Basic Properties of Metric and Normed Spaces

Computational and Metric Geometry

Instructor: Yury Makarychev

The second part of this course is about metric geometry. We will study metric spaces,
low distortion metric embeddings, dimension reduction transforms, and other topics. We
will discuss numerous applications of metric techniques in computer science.

1 Definitions and Examples

1.1 Metric and Normed Spaces
Definition 1.1. A metric space is a pair (X, d), where X is a set and d is a function from
X × X to R such that the following conditions hold for every x, y, z ∈ X.

1. Non-negativity: d(x, y) ≥ 0.

2. Symmetry: d(x, y) = d(y, x).

3. Triangle inequality: d(x, y) + d(y, z) ≥ d(x, z) .

4. d(x, y) = 0 if and only if x = y.

Elements of X are called points of the metric space, and d is called a metric or distance
function on X.

Exercise 1. Prove that condition 1 follows from conditions 2-4.

Occasionally, spaces that we consider will not satisfy condition 4. We will call such spaces
semi-metric spaces.

Definition 1.2. A space (X, d) is a semi-metric space if it satisfies conditions 1-3 and 4 0 :

4 0 . if x = y then d(x, y) = 0.

Examples. Here are several examples of metric spaces.

1. Euclidean Space. Space Rd equipped with the Euclidean distance d(x, y) = kx−yk2 .

1
 2. Uniform Metric. Let X be an arbitrary non-empty set. Define a distance function
d(x, y) on X by d(x, y) = 1 if x 6= y and d(x, x) = 0. The space (X, d) is called a
uniform or discrete metric metric space.

3. Shortest Path Metric on Graphs. Let G = (V, E, l) be a graph with positive edge
lengths l(e). Let d(u, v) be the length of the shortest path between u and v. Then
(V, d) is the shortest path metric on G.

4. Tree Metrics. A very important family of graph metrics is the family of tree metrics.
A tree metric is the shortest path metric on a tree T .

5. Cut Semi-metric. Let V be a set of vertices and S ⊂ V be a proper subset of V . Cut

semi-metric δS is defined by δS (x, y) = 1 if x ∈ S and y ∈ / S, or x ∈/ S and y ∈ S; and
δS (x, y) = 0, otherwise. In general, the space (X, d) is not a metric since d(x, y) = 0
for some x 6= y. Nevertheless, δS (x, y) is often called a cut metric.
Definition 1.3. A normed space is a pair (V, k · k), where V is a linear space (vector space)
and k · k : V → R is a norm on V such that the following conditions hold for every x, y ∈ V .
1. kxk > 0 if x 6= 0.

2. kxk = 0 if and only if x = 0.

3. kαxk = |α| · kxk for every α ∈ R.

4. kx + yk ≤ kxk + kyk (convexity).

Every normed space (V, k · k) is a metric space with metric d(x, y) = kx − yk on V .

1.2 Lebesgue Spaces Lp (X, µ)

In this section, we define Lebesgue spaces, a very important class of normed spaces. Let
(X, µ) be a measure space. We consider the set of measurable real valued functions on X.
For p ≥ 1, we define the the p–norm of a function f by
Z 1/p
p
kf kp = |f (x)| dµ(x) .
X

If the integral above is infinite (diverges), we write kf kp = ∞. Similarly, we define

kf k∞ = sup |f (x)|.

Now we define the Lebesgue space Lp (X, µ) (for 1 ≤ p ≤ ∞):

Lp (X, µ) = {f : f is measurable w.r.t. measure µ; kf kp < ∞}.

Caveat: The norm kf kp can be equal to 0 for a function f ∈ Lp (X, µ), which is not
identically equal to 0. So formally Lp (X, µ) (as defined above) is not a normed space. The

2
standard way to resolve this problem is to identify functions that differ only on a set of
measure 0. The norm k · k∞ is usually defined as
kf k∞ = ess sup |f (x)| = inf{sup |f˜(x)| : f˜(x) = f (x) almost everywhere}.
x∈X x∈X

Examples. Consider several examples of Lp –spaces.

1. Space `p . Let X = N, and µ be the counting measure; i.e. µ(S) = |S| for S ⊂ N. The
elements of `p are infinite sequences of real numbers a = (a1 , a2 , . . . ) (which we identify
with maps from N to R) s.t. kakp < ∞. The p−norm of a sequence a = (a1 , a2 , . . . )
equals
∞
!1/p
X
p
kakp = |a| .
i=1

2. Space `dp . Let X = {1, . . . , d}, and µ be again the counting measure; i.e. µ(S) = |S|
for S ⊂ N. The elements of `dp are d-tuples of real numbers a = (a1 , a2 , . . . , ad ) ∈ Rd .
The p−norm of a vector a = (a1 , a2 , . . . , ad ) equals
d
!1/p
X
kakp = |a|p .
i=1

3. Space Lp [a, b]. Let X = [a, b], and µ be the standard measure on R. The elements
of Lp [a, b] are measurable functions f : [a, b] → R with kf kp < ∞. The p–norm of a
function f equals
Z b 1/p
p
kf kp = |f (x)| dx .
a

Lemma 1.4. For every 1 ≤ p < q ≤ ∞, we have `p ⊂ `q and Lq [0, 1] ⊂ Lp [0, 1]. Both
inclusions are proper.
Proof. We consider the case when q < ∞. Let a ∈ `p . P Let I = {i : |ai | ≥ 1}. Note that I
is a finite set, as otherwise we would have that kakp ≥ i∈I |ai |p = ∞. For every i ∈
p
/ I, we
q p
have |ai | < |ai | . Therefore,
X X X X X
kakqq = |ai |q + |ai |q ≤ |ai |q + |ai |p ≤ |ai |q + kakpp < ∞.
i∈I i∈I
/ i∈I i∈I
/ i∈I

We conclude that a ∈ `q .
p q
R Now qlet f ∈R Lq [0, 1]. Let I = {x : |f (x)| ≤ 1}. Note that |f | < |f | when x ∈ / I, and
I
|f (x)| dx ≤ I 1 dx ≤ 1. Therefore,
Z 1 Z Z Z
p p p p
kf kp = |f (x)| dx = |f (x)| dx+ |f (x)| dx ≤ 1+ |f (x)|q dx ≤ 1+kf kqq < ∞.
0 I [0,1]\I [0,1]\I

We get that f ∈ Lp [0, 1].

3
Exercise 2. Prove the statement of Lemma 1.4 for q = ∞.

Exercise 3. Let (X, µ) be a measure space with µ(X) < ∞, and 1 ≤ p < q ≤ ∞. Prove
that Lq (X, µ) ⊂ Lp (X, µ). Show that on the other hand Lq (R) 6⊂ Lp (R).

1.3 Metric Embeddings

Consider two metric spaces (X, dX ), (Y, dY ) and a map f : X → Y . We say that f : X → Y
is a Lipschitz map if there is a number C such that

dY (f (x1 ), f (x2 )) ≤ CdX (x1 , x2 ) for all x1 , x2 , ∈ X.

The Lipschitz constant kf kLip of f is the minimum C such that this inequality holds.
We say that a bijective map ϕ : X → Y is an isometry if for every x1 , x2 ∈ X,
dY (ϕ(x1 ), ϕ(x2 )) = dX (x1 , x2 ). We say that an injective map ϕ : X → Y is an isomet-
ric embedding if ϕ is an isometry between X and ϕ(X) (the image of X under ϕ).
The distortion of a map f : X → Y equals kf kLip · kf −1 kLip where f −1 is the inverse map
from f (X) to X.

Exercise 4. Prove that f has distortion at most D if and only if there is a number c > 0
such that

c · dX (x1 , x2 ) ≤ dY (x1 , x2 ) ≤ c D · dX (f (x1 ), f (x2 )) for every x1 , x2 ∈ X.

Exercise 5. Prove the following statements.

1. An isometric embedding has distortion 1.

2. Let f be a Lipschitz map from X to Y and g be a Lipschitz map from Y to Z then

h = g ◦ f is a Lipschitz map from X to Z and khkLip ≤ kf kLip · kgkLip .

3. Let f be an embedding of X into Y and g be an embedding of Y into Z then the

distortion of h = g ◦ f is at most the product of distortions of f and g.

2 Embeddings into Lp spaces

Theorem 2.1. Every finite metric subspace (X, d) embeds isometrically into `n∞ for n = |X|.

Proof. Denote the elements of X by x1 , x2 , . . . , xn . Now define the embedding ϕ : X → `n∞

as follows
ϕ : x 7→ (d(x1 , x), d(x2 , x), . . . , d(xn , x)).
We claim that ϕ is an isometric embedding. That is,

kϕ(xi ) − ϕ(xj )k∞ = d(xi , xj ).

4
First, we prove that kϕ(xi ) − ϕ(xj )k∞ ≤ d(xi , xj ). We need to show that all coordinates
of the vector ϕ(xi ) − ϕ(xj ) are bounded by d(xi , xj ) in the absolute value. Indeed, the
k-th coordinate of ϕ(xi ) − ϕ(xj ) equals d(xk , xi ) − d(xk , xj ). From the triangle inequalities
for xi , xj and xk , it follows that |d(xk , xi ) − d(xk , xj )| ≤ d(xi , xj ). Now, we verify that
kϕ(xi ) − ϕ(xj )k∞ ≥ d(xi , xj ). Note that kϕ(xi ) − ϕ(xj )k∞ ≥ |d(xk , xi ) − d(xk , xj )| (the
absolute value of the k-th coordiante) for every k. In particular, this inequality holds for
k = i. That is,
kϕ(xi ) − ϕ(xj )k∞ ≥ |d(xi , xi ) − d(xi , xj )| = d(xi , xj ).

Theorem 2.2. Let p ∈ [1, ∞). Metric space `d2 (Euclidean d-dimensional space) embeds
isometrically into Lp (X, µ) for some space X.
Proof. Let X = `d2 and µ = γ be the d
. Gaussian measure on X = `2 (µ is the probability
2
measure on X with density e−kxk /2 (2π)d/2 ). Recall that the elements of Lp (`d2 , γ) are
R 1/p
functions on `d2 . Let M = `d |x1 |p dγ(x) . We construct an embedding ϕ that maps
2
every v ∈ `d2 to a function fv defined as follows:
hv, xi
fv (x) = .
M
We prove that the embedding is an isometry. Consider two vectors u and v. Let w = u − v,
and e = w/kwk2 . We have,
 hu, xi hv, xi p
Z   Z
1
p
kϕ(u) − ϕ(v)kp =  −  dγ(x) = p |hu − v, xi|p dγ(x)
`d2
 M M  M `d2
Z Z
1 p 1 p
= p |hkwk e, xi| dγ(x) = p kwk |he, xi|p dγ(x)
M `d2 M d
`2

Consider a coordinate frame in which the x1 -axis is parallel to the vector e (i.e. vector e has
coordinates (1, 0, . . . , 0)). Then |he, xi| = |x1 |. We get
!1/p
kwk2
Z
kϕ(u) − ϕ(v)kp = |x1 |p dµ(x) = kwk2 = ku − vk2 .
M d
`2

We proved that the map ϕ is an isometry.

We showed that every finite subset S of `d2 embeds isometrically into space Lp (X, µ). Can
we embed S into a “simpler” space `N p (for some N )? We will see that all spaces Lp (X, µ)
(of sufficiently large dimension) have essentially the same finite metric subspaces. Therefore,
if a metric space embeds into some Lp (X, µ), then it also embeds into `N p for some N .

Theorem 2.3. Let S be a finite subset of Lp (Z, µ), n = |S|, and N = n2 + 1. Then S

isometrically embeds into `N p .

5
Proof. Consider the linear space M
of all symmetric n ×nn matrices with zeros on the
n
diagonal. The dimension of M is 2 . Consider a map f : R → M defined as follows. The
map f sends a vector u ∈ Rn to the matrix A = (aij ) with entries aij = |ui − uj |p . Clearly,
f (v) ∈ M for every v ∈ Rn . Let B = f (Rn ) ≡ {f (u) : u ∈ Rn }. Let C be the convex closed
cone generated by B; that is, C is the closure of the following set
( m ) ( m )
X X
λ(i) b(i) : b(i) ∈ B, λ(i) ≥ 0, m ∈ N = λi f (x(i) ) : x(i) ∈ Rn , λ(i) ≥ 0, m ∈ N .
i=1 i=1

For every metric space (S, d) on a set S = {s1 , s2 , . . . , sn }, we define a matrix F S by

FijS = d(si , sj )p .
The theorem follows from the following two lemmas.
Lemma 2.4. Suppose that S = {s1 , . . . , sn } ⊂ Lp (Z, µ) then F S ∈ C.
Proof. Recall that each element si is a function from Z to R. Let σ(z) = (s1 (z), . . . , sn (z))
for every z ∈ Z. We have,
Z Z
S p p
Fij = d(si , sj ) = |si (z) − sj (z)| dµ(z) = fij (σ(z))dµ(z).
Z Z
R
Therefore, F S = Z
f (σ(z))dµ(z). Since f (σ(z)) ∈ B ⊂ C for every z ∈ Z, we conclude that
F S ∈ C.
Lemma 2.5. Consider a finite metric space S = {s1 , . . . , sn }. Suppose that F S ∈ C. Then
n
N


S embeds into `p , where N = 2 + 1.
Proof. By the Carathéodory theorem, every point in the cone C can be expressed as a sum
of at most dim M + 1 = N points in B. In particular, we can write
N
X N
X
S (k)
F = b = f (x(k) ),
k=1 k=1

for some b(1) , . . . , b(N ) ∈ B and some x(k) ∈ f −1 (b(k) ) (xi is a preimage of b(k) ). By the
definition of F S , we have
N
X N
X
p (k) (k)
d(si , sj ) = FijS = fij (x ) = (k)
|xi − xj |p . (1)
k=1 k=1

We define the embedding ϕ of S to `N

p :

(1) (2) (N )
ϕ(si ) = (xi , xi , . . . , xi ).

Note that equation (1) says that d(si , sj )p = kϕ(si ) − ϕ(sj )kpp , and therefore d(si , sj ) =
kϕ(si ) − ϕ(sj )kp . We conclude that ϕ is an isometric embedding.

6
Corollary 2.6. Suppose that S is a subset of `d2 . Then S isometrically embeds into `N
p ,
|S|


where N = 2 + 1.

Exercise 6. In our proof, we used the Carathéodory theorem for arbitrary convex sets:
every point in the convex hull of S ⊂ Rd is a convex combination of at most d + 1 points
from S. Show that if the convex hull CH(S) of S is a cone, then every point CH(S) is a
linear combination, with positive coefficients, of at
most d points in
S. Conclude that in the
statement of Theorem 2.3 we can replace N = n2 + 1 with N = n2 .

Definition 2.7. Let cp (X) be the least distortion1 with which a finite metric space (X, d)
embeds into `p .

Theorem 2.8. For every finite metric space X and every p ∈ [1, ∞], we have 1 = c∞ (X) ≤
cp (X) ≤ c2 (X).

Proof. The inequality 1 = c∞ (X) ≤ cp (X) follows from Theorem 2.1. Let f be an embedding
of X into `2 (X) with distortion c2 (X). By Corollary 2.6, there is an isometric embedding g
of f (X) ⊂ `2 into `p . Then map g ◦ f is an embedding of X into `p with distortion at most
c2 (X). We conclude that cp (X) ≤ c2 (X).

1
A simple compactness argument shows that there is an embedding with the least possible distortion.