Operators on Hilbert Spaces - user web page - AIMS
Operators on Hilbert Spaces - user web page - AIMS
Operators on Hilbert Spaces - user web page - AIMS
You also want an ePaper? Increase the reach of your titles
YUMPU automatically turns print PDFs into web optimized ePapers that Google loves.
<str<strong>on</strong>g>Operators</str<strong>on</strong>g> <strong>on</strong> <strong>Hilbert</strong> <strong>Spaces</strong><br />
Margaret Modupe Kajot<strong>on</strong>i (meg@aims.ac.za)<br />
African Institute for Mathematical Sciences (<strong>AIMS</strong>)<br />
Supervised by S<strong>on</strong>ja Mout<strong>on</strong><br />
University of Stellenbosch, South Africa<br />
June 4, 2007
Abstract<br />
The c<strong>on</strong>cepts of arbitrary vector spaces can be generalised to inner product spaces and complete<br />
inner product spaces, called <strong>Hilbert</strong> spaces. The inner product is a generalisati<strong>on</strong> of the dot<br />
product in R n . The dot product and orthog<strong>on</strong>ality are important in many applicati<strong>on</strong>s. The<br />
theory of inner product spaces and <strong>Hilbert</strong> spaces is richer than that of general normed and<br />
Banach spaces.<br />
We recall that a Banach space is a complete normed vector space. Hence, every <strong>Hilbert</strong> space is<br />
a Banach space but the c<strong>on</strong>verse is not generally true. A necessary and sufficient c<strong>on</strong>diti<strong>on</strong> for a<br />
Banach space to be a <strong>Hilbert</strong> space is that the parallelogram equality, which will be discussed in<br />
a subsequent secti<strong>on</strong>, should hold.<br />
The distinguishing features between <strong>Hilbert</strong> spaces and Banach spaces are<br />
1. Representati<strong>on</strong> of a <strong>Hilbert</strong> space as a direct sum of a closed subspace and its orthog<strong>on</strong>al<br />
complement.<br />
2. Orthog<strong>on</strong>al sets and sequences and corresp<strong>on</strong>ding representati<strong>on</strong>s of the elements of the<br />
<strong>Hilbert</strong> space.<br />
3. The Riesz representati<strong>on</strong> of bounded linear functi<strong>on</strong>als by inner products.<br />
4. The <strong>Hilbert</strong>-adjoint operator T ∗ of the bounded linear operator T.<br />
Spectral theory is a very broad but important aspect of applied functi<strong>on</strong>al analysis which extends<br />
to the eigenvector and eigenvalue theory of a single square matrix. The name was introduced<br />
by David <strong>Hilbert</strong> in his original formulati<strong>on</strong> of <strong>Hilbert</strong> space theory. It was later discovered that<br />
spectral theory could explain features of atomic spectral in quantum mechanics. In this essay, we<br />
will c<strong>on</strong>sider the spectrum of a Hermitian (or self-adjoint) operator.<br />
i
C<strong>on</strong>tents<br />
Abstract i<br />
1 Introducti<strong>on</strong> 1<br />
2 Basic C<strong>on</strong>cepts 2<br />
2.1 Inner product spaces and <strong>Hilbert</strong> spaces . . . . . . . . . . . . . . . . . . . . . . 2<br />
2.2 Properties of inner product space . . . . . . . . . . . . . . . . . . . . . . . . . 3<br />
2.3 Orthog<strong>on</strong>al complement and direct sum . . . . . . . . . . . . . . . . . . . . . . 5<br />
3 <str<strong>on</strong>g>Operators</str<strong>on</strong>g> 10<br />
3.1 Functi<strong>on</strong>als . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10<br />
3.2 Riesz Representati<strong>on</strong> of functi<strong>on</strong>als . . . . . . . . . . . . . . . . . . . . . . . . 11<br />
3.3 Sesquilinear functi<strong>on</strong>als and Riesz representati<strong>on</strong> . . . . . . . . . . . . . . . . . 12<br />
3.4 <strong>Hilbert</strong>-adjoint and self-adjoint operators . . . . . . . . . . . . . . . . . . . . . 15<br />
4 Spectral Analysis of a Self-Adjoint Operator 21<br />
5 C<strong>on</strong>clusi<strong>on</strong> and further work 25<br />
Bibliography 27<br />
ii
1. Introducti<strong>on</strong><br />
The equati<strong>on</strong> Lf = g occurs frequently in fields like mathematics, physics and engineering. Here<br />
g is a known functi<strong>on</strong> of space and time variables, L is a linear differential operator or mapping<br />
and f is a functi<strong>on</strong> which is unknown but required to satisfy some initial or boundary c<strong>on</strong>diti<strong>on</strong>s.<br />
The major problem will be that of finding f. No matter the trick used, it is not certain to work<br />
for all equati<strong>on</strong>s. Moreover, it is unlikely to obtain a general soluti<strong>on</strong> in a form explicit enough<br />
to tell everything about the system. We will be faced with questi<strong>on</strong>s like, does every equati<strong>on</strong><br />
has a soluti<strong>on</strong>? If so, is it unique? In dealing with these questi<strong>on</strong>s, we might seek inspirati<strong>on</strong><br />
from linear algebra since Lf = g resembles Ax = b, where A is a matrix and x, b are vectors.<br />
In finite dimensi<strong>on</strong>s, the soluti<strong>on</strong> exists and is unique if and <strong>on</strong>ly if the determinant of L is not<br />
equal to zero (det L = 0). However, the differences between finite and infinite dimensi<strong>on</strong>s are<br />
great. In infinite dimensi<strong>on</strong>s, determinants can <strong>on</strong>ly play a very limited role. Nineteenth-century<br />
analysts made great progress with these questi<strong>on</strong>s. In trying to present the result with simplicity<br />
and generality brought about the evoluti<strong>on</strong> of functi<strong>on</strong>al analysis.<br />
In modern view, functi<strong>on</strong>al analysis is seen as the study of spaces of functi<strong>on</strong>s over the real or<br />
complex numbers. <strong>Hilbert</strong> spaces are the most useful spaces in practical applicati<strong>on</strong>s of functi<strong>on</strong>al<br />
analysis. The noti<strong>on</strong> of the <strong>Hilbert</strong> space was initiated by D. <strong>Hilbert</strong>, a German mathematician,<br />
in 1912. Completeness is an extremely important property of <strong>Hilbert</strong> spaces as we shall see.<br />
Complete spaces posses many useful properties that are absent in incomplete spaces. <strong>Hilbert</strong><br />
spaces are useful in partial differential equati<strong>on</strong>s, quantum mechanics and signal processing.<br />
<str<strong>on</strong>g>Operators</str<strong>on</strong>g> or mappings such as L in the equati<strong>on</strong> above can be defined <strong>on</strong> <strong>Hilbert</strong> spaces. In<br />
particular, bounded linear operators can be defined <strong>on</strong> <strong>Hilbert</strong> spaces and this is what we seek<br />
to study in this essay. We shall acquaint ourselves with properties of normed and inner product<br />
spaces. We shall discuss mappings or operators from <strong>on</strong>e space to another. We need at least<br />
two sets in order to define an operator. If the two sets are vector spaces, we can introduce the<br />
c<strong>on</strong>cept of a linear operator, if the sets are normed spaces, we can c<strong>on</strong>struct a theory of bounded<br />
linear operators <strong>on</strong> such spaces. <str<strong>on</strong>g>Operators</str<strong>on</strong>g> that map members of a specified space into the real<br />
or complex numbers are called functi<strong>on</strong>als. We shall also discuss the <strong>Hilbert</strong>-adjoint operators as<br />
well as the self-adjoint, unitary and normal operators. Finally, we shall look at the spectral analysis<br />
of the Hermitian (self-adjoint) operators which is an important aspect of functi<strong>on</strong>al analysis.<br />
1
2. Basic C<strong>on</strong>cepts<br />
2.1 Inner product spaces and <strong>Hilbert</strong> spaces<br />
To make a definiti<strong>on</strong> that will be applicable to both real and complex vector spaces, we need to<br />
examine some properties of the complex plane C n . Recall that if λ = a + bi, where a, b ∈ R,<br />
then the absolute value of λ is defined by |λ| = √ a 2 + b 2 and the complex c<strong>on</strong>jugate of λ is<br />
defined by λ = a − bi. Also, |λ| 2 = λλ. For z = (z1, . . . , zn) ∈ C n , we define the norm of z by<br />
z = |z1| 2 + · · · + |zn| 2 . We want z to be n<strong>on</strong>negative hence we take the absolute values<br />
of its elements. Note that z 2 = z1z1 + · · · + znzn. Also for any w = (w1, . . . , wn) ∈ C n , the<br />
inner product of w with z is equal to the complex c<strong>on</strong>jugate of the inner product of z with w.<br />
Definiti<strong>on</strong> 2.1.1 ([Kre78], 3·1-1 (Inner product))<br />
An inner product <strong>on</strong> a vector space X is a functi<strong>on</strong> that takes each ordered pair (x, y) of elements<br />
of X to number 〈x, y〉 of the elements of K (where K is the scalar field of X). That<br />
is, with every pair of vector (x, y), there is associated a scalar 〈x, y〉 with the following properties:<br />
1. 〈x, x〉 ≥ 0 for all x ∈ X,<br />
2. 〈x, x〉 = 0 if and <strong>on</strong>ly if x = 0,<br />
3. 〈x + y, z〉 = 〈x, z〉 + 〈y, z〉 for all x, y, z ∈ X,<br />
4. 〈ay, z〉 = a〈y, z〉 for all a ∈ K, where K is the scalar field of X and y, z ∈ X,<br />
5. 〈y, z〉 = 〈z, y〉 for all y, z ∈ X.<br />
An inner product <strong>on</strong> X defines a norm <strong>on</strong> X given by x = 〈x, x〉 = 〈x, x〉 1<br />
2 and a metric <strong>on</strong><br />
X given by d(x, y) = x − y = 〈x − y, x − y〉.<br />
An inner product space (or pre-<strong>Hilbert</strong> space) is a vector space with inner product 〈x, y〉 defined<br />
<strong>on</strong> it.<br />
Definiti<strong>on</strong> 2.1.2 ([Kre78], 3·1-2 (Orthog<strong>on</strong>ality))<br />
An element x of an inner product space X is said to be orthog<strong>on</strong>al to an element y ∈ X if<br />
〈x, y〉 = 0. It is denoted by x ⊥ y and we say that x and y are orthog<strong>on</strong>al. Similarly, for subsets<br />
A, B ⊂ X, we write x ⊥ A if x ⊥ a for all a ∈ A, and A ⊥ B if a ⊥ b for all a ∈ A and all<br />
b ∈ B.<br />
Definiti<strong>on</strong> 2.1.3 (<strong>Hilbert</strong> space)<br />
A <strong>Hilbert</strong> space is a complete inner product space.<br />
2
Secti<strong>on</strong> 2.2. Properties of inner product space Page 3<br />
Example 2.1.4 ([Kre78], 3·1-3 (Examples of <strong>Hilbert</strong> spaces))<br />
1. The Euclidean space Rn is a <strong>Hilbert</strong> space with inner product defined by<br />
〈x, y〉 = x1y1 + · · · + xnyn where x = (x1, ..., xn) and y = (y1, · · · , yn). We obtain that<br />
x = 〈x, x〉 1<br />
2 = (x 2 1 + · · · + x 2 n) 1<br />
2 and the metric defined <strong>on</strong> it is given by<br />
d(x, y) = x − y = 〈x − y, x − y〉 1<br />
2 = (x1 − y1) 2 1<br />
+ · · · + (xn − yn)<br />
2 2 .<br />
2. The unitary space Cn : 〈x, y〉 = x1y1 + · · · + xnyn. 3. The <strong>Hilbert</strong> space l2 <br />
=<br />
, with inner product defined by 〈x, y〉 =<br />
∞ j=1 |xj| 2 < ∞<br />
∞ j=1 xjyj and the norm x = 〈x, x〉 1<br />
2 = ( ∞ j=1 |xj| 2 ) 1<br />
2 .<br />
2.2 Properties of inner product space<br />
The norm <strong>on</strong> an inner product space X satisfies the following properties.<br />
Theorem 2.2.1 ([Erd80], Lemma 1.4 (Parallelogram law))<br />
Proof.<br />
If x, y ∈ X, then<br />
x + y 2 + x − y 2 = 2(x 2 + y 2 ). (2.1)<br />
x + y 2 + x − y 2 = 〈x + y, x + y〉 + 〈x − y, x − y〉<br />
= x 2 + y 2 + 〈x, y〉 + 〈y, x〉 + x 2 + y 2 − 〈x, y〉 − 〈y, x〉<br />
= 2(x 2 + y 2 ).<br />
Remark 2.2.2 If a norm does not satisfy the parallelogram equality, then it cannot be obtained<br />
from an inner product space.<br />
Theorem 2.2.3 ([Erd80], Lemma 1.5 (Pythagorean theorem)) If x, y are orthog<strong>on</strong>al in an inner<br />
product space X, then<br />
Proof.<br />
Suppose x, y are orthog<strong>on</strong>al vectors in X. Then<br />
x + y 2 = x 2 + y 2 . (2.2)<br />
x + y 2 = 〈x + y, x + y〉<br />
= x 2 + y 2 + 〈x, y〉 + 〈y, x〉<br />
= x 2 + y 2 . (2.3)
Secti<strong>on</strong> 2.2. Properties of inner product space Page 4<br />
The Cauchy-Schwarz inequality gives <strong>on</strong>e of the most important inequalities in Mathematics.<br />
Before we state and prove the theorem, let us examine how we can decompose two orthog<strong>on</strong>al<br />
vectors. If x, y ∈ X, then we would like to write x as a scalar multiple of y plus a vector w<br />
orthog<strong>on</strong>al to y. This is called the orthog<strong>on</strong>al decompositi<strong>on</strong>. Now let a ∈ K (where K is the<br />
scalar field of X). Then x = ay + (x − ay). We choose a so that y is orthog<strong>on</strong>al to (x − ay).<br />
That is, we want<br />
(where y = 0). Hence we write<br />
0 = 〈x − ay, y〉 = 〈x, y〉 − ay 2 thus, a =<br />
x =<br />
<br />
〈x, y〉<br />
y + x −<br />
y2 We will use this to prove the Cauchy-Schwarz inequality.<br />
<br />
〈x, y〉<br />
y .<br />
y2 〈x, y〉<br />
y 2<br />
Theorem 2.2.4 ([Erd80], Theorem 1.1 (Cauchy-Schwarz inequality)) The Cauchy Schwarz inequality<br />
states that if x, y ∈ X, then<br />
|〈x, y〉| ≤ xy. (2.4)<br />
Proof.<br />
Let x, y ∈ X. If y = 0, then both sides of (2.4) will be equal to 0 and the inequality will hold.<br />
Suppose y = 0. C<strong>on</strong>sider the orthog<strong>on</strong>al decompositi<strong>on</strong> x = 〈x,y〉<br />
y2 y + w, where w is orthog<strong>on</strong>al<br />
to y. By the Pythagorean theorem,<br />
x 2 <br />
<br />
<br />
= <br />
〈x, y〉 <br />
<br />
y<br />
y2 <br />
2<br />
+ w 2 =<br />
|〈x, y〉|2<br />
y 2<br />
+ w2 ≥<br />
|〈x, y〉|2<br />
y 2 . (2.5)<br />
Multiplying both sides of equati<strong>on</strong> (2.5) by y 2 and then taking square roots yield the Cauchy-<br />
Schwarz inequality.<br />
Remark 2.2.5 Equati<strong>on</strong> (2.4) is an equality if and <strong>on</strong>ly if (2.5) is an equality. This will happen<br />
if and <strong>on</strong>ly if w = 0. But w = 0 if and <strong>on</strong>ly if x is a multiple of y. Hence equati<strong>on</strong> (2.4) is an<br />
equality if and <strong>on</strong>ly if <strong>on</strong>e of x or y is a scalar multiple of the other.<br />
Theorem 2.2.6 ([Axl97], 6.9 (Triangular inequality)) The triangular inequality states that the<br />
length of any side of a triangle is less than the sum of the lengths of the other two sides. If x,y<br />
∈ X, then<br />
x + y ≤ x + y. (2.6)<br />
This theorem can be used to show that the shortest path between two points is a straight line<br />
segment.
Secti<strong>on</strong> 2.3. Orthog<strong>on</strong>al complement and direct sum Page 5<br />
Proof.<br />
Let x, y ∈ X. Then<br />
and by Cauchy-Schwarz inequality,<br />
x + y 2 = 〈x + y, x + y〉<br />
= 〈x, x〉 + 〈y, y〉 + 〈x, y〉 + 〈y, x〉<br />
= 〈x, x〉 + 〈y, y〉 + 〈x, y〉 + 〈x, y〉<br />
= x 2 + y 2 + 2Re〈x, y〉<br />
≤ x 2 + y 2 + 2|〈x, y〉| (2.7)<br />
≤ x 2 + y 2 + 2xy<br />
= (x + y) 2<br />
Therefore x + y ≤ x + y.<br />
Theorem 2.2.7 ([Kre78], 3·2-2 Lemma (C<strong>on</strong>tinuity of inner product))<br />
In an inner product space, if xn → x and yn → y, then<br />
〈xn, yn〉 → 〈x, y〉. (2.8)<br />
Proof.<br />
The c<strong>on</strong>diti<strong>on</strong> 〈xn, yn〉 → 〈x, y〉 is equivalent to |〈xn, yn〉 − 〈x, y〉| → 0 as n → ∞.<br />
|〈xn, yn〉 − 〈x, y〉| = |〈xn, yn〉 − 〈xn, y〉 + 〈xn, y〉 − 〈x, y〉|<br />
≤ |〈xn, yn − y〉| + |〈xn − x, y〉| (By triangular inequality)<br />
≤ xnyn − y + xn − xy (By Schwarz inequality)<br />
Now (xn) is c<strong>on</strong>vergent and hence bounded, so that xn ≤ K for all n ∈ N. Therefore<br />
|〈xn, yn〉 − 〈x, y〉| ≤ Kyn − y + xn − xy.<br />
Since yn − y → 0 and xn − x → 0, the last expressi<strong>on</strong> tends to zero as n tends to infinity.<br />
2.3 Orthog<strong>on</strong>al complement and direct sum<br />
In a metric space X, the distance δ from an element x ∈ X to a n<strong>on</strong>empty subset M ⊂ H is<br />
given by<br />
δ = inf d(x, y). (2.9)<br />
y∈M<br />
In a normed space it becomes<br />
δ = inf x − y. (2.10)<br />
y∈M
Secti<strong>on</strong> 2.3. Orthog<strong>on</strong>al complement and direct sum Page 6<br />
Theorem 2.3.1 ([Kre78], [Red53], 3·3-1 Theorem (Minimising vector)) Let X be an inner product<br />
space and let M = ∅ be a c<strong>on</strong>vex subset which is complete (in the metric induced by the<br />
inner product). Then for every given x ∈ X, there exists a unique y ∈ M such that<br />
δ = inf<br />
ˆy∈M<br />
x − ˆy = x − y. (2.11)<br />
Proof.<br />
Existence<br />
We choose a sequence (yn) in M such that δn → δ, where δn = x − yn (by definiti<strong>on</strong> of<br />
infimum). We will show that (yn) is a Cauchy sequence and then we can make some deducti<strong>on</strong>s.<br />
Using the parallelogram law,<br />
But 1<br />
yn − ym 2 = (yn − x) − (ym − x) 2<br />
= 2yn − x 2 + 2ym − x 2 − (yn − x) + (ym − x) 2<br />
= 2δn 2 + 2δm 2 − 2 2 1<br />
2 (yn + ym) − x 2 .<br />
2 (yn + ym) ∈ M so that 1<br />
2 (yn + ym) − x2 ≥ δ2 . As n and m tends to infinity, we have<br />
yn − ym ≥ 0, which implies that (yn) is a Cauchy sequence. Now since M is complete, the<br />
sequence (yn) will c<strong>on</strong>verge to a limit say y0 ∈ M so that x − y0 ≥ δ. Also<br />
x − y0 = x − yn + yn − y0 ≤ x − yn + yn − y0<br />
= δn + yn − y0.<br />
As n tends to infinity, yn − y0 tends to zero and δn tends to δ. Hence x − y0 ≤ δ and we<br />
c<strong>on</strong>clude that x − y0 = δ.<br />
Uniqueness<br />
We assume that y1 ∈ M and y2 ∈ M both satisfy x − y1 = δ and x − y2 = δ. We show<br />
that y1 = y2. By the parallelogram equality,<br />
y1 − y2 2 = (y1 − x) − (y2 − x) 2<br />
= 2y1 − x 2 + 2y2 − x 2 − (y1 − x) + (y2 − x) 2<br />
= 2δ 2 + 2δ 2 − 2 2 1<br />
2 (y1 + y2) − x 2<br />
But 1<br />
2 (y1 + y2) ∈ M so that 1<br />
2 (y1 + y2) − x 2 ≥ δ 2 , which implies that<br />
Hence<br />
2δ 2 + 2δ 2 − 2 2 1<br />
2 (y1 + y2) − x 2 ≤ 4δ 2 − 4δ 2 .<br />
y1 − y2 ≤ 0.<br />
Clearly y1 − y2 ≥ 0, which means that y1 = y2.
Secti<strong>on</strong> 2.3. Orthog<strong>on</strong>al complement and direct sum Page 7<br />
Theorem 2.3.2 ([Kre78], 3·3-2 Lemma (Orthog<strong>on</strong>ality)) In Theorem 2.3.1, let M be a complete<br />
subspace Y and x ∈ X fixed. Then z = x − y is orthog<strong>on</strong>al to Y.<br />
Proof.<br />
Suppose by c<strong>on</strong>tradicti<strong>on</strong> that z is not orthog<strong>on</strong>al to Y. Then there would be a y1 ∈ Y such that<br />
y1 = 0 since otherwise 〈z, y1〉 = 0.<br />
Now for any scalar α,<br />
z − αy1 2 = 〈z − αy1, z − αy1〉<br />
〈z, y1〉 = β = 0. (2.12)<br />
= 〈z, z〉 − ¯α〈z, y1〉 − α[〈y1, z〉 − ¯α〈y1, y1〉]<br />
= 〈z, z〉 − ¯αβ − α[ ¯ β − ¯α〈y1, y1〉]. (2.13)<br />
Let us choose α = β<br />
〈y1,y1〉 , so that ¯α = ¯ β<br />
〈y1,y1〉 . Then the expressi<strong>on</strong> ¯ β − ¯α〈y1, y1〉 becomes zero.<br />
From Theorem 2.3.1, we have that z = x − y = δ for some y ∈ Y, so that equati<strong>on</strong> (2.13)<br />
becomes<br />
z − αy1 2 = z 2 − |β|2<br />
〈y1, y1〉 < δ2 .<br />
But this is impossible since z − αy1 = x − y2, where y2 = y + αy1 ∈ Y so that z − αy1 ≥ δ,<br />
by definiti<strong>on</strong> of δ. Hence equati<strong>on</strong> (2.12) cannot hold, and the lemma is proved.<br />
Definiti<strong>on</strong> 2.3.3 ([Erd80], Orthog<strong>on</strong>al complement)<br />
Given a subspace M of H, the orthog<strong>on</strong>al complement M ⊥ is defined by<br />
M ⊥ = {x ∈ X : 〈x, m〉 = 0 for all m ∈ M } . (2.14)<br />
Lemma 2.3.4 ([Erd80], 1·9 Lemma) For subsets A and B of a <strong>Hilbert</strong> space H, if B ⊃ A, then<br />
(i) A ⊂ A ⊥⊥ ,<br />
(ii) B ⊥ ⊂ A ⊥ ,<br />
(iii) A ⊥⊥⊥ = A ⊥ .<br />
Proof.<br />
(i) Let a ∈ A. Then a ⊥ A ⊥ and by Definiti<strong>on</strong> 2.3.3 a ∈ A ⊥⊥ . Hence A ⊂ A ⊥⊥ .<br />
(ii) Let y ∈ B ⊥ and a ∈ A. Then A ⊂ B implies that a ∈ B. Thus y ⊥ a. Hence y ∈ A ⊥ .<br />
(iii) Applying (i) to A ⊥ gives A ⊥ ⊂ A ⊥⊥⊥ . Also from (i), A ⊂ A ⊥⊥ and applying (ii) gives<br />
A ⊥ ⊃ A ⊥⊥⊥ . Hence A ⊥⊥⊥ = A ⊥ .
Secti<strong>on</strong> 2.3. Orthog<strong>on</strong>al complement and direct sum Page 8<br />
Definiti<strong>on</strong> 2.3.5 (([Kre78], 3·3-3 Definiti<strong>on</strong> (Direct Sum)) A vector space X is said to be the<br />
direct sum of two subspaces Y and Z of X if each x ∈ X has a unique representati<strong>on</strong><br />
We denote the direct sum of Y and Z by<br />
x = y + z, with y ∈ Y and z ∈ Z. (2.15)<br />
X = Y ⊕ Z. (2.16)<br />
Our main interest is to represent a <strong>Hilbert</strong> space H as a direct sum of a closed subspace Y and<br />
its orthog<strong>on</strong>al complement Y ⊥ = {z ∈ H |z ⊥ Y}. This leads us to the next theorem which is<br />
sometimes called the projecti<strong>on</strong> theorem. Before we proceed, we will state and prove two useful<br />
lemmas.<br />
Lemma 2.3.6 If A is a subspace of a <strong>Hilbert</strong> space H, then A ⊥ is closed.<br />
Proof. Clearly, A ⊥ is a vector subspace. To show that it is closed, let tn ∈ A ⊥ be a sequence<br />
c<strong>on</strong>verging to t. Then by the c<strong>on</strong>tinuity of inner product, Theorem 2.2.7, for all m ∈ A,<br />
〈t, m〉 = limn→∞〈tn, m〉 = 0 so that t ∈ A ⊥ .<br />
Lemma 2.3.7 ([Erd80], Corollary 1·8) If N is a closed subspace of a <strong>Hilbert</strong> space H, then<br />
N ⊥ = {0} ⇔ N = H. (2.17)<br />
Proof.<br />
Clearly if N = H then N ⊥ = {0}. Now suppose that N ⊥ = {0} and N = H. Take h /∈ N.<br />
Then there exists n0 ∈ N such that d(h, N) = h − n0 = 0. Moreover by Theorem 2.3.2,<br />
0 = h − n0 ⊥ N, so that N ⊥ = {0} which c<strong>on</strong>tradicts our assumpti<strong>on</strong> that N ⊥ = {0}. Hence<br />
N = H.<br />
Theorem 2.3.8 ([Erd80], Theorem 1·11) Let N be any closed subspace of a <strong>Hilbert</strong> space H.<br />
Then<br />
H = N ⊕ N ⊥ . (2.18)<br />
Proof.<br />
If N is a closed subspace of H, then by Theorem 2.3.8, N ⊕ N ⊥ is also a closed subspace of<br />
H. Also if x ∈ (N ⊕ N ⊥ ) ⊥ then x ∈ N ⊥ ∩ N ⊥⊥ so that x = {0}. Hence from Theorem 2.3.7,<br />
N ⊕ N ⊥ = H.<br />
Corollary 2.3.9 A subspace A of a <strong>Hilbert</strong> space is closed if and <strong>on</strong>ly if A = A ⊥⊥ .
Secti<strong>on</strong> 2.3. Orthog<strong>on</strong>al complement and direct sum Page 9<br />
Proof.<br />
Let A be a closed subspace of a <strong>Hilbert</strong> space H. Then by Lemma 2.3.4 (i), A ⊂ A ⊥⊥ .<br />
Let x ∈ A ⊥⊥ . Since A is a closed subspace of H, then by Theorem 2.3.8, H = A ⊕ A ⊥ and for<br />
some y ∈ A and z ∈ A ⊥ we have that x = y + z. Now since A ⊂ A ⊥⊥ , we have that y ∈ A ⊥⊥<br />
and z = x − y ∈ A ⊥⊥ . But z ∈ A ⊥ , so that z = 0. Hence x = y ∈ A and A ⊥⊥ = A.<br />
C<strong>on</strong>versely let A = A ⊥⊥ . Since A ⊥ ⊂ H, by Lemma 2.3.6 we have that (A ⊥ ) ⊥ = A is closed.
3. <str<strong>on</strong>g>Operators</str<strong>on</strong>g><br />
3.1 Functi<strong>on</strong>als<br />
Definiti<strong>on</strong> 3.1.1 ([Red53], 3·2 (Linear functi<strong>on</strong>al)) A linear functi<strong>on</strong>al <strong>on</strong> a <strong>Hilbert</strong> space H is<br />
a linear map from H to C. That is<br />
ϕ : H → C. (3.1)<br />
Definiti<strong>on</strong> 3.1.2 ([Red53], 5·2 (Bounded linear functi<strong>on</strong>al)) A linear functi<strong>on</strong>al ϕ is bounded,<br />
or c<strong>on</strong>tinuous, if there exists a c<strong>on</strong>stant k such that<br />
.<br />
The norm of a bounded linear functi<strong>on</strong>al is<br />
If y ∈ H, then<br />
is a bounded linear functi<strong>on</strong>al <strong>on</strong> H, with ϕy = y.<br />
Example 3.1.3 The functi<strong>on</strong> ϕ : L 2 (a, b) → R defined by<br />
|ϕ(x)| ≤ kx for all x ∈ H. (3.2)<br />
ϕ = sup |ϕ(x)|. (3.3)<br />
x=1<br />
ϕy(x) = 〈y, x〉 (3.4)<br />
ϕ(u) =<br />
b<br />
a<br />
ϕ u(x)dx. (3.5)<br />
is a functi<strong>on</strong>al. Where L2 <br />
(a, b) = u integrable | b<br />
a |ϕ|2 <br />
≤ +∞ and 〈ϕ, u〉 = b<br />
ϕ u(x)dx<br />
a<br />
defines an inner product <strong>on</strong> L2 . Thus for every u ∈ L2 <br />
b<br />
(a, b), ϕL2 = a ϕ2u2 (x)dx. Clearly,<br />
for ϕ = 1, ϕ is a linear functi<strong>on</strong>al since 〈ϕ, αu+βv〉 = α〈ϕ, u〉+β〈ϕ, v〉. Further more, applying<br />
the Cauchy-Schwarz inequality <strong>on</strong> L2 we have<br />
<br />
<br />
b <br />
|〈ϕ, u〉| = <br />
1 · u(x)dx<br />
≤ 1L2uL2 = |b − a|uL2. Hence ϕ is bounded.<br />
a<br />
Remark 3.1.4 One of the fundamental facts about <strong>Hilbert</strong> spaces is that all bounded linear<br />
functi<strong>on</strong>als are of the form given in equati<strong>on</strong> (3.4). This is the basis for the next theorem.<br />
10
Secti<strong>on</strong> 3.2. Riesz Representati<strong>on</strong> of functi<strong>on</strong>als Page 11<br />
3.2 Riesz Representati<strong>on</strong> of functi<strong>on</strong>als<br />
Theorem 3.2.1 ([Red53], 5·4 Theorem (Riesz representati<strong>on</strong> theorem)) Every bounded linear<br />
functi<strong>on</strong>al ϕ <strong>on</strong> a <strong>Hilbert</strong> space H can be represented in terms of the inner product<br />
ϕ(x) = 〈x, u〉, (3.6)<br />
where u depends <strong>on</strong> ϕ and is uniquely determined by ϕ and has norm ϕ = u.<br />
We want to prove that<br />
(a) ϕ has a representati<strong>on</strong> ϕ(x) = 〈x, u〉,<br />
(b) u is unique,<br />
(c) ϕ = u holds.<br />
Proof.<br />
If ϕ = 0 then (a), (b) and (c) hold, if we take u = 0. So suppose that ϕ = 0.<br />
(a) Now N(ϕ) is a closed subspace of H. Furthermore since ϕ = 0, we have that N(ϕ) = H,<br />
which implies that N(ϕ) ⊥ = 0 by Theorem 2.3.7. Hence there must be at least <strong>on</strong>e n<strong>on</strong>zero<br />
element, say u0 in N(ϕ) ⊥ .<br />
Now set z = ϕ(x)u0 − ϕ(u0)x, then apply ϕ to give ϕ(z) = ϕ(x)ϕ(u0) − ϕ(u0)ϕ(x) =<br />
0, and so z ∈ N(ϕ). Also, since u0 ∈ N(ϕ) ⊥ we have,<br />
where<br />
which implies that<br />
If we set<br />
then ϕ(x) = 〈x, u〉. Hence (a) is proved.<br />
0 = 〈z, u0〉 = 〈ϕ(x)u0 − ϕ(u0)x, u0〉<br />
= ϕ(x)〈u0, u0〉 − ϕ(u0)〈x, u0〉<br />
〈u0, u0〉 = u0 2 = 0,<br />
ϕ(x) = ϕ(u0)〈x, u0〉<br />
u02 .<br />
u = ϕ(u0)u0<br />
,<br />
u02 (b) To prove that u is unique, suppose that for all x ∈ H we have,<br />
Then<br />
ϕ(x) = 〈x, u1〉 = 〈x, u2〉.<br />
〈x, u1 − u2〉 = 0 for all x.
Secti<strong>on</strong> 3.3. Sesquilinear functi<strong>on</strong>als and Riesz representati<strong>on</strong> Page 12<br />
If we choose x such that x = u1 − u2, then<br />
〈x, u1 − u2〉 = 〈u1 − u2, u1 − u2〉 = u1 − u2 2 = 0.<br />
Hence u1 − u2 = 0, which implies that u1 = u2. Hence u is unique.<br />
(c) Let ϕ = 0. Then u = 0. From (a) with x = u,<br />
which implies that<br />
By Cauchy-Schwarz inequality and (a),<br />
u 2 = 〈u, u〉 = ϕ(u) ≤ ϕu<br />
u ≤ ϕ. (3.7)<br />
|ϕ(x)| = |〈x, u〉| ≤ xu<br />
which implies that ϕ = sup |〈x, u〉| ≤ u.<br />
x=1<br />
Hence<br />
ϕ ≤ u. (3.8)<br />
Combining equati<strong>on</strong>s (3.7) and (3.8) gives<br />
which yields (c).<br />
u = ϕ, (3.9)<br />
Remark 3.2.2 The Riesz theorem is named after Frigyes Riesz a Hungarian mathematician.<br />
This representati<strong>on</strong> is important in the theory of operators <strong>on</strong> <strong>Hilbert</strong> spaces. In particular it is<br />
important in the representati<strong>on</strong> of the <strong>Hilbert</strong>-adjoint operator of a bounded linear operator. In<br />
the mathematical treatment of quantum mechanics, the theorem can be seen as a justificati<strong>on</strong><br />
for the popular bra-ket notati<strong>on</strong>. When the theorem holds, every ket |ψ〉 has a corresp<strong>on</strong>ding bra<br />
〈ψ|.<br />
3.3 Sesquilinear functi<strong>on</strong>als and Riesz representati<strong>on</strong><br />
Definiti<strong>on</strong> 3.3.1 ([Kre78], 3·8-3 Definiti<strong>on</strong> (Sesquilinear functi<strong>on</strong>al)) Let X and Y be vector<br />
spaces over the same field K (R or C). A sesquilinear form (or sesquilinear functi<strong>on</strong>al) is defined<br />
as a mapping<br />
h : X × Y → K (3.10)<br />
such that for all x, x1, x2 ∈ X and y, y1, y2 ∈ Y and all scalars α and β,<br />
1. h(x1 + x2, y) = h(x1, y) + h(x2, y),
Secti<strong>on</strong> 3.3. Sesquilinear functi<strong>on</strong>als and Riesz representati<strong>on</strong> Page 13<br />
2. h(x, y1 + y2) = h(x, y1) + h(x, y2),<br />
3. h(αx, y) = αh(x, y),<br />
4. h(x, βy) = ¯ βh(x, y).<br />
Remark 3.3.2 Notice that h is linear in the first argument and c<strong>on</strong>jugate linear in the sec<strong>on</strong>d<br />
<strong>on</strong>e. If X and Y are real then (4) becomes h(x, βy) = βh(x, y). Clearly, h is bilinear in both<br />
arguments.<br />
Definiti<strong>on</strong> 3.3.3 ([Swa97], Definiti<strong>on</strong> 8 (Bounded sesquilinear functi<strong>on</strong>al)) A sesquilinear functi<strong>on</strong>al<br />
h is bounded if there exists K > 0 such that<br />
The norm of of h is defined to be<br />
|h(x, y)| ≤ Kxy, for every x ∈ X, y ∈ Y. (3.11)<br />
|h(x, y)|<br />
h = sup<br />
x=0 xy<br />
y=0<br />
= sup<br />
x=1<br />
y=1<br />
|h(x, y)|. (3.12)<br />
It follows that |h(x, y)| ≤ hxy for all x ∈ X, y ∈ Y. The inner product is sesquilinear<br />
and it is bounded.<br />
Lemma 3.3.4 ([Kre78], 3·8-2 Lemma (Equality))<br />
If 〈x1, y〉 = 〈x2, y〉 for all y in an inner product space X, then x1 = x2. In particular if 〈x1, w〉 = 0<br />
for all w ∈ X, then x1 = 0.<br />
Proof.<br />
For all y,<br />
〈x1 − x2, y〉 = 〈x1, y〉 − 〈x2, y〉 = 0.<br />
If we choose y = x1 − x2, then x1 − x2 2 = 0. Hence we have x1 − x2 = 0 implying that<br />
x1 = x2. In particular, if 〈x1, y〉 = 0 with y = x1 we get x1 2 = 0, giving x1 = 0.<br />
Theorem 3.3.5 ([Kre78], 3·8-4 Theorem (Riesz representati<strong>on</strong>))<br />
Let H1, H2 be <strong>Hilbert</strong> spaces and let h : H1 × H2 → K be a bounded sesquilinear form. Then h<br />
has a representati<strong>on</strong><br />
h(x, y) = 〈Sx, y〉 for x ∈ H1, y ∈ H2<br />
(3.13)<br />
where S : H1 → H2 is a bounded linear operator and S is uniquely determined and has norm<br />
S = h.
Secti<strong>on</strong> 3.3. Sesquilinear functi<strong>on</strong>als and Riesz representati<strong>on</strong> Page 14<br />
We will apply Riesz Theorem 3.2.1 in the proof of this theorem. We will show that<br />
(a) h has a representati<strong>on</strong> h(x, y) = 〈Sx, y〉,<br />
(b) h = S, and<br />
(c) S is unique.<br />
Proof.<br />
(a) Let us take a fixed x and then we will show that φ : y → h(x, y) is a linear map. Suppose<br />
φ(y) = h(x, y), then by Definiti<strong>on</strong> 3.3.1,<br />
Moreover,<br />
φ(y1 + y2) = h(x, y1 + y2)<br />
= h(x, y1) + h(x, y2)<br />
= φ(y1) + φ(y2).<br />
φ(αy) = h(x, αy)<br />
= αh(x, y)<br />
= αh(x, y)<br />
= αφ(y).<br />
Hence h(x, y) is linear.<br />
Applying Theorem 3.2.1 yields a representati<strong>on</strong> h(x, y) = 〈y, z〉. Hence<br />
h(x, y) = 〈z, y〉. (3.14)<br />
Our z ∈ H2 is unique but depends <strong>on</strong> our fixed x ∈ H1. Equati<strong>on</strong> (3.14) with variable x defines<br />
an operator S : H1 → H2 given by z = Sx. Hence equati<strong>on</strong> (3.14) becomes h(x, y) = 〈Sx, y〉,<br />
giving (3.13).<br />
Now let us show that S is linear. From equati<strong>on</strong> (3.13), and the sesquilinearity we obtain,<br />
for all y in H2, so that by Lemma 3.3.4,<br />
〈S(αx1 + βx2), y〉 = h(αx1 + βx2, y)<br />
= αh(x1, y) + βh(x2, y)<br />
= α〈Sx1, y〉 + β〈Sx2, y〉<br />
= 〈αSx1 + βSx2, y〉<br />
S(αx1 + βx2) = αSx1 + βSx2.<br />
Hence S is linear.<br />
(b) We show that S is bounded. Leaving the trivial case S = 0, we have<br />
|〈Sx, y〉|<br />
h = sup<br />
x=0 xy<br />
y=0<br />
|〈Sx, Sx〉|<br />
≥ sup<br />
x=0 xSx<br />
Sx=0<br />
Sx<br />
= sup<br />
x=0 x<br />
= S. (3.15)
Secti<strong>on</strong> 3.4. <strong>Hilbert</strong>-adjoint and self-adjoint operators Page 15<br />
Now by Cauchy-Schwarz inequality,<br />
so that<br />
|〈Sx, y〉|<br />
sup<br />
x=0 xy<br />
y=0<br />
Sxy<br />
≤ sup<br />
x=0 xy<br />
y=0<br />
= S, (3.16)<br />
h ≤ S. (3.17)<br />
Combining equati<strong>on</strong>s (3.15) and (3.17) gives h = S.<br />
(c) Let us assume that T : H1 → H2 is a linear operator such that for all x ∈ H1 and y ∈ H2,<br />
we have<br />
h(x, y) = 〈Sx, y〉 = 〈Tx, y〉.<br />
Then, by Lemma 3.3.4, Sx = Tx for all x ∈ H1. Hence S = T by definiti<strong>on</strong>.<br />
Remark 3.3.6 An important c<strong>on</strong>sequence of the Riesz representati<strong>on</strong> theorem is the existence<br />
of the adjoint of a bounded operator <strong>on</strong> a <strong>Hilbert</strong> space.<br />
3.4 <strong>Hilbert</strong>-adjoint and self-adjoint operators<br />
Definiti<strong>on</strong> 3.4.1 ([Kre78], 3·9-1 Definiti<strong>on</strong> (<strong>Hilbert</strong>-adjoint operator, T ∗ ))<br />
Given two <strong>Hilbert</strong> spaces, H1 and H2, let T : H1 → H2 be a bounded linear operator. Then the<br />
<strong>Hilbert</strong>-adjoint operator T ∗ of T is the operator T ∗ : H2 → H1 such that for all x ∈ H1 and<br />
y ∈ H2,<br />
〈Tx, y〉 = 〈x, T ∗ y〉. (3.18)<br />
Let us show that there exists such an operator T ∗ .<br />
Theorem 3.4.2 ([Kre78], 3·9-2 Theorem (Existence)) The <strong>Hilbert</strong>-adjoint operator T ∗ of T<br />
exist, is unique and is a bounded linear operator with norm<br />
Proof.<br />
C<strong>on</strong>sider<br />
T ∗ = T. (3.19)<br />
h(y, x) = 〈y, Tx〉. (3.20)
Secti<strong>on</strong> 3.4. <strong>Hilbert</strong>-adjoint and self-adjoint operators Page 16<br />
We will show that h is sesquilinear. Now h is linear in the first argument and c<strong>on</strong>jugate linear in<br />
the sec<strong>on</strong>d argument since,<br />
h(y, αx1 + βx2) = 〈y, T(αx1 + βx2)〉<br />
= 〈y, αTx1 + βTx2〉<br />
= ¯α〈y, Tx1〉 + ¯ β〈y, Tx2〉<br />
= ¯αh(y, x1) + ¯ βh(y, x2).<br />
Hence since the inner product is sesquilinear, we c<strong>on</strong>clude that h is sesquilinear. By the Cauchy-<br />
Schwarz inequality,<br />
|h(y, x)| = |〈y, Tx〉| ≤ yTx ≤ Txy<br />
which implies that |h(y,x)|<br />
xy<br />
Also,<br />
≤ T and by equati<strong>on</strong> (3.12) we have that<br />
|〈y, Tx〉|<br />
h = sup<br />
x=0 yx<br />
y=0<br />
h ≤ T. (3.21)<br />
|〈Tx, Tx〉|<br />
≥ sup<br />
x=0 Txx<br />
Tx=0<br />
= T. (3.22)<br />
Combining equati<strong>on</strong>s (3.21) and (3.22) gives h = T. From Theorem 3.3.5, substituting T ∗<br />
for S, we have<br />
h(y, x) = 〈T ∗ y, x〉, (3.23)<br />
where T ∗ : H2 → H1 is a uniquely determined, bounded linear operator with norm<br />
Combining (3.20) and (3.23), we get<br />
Taking the c<strong>on</strong>jugate gives equati<strong>on</strong> (3.18).<br />
T ∗ = h = T. (3.24)<br />
〈y, Tx〉 = 〈T ∗ y, x〉.<br />
Lemma 3.4.3 ([Kre78], 3·9-3 Lemma (Zero operator)) Let X and Y be inner product spaces<br />
and T : X → Y a bounded linear operator. Then:<br />
1. T = 0 if and <strong>on</strong>ly if 〈Tx, y〉 = 0 for all x ∈ X and y ∈ Y.<br />
2. For X a complex vector space, if T : X → X and 〈Tx, x〉 = 0 for all x ∈ X, then T = 0.<br />
Proof.<br />
1. If T = 0, then for all x ∈ X, Tx = 0, and for any u ∈ X we have,<br />
〈Tx, y〉 = 〈0, y〉 = 0〈u, y〉 = 0.
Secti<strong>on</strong> 3.4. <strong>Hilbert</strong>-adjoint and self-adjoint operators Page 17<br />
Now let<br />
〈Tx, y〉 = 0 for all x ∈ X, y ∈ Y.<br />
Then by Lemma 3.3.4, Tx = 0 for all x ∈ X and T = 0.<br />
2. If 〈Tx, x〉 = 0 for all x ∈ X, then for w = αx + y ∈ X we have<br />
〈Tw, w〉 = 〈T(αx + y), αx + y〉<br />
Now if we choose α = 1, then (3.25) becomes<br />
= |α| 2 〈Tx, x〉 + 〈Ty, y〉 + α〈Tx, y〉 + ¯α〈Ty, x〉. (3.25)<br />
〈Tw, w〉 = 〈Tx, x〉 + 〈Ty, y〉 + 〈Tx, y〉 + 〈Ty, x〉. (3.26)<br />
Now 〈Tx, x〉 and 〈Ty, y〉 are equal to zero by our assumpti<strong>on</strong>. Hence equati<strong>on</strong> (3.26) becomes<br />
If we choose α = i, then ¯α = −i and equati<strong>on</strong> (3.25) becomes<br />
〈Tx, y〉 + 〈Ty, x〉 = 0. (3.27)<br />
〈Tx, y〉 − 〈Ty, x〉 = 0. (3.28)<br />
Adding equati<strong>on</strong>s (3.27) and (3.28) gives 〈Tx, y〉 = 0 and T = 0 follows from 1.<br />
Theorem 3.4.4 ([Kre78], 3·9-4 Theorem (Properties of the <strong>Hilbert</strong> adjoint-operator)) Let H1<br />
and H2 be <strong>Hilbert</strong> spaces, S : H1 → H2 and T : H1 → H2 be bounded linear operators and β<br />
any scalar. Then<br />
1. 〈T ∗ y, x〉 = 〈y, Tx〉 for any x ∈ H1, y ∈ H2,<br />
2. (S + T) ∗ = S ∗ + T ∗ ,<br />
3. (βT) ∗ = ¯ βT ∗ ,<br />
4. (T ∗ ) ∗ = T,<br />
5. T ∗ T = TT ∗ = T 2 ,<br />
6. T ∗ T = 0 if and <strong>on</strong>ly if T = 0,<br />
7. (ST) ∗ = T ∗ S ∗ (if H2 = H1).<br />
Proof.<br />
1. By Definiti<strong>on</strong> 3.4.1, we have<br />
〈T ∗ y, x〉 = 〈x, T ∗ y〉 = 〈Tx, y〉 = 〈y, Tx〉. (3.29)
Secti<strong>on</strong> 3.4. <strong>Hilbert</strong>-adjoint and self-adjoint operators Page 18<br />
2. By the definiti<strong>on</strong> of the <strong>Hilbert</strong>-adjoint operator in Definiti<strong>on</strong> (3.4.1), for all x and y,<br />
〈x, (S + T) ∗ y〉 = 〈(S + T)x, y〉<br />
= 〈Sx, y〉 + 〈Tx, y〉<br />
= 〈x, S ∗ y〉 + 〈x, T ∗ y〉<br />
= 〈x, (S ∗ + T ∗ )y〉.<br />
Thus it follows from Lemma 3.3.4 that, (S + T) ∗ y = (S ∗ + T ∗ )y for all y ∈ H2, so that<br />
(S + T) ∗ = S ∗ + T ∗ .<br />
3. By Definiti<strong>on</strong> (3.4.1),<br />
〈(βT) ∗ y, x〉 = 〈y, (βT)x〉<br />
= 〈y, β(Tx)〉<br />
= ¯ β〈y, Tx〉<br />
= ¯ β〈T ∗ y, x〉<br />
= 〈 ¯ βT ∗ y, x〉.<br />
Hence by Lemma 3.4.3, (βT) ∗ y = ¯ βT ∗ y for all y ∈ H2, which implies that (βT) ∗ = ¯ βT ∗ .<br />
4. By Definiti<strong>on</strong> 3.4.1 and from 1 we have, 〈(T ∗ ) ∗ x, y〉 = 〈x, T ∗ y〉 = 〈Tx, y〉 so that 〈((T ∗ ) ∗ −<br />
T)x, y〉 and by Lemma 3.4.3, we have (T ∗ ) ∗ = T.<br />
5. We know that T ∗ T : H1 → H1 and TT ∗ : H2 → H2. By the Cauchy-Schwarz inequality<br />
and by the definiti<strong>on</strong> of the <strong>Hilbert</strong>-adjoint operator in Definiti<strong>on</strong> (3.4.1) we have<br />
Tx 2 = 〈Tx, Tx〉 = 〈T ∗ Tx, x〉 ≤ T ∗ Txx ≤ T ∗ Tx 2 .<br />
Taking the supremum over all x of norm 1 we obtain T 2 ≤ T ∗ T. Now by Theorem 3.4.2,<br />
we have T ∗ T ≤ T ∗ T = T 2 . Hence T ∗ T = T 2 . We substitute T ∗ for T to get<br />
T ∗∗ T ∗ = T ∗ 2 = T 2 , But by 4 we have (T ∗ ) ∗ = T, so that TT ∗ = T 2 .<br />
6. From 5, if T ∗ T = 0, then T = 0 and c<strong>on</strong>versely if T = 0, then T ∗ T = 0.<br />
7. By Definiti<strong>on</strong> 3.4.1, 〈x, (ST) ∗ y〉 = 〈(ST)x, y〉 = 〈Tx, S ∗ y〉 = 〈x, T ∗ S ∗ y〉. Hence by Lemma<br />
3.3.4 we obtain (ST) ∗ y = T ∗ S ∗ y for all y ∈ H2, so that (ST) ∗ = T ∗ S ∗ .<br />
Definiti<strong>on</strong> 3.4.5 ([Kre78], 3·10-1 (Self-adjoint operator)) A bounded linear operator T : H → H<br />
<strong>on</strong> a <strong>Hilbert</strong> space H is said to be self-adjoint or Hermitian if<br />
Equivalently, a bounded linear operator T is said to be self-adjoint if<br />
T ∗ = T. (3.30)<br />
〈x, Ty〉 = 〈Tx, y〉 for all x, y ∈ H. (3.31)
Secti<strong>on</strong> 3.4. <strong>Hilbert</strong>-adjoint and self-adjoint operators Page 19<br />
Example 3.4.6 A linear map <strong>on</strong> R n with matrix A is a self-adjoint if and <strong>on</strong>ly if A is symmetric<br />
(A = A T ). A linear map <strong>on</strong> C n with matrix A is self-adjoint if and <strong>on</strong>ly if A is Hermitian<br />
(A = A ∗ ).<br />
Remark 3.4.7 Self-adjoint operators <strong>on</strong> <strong>Hilbert</strong> spaces are used in quantum mechanics to represent<br />
physical observables such as positi<strong>on</strong>, momentum, angular momentum and spin. An<br />
important example is the Hamilt<strong>on</strong>ian operator given by<br />
Hψ = − 2<br />
2m ∇2 ψ + V ψ.<br />
Definiti<strong>on</strong> 3.4.8 ([Kre78], 3·10-1 (Unitary operator)) A bounded linear operator T : H → H<br />
<strong>on</strong> a <strong>Hilbert</strong> space H is said to be unitary if T is bijective and<br />
Hence<br />
TT ∗ = T ∗ T. (3.32)<br />
T ∗ = T −1 . (3.33)<br />
Definiti<strong>on</strong> 3.4.9 ([Kre78], 3·10-1 (Normal operators)) A bounded linear operator T : H → H<br />
<strong>on</strong> a <strong>Hilbert</strong> space H is said to be normal if<br />
TT ∗ = T ∗ T. (3.34)<br />
Remark 3.4.10 If T is self-adjoint or unitary, then T is normal, but the c<strong>on</strong>verse is not generally<br />
true. For example if I : H → H is the identity operator, then T = 2iI is normal since T ∗ = −2iI,<br />
so that TT ∗ = T ∗ T = 4I but T ∗ = T and T ∗ = T −1 = − 1<br />
2 iI.<br />
Theorem 3.4.11 ([Kre78], 3·10-3 Theorem (Self-Adjointness)) Let T : H → H be a bounded<br />
linear operator <strong>on</strong> a <strong>Hilbert</strong> space H. Then<br />
1. If T is self-adjoint, then 〈Tx, x〉 ∈ R for all x ∈ H.<br />
2. If H is complex and 〈Tx, x〉 ∈ R for all x ∈ H, then the operator T is self-adjoint.<br />
Proof.<br />
1. If T is self-adjoint, then for all x,<br />
〈Tx, x〉 = 〈x, Tx〉. (3.35)<br />
By definiti<strong>on</strong>, 〈Tx, y〉 = 〈x, T ∗ y〉 and since T is self-adjoint, we have<br />
Combining equati<strong>on</strong>s (3.35) and (3.36) gives<br />
〈Tx, x〉 = 〈x, Tx〉. (3.36)<br />
〈Tx, x〉 = 〈Tx, x〉.<br />
Hence 〈Tx, x〉 is equal to its complex c<strong>on</strong>jugate which implies that it is real.
Secti<strong>on</strong> 3.4. <strong>Hilbert</strong>-adjoint and self-adjoint operators Page 20<br />
2. If 〈Tx, x〉 ∈ R for all x ∈ H, then<br />
Hence<br />
〈Tx, x〉 = 〈Tx, x〉 = 〈x, T ∗ x〉 = 〈T ∗ x, x〉.<br />
0 = 〈Tx, x〉 − 〈T ∗ x, x〉 = 〈(T − T ∗ )x, x〉<br />
and by Lemma 3.4.3, T − T ∗ = 0. Therefore T = T ∗ .<br />
Theorem 3.4.12 ([Kre78], 3·10-5 Theorem (Sequence of self-adjoint operators)) Let Tn be a<br />
sequence of bounded self-adjoint linear operators Tn : H → H <strong>on</strong> a <strong>Hilbert</strong> space H. If Tn<br />
c<strong>on</strong>verges to T, then T is a bounded self-adjoint linear operator.<br />
Proof.<br />
If Tn → T, then Tn − T → 0. Now by Theorem 3.4.4 and Theorem 3.4.2 we have that<br />
Tn ∗ − T ∗ = (Tn − T) ∗ = Tn − T, so that<br />
T − T ∗ ≤ T − Tn + Tn − T ∗<br />
n + T ∗<br />
n − T ∗ <br />
= T − Tn + Tn − T = 2Tn − T.<br />
As n tends to infinity, Tn − T tends to zero. Hence T − T ∗ = 0 which implies that T ∗ = T.<br />
Hence T is self-adjoint.
4. Spectral Analysis of a Self-Adjoint<br />
Operator<br />
Definiti<strong>on</strong> 4.0.13 ([Aup91], 2·2·4 Definiti<strong>on</strong> (Spectrum of T)) Let L(H) denote the set of all<br />
bounded linear operators <strong>on</strong> H and let T ∈ L(H). We define the spectrum of T as the set of<br />
λ ∈ C such that T − λI is not invertible in L(H). It is denoted by Sp T. So λ ∈ SpT if and<br />
<strong>on</strong>ly if at least <strong>on</strong>e of the following c<strong>on</strong>diti<strong>on</strong>s hold.<br />
1. the range of T − λI is not all of H, that is T − λI is not surjective.<br />
2. T − λI is not injective.<br />
Hence<br />
SpT = {λ ∈ C : T − λI is not invertible in L(H)} .<br />
Remark 4.0.14 The kernel and range of T are denoted by N(T) and R(T) respectively. If 2<br />
holds, then λ is called the eigenvalue of T and N(T − λI) is the corresp<strong>on</strong>ding eigenspace which<br />
is the set of all x ∈ X such that Tx = λx, where x = 0 is called an eigenvector corresp<strong>on</strong>ding<br />
to the the eigenvalue λ.<br />
Definiti<strong>on</strong> 4.0.15 ([Aup91]) The spectral radius of T ∈ L(H) is given by<br />
ρ(T) = max {|λ| : λ ∈ SpT} . (4.1)<br />
Theorem 4.0.16 ([Aup91], Theorem 2·3·1) Let H be a <strong>Hilbert</strong> space and let T ∈ L(H). Then<br />
1. N(T) = R(T ∗ ) ⊥ and N(T ∗ ) = R(T) ⊥ ,<br />
2. I + T ∗ T is invertible in L(H).<br />
Proof.<br />
1. If we choose x ∈ N(T), then for all y ∈ H,<br />
〈0, y〉 = 〈Tx, y〉 = 〈x, T ∗ y〉 = 0,<br />
so that x ∈ R(T ∗ ) ⊥ . Now applying the adjoint ∗ gives<br />
N(T ∗ ) = R(T ∗∗ ) ⊥ ,<br />
21
ut by Theorem 3.4.4 (4), T ∗∗ = T which implies that N(T ∗ ) = R(T) ⊥ .<br />
2.To show that I + T ∗ T is invertible in L(H), we show that I + T ∗ T is<br />
(i) injective and<br />
(ii) show that it is surjective by showing that the range of I + T ∗ T is the whole of H.<br />
(i) C<strong>on</strong>sider T ∗ T and I + T ∗ T which are self-adjoint.<br />
(I + T ∗ T)x 2 = 〈(I + T ∗ T)x, (I + T ∗ T)x〉<br />
= 〈(I + T ∗ T)(I + T ∗ T)x, x〉<br />
= 〈(I + T ∗ T) 2 x, x〉<br />
= 〈I + 2T ∗ T + (T ∗ T) 2 x, x〉<br />
= 〈(x + (2T ∗ T)x + (T ∗ T) 2 )x, x〉<br />
= 〈x, x〉 + 2〈T ∗ (Tx), x〉 + 〈(T ∗ T) 2 x, x〉<br />
= x 2 + 2〈Tx, Tx〉 + 〈T ∗ Tx, T ∗ Tx〉<br />
Page 22<br />
(4.2)<br />
= x 2 + 2Tx 2 + T ∗ Tx 2 ≥ x 2 . (4.3)<br />
It follows that if (I + T ∗ T)x = 0, then x = 0. Hence we c<strong>on</strong>clude that N(I + T ∗ T) = {0}, so<br />
that I + T ∗ T is injective.<br />
(ii) We will first show that R(I + T ∗ T) is closed. Let ((I + T ∗ T)xn) be a Cauchy sequence in<br />
R(I + T ∗ T). Then given ɛ ≥ 0, there exist N such that for n, m ≥ N we have (I + T ∗ T)(xn) −<br />
(I + T ∗ T)(xm) ≤ ɛ. Now from inequality (4.3),<br />
ɛ ≥ (I + T ∗ T)(xn) − (I + T ∗ T)(xm) = (I + T ∗ T)(xn − xm) ≥ xn − xm,<br />
which implies that xn−xm ≤ ɛ. Hence xn is also a Cauchy sequence. Now since H is complete,<br />
xn → x ∈ H. Since I + T ∗ T is bounded, then it is c<strong>on</strong>tinuous. Hence lim (I + T<br />
n→∞ ∗ T)(xn) =<br />
(I + T ∗ T)( lim (xn)) = (I + T<br />
n→∞ ∗ T)x. Hence (I + T ∗ T)x ∈ R(I + T ∗ T) and we c<strong>on</strong>clude that<br />
R(I + T ∗ T) is closed.<br />
Now by direct sum, Theorem 2.3.8, H = R(I + T ∗ T) ⊕ R(I + T ∗ T) ⊥ . But by 1 we have<br />
that R(I + T ∗ T) ⊥ = N(I + T ∗ T) = {0} . Therefore H = R(I + T ∗ T). Hence I + T ∗ T is<br />
surjective.<br />
We c<strong>on</strong>clude from (i) and (ii) that I + T ∗ T is bijective and hence invertible.<br />
Lemma 4.0.17 If S ∈ L(H) where H a <strong>Hilbert</strong> space, S = S ∗ and λ is an eigenvalue of S, then<br />
λ ∈ R.<br />
Proof.<br />
If λ is an eigenvalue of S, then by Remark 4.0.14, S−λI is not injective. Hence N(S−λI) = {0}.<br />
Hence for x ∈ N(S − λI), Sx = λx. Now since S is self-adjoint, then by Theorem 3.4.11 (1),<br />
〈Sx, x〉 ∈ R for all x ∈ H and 〈x, Sx〉 = 〈x, λx〉 = λx 2 ∈ R. Hence λ = λ and we c<strong>on</strong>clude<br />
that λ ∈ R.
Page 23<br />
Lemma 4.0.18 Let S ∈ L(H) where H is a <strong>Hilbert</strong> space. Suppose that Sλ = λI − S with<br />
λ = α + iβ and S = S ∗ . If Sλx ≥ |β|x, with β = 0, then Sλ is invertible.<br />
Proof.<br />
(i) To show that it is injective, let Sλx = 0, then |β|x ≤ 0 which implies that x = 0 for<br />
β = 0. Hence N(Sλ) = {0} and we c<strong>on</strong>clude that Sλ is injective.<br />
(ii) Let us first show that R(Sλ) is closed. If (Sλ(xn)) is a Cauchy sequence so is (xn).<br />
Now since H is complete, xn → x ∈ H. Since Sλ is bounded, then it is c<strong>on</strong>tinuous. Hence<br />
lim<br />
n→∞ (Sλ(xn)) = (Sλ) lim (xn) = (Sλ)x. Hence (Sλ)x ∈ R(Sλ) and we c<strong>on</strong>clude that R(Sλ) is<br />
n→∞<br />
closed.<br />
Now by direct sum, Theorem 2.3.8, H = R(Sλ) ⊕ R(Sλ) ⊥ . By Theorem 4.0.16 (1), we have that<br />
R(Sλ) ⊥ = N(Sλ ∗ ). We will show that Sλ ∗ is injective and deduce that R(Sλ) ⊥ = N(Sλ ∗ ) = {0}.<br />
Now by Theorem 4.0.16 (1), R(Sλ) ⊥ = N(Sλ ∗ ). For β = 0, N((αI + iβI − S) ∗ ) = N(αI −<br />
iβI − S) = {0} otherwise by Lemma 4.0.17, α − iβ would be an eigenvalue. But α − iβ /∈ R,<br />
so that α − iβ cannot be an eigenvalue. This implies that N(Sλ ∗ ) = R(Sλ) ⊥ = {0}. Hence<br />
H = R(Sλ), showing that Sλ is surjective. We c<strong>on</strong>clude that Sλ is bijective hence invertible.<br />
Theorem 4.0.19 ([Kre78], 7·5-5 (Theorem)) If T is a bounded linear operator <strong>on</strong> a complex<br />
Banach space, then<br />
ρ(T) = lim T<br />
n→∞ n 1<br />
n . (4.4)<br />
Remark 4.0.20 The proof of this theorem is not trivial and falls outside the scope of this essay.<br />
Corollary 4.0.21 ([Aup91], Corollary 2·3·2) Let H be a <strong>Hilbert</strong> space and let S be a self-adjoint<br />
operator <strong>on</strong> H. Then<br />
1. S 2 = S 2 and c<strong>on</strong>sequently S = ρ(S).<br />
2. Sp S ⊂ R.<br />
Proof.<br />
1. From Theorem 3.4.4 (5), S ∗ S = S 2 . But S is self-adjoint. Hence S 2 = S 2 and by<br />
inducti<strong>on</strong> we have S 2n<br />
= S2n . Now by Theorem 4.0.19, we have that ρ(S) = lim S<br />
n→∞ n 1<br />
n =<br />
S.<br />
2. Let λ = a + iβ with α, β ∈ R and let Sλ = λI − S for all λ ∈ C. Then<br />
Sα + iβI = αI − S + iβI<br />
= (α + iβ)I − S<br />
= λI − S.
Therefore Sα + iβI = Sλ. Thus for all x ∈ H and S self-adjoint we have,<br />
Sλx 2 = 〈Sλx, Sλx〉<br />
= 〈(Sα + iβI)x, (Sα + iβI)x〉<br />
= 〈Sαx, Sαx〉 + 〈iβx, iβx〉 + 〈Sαx, iβx〉 + 〈iβx, Sαx〉<br />
= 〈Sαx, Sαx〉 + 〈iβx, iβx〉 − iβ〈Sαx, x〉 + iβ〈Sαx, x〉<br />
= 〈Sαx, Sαx〉 + |β| 2 x 2<br />
= Sαx 2 + |β| 2 x 2 ≥ |β| 2 x 2 .<br />
Page 24<br />
Hence Sλx ≥ |β|x, and by Lemma 4.0.18, we c<strong>on</strong>clude that for β = 0, Sλ = λI − S is<br />
invertible and hence Sp S ⊂ R.
5. C<strong>on</strong>clusi<strong>on</strong> and further work<br />
The parallelogram equality gives <strong>Hilbert</strong> spaces an edge over Banach spaces. Moreover orthog<strong>on</strong>ality,<br />
c<strong>on</strong>tinuity of the inner product and completeness are also important features of <strong>Hilbert</strong><br />
spaces.<br />
<str<strong>on</strong>g>Operators</str<strong>on</strong>g> <strong>on</strong> <strong>Hilbert</strong> spaces is a basis for a comprehensive study of the spectral theory. In<br />
particular, the Hermitian or self-adjoint operators are very important in many applicati<strong>on</strong>s. In<br />
this essay, we have seen that the minimising vector theorem, Riesz representati<strong>on</strong> of functi<strong>on</strong>als<br />
and direct sum play important roles in this study. Given an inner product space X and M is<br />
a n<strong>on</strong>empty complete c<strong>on</strong>vex subspace of X , the Minimising Vector Theorem (Theorem 2.3.1)<br />
enables us to find a unique point y ∈ M that is closer to x ∈ X than any other point in M. The<br />
Riesz theorem, Theorem 3.2.1 is important in the representati<strong>on</strong> of <strong>Hilbert</strong>-adjoint operators and<br />
also in quantum physics. The direct sum theorem, Theorem 2.3.8, makes it possible to represent a<br />
<strong>Hilbert</strong> space H as the sum of any closed subspace and its orthog<strong>on</strong>al complement (Lemma 2.3.7).<br />
The algebra of all n × n complex matrices is a special case of an algebra of bounded linear<br />
operators <strong>on</strong> a <strong>Hilbert</strong> space and so ”Hermitian matrices” corresp<strong>on</strong>d to ”self-adjoint operators”.<br />
Moreover the spectrum of an n × n matrix (c<strong>on</strong>sidered as an operator) c<strong>on</strong>sists of exactly all<br />
the eigenvalues of the matrix whereas the spectrum SpT of the self-adjoint operator T c<strong>on</strong>tains<br />
other elements which are not eigenvalues of T. An element, λ ∈ SpT of the spectrum of the<br />
self-adjoint operator is an eigenvalue of T if and <strong>on</strong>ly if T − λI is not injective. It is also well<br />
known that the eigenvalues of a Hermitian matrix are real numbers. In this essay, we have generalised<br />
this idea by showing that the spectrum of a self-adjoint operator <strong>on</strong> a <strong>Hilbert</strong> space also<br />
c<strong>on</strong>sists entirely of real values (Theorem 4.0.21).<br />
So far, we have studied a small aspect of the operators <strong>on</strong> <strong>Hilbert</strong> spaces. However, there<br />
are other interesting facts about the self-adjoint operators like the bounded resoluti<strong>on</strong> of the<br />
identity. Moreover, other aspects of the spectral theory includes<br />
1. Integral equati<strong>on</strong>s, Fredholm theory, compact operators,<br />
2. Sturm-Liouville theory, hydrogen atom,<br />
3. Isospectral theory, Lax pairs,<br />
4. Atiyah-Singer index theorem.<br />
These are interesting facts about the spectral theory that can be studied in future.<br />
25
Acknowledgements<br />
Above all, I thank the God Almighty for His faithfulness. It is the Lord’s doing and it is marvellous<br />
in my eyes.<br />
I would like to appreciate the founder of <strong>AIMS</strong>, professor Neil Turok and its director, professor<br />
Fritz Hahne for giving me this great opportunity to study at <strong>AIMS</strong>.<br />
I wish to express my profound gratitude to my supervisor, Dr S. Mout<strong>on</strong> for her guidance and<br />
interest in my work. I also thank my tutor, Dr Laure Gouba, my essay tutor, Mr Christian<br />
Rivasseau, and all the tutors at <strong>AIMS</strong>. I really appreciate your zeal, which has spurred me to<br />
the peak. Many thanks also goes to Mr Igsaan Kamalie, Mr Jan Groenewald and all staff and<br />
lecturers at <strong>AIMS</strong>.<br />
To my parents, Mr and Mrs R.O. Kajot<strong>on</strong>i, I pray that the good Lord will grant you l<strong>on</strong>g life<br />
to reap the fruits of your labour. I also appreciate the c<strong>on</strong>tributi<strong>on</strong>s of my brothers Kayode<br />
and Babatunde, my sister and her husband, Mr and Mrs Ojerinde and my little niece Eniola.<br />
Furthermore, I appreciate the c<strong>on</strong>tributi<strong>on</strong>s of Mr and Mrs Abiodun Kajot<strong>on</strong>i, Mr and Mrs A.<br />
Eniolorunda and Iyabo Bello for her relentless effort.<br />
I own my fiance, Raphael olushola Folorunsho many thanks for his love, prayers, patience and<br />
c<strong>on</strong>cern.<br />
Finally, I want to appreciate the Nigerian students, juststudents2006, all <strong>AIMS</strong> alumni and others<br />
too numerous to menti<strong>on</strong>. God bless you all.<br />
26
Bibliography<br />
[Aup91] Bernard Aupetit, A primer <strong>on</strong> spectral theory, Springer, New York, 1991.<br />
[Axl97] Sheld<strong>on</strong> Axler, Linear algebra d<strong>on</strong>e right, Undergraduate text in Mathematics, Springer,<br />
San Francisco State University, 1997.<br />
[Erd80] John Erdos, <str<strong>on</strong>g>Operators</str<strong>on</strong>g> <strong>on</strong> hilbert spaces, King’s College L<strong>on</strong>d<strong>on</strong>, 1980.<br />
[Kre78] Erwin Kreyszig, Introductory functi<strong>on</strong>al analysis with applicati<strong>on</strong>s, John Wiley and s<strong>on</strong>s,<br />
New York, 1978.<br />
[Red53] B. Daya Reddy, Introductory functi<strong>on</strong>al analysis with applicati<strong>on</strong>s to boundary value<br />
problems and finite elements, Text in Applied Mathematics, no. 27, Springer, University<br />
of Cape Town, 1953.<br />
[Swa97] Charles Swartz, An introducti<strong>on</strong> to functi<strong>on</strong>al analysis, M<strong>on</strong>ographs and Textbooks, no.<br />
157, Marcel Dekker, Inc. New York, 1997.<br />
27