Scribd
Upload
68 views

Closed Convex Subset

This document summarizes key concepts from a lecture on elementary Hilbert space theory: 1) A subspace of a Hilbert space is a subset that is closed under addition and scalar multiplication. A closed subspace is also a closed set with respect to the topology induced by the metric. 2) A convex set contains the segments between any two points. Subspaces and translates of convex sets are also convex. 3) Two elements are orthogonal if their inner product is 0. The set of elements orthogonal to a given element x is a closed subspace.

Uploaded by

Anubhab Pahari
Copyright
© © All Rights Reserved
Available Formats
Download as PDF, TXT or read online on Scribd
68 views

Closed Convex Subset

This document summarizes key concepts from a lecture on elementary Hilbert space theory: 1) A subspace of a Hilbert space is a subset that is closed under addition and scalar multiplication. A closed subspace is also a closed set with respect to the topology induced by the metric. 2) A convex set contains the segments between any two points. Subspaces and translates of convex sets are also convex. 3) Two elements are orthogonal if their inner product is 0. The set of elements orthogonal to a given element x is a closed subspace.

Uploaded by

Anubhab Pahari
Copyright
© © All Rights Reserved
Available Formats
Download as PDF, TXT or read online on Scribd
You are on page 1/ 3

Math 642 Lecture #2: Elementary Hilbert Space Theory, Part II

Homework: 5, 11, 14, 15, 16


Inner Products and Linear Functionals (Continued)
Subspaces. A (linear) subspace M of a vector space V is a nonempty subset M of V
such that for all x, y ∈ M and α ∈ C there holds

x + y ∈ M, αx ∈ M.

A closed subspace M of a Hilbert space H is a subspace that is a closed set relative to


the topology induced by the metric on H.
Proposition. If M is a subspace of H, then so is its closure M .
Proof. Let x, y ∈ M , and α ∈ C.
There are sequences {xn }, {yn } in M such that xn → x and yn → y, i.e., d(xn , x) → 0
and d(yn , y) → 0.
By the triangle inequality,

d(xn + yn , x + y) = kxn + yn − x − yk
≤ kxn − xk + kyn − xk
= d(xn , x) + d(yn , y) → 0.

By properties of the inner product,

d(αxn , αx) = kαxn − αxk = kα(xn − x)k


q  p
= α(xn − x), α(xn − x) = |α|2 (xn − x, xn − x)
= |α| kxn − xk = |α|d(xn , x) → 0.

Thus, M is a closed subspace. ////


Convex Sets. A nonempty subset E of a vector space V is called convex if for all
x, y ∈ E the points
zt = (1 − t)x + ty ∈ E for all t ∈ [0, 1].
Convexity requires that E contains the segments between any two points.
Obviously, any subspace M of V is convex.
Furthermore, if E is a convex set, then any of its translates

E + x = {y + x : y ∈ E}

is also convex.
Orthogonality. Two elements x, y ∈ H are said to be orthogonal if (x, y) = 0; this is
written
x ⊥ y.
Since (x, y) = 0 implies and is implied by (y, x) = 0, the relation ⊥ is symmetry.
For an element x of H, let x⊥ denote the set of points y ∈ H that are orthogonal to x:

x⊥ = {y ∈ H : x ⊥ y}.

The set x⊥ is a subspace of H since

x ⊥ y and x ⊥ y 0 ⇒ x ⊥ (y + y 0 ), and x ⊥ (αy) for all α ∈ C.

Moreover, x⊥ is precisely the set of points y where the (uniformly) continuous functional
y → (x, y) is zero, i.e., x⊥ is the preimage of 0 for the map y → (x, y).
This implies that x⊥ is a closed subspace of H.
For a subspace M of H, set

M ⊥ = {y ∈ H : x ⊥ y for all x ∈ M }.

Observe that \
M⊥ = x⊥ .
x∈M
⊥ ⊥
Since each x is a closed subspace, then M is the intersection of closed subspaces, and
hence is a closed subspace of H.
The Parallelogram Law. For all x, y ∈ H, there holds
kx + yk2 + kx − yk2 = 2kxk2 + 2kyk2 .

Proof. The Axioms of an inner product establish this identity.


Indeed,

kx + yk2 + kx − yk2 = (x + y, x + y) + (x − y, x − y)
= (x, x) + (x, y) + (y, x) + (y, y) + (x, x) − (x, y) − (y, x) + (y, y)
= 2kxk2 + 2kyk2 ,

which gives the desired equation. ////


Theorem. Every closed, convex set E in a Hilbert space H contains a unique element
of smallest norm, i.e., there is exactly one x0 ∈ E such that kx0 k ≤ kxk for all x ∈ E.
Proof. Let δ = inf{kxk : x ∈ E}.
For any x, y ∈ E, the parallelogram law applied to x/2 and y/2 gives
x + y 2

2 2 2
(1/4)kx − yk = (1/2)kxk + (1/2)kyk − .
2
Since E is convex, the linear combination (1/2)(x + y) belongs to E.
By the definition of δ, it follows that

kx − yk2 ≤ 2kxk2 + 2kyk2 − 4δ 2 .


From this follows the uniqueness: if kxk = kyk = δ, then

kx − yk2 ≤ 0 ⇒ x = y.

Now for the existence: from the definition of δ, there is a sequence {yn } in E such that
kyn k → δ as n → ∞.
Replace x and y by yn and ym in kx − yk2 ≤ 2kxk2 + 2kyk2 − 4δ 2 to get

kyn − ym k2 ≤ 2kyn k2 + 2kym k2 − 4δ 2 .

for all n and all m.


As the left-hand side of this inequality goes to 0 as n → ∞ and m → ∞, the sequence
{yn } is Cauchy.
Completeness of H implies there is x0 ∈ H such that yn → x0 , i.e., kyn − x0 k → 0 as
n → ∞.
Since yn ∈ E and E is closed, the limit x0 of {yn } also belongs to E.
Continuity of the norm function (as shown in Lecture #1) now implies that

kx0 k = lim kyn k = δ.


n→∞

Thus E has a unique element of smallest norm. ////


Corollary (Conway p. 8). If E is a closed, convex subset of H and y is an element
of H, then there exists a unique x0 ∈ E such that ky − x0 k = inf{ky − xk : x ∈ E}.
Remark. The quantity inf{ky − xk : x ∈ E} is called the distance between the element
y and the closed, convex subset E; this distance is denoted by d(y, E).
Proof. The translate E − y = {x − y : x ∈ E} is also a closed, convex subset of H.
By the Theorem, there is a unique element h0 in E − y of smallest norm: kh0 k ≤ khk for
all h ∈ E − y.
In general, for each h ∈ E − y there is an x ∈ E such that h = x − y.
By the definition of E − y and the uniqueness of h0 , there is a unique x0 ∈ E such that
h0 = x0 − y.
Thus kh0 k ≤ khk for all h ∈ E − y becomes

ky − x0 k ≤ ky − xk for all x ∈ E.

Taking the infimum of both sides and recognizing that ky − x0 k = ky − xk when x = x0


yields ky − x0 k = inf{ky − xk : x ∈ E}. ////

You might also like