subset X is limit point compact if every infinite subset B⊂AB \subset A has a limit point in AA.
  • A cover of A⊂XA \subset X is a collection of sets {Uα}α∈I\left \{ U_\alpha \right\}_{\alpha \in I} satisfying

    A⊂⋃α∈IUα.A \subset \bigcup_{\alpha \in I}U_\alpha.
  • A cover is an open cover if each element of it is open.
  • A⊂XA \subset X is compact if every open cover of AA has a finite subcover.

Example.

Show that (0,1)⊂R(0, 1) \subset \R is not compact.

Proposition.

Let (X,d)(X, d) be a metric space. A⊂XA \subset X is sequentially compact if and only if it is limit point compact.

Proposition.

Let (X,d)(X, d) be a metric space and K⊂XK \subset X be a compact set. Then KK is closed.

Example.

Prove \cref{prop:2_seq_lim_cpt} and \cref{prop:2_cpt_cl}.

Proposition.

Let (X,d)(X, d) be a metric space and K⊂XK \subset X be a compact set. A⊂KA \subset K is compact if it is closed.

Proof. Let {Uα}α∈I\{U_\alpha\}_{\alpha \in I} be an open cover of AA. Then we note that {Uα}α∈I∪(K∖A)\{U_\alpha\}_{\alpha \in I} \cup (K \setminus A) is indeed an open cover of KK and there is an finite subcover since KK is compact so we are done.

Example.

Indeed, if KK is compact and AA is closed then K∩AK \cap A is compact.

Definition.

Let (X,d)(X, d) be a metric space, A⊂XA \subset X and {Uα}α∈I\{ U_\alpha\}_{\alpha \in I} be an open cover of AA. r>0r > 0 is a Lebesgue number of {Uα}α∈I\{ U_\alpha\}_{\alpha \in I} if for all B⊂AB \subset A with diam⁡B<r,\diam B < r, there exists β∈I\beta \in I so that

B⊂Uβ.B \subset U_\beta.

Theorem.

Let (X,d)(X, d) be a metric space and A⊂XA \subset X. If AA is sequentially compact, then there exists the Lebesgue number of each open cover of AA.

Proof. Let {Uα}α∈I\{ U_\alpha\}_{\alpha \in I} be an open cover of AA and r>0r > 0 be its Lebesgue number. \begin{steplist} \item Suppose not, then there exists a sequence (xn)⊂A(x_n) \subset A such that

B(xn,1n)⊄UαB\left(x_n, \frac{1}{n}\right) \not\subset U_\alpha

for all α∈I\alpha \in I.

\item Since AA is sequentially compact, there exists a subsequence xnk∈Ax_{n_k} \in A of xnx_n such that

lim⁡k→∞xnk=x∈A.\lim_{k \to \infty} x_{n_k} = x \in A.

Let {Uα}α∈I\left \{ U_\alpha \right\}_{\alpha \in I} be an open cover of AA then we have for r1>0r_1 > 0,

x∈B(x,r1)⊂Uβx \in B(x, r_1) \subset U_\beta

for some β∈I\beta \in I.

Further we have k0k_0 such that

k≥k0  ⟹  xnk∈B(x,r12)and1nk<r12.k \geq k_0 \implies x_{n_k} \in B\left(x, \frac{r_1}{2}\right) \quad \text{and} \quad \frac{1}{n_k} < \frac{r_1}{2}.

\item Consequently we obtain

B(xnk,1nk)⊂B(xnk,r12)⊂B(x,r1)⊂Uβ.B\left(x_{n_k}, \frac{1}{n_k}\right) \subset B\left(x_{n_k}, \frac{r_1}{2}\right) \subset B(x, r_1) \subset U_\beta.

(contradiction) \end{steplist}

Definition.

Let (X,d)(X, d) be a metric space. A⊂XA\subset X is totally bounded if for all r>0r>0, there exists finite set {x1,x2,⋯ ,xn}⊂X\left\{x_1, x_2, \cdots, x_n \right\} \subset X such that

A⊂⋃i=1nB(xi,r).A \subset \bigcup_{i=1}^n B(x_i, r).

Proposition.

Let (X,d)(X, d) be a metric space. If A⊂XA\subset X is totally bounded then it is bounded.

Example.

On the metric space (R,d)(\R, d) with the discrete metric

d(x,y)={1(x≠y)0(x=y),d(x, y) = \begin{cases} 1 \quad& (x \neq y)\\ 0 & (x = y) \end{cases},

Show that Z\Z is bounded but not totally bounded.

Theorem.

Let (X,d)(X, d) be a metric space. If A⊂XA \subset X is sequentially compact, then it is totally bounded.

Proof. Let AA be sequentially compact and suppose not. \begin{steplist} \item Let r>0r > 0 be given. Then we can construct a sequence (xn)⊂A(x_n) \subset A by taking some x1∈Ax_1 \in A and choosing

xn+1∈A∖⋃k=1nB(xk,r)x_{n+1} \in A \setminus \bigcup_{k=1}^n {B(x_{k}, r)}

for all n≥1n \geq 1. We note that A∖⋃k=1nB(xk,r)A \setminus \bigcup_{k=1}^n {B(x_{k}, r)} is nonempty since AA is not totally bounded.

\item Since AA is sequentially compact, there exists a convergent subsequence (xnk)(x_{n_k}) of (xn)(x_n). Hence, there exists n0n_0 so that

n≥n0  ⟹  d(xn,x)<r2  ⟹  xn∈B(x,r2)n \geq n_0 \implies d(x_n, x) < \frac{r}{2} \implies x_n \in B\left(x, \frac{r}{2}\right)

but it is indeed unable by our construction. (contradiction) \end{steplist}

Theorem.

Let (X,d)(X, d) be a metric space. A⊂XA \subset X is sequentially compact if and only if it is complete and totally bounded.

Proof. This proof is left as an exercise.

% \subsection{Compactness in Metric Spaces}

The following theorem is one of the most important theorem in this course.

Theorem.

Let (X,d)(X, d) be a metric space. A⊂XA \subset X is compact if and only if it is sequentially compact.

To prove the theorem, we first introduce the following lemma.

Lemma.

Let (xn)⊂A(x_n) \subset A be a sequence such that every subsequence of it diverges. For all nn, there exists rn>0r_n > 0 such that

B(xn,rn)∩{xm∣xm≠xn}=∅.B(x_n, r_n) \cap \{x_m \mid x_m \neq x_n \} = \emptyset.

Proof. If not, then there exists a subsequence (xnk)⊂A(x_{n_k}) \subset A such that

d(xnk,xk0)<1kd(x_{n_k}, x_{k_0}) < \frac{1}{k}

for all kk and some k0k_0, implying xnk→xk0x_{n_k} \to x_{k_0}. (contradiction)

Proof. [Proof of the theorem] (  ⟹  )(\implies) \begin{steplist} \item Suppose not. Then there exists a sequence (xn)⊂A(x_n) \subset A such that every subsequence of it does not converge. Without the loss of generality, we assume xi≠xjx_i \neq x_j if i≠ji \neq j.

\item \cref{lem:2_BW1} yields that there is no limit point of {xn∣n∈N}\{x_n \mid n \in \n \}, so we observe {xn∣n∈N}\{x_n \mid n \in \n \} is closed. Note that {xn∣n∈N}\{x_n \mid n \in \n \} is compact since it is closed subset of the compact set AA.
However we readily check

{B(xn,rn)}n∈N\{ B(x_n, r_n) \}_{n \in \n}

is an open cover of {xn∣n∈N}\{x_n \mid n \in \n \} with no finite subcover. (contradiction) \end{steplist}

Proof. (  ⟸  )(\impliedby) \begin{steplist} \item Let {Uα}α∈I\{U_\alpha\}_{\alpha \in I} be an open cover of AA. By \cref{thm:2_S_T} and \cref{thm:2_Lebesgue}, we observe AA is totally bounded and there exists a Lebesgue number r>0r > 0 of {Uα}α∈I\{U_\alpha\}_{\alpha \in I}. Hence there exists a finite set

{y1,y2,⋯ ,ym}\{y_1, y_2, \cdots, y_m \}

so that

A⊂⋂k=1mB(ym,r).A \subset \bigcap_{k=1}^m B(y_m, r).

\item Moreover for each B(yk,r)B(y_k, r) there exists α(yk)∈I\alpha(y_k) \in I so that

B(yk,r)⊂Uα(yk)B(y_k, r) \subset U_{\alpha(y_k)}

whence

{Uα(y1),Uα(y2),⋯ ,Uα(m)}\{U_{\alpha(y_1)}, U_{\alpha(y_2)}, \cdots, U_{\alpha(m)}\}

is an open cover of AA. \end{steplist}

% \subsection{Compactness in Finite-Dimensional Vector Spaces}

Theorem (Heine-Borel).

Let (X,∥⋅∥)(X, \|\cdot\|) be a finite-dimensional complete normed space. A⊂XA \subset X is compact if and only if AA is closed and bounded.

Proof. (  ⟹  \implies) Obvious.

(  ⟸  \impliedby) \begin{steplist} \item By \cref{thm:2_B-W}, it suffices to show that AA is sequentially compact. Let dim⁡X=d\dim X = d and {e1,⋯ ,ed}\{e_1, \cdots, e_d\} be a basis for XX. Let (xn)⊂A(x_n) \subset A be a sequence then we have a representation

xn=ξ1(n)e1+⋯+ξd(n)ed.x_n = \xi_1^{(n)}e_1 + \cdots + \xi_d^{(n)}e_d.

\item Note that ∥xn∥\|x_n\| is bounded and we have the inequality

c∑i=1d∣ξi(n)∣≤∥∑i=1dξi(n)ei(n)∥=∥xn∥<∞c \sum_{i=1}^d |\xi_i^{(n)}| \leq \left\|\sum_{i=1}^{d}\xi_i^{(n)} e_i^{(n)}\right\| = \|x_n\| < \infty

for some c>0c > 0. Hence for a fixed ii, the sequence (ξi(n))(\xi_i^{(n)}) is bounded so there exists a convergent subsequence (ξi(nk))→ξi(\xi_i^{(n_k)}) \to \xi_i. \item Consequently we conclude there exists a subsequence (xnk)(x_{n_k}) of (xn)(x_n) so that

xnk→∑i=1dξieix_{n_k} \to \sum_{i=1}^{d} \xi_i e_i

that is, AA is sequentially compact. \end{steplist}

\newpage

Summarising what we have leant, it is helpful to refer the following diagram. Note that if the space is not finite-dimensional complete normed, closed and bounded does not imply the compactness.

\begin{figure}[!h] \centering \begin{tikzcd}[math mode=false] {complete} \ar[d] & & & {closed} \ar[d]
{+} \ar[r] & {sequentialy compact} \ar[r, leftrightarrow] \ar[ul] \ar[dl] & {compact} \ar[ur] \ar[dr] & {+} \ar[l, dashrightarrow]
{totally bounded} \ar[u] & & & {bounded} \ar[u] \end{tikzcd} \end{figure}

% \subsection{Finite Intersection Property}

Definition.

Let (X,d)(X, d) be a metric space. A collection {Vα}α∈I\{ V_\alpha \}_{\alpha \in I} of the sets in XX has a finite intersection property if the any finite intersection of its element is nonempty.

Theorem.

Let (X,d)(X, d) be a metric space. A⊂XA \subset X is compact if and only if for any collection {Vα}α∈I\{ V_\alpha \}_{\alpha \in I} with finite intersection property, the intersection

A∩⋂α∈IVαA \cap \bigcap_{\alpha \in I} V_\alpha

is nonempty.

Proof. (  ⟹  \implies) Suppose not then A∩Vβ=∅A \cap V_\beta = \emptyset for some β∈I\beta \in I. Let Uα=X∖VαU_\alpha = X \setminus V_\alpha then indeed the collection {Uα}α∈I\{ U_\alpha \}_{\alpha \in I} is the open cover of AA. Since AA is compact, there exists its finite subcover and this violates the finite intersection property of {Vα}α∈I\{V_\alpha \}_{\alpha \in I}.

(  ⟸  \impliedby) Indeed, the similar argument works so we leave it for an exercise for the readers.

Corollary.

Let (X,d)(X, d) be a metric space and (Kn)⊂X(K_n) \subset X be a decreasing sequence of nonempty compact sets in XX, then ⋂n=1∞Kn\bigcap_{n=1}^\infty K_n is nonempty.

Proof. Obvious.

% \subsection{Connected Sets}

Connectedness

Definition.

Let (X,d)(X, d) be a metric space.

  • A⊂XA \subset X is disconnected if there exists disjoint open sets UU and VV such that separate AA, that is,

    U∩A≠∅,V∩A≠∅andA⊂(U∪V).U \cap A \neq \emptyset, \quad V \cap A \neq \emptyset \quad \text{and} \quad A \subset (U \cup V).

    We say (U,V)(U, V) a separation of AA.

  • A⊂XA \subset X is connected if it is not disconnected.

Example.

Show that Q⊂R\Q \subset \R is disconnected.