Professional Documents
Culture Documents
Introduction
These are notes for a Kings College course to fourth year undergraduates and MSc
students. They cover the theoretical development of operators on Hilbert space up
to the spectral theorem for bounded selfadjoint operators. I have tried to make the
treatment as elementary as possible and to include only what is essential to the
main development. The proofs are the simplest and most intuitive that I know. The
exercises are culled from various sources; some of them are more or less original. They
are designed to be worked as the course progresses and, in some cases, they anticipate
results and techniques that come in the later theorems of the course.
It should be emphasized that these notes are not complete. Although the theoretical
development is covered rather fully, examples, illustrations and applications which are
essential for the understanding of the subject, are left to be covered in the lectures.
There are good reasons for doing this. Experience has shown that audiences lose
concentration if they are provided with comprehensive notes which coincide with the
lectures. Also, in many cases examples and such are best treated in a less formal
way which is more suited to oral presentation. In this way it is possible to cater for
dierent sections of an audience with a mixed background. A formal proof may be
indicated to some while others may have to take certain statements on trust. This is
especially the case when integration spaces are involved.
I would like to thank the many students and colleagues who have pointed out errors
and obscurities in earlier versions of these notes. A few proofs contain some sentences
in square brackets. These indicate explanations that I consider rather obvious and
should be superuous to a formal proof but were added in response to some query.
For the benet of a wider audience, here is a brief indication of what might be covered
to supplement the notes and also a few comments.
Section 1. Examples of inner product spaces :
2
n
(= C
n
) and
2
. Continuous func-
tions on [a, b] with 'f, g` =
b
a
f(t)g(t) dt and problems with extending to larger classes
of functions (equivalence classes as elements of the space). Completeness of
2
n
(= C
n
)
and
2
(and continuous functions on [a, b] not complete).
L
2
[a, b]. Some brief discussion of the Lebesgue integral. The following statement to
be known or accepted: there is a denition of the integral such that the (equivalence
classes) of all functions f such that
b
a
[f(t)[
2
dt exists and is nite forms a Hilbert
space with the inner product 'f, g` =
b
a
f(t)g(t) dt (that is, it is complete). Some
more general L
2
spaces might be mentioned (e.g. L
2
(S) where S = [a, b] [a, b] or
some other subset of R
n
).
Examples of normed spaces which cannot be Hilbert spaces because they do not
satisfy the parallelogram law (C[a, b],
p
n
(n = 2) for p = 2).
Examples of normed spaces with closed convex sets where the distance from a point
is not attained uniquely (e.g. the unit ball in
1
2
with the point (1, 1)) or not attained
at all (e.g. the space X = f C[0, 1] : f(0) = 0, the set f X :
1
0
f(t) dt = 1
and the zero function as the point).
Some indications of the applications of the minimum distance theorem, e.g. to ap-
proximation theory and optimal control theory.
2 Operators on Hilbert space
Section 2. This section is supplemented by specic examples of operators on
2
and
L
2
[a, b]. These include diagonal operators, shifts (forward, backward and weighted)
on
2
, the bilateral shift on
2
(Z) and the following operators on L
2
[a, b].
Multiplication operator : (M
where (x) = x on L
2
[0, 1] should be identied as [0, 1] and the
fact that M
2
'x, x` + ('x, y` +'y, x`) +'y, y` 0 .
In the case that 'x, y` is real, we have that the discriminant (b
2
4ac) of this
quadratic function of is negative which gives the result.
In general, put x
1
= e
i
x where is the argument of the complex number 'x, y`.
Then 'x
1
, y` = e
i
'x, y` = ['x, y`[ is real and 'x
1
, x
1
` = 'x, x`. Applying the above to
x
1
, y gives the required result.
[Alternatively, put =
y,x
x,x
in 'x + y, x + y` 0.]
Theorem 1.2
|x| = 'x, x`
1/2
is a norm on H.
Proof. The facts that |x| 0, |x| = 0 x = 0 and |x| = 'x, x`
1
2
= [[.|x|
are all clear from the equivalent properties of the inner product. For the triangle
inequality,
|x + y|
2
= |x|
2
+|y|
2
+'x, y` +'y, x`
|x|
2
+|y|
2
+ 2['x, y`[
|x|
2
+|y|
2
+ 2|x|.|y| using (1.1)
= (|x| +|y|)
2
.
Lemma 1.3 (Polarization identity)
'x, y` =
1
4
|x + y|
2
|x y|
2
+ i|x + iy|
2
i|x iy|
2
.
John Erdos 5
Proof.
|x + y|
2
= |x|
2
+|y|
2
+'x, y` +'y, x`
|x y|
2
= |x|
2
+|y|
2
+'x, y` +'y, x`
i|x + iy|
2
= i|x|
2
+ i|y|
2
+'x, y` 'y, x`
i|x iy|
2
= i|x|
2
+i|y|
2
+'x, y` 'y, x` .
Adding the above gives the result.
Lemma 1.4 (Paralellogram law)
|x + y|
2
+|x y|
2
= 2|x|
2
+ 2|y|
2
.
Proof.
|x + y|
2
= |x|
2
+|y|
2
+'x, y` +'y, x`
|x y|
2
= |x|
2
+|y|
2
'x, y` 'y, x` .
Adding the above gives the result.
Denition x is said to be orthogonal to y if 'x, y` = 0; we write x y.
Lemma 1.5 (Theorem of Pythagoras)
'x, y` = 0 =|x|
2
+|y|
2
= |x + y|
2
.
Proof. Obvious.
Denition. If H is an inner product space and (H, | | ) is complete then H is called
a Hilbert space.
A set C (in a vector space) is convex if
x, y C =x + (1 )y C whenever 0 1 .
In a metric space, the distance from a point x to a set S is
d(x, S) = inf|x s| : s S.
Theorem 1.6 If K is a closed convex set in a Hilbert space H and h H then there
exists a unique k K such that
d(h, K) = |h k|.
Proof. Let C = K h = k h : k K. Note that C is also closed and convex,
d(h, K) = d(0, C) and if c = k h C is of minimal norm then k is the required
element of K. Therefore it is sucient to prove the theorem for the case h = 0.
6 Operators on Hilbert space
Let d = d(0, C) = inf
cC
|c|. The |c| d for all c C. Choose a sequence (c
n
) such
that (|c
n
|) d. Using the parallelogram law (Lemma 1.4),
|c
n
+ c
m
|
2
+|c
n
c
m
|
2
= 2|c
n
|
2
+ 2|c
m
|
2
.
But, since C is convex,
c
n
+c
m
2
C and so |
c
n
+c
m
2
| d; that is |c
n
+ c
m
|
2
4d
2
.
Therefore
0 |c
n
c
m
|
2
= 2|c
n
|
2
+ 2|c
m
|
2
|c
n
+ c
m
|
2
2(|c
n
|
2
+|c
m
|
2
) 4d
2
0 ()
as n, m . It follows easily that (c
n
) is a Cauchy sequence. [ Since (|c
n
|) d,
given > 0, there exists n
0
such that for n > n
0
, 2(|c
n
|
2
d
2
) <
2
2
. Then (*) shows
that for n, m > n
0
, |c
n
c
m
| < .] Since H is complete and C is closed, (c
n
) converges
to an element c C and |c| = lim
n
|c
n
| = d.
To prove uniqueness, suppose also that c
C with |c
) shows that
0 |c c
|
2
2|c|
2
+ 2|c
|
2
|c + c
|
2
2d
2
+ 2d
2
4d
2
= 0
and so c = c
.
Lemma 1.7 If N is a closed subspace of a Hilbert space H and h H then
d(h, N) = |h n
0
| if and only if 'h n
0
, n` = 0 for all n N .
Proof. Suppose d(h, N) = |hn
0
|. Write z = hn
0
. Then for all non-zero n N,
|z|
2
z
'z, n`n
|n|
2
2
= |z|
2
2['z, n`[
2
|n|
2
+
['z, n`[
2
|n|
2
= |z|
2
['z, n`[
2
|n|
2
so 'z, n` = 0.
Conversely if h n
0
N then, by Pythagoras (Lemma 1.5) for all n N,
|h n|
2
= |h n
0
+ n
0
n|
2
= |h n
0
|
2
+|n
0
n|
2
|h n
0
|
2
.
Hence inf
nN
|h n| is attained at n
0
.
Note that the above proof is putting a geometrical argument into symbolic form. The
quantity
z,nn
n
2
is the resolution of the vector z in the direction of n.
In these notes the term subspace (of a Hilbert space) will always mean a closed
subspace. The justication for this is that the prex sub refers to a substructure;
so the subspace should be a Hilbert space in its own right, that is, it should be
John Erdos 7
complete. But it is an easy fact that a subset of a complete space is complete if and
only if it is closed.
Denition. Given a subset S of H the orthogonal complement S
is dened by
S
= (0) N = H
Proof. Clearly, if N = H then N
= (0).
Lemma 1.9 For subsets of a Hilbert space H
(i) S
is a closed subspace,
(ii) S
1
S
2
=S
1
S
2
,
(iii) S S
,
(iv) S
= S
,
(v) S S
= (0).
Proof. (i) Clearly S
be a
sequence converging to t. Then, by the continuity of the inner product, for all s S,
't, s` = lim
n
't
n
, s` = 0
so t S
yields S
then 'x, x` = 0 so x = 0.
Lemma 1.10 If M and N are orthogonal subspaces of a Hilbert space then M N
is closed.
Proof. Note that since N M, we have that N M = (0) and the sum M + N is
automatically direct. Let z
n
M N such that (z
n
) z. We need to show that
z MN. Now z
n
= x
n
+y
n
with x
n
N and y
n
M. Therefore, using Pythagoras
(Lemma 1.5) since M N,
|z
n+p
z
n
|
2
= |x
n+p
x
n
|
2
+|y
n+p
y
n
|
2
.
As (z
n
) is convergent, it is a Cauchy sequence. If follows easily from the above that
both (x
n
) and (y
n
) are Cauchy sequences so, since H is complete, (x
n
) and (y
n
) both
converge. Call the limits x and y. Then, since M and N are closed subspaces, x M
and y N. Thus z = lim(x
n
+ y
n
) = x + y M N.
8 Operators on Hilbert space
Theorem 1.11 If N is a subspace of a Hilbert space H then N N
= H.
Proof. From above, N N
then
x N
= H.
Corollary 1.12
(i) If N is a subspace then N
= N.
(ii) For any subset S of a Hilbert space H, S
. Since H = N N
, if x N
then
x = s + t with s N and t N
, so t = 0 and
x = s N.
(ii) Clearly S
= M.
John Erdos 9
Exercises 1
1. For a Hilbert space H, show that the inner product, considered as a map from HH
to C, is continuous.
2. Let H
1
and H
2
be Hilbert spaces. Let H be the set of ordered pairs H
1
H
2
with
addition and multiplication dened (in the usual way) as follows:
(h
1
, h
2
) + (g
1
, g
2
) = (h
1
+ g
1
, h
2
+ g
2
)
(h
1
, h
2
) = (h
1
, h
2
).
Show that the inner product dened by
'(h
1
, h
2
), (g
1
, g
2
)` = 'h
1
, g
1
` +'h
2
, g
2
`
satises the axioms for an inner product and H with this inner product is a Hilbert
space. [H is called the (Hilbert space) direct sum of H
1
and H
2
. One writes H =
H
1
H
2
.]
3. Prove that in the CauchySchwarz inequality ['x, y`[ |x||y| the equality holds i
the vectors x and y are linearly dependent.
4. For which real does the function f(t) = t
belong to
(i) L
2
[0, 1] (ii) L
2
[1, ] (iii) L
2
[0, ] ?
5. Let x and y be vectors in an inner product space. Given that | x+y |=| x | + | y |,
show that x and y are linearly dependent.
6. Let W[0, 1] be the space of complex-valued functions which are continuously dier-
entiable on [0, 1]. Show that,
'f, g` =
1
0
f(t)g(t) + f
(t)g
(t) dt
denes an inner product on W[0, 1].
7. Prove that in a complex inner product space the following equalities hold:
'x, y` =
1
N
N
k=1
|x + e
2ik/N
y|
2
e
2ik/N
for N 3,
'x, y` =
1
2
2
0
|x + e
i
y|
2
e
i
d .
[This generalizes the polarization identity.]
8. Let M and N be closed subspaces of a Hilbert space. Show that
(i) (M + N)
= M
(ii) (M N)
= M
+ N
.
9. Show that the vector subspace of
2
spanned by sequences of the form (1, ,
2
,
3
, . . .),
where 0 < 1, is dense in
2
.
A challenging but not very important exercise :
10. Show that, for any four points of a Hilbert space,
| x z | . | y t || x y | . | z t | + | y z | . | x t | .
10 Operators on Hilbert space
2 Linear Operators.
Some of the results in this section are stated for normed linear spaces but they will
be used in the sequel only for Hilbert spaces.
Lemma 2.1 Let X and Y be normed linear spaces and let L : X Y be a linear
map. Then the following are equivalent :
1. L is continuous;
2. L is continuous at 0;
3. there exists a constant K such that |Lx| K|x| for all x X.
Proof. 1 implies 2 is obvious. If 2 holds, take any > 0. Continuity at 0 shows that
there is a corresponding > 0 such that |LX| < whenever |x| < . Take some c
with 0 < c < . Then for any x = 0,
cx
x
= c < and so
cx
|x|
= c
|Lx|
|x|
< .
This shows that |Lx| < K|x| where K =
c
.
If 3 holds, to show continuity at any point x
0
, note that
|Lx Lx
0
| = |L(x x
0
)| K|x x
0
| .
Therefore, given any > 0, let =
K
. Then if |xx
0
| < we have |LxLx
0
| < .
The set of all continuous (bounded) linear maps X Y is denoted by B(X, Y ).
When X = Y we write B(X).
For L B(X, Y ), dene |L| = sup
x=0
Lx
x
.
Exercise. | | is a norm on B(X, Y ) and
|L| = sup
x=0
|Lx|
|x|
= sup
x1
|Lx| = sup
x=1
|Lx|.
If Y is complete then so is B(X, Y )
When Y = C then B(X, C) is called the dual of X and denoted by X
(sometimes by
X
, y) = (x, y) + (x
, y)
(x, y + y
) = (x, y) +
(x, y
) .
A bilinear form is said to be bounded if, for some constant K, [(x, y)[ K|x|.|y|
for all x, y H.
Theorem 2.5 (Riesz) Every bounded bilinear form on H is of the form
(x, y) = 'Ax, y`
for some A B(H).
12 Operators on Hilbert space
Proof. Consider x xed for the moment. Then (x, y) is conjugate linear in y, so
that (x, y) is linear in y. Using Theorem 2.2 we have that there is a (unique) h H,
(x, y) = 'y, h` , that is (x, y) = 'h, y` .
One can nd such an h corresponding to each x H. Dene a function H H by
Ax = h. Then A is linear since , for all x, x
, y,
'A(x + x
), y` = ((x + x
), y) = (x, y) + (x
, y) = 'Ax, y` +'Ax
, y`
so A(x + x
) = Ax + Ax
[since A(x + x
) Ax Ax
= (0)]. Similarly
A(x) = Ax. Also,
|Ax| = sup
y=0
'Ax, y`
|y|
= sup
y=0
[(x, y)[
|y|
K|x|
so A is continuous.
Denition The adjoint. Let A B(H). Then (x, y) = 'x, Ay` is a bounded bilinear
form on H so, by Theorem 2.5 there is an operator A
= A, (ii) (A)
=
A
,
(iii) (A + B)
= A
+ B
, (iv) (AB)
= B
,
(v) |A| = |A
|.
Note. Bilinear forms could have been dened as maps from H / to C where
H and / are dierent Hilbert spaces. All the above can be done with essentially no
change; (the adjoint of A B(H, /) is then an operator in B(/, H)).
Denition.
If A = A
= A
= U
.
The orthogonal projection P onto N is dened by Ph = x (where h = x + y is the
decomposition above). Note that then y = (I P)h and I P is the projection onto
N
.
In this course we shall not consider projections that are not orthogonal and usually
call these operators projections.
Lemma 2.6 Let N be a closed subspace of H and let P be the orthogonal projection
onto N. Then
(i) P is linear,
(ii) |P| = 1 (unless N = 0),
(iii) P
2
= P,
(iv) P
= P.
Also, if E B(H) satises E = E
2
= E
= x
+ y
are the
unique decompositions of h and h
with x, x
N and y, y
. Then h + h
=
(x + x
) + (y + y
) is the decomposition of h + h
and
P(h + h
) = x + x
= Ph + Ph
.
(ii) If h = x + y with x N and y N
,
|Ph|
2
= |x|
2
|x|
2
+|y|
2
= |h|
2
and so |P| 1. But if 0 = h N then Ph = h and so |P| = 1.
(iii) If h N [then h = h + 0 is the decomposition of h and] Ph = h. But for any
h H, Ph N so P(Ph) = Ph, that is, P
2
= P.
(iv) If h = x + y and h
= x
+ y
with x, x
N and y, y
.
'Ph, h
` = 'x, x
+ y
` = 'x, x
`
since x N and y
. Similarly 'h, Ph
` = 'x, x
` and so P = P
.
Finally, if E B(H) satises E = E
2
= E
let N = x : Ex = x. Then N =
ker(I E), so N is closed. For any h H, write
h = Eh + (I E)h.
Then Eh N since E(Eh) = E
2
h = Eh and (I E)h N since if x N, Ex = x
and
'(I E)h, x` = '(I E)h, Ex` = 'E
(I E)h, x` = '(E
2
E)h, x` = 0.
This shows that E is the projection onto N.
14 Operators on Hilbert space
Lemma 2.7 If P is the orthogonal projection onto a subspace N then for all h H,
d(h, N) = |(I P)h|.
Proof. For any h H we have Ph N and '(I P)h, n` = 0 for all n N.
Therefore from Lemma 1.7
d(h, N) = |h Ph| = |(I P)h| .
Lemma 2.8 Let A B(H) and P be the orthogonal projection onto a subspace N.
(i) N is invariant under A AP = PAP.
(ii) N
= (PAP)
PA = PAP since A = A
. If these
equalities hold then PA = AP.
John Erdos 15
Exercises 2
1. Let X B(H). Show that :
(i) X is selfadjoint 'Xx, x` is real for all x H,
(ii) X is normal | Xx |=| X
x | for all x H,
(iii) X is unitary | Xx |=| X
= (0).]
Note that the selfadjointness condition is needed consider the operator S on
2
dened by S(
1
,
2
,
3
, . . .) = (0,
1
,
2
,
3
, . . .).
4. The operators D and W on
2
are dened by
D(
1
,
2
,
3
, . . .) = (
1
1
,
2
2
,
3
3
, . . .) ,
W(
1
,
2
,
3
, . . .) = (0,
1
1
,
2
2
,
3
3
, . . .) ,
where (
n
) is a bounded sequence of complex numbers. Show that W and D are
bounded operators and nd their adjoints.
5. Given that X B(H) is invertible (that is, there exists X
1
B(H) such that
XX
1
= X
1
X = I) prove that X
is invertible and (X
)
1
= (X
1
)
.
6. Find the adjoint of the operator V dened on L
2
[0, 1] by (V f)(x) =
x
0
f(t) dt .
7. Let T : L
2
[0, 1] L
2
[0, 1] be dened by
(Tf)(x) =
2xf(x
2
) .
Find the adjoint of T and deduce that T is unitary.
8. Let E, F be the orthogonal projections onto subspaces M and N respectively. Prove
that,
(i) EF = F N M E F is an orthogonal projection,
(ii) EF = 0 N M
E + F is an orthogonal projection,
(iii) EF = FE E + F FE is an orthogonal projection.
9. The operator A B(H) satises Ax = x for some x H. Prove that A
x = x + y
where y x. If, further, |A| 1, show that A
x = x.
Suppose that E
2
= E and |E| = 1. Use the above to show that ran(E) = ran(E
)
and ker(E) = ker(E
2
,
1
3
3
, . . .) ,
and let M and N be linear subspaces of l
2
l
2
dened by
M = (x, 0) : x l
2
N = (x, Dx) : x l
2
.
Observe that M is closed and use the continuity of D to show that N is also closed.
Show that M N = (0) and that the algebraic direct sum of M and N is dense in
l
2
l
2
but is not equal to l
2
l
2
(and so it is not closed).
John Erdos 17
3 Orthonormal Sets.
Denition. A set o of vectors of H is said to be orthonormal if
1. |x| = 1 for all x o,
2. 'x, y` = 0 if x = y and x, y o.
Lemma 3.1 (Bessels inequality) If x
i
: 1 i n is a nite orthonormal set then,
for any h H, writing
i
= 'h, x
i
`,
n
i=1
[
i
[
2
|h|
2
.
(Note that the case n = 1 is the Cauchy-Schwartz inequality)
Proof. Let h H. Then
|h
n
i=1
i
x
i
|
2
= 'h
n
i=1
i
x
i
, h
n
i=1
i
x
i
`
= |h|
2
'h,
n
i=1
i
x
i
` '
n
i=1
i
x
i
, h` +
n
i,j=1
j
'x
i
, x
j
`
= |h|
2
'h,
n
i=1
i
x
i
` '
n
i=1
i
x
i
, h` +
n
i=1
[
i
[
2
= |h|
2
i=1
[
i
[
2
0.
Lemma 3.2 If x
i
: i = 1, 2, 3 is an orthonormal sequence then, for any h H,
writing
i
= 'h, x
i
`,
i=1
i
x
i
converges to a vector h
, x
i
` = 0 for all i.
Proof. Put h
r
=
r
i=1
i
x
i
. Then
|h
r+p
h
r
|
2
=
r+p
i=r+1
i
x
i
2
=
r+p
i=r+1
[
i
[
2
.
Now
n
i=1
[
i
[
2
|h|
2
for all n and so
i=1
[
i
[
2
is convergent and so is a Cauchy
series. Hence, given any > 0 there exists n
0
such that for r > n
0
, p > 0 we have
r+p
i=r+1
[
i
[
2
<
2
; that is, |h
r+p
h
r
| < . Therefore (h
r
) is a Cauchy sequence and,
since H is complete, it is convergent. Call its limit h
.
For any xed i, 'h h
r
, x
i
` = 0 for all r > i. Now let r . Then it follows from
the continuity of the inner product that 'h h
, x
i
` = 0 for all i.
18 Operators on Hilbert space
Theorem 3.3 Let x
i
: i = 1, 2, 3 be an orthonormal sequence. The the following
statements are equivalent.
(i) x
i
: i = 1, 2, 3 is maximal (that is, it is not a proper subset of any or-
thonormal set).
(ii) If
i
= 'h, x
i
` = 0, for all i then h = 0.
(iii) (Fourier expansion) For all h H we have h =
i=1
i
x
i
.
(iv) (Parsevals relation) For all h, g H we have 'h, g` =
i=1
i
.
(v) (Bessels equality) For all h H we have |h|
2
=
i=1
[
i
[
2
.
(In the above,
i
= 'h, x
i
` and
i
= 'g, x
i
`.)
Proof. (i) = (ii). If (ii) is false then adding
h
h
to the set x
i
: i = 1, 2, 3 gives
a larger orthonormal set, contradicting (i).
(ii) = (iii). Let h
i=1
i
x
i
(this exists, by Lemma 3.2). Then 'h h
, x
i
` = 0
for all i and so h = h
by (ii)
(iii) = (iv). Let h
r
=
r
i=1
i
x
i
and g
s
=
s
i=1
i
x
i
. Then
'h
r
, g
s
` =
min[r,s]
i=1
i
.
Let r and s . Using the continuity of the inner product, it follows that
'h, g` =
i=1
i
.
(iv) = (v). Put g = h in (iv).
(v) = (i). If x
i
: i = 1, 2, 3 is not maximal and can be enlarged by adding z
then 'z, x
i
` = 0 for all i but also
1 = |z|
2
=
i=1
['z, x
i
`|
2
= 0
which give a contradiction.
Denition. A maximal orthonormal sequence is called an orthonormal basis.
Clearly the concept of an orthonormal basis refers to a set of vectors so that any
permutation of such a set is still an orthonormal basis. It follows that the series giving
the fourier expansion of a vector can be re-arranged arbitrarily without altering its
convergence or its sum. Such a series is said to be unconditionally convergent.
John Erdos 19
Theorem 3.4 (Gram-Schmidt process) Let y
i
: i = 1, 2, 3 be a sequence of
vectors of H. Then there exists an orthonormal sequence x
i
: i = 1, 2, 3 such
that, for each integer k
spanx
1
, x
2
, x
3
x
k
spany
1
, y
2
, y
3
y
k
.
If y
i
: i = 1, 2, 3 is a linearly independent set, then the above inclusion is an
equality for each k.
Proof. First consider the case when y
i
: i = 1, 2, 3 is a linearly independent set.
Dene
u
1
= y
1
x
1
=
u
1
|u
1
|
,
u
2
= y
2
'y
2
, x
1
`x
1
x
2
=
u
2
|u
2
|
,
.
.
.
.
.
.
u
n
= y
n
n1
i=1
'y
n
, x
i
`x
i
x
n
=
u
n
|u
n
|
.
Easy inductive arguments show that for each k, x
k1
spany
1
, y
2
, y
k1
and,
since y
i
: i = 1, 2, 3 is linearly independent, that u
k
= 0. A further easy
induction shows that for r < n we have 'u
n
, x
r
` = 0 and so it follows easily that
x
i
: i = 1, 2, 3 is an orthonormal sequence. In the general case, the same
construction applies except that whenever y
r
is a linear combination of y
1
, y
2
, , y
r1
this element is ignored.
It is clear in general that x
r
is a linear combination of on y
1
, y
2
, , y
r1
and so the
inclusion
spanx
1
, x
2
, x
3
x
k
spany
1
, y
2
, y
3
y
k
.
is obvious. When y
i
: i = 1, 2, 3 is a linearly independent set we have equality,
since both sides have dimension k.
For the rest of this course we shall often need to assume that the Hilbert space we
consider has a countable orthonormal basis. This is true for all spaces considered in
the applications. The restriction is a rather technical matter and could be avoided
but this would entail a discussion of uncountable sums. A few statements would also
need modication.
The appropriate way to state this restriction is to say that the Hilbert space we
consider is separable. For our purposes one could say that a Hilbert space is separable
if it has a countable orthonormal basis, and take this as the denition of separability.
However, this is a more general notion: recall that a metric space is dened to be
separable if it has a countable, dense subset. The proposition which follows connects
these ideas.
20 Operators on Hilbert space
Proposition
1. In a separable Hilbert space, every orthonormal set is countable.
2. A Hilbert space is separable if and only if it has a (countable) orthonormal basis.
We shall not prove this in detail, but here is a sketch. For each element x
of an
orthonormal set x
A
, let B
, radius
2
2
. Since these
balls are disjoint and since every open set must contain at least one element of a
dense subset, it is clear that if the space is separable the orthonormal set must be
countable. For 2, applying the Gram-Schmidt process to a countable dense subset
results in an orthonormal sequence that is easily proved to be a basis. Conversely, if
x
n
: n = 1, 2, 3, is a countable orthonormal basis, tedious but routine arguments
show that the set
S =
N
n=1
r
n
x
n
: N nite , r
n
rational
is countable. It is clearly dense because the closure of S is a subspace containing an
orthonormal basis.
John Erdos 21
Exercises 3
1. Let N
i
be a sequence of mutually orthogonal subspaces of a Hilbert space H and
let E
i
be the sequence of projections onto N
i
. Show that for each x H,
(i) for a nite subset N
i
n
i=1
of N
i
,
n
i=1
| E
i
x |
2
| x |
2
.
(ii)
i=1
E
i
x converges to some h H which satises (x h) N
i
for each i.
Show further that the following are equivalent :
(a) N
i
is not a proper subset of any orthogonal set of subspaces of H,
(b) h N
i
for all i h = 0,
(c) for each x H, x =
i=1
E
i
x,
(d) for each x, y H, 'x, y` =
i=1
'E
i
x, E
i
y`,
(e) for each x H, | x |
2
=
i=1
| E
i
x |
2
.
[Hints : You will need Q. 8 (ii) of Sheet 2 or its equivalent. Note that under the given
conditions, since E
i
= E
i
= E
2
i
, 'E
i
x, E
j
y` = 'E
j
E
i
x, y` = 0 if i = j.]
2. Find the rst three functions obtained by applying the Gram- Schmidt process to
the elements t
n
: n = 0, 1, . . . of L
2
[1, 1]. [Note: apart from constant factors, this
process yields the Legendre polynomials.] Use your results and the theory developed
in lectures to nd a, b and c which minimises the quantity
1
1
[t
4
a bt ct
2
[
2
dt .
3. Let N be a subspace of L
2
[0, 1] with the property that for some xed constant C and
each f N,
[f(x)[ C|f| almost everywhere .
Prove that N is nite dimensional.
[Hint: for any orthonormal subset f
1
, f
2
, . . . f
n
, of N, evaluate, for any xed y, the
norm of g where
g(x) =
n
i=1
f
i
(y)f
i
(x) .
Deduce that
n
i=1
[f
i
(y)[
2
C
2
and integrate this relation with respect to y.]
22 Operators on Hilbert space
4 Compact Operators.
Denition. An operator K B(H) is said to be compact if for every bounded set o
of vectors of H the set Ks : s o is compact.
Equivalently :
Denition. An operator K B(H) is said to be compact if for every bounded
sequence (x
n
) of vectors of H the sequence (Kx
n
) has a convergent subsequence.
We shall denote the set of all compact operators on H by /(H).
Denition. The rank of an operator is the dimension of its range.
Note that every operator of nite rank is compact. This is an immediate consequence
of the Bolzano-Weierstrass theorem which states that every bounded sequence in C
n
has a convergent subsequence. Note also that the identity operator on a Hilbert space
H is compact if and only if H is nite-dimensional.
Theorem 4.1 /(H) is an ideal of B(H).
Proof. We need to show that, if A, B /(H) and T B(H) then A, A + B, TA
and AT are all in /(H). That is, for any a bounded sequence (x
n
), we must show
that (Ax
n
), ([A + B]x
n
), (TAx
n
) and (ATx
n
) all have convergent subsequences.
Since A is compact, (Ax
n
) has a convergent subsequence (Ax
n
i
). Then clearly (Ax
n
i
)
is a convergent subsequence of (Ax
n
) showing that A is compact. Also, (x
n
i
) is a
bounded sequence and so, since B is compact, (Bx
n
i
) has a convergent subsequence
(Bx
n
i
j
). Then ([A+B]x
n
i
j
) is a convergent subsequence of ([A+B]x
n
), showing that
A + B is compact.
Again, since T B(H), T is continuous and so (TAx
n
i
) is a convergent subsequence
of (TAx
n
) showing that TA is compact. The proof for AT is slightly dierent. Here,
since (x
n
) is bounded and |Tx
n
| |T|.|x
n
| we have that (Tx
n
) is bounded and
so, since A is compact, (ATx
n
) has a convergent subsequence, showing that TA is
compact.
A consequence of the above theorem is that, it H is innite-dimensional then and
T B(H) has an inverse T
1
B(H) then T is not compact.
Theorem 4.2 /(H) is closed.
Proof. Let (K
n
) be a sequence of compact operators converging to K. To show that
K is compact, we need to show that if (x
i
) is a bounded sequence the (Kx
i
) has a
convergent subsequence.
Let (x
1
i
) be a subsequence of (x
i
) such that (K
1
x
1
i
) is convergent,
let (x
2
i
) be a subsequence of (x
1
i
) such that (K
2
x
2
i
) is convergent,
let (x
3
i
) be a subsequence of (x
2
i
) such that (K
3
x
3
i
) is convergent,
and continue in this way.
[The notation above is slightly unusual and is adopted to avoid having to use sub-
scripts on subscripts on .]
Let z
i
= x
i
i
. Then (z
i
) is a subsequence of (x
i
). Also, for each n, apart from the rst
n terms, (z
i
) is a subsequence of (x
n
i
) and so (K
n
z
i
) is convergent.
John Erdos 23
We now show that (Kz
i
) is convergent by showing that it is a Cauchy sequence. For
all i, j, n we have
|Kz
i
Kz
j
| = |(K K
n
)z
i
+ K
n
z
i
K
n
z
j
(K K
n
)z
j
|
|K K
n
|(|z
i
| +|z
j
|) +|K
n
(z
i
z
j
)| .
Let > 0 be given. Since (K
n
) K we can nd n
0
such that |K K
n
| <
4c
for
n > n
0
where c satises |x
i
| c for the bounded sequence (x
i
). Choose one xed
such n. Now, since (K
n
z
i
) converges, it is a Cauchy sequence and so there is an
i
0
such that for i > i
0
, j > i
0
we have |K
n
z
i
K
n
z
j
| <
2
. Combining these with
the displayed inequality shows that for i > i
0
, j > i
0
, |Kz
i
Kz
j
| < so (Kz
i
) is
convergent as required.
Example. The operator K on L
2
[a, b] dened by
(Kf)(x) =
b
a
k(x, t)f(t) dt ,
where
b
a
b
a
[k(x, t)[
2
dxdt = M
2
< , is compact.
We have already seen that operators of the above type are continuous with |K| M
(Recall that k(x, t) is called the kernel of the integral operator K). We shall show
that K is the norm limit of a sequence of nite rank operators. Note that if k(x, t) is
of the form u(x)v(t) then
(Rf)(x) =
b
a
u(x)v(t)f(t) dt = 'f, v`u = (v u)f
is a rank one operator.
Let S be the square [a, b] [a, b]. We shall apply Hilbert space theory to L
2
(S) which
is a Hilbert space of functions of 2 variables with the inner product
', ` =
b
a
b
a
(x, t)(x, t) dxdt .
Let (u
i
) be an orthonormal basis of L
2
[a, b]. Then (u
i
(x)u
j
(t))
i,j=1
is an orthonormal
basis of L
2
(S). Indeed,
'u
i
(x)u
j
(t), u
k
(x)u
l
(t)` =
b
a
b
a
u
i
(x)u
j
(t)u
k
(x)u
l
(t) dxdt
=
b
a
u
i
(x)u
k
(x) dx
b
a
u
j
(t)u
l
(t) dt = 0
unless i = k and j = l, in which case the integral is 1. Thus (u
i
(x)u
j
(t))
i,j=1
is an
orthonormal sequence. To show that it is a basis, suppose (x, t) u
i
(x)u
j
(t) for all
i, j. Then
0 =
b
a
b
a
(x, t)u
i
(x)u
j
(t) dxdt =
b
a
b
a
(x, t)u
i
(x) dx
u
j
(t) dt .
This shows that, for each i, the function
b
a
(x, t)u
i
(x) dx of t is orthogonal to u
j
(t)
for each j. Therefore, since (u
j
) is a basis of L
2
[a, b], it is (equivalent to) the zero
function. Then, for xed t the function (x, t) is orthogonal to u
i
(x) for each i and
so it is zero.
24 Operators on Hilbert space
Returning to the operator K, note that k L
2
(S). Therefore, by Theorem 3.3 (iii)
it has a fourier expansion using the basis (u
i
u
j
) of the type
k(x, t) =
i,j=1
ij
u
i
(x)u
j
(t) .
Thus, writing k
n
(x, t) =
n
i,j=1
ij
u
i
(x)u
j
(t) and
(K
n
f)(x) =
b
a
k
n
(x, t)f(t) dt ,
we have that K
n
is a nite rank operator (of rank at most n
2
). Note that K K
n
is
an integral operator (of the same type as K) with kernel k(x, t) k
n
(x, t). Thus
|K K
n
|
2
b
a
b
a
[k(x, t) k
n
(x, t)[
2
dxdt = |k k
n
|
2
L
2
(S)
and the right hand side 0. Therefore Theorem 4.2 shows that K is compact.
Lemma 4.3 Let K be a compact operator on H and suppose (T
n
) is a bounded se-
quence in B(H) such that, for each x H the sequence (T
n
x) converges to Tx, where
T B(H). Then (T
n
K) converges to TK in norm.
Briey, the above can be rephrased as :
If K /(H) and |T
n
x Tx| 0 for all x H then |T
n
K TK| 0.
In words : multiplying by a compact operator on the right converts a pointwise
convergent sequence of operators into a norm convergent one.
Proof. Since (T
n
) is a bounded sequence, |T
n
| C for some constant C. Then for
all x H, |Tx| = lim
n
|T
n
x| C|x| and so |T| C.
Let K be compact and suppose that |TKT
n
K| 0. Then there exists some > 0
and a subsequence (T
n
i
K) such that |TK T
n
i
K| > . Choose unit vectors (x
n
i
)
of H such that |(TK T
n
i
K)x
n
i
| > . [That this can be done follows directly from
the denition of the norm of an operator.] Using the fact that K is compact, we can
nd a subsequence (x
n
j
) of (x
n
i
) such that (Kx
n
j
) is convergent. Let the limit of this
sequence be y. Then for all j
< |(TK T
n
j
K)x
n
j
| |(T T
n
j
)(Kx
n
j
y)| +|(T T
n
j
)y| .
Now, using the convergence of (Kx
n
j
) to y, there exists n so that, for n
j
> n,
|Kx
n
j
y| <
8C
. Also, using the convergence of (T
n
j
) to T, there exists m so that,
for n
j
> m, |(T T
n
j
)y| <
4
. Then, for j > max[n, m] the right hand side of the
displayed inequality is less than
2
, and this contradiction shows that the supposition
that |TK T
n
K| 0 is false.
John Erdos 25
The theorem below is true for all Hilbert spaces, but we shall only prove it for the
case when the space is separable.
Theorem 4.4 Every compact operator on H is a norm limit of a sequence of nite
rank operators.
Proof. Let x
i
be an orthonormal basis of H. Dene P
n
by
P
n
h =
n
i=1
'h, x
i
`x
i
.
[Note that P
n
is the projection onto span x
1
, x
2
, , x
n
. Also, P
n
could be written
as P
n
h =
n
i=1
x
i
x
i
.] From Theorem 3.3(iii), for all x H, P
n
x converges to x
(that is, P
n
converges pointwise to the identity operator I). Now, if K is any compact
operator, P
n
K is of nite rank and, from Theorem 4.3 (P
n
K) converges to K in norm.
26 Operators on Hilbert space
Exercises 4
1. Let T be the operator on l
2
l
2
dened by T(x, y) = (0, x). Show that T
2
= 0 and
that T is not compact.
2. Let (x
n
) be an orthonormal sequence in a Hilbert space H and let (
n
) be a bounded
sequence of complex numbers. Prove that the operator A dened by
Ax =
n=1
n
'x, x
n
`x
n
is bounded with
| A | sup
n
[
n
[ .
Hence prove that if lim
n
(
n
) = 0 then A is compact.
Show that, when m = n,
| Ax
m
Ax
n
|
2
= [
m
[
2
+[
n
[
2
.
Hence prove that, conversely if lim
n
(
n
) = 0 then A is not compact.
3. Given that K
Kx
n
) is convergent, prove that (Kx
n
) is a Cauchy sequence.]
4. Let K be a compact operator. Using the hints below, prove that for any orthonormal
sequence x
n
, (Kx
n
) 0 as n
Hints: Observe that, for any vector z, 'x
n
, z` 0. [A result of the course states that
['x
n
, z`[
2
is convergent.] Apply this, with z = K
x, x` = 'Ax, x`.
Lemma 5.3
|A| = sup
x1
['Ax, x`[ .
Proof. Let k = sup
z1
['Az, z`[. Then ['Ax, x`[ k|x|
2
for all x and, from the
Cauchy-Schwatrz in equality, k |A|. Since
|A| = sup
x1
|Ax| = sup
x1
sup
y1
['Ax, y`[ ,
to show that |A| k, we need to show that ['Ax, y`[ k whenever |x| 1 and
|y| 1. It is sucient to prove this when 'Ax, y` is real, since if ['Ax, y`[ = e
i
'Ax, y`
28 Operators on Hilbert space
then applying the result for the real case for 'Ax
, y` where x
= e
i
x, proves the
general result.
Now, using the polarization identity (Lemma 2.3) and the paralellogram law (Lemma 1.4),
4'Ax, y` = 'A(x + y), (x + y)` 'A(x y), (x y)`
+i['A(x + iy), (x + iy)` 'A(x iy), (x iy)`]
k|x + y|
2
+|x y|
2
= k(2|x|
2
+ 2|y|
2
) 4k ,
(the expression in square brackets being zero since 'Ax, y` is real).
Note that sup
x1
['Ax, x`[ = sup
x=1
['Ax, x`[. We write
m = inf
x=1
'Ax, x` and M = sup
x=1
'Ax, x` .
Corollary 5.4 For all T B(H)
|T
T| = |T|
2
.
Proof. Since T
T is selfadjoint,
|T
T| = sup
x1
['T
(A I), we have
| (I A)x |=| (iI X)x | [[|x| .
If is real with > M, we have that for |x| = 1,
|(I A)x| = sup
y1
['(I A)x, y`[ '(I A)x, x` M
so that (dividing by |x|) we have |(I A)x| ( M)|x| for all x. A similar
proof holds when < m.
Theorem 5.5
(i) (A) [m, M] ,
(ii) m (A) and M (A) .
John Erdos 29
Proof. (i) Suppose [m, M] and let d = dist(, [m, M]). Let x H be any
unit vector and write = 'Ax, x`. Then '(I A)x, x` = 'x, (I A)x` = 0 and
|(I A)x|
2
= |[I I + (I A)]x|
2
= '[I I + (I A)]x, [I I + (I A)]x`
= [ [
2
|x|
2
+ (
)'(I A)x, x`
+( )'x, (I A)x` +|(I A)x|
2
[ [
2
d
2
.
It follows that |(I A)x| d|x| [apply the above for
x
x
]. Hence I A is
injective and, by Lemma 5.1, it has closed range. Further, if 0 = z ran(I A)
then 0 = '(I A)x, z` = 'x, (
, [m, M]),
we have |
((I K)x
k
i
+ Kx
k
i
)
and so (x
k
i
)
y
. Since (x
k
i
) is a sequence of unit vectors, y = 0. But then,
(I K)y = lim
i
(I K)x
k
i
=
y
y = 0 .
This contradicts the fact that is not an eigenvalue, so (a) is established.
(b) ran(I K) = H.
Let H
n
= ran(I K)
n
and write H
0
= H. It follows from (a) using Lemma 5.1 that
(H
n
) is a sequence of closed subspaces. Also
(I K)H
n
= H
n+1
H
0
H
1
H
2
H
3
.
Note that, if y H
n
then Ky = ((K I)y + y) H
n
so that K(H
n
) H
n
.
We now use the compactness of K to show that the inclusion H
n
H
n+1
is not
always proper. Suppose, on the contrary that
H
0
H
1
H
2
H
3
.
Using Lemma 1.7, for each n we can nd a unit vector x
n
such that x
n
H
n
and
x
n
H
n+1
. We show that (Kx
n
) cannot have a Cauchy subsequence. Indeed, if
m > n
Kx
n
Kx
m
= (K I)x
n
+ x
n
Kx
m
= x
n
+ [(K I)x
n
Kx
m
]
= x
n
+ z
where z H
n+1
[Kx
m
H
m
H
n+1
follows from m > n]. Thus
|Kx
n
Kx
m
|
2
= [[
2
+|z|
2
[[
2
32 Operators on Hilbert space
and so (Kx
n
) has no convergent Cauchy subsequence. Therefore, the inclusion is not
always proper. Let k be the smallest integer such that H
k
= H
k+1
. If k = 0 then
choose x H
k1
` H
k
. Then (I K)x H
k
= H
k+1
and so, for some y,
(I K)x = (I K)
k+1
y = (I K)z
where z = (I K)
k
y H
k
. Now x H
k
so x z = 0 and
(I K)(x z) = 0
contradicting the fact that is not an eigenvalue. Therefore k = 0, that is ran(I
K) = H
1
= H
0
= H.
(c) Completing the proof. This is done exactly as in Theorem 5.5 (i). For any
y H, there is a unique x H such that y = (I K)x. Dene (I K)
1
y = x.
Then |y| c|x| so
|(I K)
1
y| = |x|
1
c
|y|
showing that (I K)
1
B(H) (i.e. it is continuous). Thus (K).
Lemma 6.2 If x
n
are eigenvectors of K corresponding to dierent eigenvalues
x
n
, then x
n
is a linearly independent set.
Proof. This is exactly as in an elementary linear algebra course. Suppose the state-
ment is false and k is the rst integer such that x
1
, x
2
, , x
k
is linearly dependent.
Then
k
i=1
i
x
i
= 0 and
k
= 0. Also, by hypothesis Kx
i
=
i
x
i
with the
i
s all
dierent. Now x
k
=
k1
i=1
i
x
i
(where
i
=
i
/
k
) and so
0 = (
k
I K)x
k
=
k1
i=1
(
k
i
)
i
x
i
showing that x
1
, x
2
, , x
k1
is linearly dependent, contradicting the denition of k.
Theorem 6.3 (K)`0 consists of eigenvalues with nite-dimensional eigenspaces.
The only possible point of accumulation of (K) is 0.
Proof. Let be any non-zero eigenvalue and let N = x : Kx = x be the
eigenspace of . If N is not nite-dimensional, we can nd an orthonormal sequence
(x
n
) of elements of N [apply the Gram-Schmidt process (Theorem 3.4) to any linearly
independent sequence]. Then
|Kx
n
Kx
m
|
2
= |x
n
x
m
|
2
= 2[[
2
which is impossible, since K is compact.
To show that (K) has no points of accumulation other than (possibly) 0, we show
that C : [[ > (K) is nite for any > 0. Suppose this is false and there
is a sequence of distinct eigenvalues (
i
) with [
i
[ > for all i. Then we have vectors
x
i
with Kx
i
=
i
x
i
.
Let H
n
= spanx
1
, x
2
, , x
n
. Then, since x
n
is a linearly independent set, we
have the proper inclusions
H
1
H
2
H
3
H
4
.
John Erdos 33
It is easy to see that K(H
n
) H
n
and (
n
I K)H
n
H
n1
. Choose, as in The-
orem 6.1 a sequence of unit vectors (y
n
) with y
n
H
n
and y
n
H
n1
. Then, for
n > m,
Ky
n
Ky
m
=
n
y
n
[(
n
I K)y
n
Ky
m
] .
Since (
n
I K)y
n
H
n1
and Ky
m
H
m
H
n1
, the vector in square brackets is
in H
n1
. Therefore, since y
n
H
n1
,
|Ky
n
Ky
m
| > [
n
[ >
showing that (Ky
n
) has no convergent subsequence.
Corollary 6.4 The eigenvalues of K are countable and whenever they are put into
a sequence (
i
) we have that lim
i
= 0.
Proof. [The set of all eigenvalues is (possibly) 0 together with the countable union
of the nite sets of eigenvalues >
1
n
, (n = 1, 2, ).
If > 0 is given then, since : an eigenvalue of K, [[ is nite, we have that
[
i
[ < for all but a nite number of values of i. Hence (
i
) 0. ]
Corollary 6.5 If A is a compact selfadjoint operator then |A| equals its eigenvalue
of largest modulus.
Proof. This is immediate from Theorem 5.5 (ii).
The Fredholm alternative. For any scalar , either
(I K)
1
exists
or the equation
(I K)x = 0
has a nite number of linearly independent solutions.
(Fredholm formulated this result for the specic operator (Kf)(x) =
b
a
k(x, t)f(t) dt .
In fact, he said : EITHER the integral equation
f(x)
b
a
k(x, t)f(t) dt = g(x)
has a unique solution, OR the associated homogeneous equation
f(x)
b
a
k(x, t)f(t) dt = 0
has a nite number of linearly independent solutions.)
We now turn to compact selfadjoint operators. For the rest of this section A will
denote a compact selfadjoint operator.
Note that every eigenvalue of of A is real. This is immediate from Theorem 5.5, but
can be proved much more simply since if Ax = x, where x is a unit eigenvector,
x, x` = 'Ax, x` = .
Lemma 6.6 Distinct eigenspaces of A are mutually orthogonal.
34 Operators on Hilbert space
Proof. Let x and y be eigenvectors corresponding to distinct eigenvalues and .
Then,
'x, y` = 'Ax, y` = 'x, A
n
'x
n
, y` = 0 and so A(M) M. Therefore A with its domain restricted to M
is a compact selfadjoint operator on the Hilbert space M. Clearly this operator is
selfadjoint ['Ax, y` = 'x, Ay` for all x, y H so certainly for all x, y M]. Also
it cannot be have a no-zero eigenvector [for then M N
k
= (0) for some k > 0].
Therefore, by Corollary 6.5, it is zero. But then M N
0
. But also M N
0
and so
M = (0). Therefore (x
n
) is a basis.
Corollary 6.8 Then there is an orthonormal basis x
n
of H such that, for all h,
Ah =
n=1
n
'h, x
n
`x
n
.
Proof. Let (x
n
) be the basis found in the Theorem and let
n
= 'Ax
n
, x
n
` (this is
merely re-labeling the eigenvalues. The from Theorem 3.3 (iii), for any h H,
h =
n=1
'h, x
n
`x
n
.
Acting on this by A, since A is continuous and Ax
n
=
n
x
n
we have that
Ah =
n=1
n
'h, x
n
`x
n
.
Theorem 6.9 If A is a compact selfadjoint operator on a Hilbert space H then there
is an orthonormal basis x
n
of H such that
A =
n=1
n
(x
n
x
n
)
where the series is convergent in norm.
Proof. Let x
n
be the basis found as above so that Ax
n
=
n
x
n
and
Ah =
n=1
n
'h, x
n
`x
n
.
John Erdos 35
Note that (
n
) 0. Let
A
k
=
k
n=1
n
(x
n
x
n
) .
We need to show that |A A
k
| 0 as k .
Now
(A A
k
)h =
n=k+1
n
'h, x
n
`x
n
.
and, using Theorem 3.3 (v)
|(A A
k
)h|
2
=
n=k+1
[
n
'h, x
n
`[
2
sup
nk+1
[
n
[
2
n=k+1
['h, x
n
`[
2
sup
nk+1
[
n
[
2
n=1
['h, x
n
`[
2
= sup
nk+1
[
n
[
2
|h|
2
.
Thus |(AA
k
)| sup
nk+1
[
n
[, and so since (
n
) 0, we have that |AA
k
| 0
as k .
Alternatively, Theorem 4.4 may be used to prove the above result. Let x
n
and A
k
and A be as above and let
P
k
=
k
n=1
(x
n
x
n
) .
Then, since x
n
is a basis, Theorem 3.3 (iii) shows that (P
k
) converges pointwise to
the identity operator I. Since A
k
= AP
k
, Theorem 4.4 shows that (A
k
) converges to
A in norm.
36 Operators on Hilbert space
Exercises 6
1. Let K be a compact operator on a Hilbert space H and let = 0 be an eigenvalue
of K. Show that I K has closed range. [Hint : let N = ker(I K) and let
M = N
x
0
f(t) dt
.
Hints: Use Corollary 5.4 and the fact that the norm of the compact selfadjoint opera-
tor V
V is given by its largest eigenvalue. Now use the result of Exercises 2 Question
6 to show that if f satises V
V f = f then it satises
+ f = 0
f(1) = 0, f
(0) = 0.
[You may assume that any vector in the range of V
n=1
|Tx
n
|
2
converges. Prove that
(i) T is compact,
(ii)
n=1
|Ty
n
|
2
converges for every orthonormal basis y
n
of H and for the sum
is the same for every orthonormal basis.
Note : an operator satisfying the above is called a Hilbert-Schmidt operator.
Hints: (i) write h H as a Fourier series, h =
i=1
i
x
i
where
i
= 'h, x
i
`. Dene
T
n
h =
n
i=1
i
Tx
i
and show that
|(T T
n
)h|
2
n+1
[
i
[.|Tx
i
|
|h|
2
.
n+1
|Tx
i
|
2
.
(ii) Take an orthonormal basis
k
of H consisting of eigenvectors of the compact
operator T
T. Observe that if T
T
k
=
k
k
then
k
= 'T
T
k
,
k
` = |T
k
|
2
0.
Now use the spectral theorem for T
n=1
|Tx
n
|
2
converges then
n=1
|Tx
n
|
2
=
n=1
'T
Tx
n
, x
n
` =
k=1
k
.
Note that for a double innite series with all terms positive, the order of summation
may be interchanged.
John Erdos 37
7 The Sturm-Liouville problem.
In this section we shall discuss the dierential operator
Ly =
d
dx
p(x)
dy
dx
+ q(x)y
acting on functions y dened on a closed bounded interval [a, b]. We shall assume
that p(x) > 0 and q(x) real for a x b.
We make further assumptions that may be summarized, broadly speaking, by saying
that everything makes sense. Specically we need L to act on functions that are
twice dierentiable and whose second derivatives are in L
2
[a, b]. We also need to have
that p is dierentiable with p
p(x)
dy
dx
+ q(x)y = f(x) ()
subject to boundary conditions
1
y(a) +
2
y
(a) = 0
1
y(b) +
2
y
(b) = 0
()
Where
1
,
2
,
1
and
2
are real and
1
2
= 0,
1
2
= 0.
Note. The following calculation is of interest because it shows that L satises a symmetry
condition that would, for a bounded operator, make it self-adjoint. However, it will not be
used in the sequel. If L is restricted to act on the set of functions that satisfy the boundary
conditions, then 'Ly, z` = 'y, Lz`. Indeed,
'Ly, z` +'y, Lz` =
b
a
d
dt
p(t)
d z(t)
dt
y
d
dt
p(t)
dy
dt
dt +
b
a
(qy z +yq z) dt
=
p(t)
d z
dt
y(t) p(t)
dy
dt
z(t)
b
a
+
b
a
p(t)
dy
dt
d z
dt
d z
dt
.
dy
dt
dt
and, as the integrals on the right of each line are 0, this will vanish if
p(b)
(b)y(b) z(b)y
(b)
= p(a)
(a)y(a) z(a)y
(a)
.
But z
(a)y(a) z(a)y
(a) = 0 ,
y(a) +y
(a) = 0 ,
which has the non-trivial solution (, ) = (
1
,
2
) when y and z satisfy the boundary
conditions (). So z
(a)y(a) z(a)y
(b)y(b) z(b)y
(b) = 0. Therefore
'Ly, z` = 'y, Lz`.
We shall be looking for eigenvalues and eigenfunctions of L that satisfy the conditions
(); that is, for scalars and corresponding functions f that satisfy () and the equation
Lf = f. We make the additional assumption that = 0 is not an eigenvalue of the system.
This is quite a reasonable assumption, since if it fails then the problem (*), subject to ()
does not have a unique solution [an arbitrary multiple of the eigenfunction corresponding
to = 0 could be added to any solution to obtain another solution].
38 Operators on Hilbert space
Theorem 7.1 (Existence of the Greens function.) Under the assumptions stated above,
the problem (*), subject to () has the solution
y(x) =
b
a
k(x, t) f(t) dt
where k(x, t) is real-valued and continuous on the square [a, b] [a, b].
Proof. From the elementary theory of the initial value problem for linear dierential
equations, (also from Questions 1,2 and 3 of Exercises 6) we have that there is a unique
function u such that Lu = 0, u(a) =
2
, u
(a) =
1
. It follows easily that every solution
of Ly = 0,
1
y(a) +
2
y
(b) =
1
. The assumption that 0 is not an eigenvalue
implies that u and v are linearly independent [if u were a multiple of v then it would be an
eigenfunction].
Let
k(x, t) =
l u(x) a x t
mv(x) t x b
where (for xed t) l, m are constants to be chosen. [Our motivational work suggests that
we require k(x, t) to be continuous and p(x).
x
k(x, t) to have a unit discontinuity at x = t.]
Choose l, m such that
m.v(t) l.u(t) = 0,
p(t)[m.v
(t) l.u
(t)] = 1.
Solving for l, m gives
l =
v(t)
m =
u(t)
where = p(t)(v
u u
+u
(pv
) v(pu
(pu
) = quv +vqu = 0 ,
so is a constant (i.e. also independent of t). [One can see, independently of the theory
of Jacobeans, that = 0 since otherwise, at some point t
0
.u(t
0
) +.v(t
0
) = 0,
.u
(t
0
) +.v
(t
0
) = 0
has a non-trivial solution (, ). Then y = .u+.v is a solution of Ly = 0, y(t
0
) = y
(t
0
) = 0
and so .u + .v is identically 0, contradicting the linear independence of u and v.] Hence
we have that
k(x, t) =
u(x).v(t)
a x t ,
u(t).v(x)
t x b .
John Erdos 39
To complete the proof, we just verify directly that
y(x) =
b
a
k(x, t) f(t) dt
is the required solution. First note that when x = a we have x t throughout the range of
integration and so
y(a) =
1
b
a
u(a).v(t) f(t) dt
and, since y
(x) =
b
a
k(x, t)
x
f(t) dt,
y
(a) =
1
b
a
u
(a).v(t) f(t) dt .
Therefore
1
y(a) +
2
y
1
y(b) +
2
y
(b) = 0.
We now substitute into the equation. For notational convenience we substitute y(x)
(remembering that is constant). Since u(x), v(x) can be taken outside the integration,
we obtain
y(x) =
b
a
k(x, t) f(t) dt = v(x)
x
a
u(t) f(t) dt +u(x)
b
x
v(t) f(t) dt
(y(x))
= v(x)u(x)f(x) +v
(x)
x
a
u(t) f(t) dt
u(x)v(x)f(x) +u
(x)
b
x
v(t) f(t) dt
(p(x)(y(x))
= (pv
x
a
u(t) f(t) dt +pv
uf + (pu
b
x
v(t) f(t) dt pu
vf .
Therefore
(p(x)(y(x))
+qy = [(pv
+qv]
x
a
u(t) f(t) dt
+[(pu
+qu]
b
x
v(t) f(t) dt +pf(v
u u
v)
= f.
since u and v are solutions of Ly = 0.
40 Operators on Hilbert space
We can now apply the results of Section 6 to draw conclusions about the eigenfunctions
and eigenvalues of the Sturm-Liouville system (*), ().
Dene the operator K by
(Kf)(x) =
b
a
k(x, t) f(t) dt .
Since k is continuous on [a, b] [a, b] it is clear that
[k[
2
< and so, as shown in Section
4, K is compact.
Let T be the set of functions y such that Ly exists as a function in L
2
[a, b] and satises
the boundary conditions (). (There is a little technical hand waving here. A more precise
statement is: y T (which is an equivalence class of functions) if there is a representative
y such that y is dierentiable and p.(y)
L
2
[a, b]. Informally T is the all y L
2
[a, b] that qualify as solutions of (*),() for
some right hand side.) If y T and f = Ly then Theorem 7.1 shows that Kf = K(Ly) = y,
that is KL acts like the identity on T.
In the other order, it follows from the proof of Theorem 7.1, that for every f L
2
[a, b]
Kf T. (The verication that Kf is a solution of Ly = f explicitly shows this. Naturally,
for the most general f, the dierentiation of expressions like
x
a
u(t) f(t) dt one requires the
relevant background from Lebesgue integration.) Also, from Theorem 7.1, LKf = f. Thus
LK = I.
Note that L fails to be an inverse of K since it is not dened on the whole of L
2
[a, b], the
Hilbert space in question. Indeed, since K is compact, it cannot be invertible. However, L
is dened on a dense subset.
We use these notations and observations in the statements a proofs below.
Theorem 7.2 (i) The operator K does not have = 0 as an eigenvalue.
(ii) is an eigenvalue of K if an only if =
1
= ,
and is an eigenvalue of (*),().
Conversely, if is an eigenvalue of (*),(), by assumption = 0. We then have L =
and T. Then
KL = = K
and so K = where =
1
is an eigenvalue of K.
The consequences are immediate deductions from the results of Section 6 (principally 6.3,
6.4 and 6.7). Note that in this case the set of eigenvalues of K cannot be nite because this
would imply (by Corollary 6.8) that K vanishes on a non-zero (in fact, innite-dimensional)
subspace.
The most important result arising from this is consequence 2, since, for example, this is
what justies the expansions that are required in solving partial dierential equations by
the method of separation of variables.
Note. The assumption that 0 is not an eigenvalue of the system is not an essential restric-
tion. For any constant c, the eigenfunctions of L and L+c are the same and the eigenvalues
of L + c are + c whenever is an eigenvalue L. It is a fact that, by adding a suitable
constant to q we can always ensure that = 0 is not an eigenvalue of the system. For
example, if the boundary conditions are y(a) = y(b) = 0, choose c so that c + q does not
change sign in [a, b]; for deniteness, assume c + q(t) < 0 for a t b. Let u be the
(unique) solution of
L
c
y =
d
dt
p(t)
dy
dt
+ [c +q(t)]y = 0, y(a) = 0, y
(a) = 1 .
Any solution of L
c
y = 0, y(a) = 0 is a multiple of u so to show that 0 is not an eigenvalue
we must show that u(b) = 0.
Since u
(a) > 0 and u(a) = 0, it follows that u is strictly positive in some interval (a, a+).
Suppose z is the smallest zero of u that is > a (if any). Then u
vanishes
between a and z. Hence 0 is not an eigenvalue of L
c
y = 0, y T.
Similar, but more complicated arguments can be used for other boundary conditions (see
Dieudonne, Foundations of modern analysis, Chapter XI, Section 7).
The trigonometric functions form an orthonormal basis of L
2
[, ]. This fact
can be deduced from the work of the present section. [The trigonometric functions are
actually the eigenfunctions of y
() = y
(), and such systems are not covered by our work; however the device
below gives us the result.]
The eigenfunctions of the system y
(0) = y
(0) = y
f(t) sin nt dt =
f(t) sin nt dt +
0
f(t) sin nt dt =
0
[f(t) f(t)] sin nt dt .
Therefore f(t) = f(t) (almost everywhere) for 0 t , showing that f is an even
function on [, ]. A similar calculation with cosines shows that f is also an odd function
on [, ]. Thus f = 0 (almost everywhere) and the fact is proved.
John Erdos 43
Exercises 7
1. Let X be a Banach space (that is, a normed linear space that is complete). A series
n
x
n
in X is said to be absolutely convergent if the series
n
| x
n
| of real numbers is
converegent. Prove that an absolutely convergent series in a Banach space is convergent.
[Hint: prove that the sequence of partial sums is Cauchy.]
Existence theory for linear initial value problems using operator theory.
2. Let k(x, t) be bounded and square integrable over the square [a, b] [a, b], (in particular
this will hold if k is continuous on [a, b] [a, b]). Dene K : L
2
[a, b] L
2
[a, b] by
(Kf)(x) =
x
a
k(x, t)f(t) dt.
Prove that | K | (b a)M where M is a bound for k in [a, b] [a, b].
Let k
n
be dened inductively by k
1
= k and k
n
(x, t) =
x
t
k(x, s)k
n1
(s, t) ds. Prove that
(K
n
f)(x) =
x
a
k
n
(x, t)f(t) dt.
Show by induction that
[k
n
(x, t)[ M
n
[x t[
n1
(n 1)!
.
Using this result and the formal binomial expansion of (I K)
1
, deduce from Question 1
that (I K) has an inverse in B(H). [Hint : after proving absolute convergence, verify by
multiplication that the sum of the formal expansion is the inverse of (I K).] For each
= 0, observe that
K
y
(n)
(x) +p
1
(x)y
(n1)
(x) +. . . +p
n
(x)y(x) = f(x)
y(0) = y
(0) = . . . = y
(n1)
(0) = 0
where p
i
and f are continuous. By putting u(x) = y
(n)
(x), show that, for any b > 0, this
problem reduces to
() (I K)u = f
where K is an operator on L
2
[0, b] of the type considered in Question 6.
[Hint : show that y
(nr)
(x) =
x
0
(xt)
r1
(r1)!
u(t) dt.]
Prove that () has a unique solution in L
2
[0, b] for each b > 0. By quoting appropriate
theorems show that this solution is continuous (strictly, that the equivalence class of this
solution contains a continuous function). Deduce that there is a unique continuous function
with n continuous derivatives that satises () in [0, ).
Note that the general initial value problem
()
y
(n)
(x) +p
1
(x)y
(n1)
(x) +. . . +p
n
(x)y(x) = f(x)
y(0) = a
0
, y
(0) = a
1
, . . . = y
(n1)
(0) = a
n1
44 Operators on Hilbert space
can be transformed into the form (*) by changing the dependent variable from y to z where
z(x) = y(x)
n1
k=0
a
k
x
k
and thus () also has a unique solution.
4. Find a Greens function for the system
y
= f, y(0) = y(1) = 0 .
Check your answer by verifying that it gives x(x 1) as the solution when f = 2.
Evaluate the eigenvalues and eigenfunctions of
y
= y, y(0) = y(1) = 0 ,
and consequently nd an orthonormal basis of L
2
[0, 1].
5. Repeat the above question with the system
y
= f, y(0) +y
(0) = y(1) y
(1) = 0 .
John Erdos 45
8 The Spectral analysis of selfadjoint operators.
In this section A will always denote a selfadjoint operator with spectrum = (A) where
(A) [m, M] as dened in Section 6.
Theorem 8.1 (Spectral Mapping Theorem) Let T B(H). For any polynomial p,
(p(T)) = p() : (T) .
Proof. Let p be any polynomial. By the elementary scalar remainder theorem, x is a
factor of p(x) p(), that is, p(x) p() = (x )q(x) for some polynomial q. Therefore,
p(T) p()I = (T I)q(T) .
Now if p() (p(T)) then p(T) p()I has an inverse X and so
I = X.[p(T) p()I] = X.q(T).(T I) = [p(T) p()I].X = (T I).q(T).X .
Therefore (T I) has both a left inverse (namely X.q(T)) and a right inverse (q(T).X).
An easy algebraic argument shows that these are equal and (T I) has an inverse, that
is, (T). That is, if (T) then p() (p(T)).
Conversely, if k (p(T)) then the polynomial p(x) k factors over the complex eld into
linear factors: p(x) k = (x
1
)(x
2
)(x
3
) (x
n
) . Where
1
,
2
,
3
n
are
the roots of p(x) = k. Then
p(T) kI = (T
1
I)(T
2
I)(T
3
I) (T
n
I) ,
and these factors clearly commute. If each T
i
I has an inverse then the product of all
these would be an inverse of p(T) kI. Therefore T
i
I fails to have an inverse for at
least one root
i
of p(x) = k. That is k = p(
i
) for some
i
(T).
Corollary 8.2 If A is a selfadjoint operator then, for any polynomial p,
|p(A)| = sup[p()[ : (A)
Proof. If p is real, p(A) is selfadjoint. We know from Theorem 5.5 that for selfadjoint
operators, the norm equals the spectral radius. Therefore, using the result of the theorem,
|p(A)| = sup[k[ : k (p(A)) = sup[p()[ : (A) .
For the general case, since ( p.p)(A) = p(A).p(A) = p(A)
f(A) = f(A)
.
These follow easily from the denitions since, if (p
n
) and (q
n
) are sequences of polynomials
converging uniformly on [m, M] to f and g respectively, then
(f +g)(A) = lim
n
(p
n
+q
n
)(A) = lim
n
p
n
(A) + lim
n
q
n
(A) = f(A) +g(A)
and the other statements are easily proved in the same way.
Finally,
|f(A)| = | lim
n
p
n
(A)| = lim
n
|p
n
(A)| = lim
n
sup
[p
n
()[ = sup
[f()[ .
The extension of the functional calculus. We wish to extend the functional calculus
to limits of pointwise monotonically convergent sequences of continuous functions. (The
immediate goal is to attach a meaning to
[,t]
(A). Note that
[,t]
is a real function
which is equal to its square. Our hope is that E
t
=
[,t]
(A) will satisfy E
t
= E
2
t
= E
t
;
that is, that E
t
is a projection.)
We rst need a denition some technical results.
Denition. An operator A is said to be positive (in symbols A 0) if 'Ax, x` 0 for all
x H. We write A B to mean AB 0.
Note that, from the polarization identity, Theorem 2.3 and one of the exercises, any positive
operator is selfadjoint. Also, it follows from Theorem 5.5 that A 0 if and only if (A)
R
+
.
John Erdos 47
Lemma 8.4 Let A be a positive operator. Then
(i) ['Ax, y`[ 'Ax, x`.'Ay, y`
(ii) |Ax| |A|'Ax, x`.
Proof. (i) This is proved in exactly the same way as Theorem 1.1 from the fact that
'A(x +y), (x +y)` 0 for all .
(ii) The result is obvious if Ax = 0. If Ax = 0, using (i) with y = Ax we have
|Ax|
4
= ['Ax, Ax`[
2
'Ax, x`.'A
2
x, Ax`
'Ax, x`.|A
2
x|.|Ax|
'Ax, x`.|A|.|Ax|
2
and the result follows on dividing by |Ax|
2
.
Theorem 8.5 Let (A
n
) be a decreasing sequence of positive operators. Then (A
n
) converges
pointwise (strongly) to an operator A such that 0 A A
n
for all n.
Proof. Note that for each n, using Lemma normsa, |A
n
| = sup
x=1
'A
n
x, x` sup
x=1
'A
1
x, x` =
|A
1
|. The hypothesis shows that, for each x H, the sequence ('A
n
x, x`) is a decreasing
sequence of positive real numbers and hence is convergent. If m > n then A
m
A
n
0
and from Lemma 8.4 (ii),
|(A
m
A
n
)x|
2
|A
m
A
n
|.'(A
m
A
n
)x, x`
2.|A
1
|.['A
m
x, x` 'A
n
x, x`]
and this shows that (A
n
) is a Cauchy sequence, and so convergent. Call the limit Ax. It is
routine to show that A is linear, bounded, selfadjoint and 0 A A
n
for all n. [E.g. to
show that A is selfadjoint, we write
'Ax, y` = lim
n
'A
n
x, y` = lim
n
'x, A
n
y` = 'x, Ay` .
Lemma 8.6 If the sequences (A
n
), (B
n
) converge pointwise (strongly) to an operators A, B
respectively (and if |A
n
| is bounded) then (A
n
.B
n
) AB pointwise.
[Note that pointwise convergence is convergence in a topology on B(H) called the strong
operator topology; hence the alternative terminology. Note also that by a theorem of
Banach space theory (the Uniform Boundedness Theorem) the condition in brackets is
implied by the convergence of (A
n
).]
Proof. For any x H we have
|(A
n
B
n
AB)x| = |(A
n
B
n
A
n
B+A
n
BAB)x| |A
n
|.|(B
n
B)x| +|(A
n
A)Bx|
and the right hand side 0 from the pointwise convergence of (A
n
) and (B
n
) [at the points
Bx and x respectively].
Denition. The extended functional calculus. Let be a positive function on [m, M]
that is the pointwise limit of a decreasing sequence (f
n
) of positive functions C[m, M].
The operator (A) is dened as the pointwise limit of the sequence (f
n
(A)).
It is a fact that (A) is well dened (that is, it depends only on and not on the approxi-
mating sequence).
48 Operators on Hilbert space
Lemma 8.7 Let , be a positive functions on [m, M] that are the pointwise limits of a
decreasing sequences of positive functions C[m, M]. Then
(i) (A) +(A) = ( +)(A) .
(ii) (A).(A) = (.)(A) .
(iii) If X commutes with A then X commutes with (A) .
Proof. (i) Let (f
n
) and (g
n
) be decreasing sequences of functions that are continuous on
[m, M] and converge pointwise [m, M] to and respectively. Then (f
n
+g
n
) is a decreasing
sequence of continuous functions converging pointwise to + and
( +)(A) = lim
n
(f
n
+g
n
)(A) = lim
n
f
n
(A) + lim
n
g
n
(A) = (A) +(A)
where the limits indicate pointwise convergence in H.
(ii) As in (i) (f
n
.g
n
) is a decreasing sequence of continuous functions converging pointwise
to . and
(.)(A) = lim
n
(f
n
.g
n
)(A) = lim
n
f
n
(A). lim
n
g
n
(A) = (A).(A),
using Lemma 8.6, where the limits indicate pointwise convergence in H.
(iii) If X commutes with A then X commutes with f(A) for every f C[m, M] since if
(p
n
) is a sequence of polynomials converging uniformly to f on [m, M],
X.f(A) = lim
n
X.p
n
(A) = lim
n
p
n
(A).X = f(A).X .
Now if (f
n
) is a decreasing sequences of functions in C[m, M] that converge pointwise [m, M]
to then, using Lemma 8.6,
X.(A) = lim
n
X.f
n
(A) = lim
n
f
n
(A).X = (A).X .
For every real we dene the operator E
as follows: let
f
n,
=
1 t <
1 n(t ) t +
1
n
0 t > +
1
n
Then (f
n,
) is is a decreasing sequence of continuous functions converging pointwise to
[,]
. Now let E
=
[,]
(A). If < m then for all t [m, M] we have f
n,
(t) = 0 for
all suciently large n and so E
= 0. Similarly, E
= I for M.
Note that
E
.E
= E
where = min[, ] .
It follows that E
= E
2
. Also E
0 and so E
= E
.
A family of projections with these properties is called a bounded resolution of the
identity.
We call the family E
M
m
f() dE
of a function with respect to a bounded resolution of the identity exists and is equal to T
if, given any > 0 there exists a partition
0
<
1
<
2
<
n1
<
n
John Erdos 49
with
0
< m and
n
> M such that, for any (
i1
,
i
],
T
n
i=1
f(
i
)(E
i
E
i1
)
< .
For the proof of the Spectral Theorem, we need the following lemmas.
Lemma 8.8 Let
i
: 1 i n be orthogonal projections such that I =
n
i=0
i
and
i
.
j
= 0 when i = j. If X B(H) commutes with each
i
then
|X| = max
1in
|
i
X
i
| .
Proof. Let h H. Then
|h|
2
= |
n
i=0
i
h|
2
=
i=0
i
h,
n
j=0
j
h
=
n
i=0
|
i
h|
2
since the cross terms are zero. Therefore, since X
i
= X
2
i
=
i
X
i
=
i
X
|Xh|
2
=
n
i=0
|
i
Xh|
2
=
n
i=0
|
i
X
i
.
i
h|
2
i=0
|
i
X
i
|
2
.|
i
h|
2
max
1in
|
i
X
i
|
2
.
n
i=0
|
i
h|
2
= max
1in
|
i
X
i
|
2
. |h|
2
.
Thus |X| max
1in
|
i
X
i
| . But the opposite inequality is clear, since for each i we
have |
i
X
i
| |
i
|.|X|.|
i
| = |X|.
Corollary 8.9
M
m
f() dE
sup
mtM
[f(t)[ .
Proof. From the lemma,
i=1
f(
i
)(E
i
E
i1
)
= max
1in
|f(
i
)(E
i
E
i1
)| = max
1in
[f(
i
)[ sup
mtM
[f(t)[ .
Since the integral is approximated arbitrarily closely in norm by these sums, the result
follows.
Lemma 8.10 For > ,
(E
) (E
)A(E
) (E
) .
50 Operators on Hilbert space
Proof. First note the general fact that if S T then for any operator X we have X
SX
X
TXh, h` .
Next we claim that AE
. Let f
,n
be as in the denition of E
and dene g
n
(t) =
( t)f
,n
+
1
n
. It is easy to see that g
n
0. [This is obvious when t , and also
when t +
1
n
, for then f
,n
(t) = 0; for t (, +
1
n
) we have t >
1
n
so, since
0 f
,n
(t) 1, it follows that g
n
(t) 0.] Also (g
n
) is decreasing (this is an elementary but
tedious exercise) and so the pointwise (strong) limit of (g
n
(A)) exists and is positive. But
(g
n
(t)) is pointwise convergent to (, ](t) t(, ](t). Therefore
E
AE
0
proving the claim.
Note that E
= AE
AE
= AE
2
.
We now use that (I E
) = (I E
)A(E
) = (I E
)E
AE
(I E
) (I E
)E
(I E
) = (E
) .
The fact that (E
) (E
)A(E
such that
A =
M
m
dE
.
Proof. Let E
i
= E
i
E
i1
. From Lemma 8.10 we have that
i1
i
i
A
i
i
i
.
Hence, for any
i
[
i1
,
i
],
(
i1
i
)
i
i
A
i
i
(
i
i
)
i
.
Note that when |h| = 1, since
i
is an orthogonal projection, 0 '
i
h, h` = |
i
h|
2
1 .
Therefore, using Theorem 5.3
|
i
A
i
i
| max[[
i
i
[, [
i1
i
[] < .
Observe that
i
: 1 i n satises the hypotheses of Lemma 8.8 and that X =
(A
n
i=1
i
) commutes with each
i
. Therefore, applying Lemma 8.8 we have that
|(A
n
i=1
i
)| = max
1in
|
i
A
i
i
| <
and this is exactly what is required.
i
.
j
= 0 when i = j.
John Erdos 51
Corollary 8.12 If f is continuous on [m, M] then
f(A) =
M
m
f() dE
.
Proof. For any integer k, choose n,
1
,
2
n
,
0
,
1
, . . . ,
n
as in the Theorem so that
(A
n
i=1
i
)
<
1
k
and write 1
k
=
n
i=1
i
. Then lim
k
1
k
= A. Therefore, for any integer r,
A
r
= lim
k
1
r
k
.
But
1
r
k
=
i=1
r
=
n
i=1
r
i
i
,
and the right hand side is the approximating sum to
M
m
r
dE
. Therefore
A
r
=
M
m
r
dE
.
and, by taking linear combinations,
p(A) =
M
m
p() dE
such that
sup
mM
[f() p()[ < .
Then, from Theorem 8.3 |f(A) p(A)| < and
f(A)
M
m
f() dE
f(A)
M
m
p() dE
M
m
p() dE
M
m
f() dE
= |f(A) p(A)| +
M
m
p() f() dE
< 2 ,
the last step using Corollary 8.9. Since is arbitrary, it follows that
f(A) =
M
m
f() dE
.
52 Operators on Hilbert space
Exercises 8
1. (More spectral mapping results.) If X B(H) show that
(i) (X
) = :
(X),
(ii) if X is invertible then (X
1
) =
1
: (X).
Deduce that every member of the spectrum of a unitary operator has modulus 1.
2. For any selfadjoint operator A, prove that ker A = (ranA)
.
3. Let A and B be selfadjoint operators. Show that if there exists an invertible operator T
such that T
1
AT = B then there exists a unitary operator U such that U
AU = B (that
is, if A and B are similar then they are unitarily equivalent).
[Hint: use the polar decomposition of T.]
4. Suppose X B(H) is selfadjoint and | X | 1. Observe that X + i
I X
2
can be
dened and prove that it is unitary. Deduce that any operator T can be written as a linear
combination of at most 4 unitary operators. [First write T = X+iY with X, Y selfadjoint.]
5. (i) Use results on the spectrum show that A kI (where k R) if and only if for all
(A), k. [Note that A is selfadjoint, since 'Ax, x` is real make sure you know how
this follows!] Deduce that if A I then A
n
I for every positive integer n. [Alternatively
factorise A
n
I.]
(ii) If B and C commute and B C 0 then prove that B
n
C
n
for every positive
integer n. [Factorise B
n
C
n
.]
6. Let U be both selfadjoint and unitary. Prove that (U) = 1, 1 (unless U = I).
[Use Question 1.] Use the spectral theorem to nd an orthogonal projection E such that
U = 2EI; (alternatively, if you are given the result it is trivial to verify that E =
1
2
(U +I)
is a suitable E). Note that a self adjoint isometry must be unitary [use Question 2]. Deduce
that the only positive isometry is I.
[Denition: V is an isometry if |V h| = |h| for all h H.]