Lecture 2 (2-10-03)

1.(a) C(X) is a vector space, dening

(f +g)(x) := f(x) +g(x)
(f)(x) := f(x)
for f, g C(X), C, x X.
(b) Clearly f

0 and f

= 0 i f(x) = 0 for all x X, that is,

i f = 0 in C
b
(X). It is also obvious that f

= || f

, for any
C. We are left with showing that the triangle inequality holds. But,
since |f(x) + g(x)| |f(x)| + |g(x)|, for any f, g C
b
(X), x X, we also
have that f +g

+g

.
(c) Let (f
n
)
nN
be a Cauchy sequence in C
b
(X). Then for any > 0, there
is N N such that for any n, m > N,
f
n
f
m

< |f
n
(x) f
m
(x)| < , for all x X. (1)
In particular, the sequence (f
n
(x))
nN
is Cauchy, for any x X, and there-
fore converges in C. Let f(x) := limf
n
(x). Letting m in (1), we have
that for all n > N,
|f
n
(x) f(x)| < , for all x X f
n
f

< .
We then have that f is bounded, since, for some xed n > N and any x X,
|f(x)| |f(x) f
n
(x)| +|f
n
(x)| +f
n

< M,
and continuous, since f
n
is and
|f(x) f(y)| |f(x) f
n
(x)| +|f
n
(x) f
n
(y)| +|f(y) f
n
(y)|.
We conclude that f C
b
(X) and f
n
f in C
b
(X), and hence C
b
(X) is
complete.
(d) Let f, g, h C
b
(X). Then,
(fg)(x) = (gf)(x)
(fg)h(x) = f(x)g(x)h(x) = f(gh)(x)
1
(f +g)h(x) = f(x)h(x) +g(x)h(x) = (fh +gh)(x)
that is, multiplication is commutative, associate and distributive with re-
spect to addition. Therefore C
b
(X) is a commutative algebra.
(e) We have that, for any f, g C
b
(X),
f g

= sup
xX
|f(x)g(x)| = sup
xX
|f(x)| |g(x)| sup
xX
|f(x)| sup
xX
|g(x)| = f

.
(f) For f, g C
b
(X), C:
(fg)

(x) = (fg)(x) = f(x)g(x) = f

(x)g

(x) = g

(x)f

(x)
f

(x) = f(x) = f(x)

(f)

(x) = f(x) = f

(x).
(g) Just note that |f

(x) f(x)| = |f(x)f(x)| = |f(x)|

2
, and therefore
f

= sup
xX
|f(x)|
2
=
_
sup
xX
|f(x)|
_
2
= f
2

.
(h) Need to show that C(X) C
b
(X), that is that every continuous function
on a compact space X is bounded. But the image of a compact set under
a continuous function is compact, that is, for any f C(X), f(X) C is
compact and hence bounded.
2.(a) M
n
(C) is a vector space with
a +b := (a
ij
+b
ij
) , a := (a
ij
)
for a = (a
ij
) , b = (b
ij
) M
n
(C), C, and it is clearly an algebra with
respect to the usual matrix multiplication
(ab)
ij
:=
n

k=1
a
ik
b
kj
, i, j = 1, ..., n.
(b) Write (ab)

= (c
ij
), where
c
ij
=
n

k=1
a
jk
b
ki
=
n

k=1
a

kj
b

ik
= (b

)
ij
2
and we have (ab)

= b

, for all a, b M
n
(C). Also,
(a)

ij
= a
ji
= a

ij
a

ij
= (a
ji
)

= a
ij
= a
ij
.
(c) (i) We clearly have a 0, for a M
n
(C), and a = 0 i az = 0 for all
z C, that is, i a = 0. Also, since (a)z = || az, for any a M
n
(C),
z C
n
, C, we have a = || a. The triangle inequality follows from
the fact that (a +b)z az +bz, for any a, b M
n
(C), z C
n
.
To check the last assertion, write z
2
= (z, z), where (z, w) :=

n
k=1
z
k
w
k
is the usual inner product in C
n
. It is easy to check that for any a M
n
(C),
z, w C
n
, (az, w) = (z, a

w), with a

dened as in (b). Then,

az
2
= (az, az) = (z, a

az)
and if is an eigenvalue of a

a and v an eigenvector, with v = 1, we have

a
2
av
2
= (v, v) = a
2
max
(and in particular we get that R
+
). On the other hand, let {v
k
} be a
basis of orthonormal eigenvectors of a

a, with respect to eigenvalues

k
, k =
1, ..., n (such a basis exists since a

a is self-adjoint). Writing z =

n
k=1
c
k
v
k
,
we have
az
2
=
n

j=1
n

k=1
c
k
c
j
(v
k
,
j
v
j
) =
n

k=1
|c
k
|
2

k
max
k
z
and therefore
a
2
max
k
.
(ii) Let (a
k
)
kN
be a Cauchy sequence in M
n
(C). Then for any > 0,
there exists N N such that for any k, m > N,
a
k
a
m
< |a
k
z a
m
z| < , for all z C
n
, z = 1. (2)
In particular, the sequence (a
k
z)
kN
is Cauchy, for any z C
n
, z = 1,
and therefore converges in C. For z = 1, let az := lima
k
z; this denes
a M
n
(C). Letting m in (2), we have that for all k > N,
|a
k
z az| < , for all z C
n
, z = 1 a
k
a < .
3
We conclude that a
k
a in M
n
(C), and hence that M
n
(C) is complete.
(iii) Have, for any z C
n
, abz a bz a b z, and therefore
ab = sup
z=1
abz a b.
(iv) For a M
n
(C), z C
n
,
az
2
= (az, az) = (z, a

az) za

az z
2
a

a
and therefore
a
2
a

a a

a (3)
and we have then that
a a

.
Since a

= a, equality follows.
(v) From a = a

and (3)
a
2
a

a a

a = a
2
and therefore a
2
= a

a, for all a M
n
(C).
3.(a) B(H) is a vector space with
(a +b)z := az +bz, (a)z := az
for a, b B(H), C, z H, and an algebra endowed with composition
of operators.
(b) Let a, b B(H), z, w H. Have that
(z, b

w) = (bz, a

w) = (abz, w) (ab)

= b

;
(z, a

w) = (a

z, w) = (w, a

z) = (aw, z) = (z, aw) a

= a;
(z, (a)

w) = (az, w) = (az, w) = (z, a

w) (a)

= a

.
(c) (i) Showing that is indeed a norm is done in exactly the same way as
in Ex.2.
Note It is no longer true that a
2
coincides with the norm of the largest
eigenvalue of a

a; in fact, a

a may not have any eigenvalues (see Exercise

4
4, Lecture 5). We have however that a
2
= r(a

a), where r(a

a) is the
spectral radius of a

a (see Exercise 1, Lecture 5).

(ii) Let (a
n
)
nN
be a Cauchy sequence in B(H). Then for any > 0,
there exists N N such that for any n, m > N,
a
n
a
m
< |a
n
z a
m
z| < , for all z H, z = 1. (4)
In particular, the sequence (a
n
z)
nN
is Cauchy in H, for any z H, z = 1,
and it converges since H is complete. For z = 1, let az := lima
n
z; this
denes a linear operator a : H H. Letting m in (4), we have that
for all n > N,
|a
n
z az| < , for all z H, z = 1 a
n
a < .
In particular, since a
n
is bounded, we see that a is also bounded, and we
conclude that a
n
a in B(H), and hence that B(H) is complete.
(iii) (Exactly as in Ex.2.)
(iv) (Exactly as in Ex.2.)
(v) (Exactly as in Ex.2.)
4.(a) For any L
2
(X; ),
(f)
2
2
=
_
X
d(x)|f(x)|
2
(x)|
2
(f

)
2

2
2
and hence (f) is bounded, with (f) f

. To prove equality, let

L < f

be arbitrary. Then, by denition of

,
({x X : |f(x)| > L}) > 0
and since is -nite, there is a measurable Y {x X : |f(x)| > L} with
0 < (Y ) < . Let
Y
L
2
(X; ) be the characteristic function of Y . We
have that
(f)
Y

2
2
=
_
X
d(x)|f(x)|
2

Y
(x)|
2
L
2

2
2
.
We conclude that (f) L, for all L < f

and therefore that (f) =

f

.
5
(b) It is clear that is linear and that (fg) = (f)(g), for any f, g
C
b
(X). Moreover,
((f

), ) =
_
X
d(x)(f

)(x)(x) =
_
X
d(x)(x)f(x)(x) = (, (f)),
for all , L
2
(X; ), and hence (f)

= (f

).
(c) We show rst that f

= f

for all f C
b
(X) i (U) > 0, for any
open U X, U = . Let then U X be open, U = , and V U be closed.
By Urysohns lemma, there exists f C
b
(X) such that 0 f 1, f = 1 on
V and suppf U. Then f

= 1 and if f

= f

, we conclude that
(U) > 0 (if (U) = 0 then f

= 0, since f = 0 on X\U). Conversely,

assume that any non-empty open set in X has non-zero measure and let
f C
b
(X). (Note that one always has f

.) For any > 0, the

set
U

:= {x X : |f(x)| > f

}
is open and U

= . Hence (U

) > 0 and therefore we have that f

>
f

,
1
for any > 0, that is, f

.
We can now see that this condition is equivalent to supp() = X. Recall
that the support of a measure is the smallest closed subset A X such
that (E) = (E A), for all measurable E X. Note that in this case,
(X\A) = 0 and X\A is open. Hence, if every non-empty open set in X has
positive measure, we conclude that X\A = , that is, supp = X. Now if
supp = X, we have that for any open U X, U = , there is a measurable
set E X such that (EU) > 0, since (E) = (EU) +(E (X\U))
and X\U is not the support of . Therefore (U) (E U) > 0.
2
(d) We have in this case (f) = f

= f

and therefore (f) = 0 i

(f) = 0 i f

= 0 i f = 0, that is, is injective.

(e) Let be a measure such that the condition in (c) is satised and let
A := (C
b
(X)). From (b), we have that is a -homomorphism, that is, A
is a -subalgebra of B(H), and, since is bijective, A and C
b
(X) are iso-
morphic as -algebras. Now, since is isometric, A is closed and C
b
(X)

= A
1
Note that f

is the smallest c such that ({x X : |f(x)| > c}) = 0.

2
It can be easily seen that the Lebesgue measure on R
n
satises this condition and
it can in fact be shown that such a measure always exists on arbitrary locally compact
spaces.
6
as C

-algebras.
7
Lecture 3 (9-10-03)
1.(a) Let f, g L
1
(R). We have
_
R
dx|f g(x)| =
_
R
dx

_
R
dy f(x y)g(y)

_
R
dx
_
R
dy |f(x y)| |g(y)|.
Since (x, y) f(x y)g(y) is measurable on R
2
, and
_
R
dy
_
R
dx|f(x y)||g(y)| =
_
R
dy |g(y)|
_
R
dx|f(x y)|
=
_
R
dy |g(y)|
_
R
dz |f(z)| = g
1
f
1
< ,
we have from Fubinis theorem that f g L
1
(R) and
f g
1
g
1
f
1
. (5)
It is easy to check that L
1
(R) is an algebra, and, noting that the Lebesgue
measure on R is invariant under translations and transformations x x,
g f (x) =
_
R
dy g(x y)f(y) =
_
R
dz g(z)f(x z) = f g (x),
that is, L
1
(R) is commutative. Since it is complete and (5) holds, we have
nally that L
1
(R) is a commutative Banach algebra.
(b) First note that any functional w (L
1
(R)) is bounded, with w 1
1
,
and therefore (L
1
(R)) (L
1
(R))

. There is an isomorphism (L
1
(R))

=
L

(R)
2
, and therefore for each w (L
1
(R)) there exists a uniquely
dened w L

(R) such that

w(f) =
_
R
dxf(x) w(x). (6)
We will show that w(x) = e
itx
, for some t R and that the correspondence
t e
itx
denes an isomorphism between R and (L
1
(R)).
1
This is true for unital algebras, and the result on the non-unital case follows easily;
see also Exercise 5, Lecture 6.
2
See for instance Rudin, Real and complex analysis, page 127.
8
Let f, g L
1
(R) and w (L
1
(R)), w L

(R) be given as in (6). We

have that
w(f g) =
_
R
dx
_
R
dyf(x y)g(y) w(x)
=
_
R
dy g(y)
_
R
dxf(x y) w(x)
=
_
R
dy g(y)w(f
y
),
where f
y
(x) := f(x y) L
1
(R). On the other hand,
w(f)w(g) =
_
R
dy g(y) w(y)w(f),
and since w is multiplicative, we have that
w(y)w(f) = w(f
y
), a.e.(dy). (7)
Now w : L
1
(R) C is bounded and for each xed f L
1
(R), the map
y R f
y
L
1
(R) is continuous. Taking f such that w(f) = 0, we
conclude that w is continuous on R.
1
We now show that w(x + y) = w(x) w(y). Note rst that for any f
L
1
(R) such that w(f) = 0, w(f
y
)w(f)
1
= w(y) does not depend on f,
and therefore w(f
y
)w(f)
1
= w(g
y
)w(g)
1
, for all y R, f, g L
1
(R) such
that w(f) = 0, w(g) = 0. In particular, since w is multiplicative, we get
w(f
y
g) = w(f g
y
). Also, for any such f, w(f
y
) = 0 for any y R, since
w(f)
2
= w(f f) = w(f
y
f
y
) = w(f
y
)w(f
y
) = 0. In this case, we have
w(x +y) = w(f
x+y
)w(f)
1
= w((f
x
)
y
))w(f
x
)
1
w(f
x
)w(f)
1
, a.e.,
and therefore that
w(x +y) = w(x) w(y) a.e, (8)
and since w is continuous, equality holds for all x, y R.
Notes 1. Property (8) could have been derived directly from the multiplica-
tivity of w: note that for any f, g L
1
(R),
w(f g) =
_
R
dx
_
R
dyf(x y)g(y) w(x)
=
_
R
dx
_
R
dyf(x)g(y) w(x +y)
1
That is, w has a continuous representative and we can therefore assume that w C(R).
9
and on the other hand
w(f)w(g) =
_
R
dx
_
R
dy f(x)g(y) w(x) w(y).
We then have that for any g L
1
(R),
_
R
dy g(y) ( w(x +y) w(x) w(y)) = 0 a.e(dx),
and hence that w(x +y) = w(x) w(y) a.e.
2. It is true that a measurable function satisfying (8) is necessarily con-
tinuous; the direct proof of this result is however far from trivial. This is
the reason we proceed in this fashion.
Now we know that w is bounded and we have that, for any n Z,
w(nx) = w(x)
n
(note that w(0) = 1 and therefore w(x) = w(x)
1
). We conclude that
| w(x)| = 1 for all x R. We have then shown that w is a continuous
map R T := {z C : |z| = 1}, which satises w(x + y) = w(x) w(y).
1
At this point, we could use the known result that any character of R is
of the form e
itx
for some t R; we give here a proof of this fact. Let
V := {e
i
: || < /2} T; since w : R T is continuous, we can take
an open U R such that U w
1
(V ) and can assume that U = (L, L),
L > 0. Now arg(e
i
) = denes a homeomorphism arg : V (/2, /2)
and for x, y U such that x+y U, write w(x) = e
ix
V , w(y) = e
iy
V .
Then w(x +y) = w(x) w(y) = e
i(x+y)
V and we have
arg( w(x +y)) =
x
+
y
= arg( w(x)) +arg( w(y)).
Therefore, arg w is a continuous function on R such that (arg w)(nx) =
n arg w(x), for any x U, n Z with nx U and by density, we have
that arg w(x) = cx for all x U. Since arg is a homeomorphism, we have
then that w(x) = e
cx
, for x U, and since | w(x)| = 1, we get c = it, for
some t R. But U = (L, L) R generates R as an additive group and
hence, from (8), we have w(x) = e
itx
, for all x R.
We conclude nally that any w (L
1
(R)) is given by
w(f) =
_
R
dxf(x) e
itx
, (9)
1
Such a map is called a character of R.
10
for some t R. It is easy to check that, for any t R, w
t
: L
1
(R) C
dened by (9) yields an element of (L
1
(R)) and that w
t
= w
s
i t = s. We
thus have a well-dened, bijective map
: R (L
1
(R)), t w
t
w
t
(f) =
_
R
dxf(x)e
itx
, f L
1
(R).
We now check that this map is a homeomorphism. If t
n
t, with t
n
, t R
then, for any f L
1
(R), w
tn
(f) w
t
(f), by the Lebesgue Dominated
Convergence Theorem, and therefore w
tn
w in the weak*-topology. We
conclude that is continuous. To check that it is bicontinuous, let w
t
i
be
a net in (L
1
(R)) such that w
t
i
w
t
in the weak

-topology. Then for any

f L
1
(R) we have that
_
R
dxf(x)e
it
i
x

_
R
dxf(x)e
itx
.
Let f(x) := e
x
, where is the characteristic function of (, 0). Then
f L
1
(R) and, for any s R,
_
R
dxf(x)e
isx
=
_
0

dxe
(is+1)x
=
1
is + 1
.
Hence,
1
it
i
+ 1

1
it + 1
t
i
t
and
1
is also continuous.
(c) The Gelfand transform is the map
L
1
(R) C((L
1
(R)), f

f
where

f() = (f). Identifying (L
1
(R)) with R as in (b), we then have
that

f(t) =
_
R
dxf(x)e
itx
, which is nothing else but the Fourier transform
of f.
2.(a) The space L
1
(Z) is the space of sequences (a
n
)
nZ
of complex numbers
such that

nZ
|a
n
| < ; it is a Banach space endowed with the norm
a
1
=

nZ
|a
n
|.
11
Dene convolution in L
1
(Z) by (a b)
n
:=

mZ
a
nm
b
m
, for a, b
L
1
(Z). We have

nZ

mZ
a
nm
b
m

nZ

mZ
|a
nm
||b
m
| =

mZ
|b
m
|

nZ
|a
nm
| = a
1
b
1
and therefore a b L
1
(Z), with a b
1
a
1
b
1
. It is clear that L
1
(Z)
equipped with is a commutative algebra, and, since it is complete, it is a
Banach algebra. It is easy to check that the element e
0
L
1
(Z) such that
(e
0
)
n
= 0, for n = 0 and (e
0
)
0
= 1 is an identity for and L
1
(Z) is unital.
(b) We will show that (L
1
(Z))

= T, where T = {z C : |z| = 1}.

Let e
k
L
1
(Z), for k Z, be such that (e
k
)
n
=
kn
. For any a L
1
(Z)
we can write a =

nZ
a
n
e
n
, and for any w (L
1
(Z)) we have that
w(a) =

nZ
a
n
w(e
n
).
1
(10)
It is easy to check that e
k
e
m
= e
m+k
, for all k, m Z. In particular,
e
k
= (e
1
)
k
and, since w is multiplicative, we get
w(a) =

nZ
a
n
(w(e
1
))
n
.
We conclude that each w (L
1
(Z)) is uniquely determined by w(e
1
) (in
the sense that if w, w

(L
1
(Z)) and w(e
1
) = w

(e
1
), then w = w

). We
now show that w(e
1
) T. Since e
k
= 1, for any k Z and w = 1
(because w (L
1
(Z)), we have
|w(e
k
)| we
k
= 1, for all k Z.
Noting that w(e
1
e
1
) = w(1) = 1, we have also that |w(e
1
)| = |w(e
1
)|
1

1 and therefore |w(e

1
)| = 1 and w(e
1
) T. We thus have a well-dened,
injective map
: (L
1
(Z)) T, w w(e
1
).
On the other hand, for each z T one can dene
w
z
(a) =

nZ
a
n
z
n
, a L
1
(Z)
1
Note that we showed that any w (L
1
(Z))

is given by w(a) =

an wn, where
w L

(Z); compare with Exercise 1.

12
and it is easy to check that w
z
(L
1
(Z)) with w
z
(e
1
) = z. Hence is also
surjective. Now, directly from the denition of the weak- topology, we have
that is continuous
1
. Since (L
1
(Z)) is compact (and T is Hausdor), we
conclude that is a homeomorphism.
(c) The Gelfand transform is the map
L
1
(Z) C((L
1
(Z)), a a,
where a(w) = w(a), w (L
1
(Z). Under the isomorphism given in (b) we
have that z T w
z
(L
1
(Z)) and we can write a C(T) such that
a(z) = w
z
(a) =

nZ
a
n
z
n
.
Now if a(n) denotes the n-th Fourier coecient of a, we have
a(n) =
1
2
_
2
0
a(e
it
)e
int
dt =
1
2

mZ
a
m
_
2
0
e
i(mn)t
dt = a
n
.
The Gelfand transform is then the map that to each a L
1
(Z) associates
the continuous function a on T whose Fourier coecients are given by a.
Note Functions on T can be identied with periodic functions on R: if
f L
1
(T), then

f(t) := f(e
it
) denes

f L
1
(R) which is periodic and
the converse is also true. In this light, given g L
1
(R), periodic, such
that a
g
:= ( g(n))
nN
L
1
(Z) (where g(n) denotes the nth Fourier coef-
cient of g), we have that the Gelfand transform a
g
C(R) is given by
a
g
(t) =

nZ
g(n)e
int
, the Fourier series of g.
3.(a) First note that a basis of neighbourhoods of w
0
(A) on the weak

-
topology is given by
U
a,
:= {w (A) | |w(a) w
0
(a)| < },
a A, > 0. Now let w
1
, w
2
(A) be such that w
1
= w
2
; then
there is a A such that w
1
(a) = w
2
(a). Let < |w
1
(a) w
2
(a)|/2 and
U
i
:= {w (A) : |w(a) w
i
(a)| < }, for i = 1, 2. It is clear that U
i
is a
neighbourhood of w
i
and U
1
U
2
= and therefore (A) is Hausdor.
1
Strictly speaking, we know only that (L
1
(Z)) C, w w(e1) is continuous, but
since (L
1
(Z)) is in fact mapped to T, this yields the continuity of .
13
One could also use the fact that a space is Hausdor i no lter on it
has more than one limit point. Now if (

)
I
is a lter in (A) such that

and

then, by the denition of the weak

-topology, for any

a A,
|

(a) (a)| 0
|

(a)

(a)| 0
hence (a) =

(a) for any a A, that is =

and (A) is Hausdor.

Alternatively, one could also use the fact that the map
a
: (A) C
is continuous with the weak

-topology. Considering neighbourhoods U

i
C
of w
i
(a), i = 1, 2, such that U
1
U
2
= , one gets neighbourhoods
1
a
(U
i
)
(A) of w
i
, i = 1, 2, with
1
a
(U
1
)
1
a
(U
2
) = .
(b) Have w(a) = w(a.1) = w(a)w(1), for all a A. Since w is non-zero,
there is a A with w(a) = 0, and hence w(1) = 1.
(c) Let w
n
(A), n N, be such that w
n
w in the weak*-topology, that
is, we have w
n
(a) w(a), for all a A. Then w A

and we need only show

that w is multiplicative and non-zero. Let a, b A. Then w
n
(ab) w(ab)
and also, w
n
(a)w
n
(b) w(a)w(b), we have that w(ab) = w(a)w(b). Since
w
n
(1) = 1 for all n N, have that w(1) = 1 and therefore w (A).
(d) If A is non-unital then (A) is not necessarily closed (one may now have
w
n
(A) and w
n
0 / (A)). We give two simple examples:
A = {f C([1, 1]) : f(0) = 0}. It is easy to check that A is a C

-
subalgebra of C([1, 1]), without unit, and that for any t [1, 1], with
t = 0, w
t
(f) := f(t) denes an element w
t
(A). Now let w
n
:= w
1/n
; for
any f A we have
w
n
(f) = f(1/n) f(0) = 0
and hence w
n
0 in the weak*-topology. But 0 / (A) and (A)) is not
closed.
A = C
0
(X), where X is a locally compact space that is not compact. If
(A) were closed, then by the Banach-Alaoglu theorem it would be com-
pact. But we know that (A)

= X (see Theorem 6.11 and Corollary 6.8)
hence (A) is not closed.
14
Lecture 5 (23-10-03)
1. For a A, we have that, since A is commutative and from the C

-axiom
b

b = b
2
, for any b A,
a
4
= a

a
2
= (a

a)

a = (a
2
)

a
2
= a
2

2
that is, a
2
= a
2
. By induction on p N, we have a
n
= a
n
, for
n = 2
p
. Therefore,
r(a) = lima
n

1/n
= lima
2
p

1/2
p
= a.
2. We have that f C(X) is invertible i f(x) = 0, for all x X (in this
case f
1
= 1/f C(X)). Hence, (f) i f 1 is not invertible i
there exists x X such that (f 1)(x) = 0 i there exists x X such
that f(x) = . Conclude that (f) = {f(x) : x X}.
3. We have that A M
n
(C) is invertible i det(A) = 0. Therefore, (A)
i det(A1) = 0, that is, i is an eigenvalue of A.
4. Examples of a bounded operator on a Hilbert space with no eigenvalues:
(1) Let H be a (separable) Hilbert space with an orthonormal basis {e
n
}
nN
and dene a linear operator s
r
: H H by
s
r
(e
n
) = e
n+1
, n N.
This is an injective bounded operator, with s
r
(x) = x, for all x H.
In particular, 0 is not an eigenvalue. Let C, = 0, and x =

x
n
e
n
be
such that s
r
x = x; have that
x
n
= x
n+1
, x
1
= 0
and therefore x
n
= 0, for all n N, and is not an eigenvalue. Conclude
that s
r
does not have any eigenvalues.
A particular instance of this example is the so-called right-shift operator
s
r
: l
2
l
2
such that
s
r
(x
1
, x
2
, ..., x
n
, ...) = (0, x
1
, x
2
, ..., x
n
, ...),
which as we have seen has no eigenvalues.
1
1
But one can show that (sr) = [0, 1].
15
(2) Let H := L
2
(X; ), where X is a compact topological space, f C(X)
and dene
M
f
: L
2
(X; ) L
2
(X; ), g fg.
Then
M
f
(g) = g f(x)g(x) = g(x) a.e. (f(x) )g(x) = 0 a.e.
and one can see that is an eigenvalue i (f
1
()) = 0. In particular, if
f is injective and ({x}) = 0, for x X, then M
f
has no eigenvalues. For
instance, if X R
n
, is the Lebesgue measure, and f(x) := x, then
M
f
(g) = g (x )g(x) = 0, a.e g(x) = 0, a.e.
and therefore g = 0 in L
2
(X). Hence M
f
has no eigenvalues (note that M
f
is self-adjoint).
Examples of a bounded operator on a Hilbert space with eigenvalues and
other parts of the spectrum:
(3) Let H be a (separable) Hilbert space with an orthonormal basis {e
n
}
nN
and (
n
)
nN
be an arbitrary bounded sequence of complex numbers. Setting
Ae
n
:=
n
e
n
one denes a bounded operator A B(H). By denition,
n
is an eigen-
value, for all n N, and one can easily see that there are no other eigen-
values. On the other hand, since (A) is compact, in particular it is closed
and (
n
)
nN
(A). If / (
n
)
nN
then ((
n
)
1
)
nN
is a bounded
sequence and denes an inverse for AI. Hence
(
n
)
nN
= (A).
For instance, take H = l
2
, A : l
2
l
2
such that
A(x) = (x
n
/n)
nN
.
The eigenvalues of A are given by = 1/n, n N and
(A) = {1/n | n N} {0}.
(one can easily see that A is not invertible in B(l
2
)).
16
Also, taking (
n
) as an enumeration of the rationals on some compact
X R
n
, we have (A) = X and we show in particular that any compact
set is the spectrum of a bounded operator.
(4) Dene the left-shift operator s
l
: l
2
l
2
such that
s
l
(x
1
, x
2
, ..., x
n
, ...) = (x
2
, x
3
, ..., x
n+1
, ...).
From s
l
(x) = x we get that x
n
=
n1
x
1
, n N. Therefore is an eigen-
value of s
l
i || < 1. On the other hand, we know that (s
l
) is compact,
in particular it is closed and hence { | || = 1} (s
l
).
The spectrum of an operator A B(H) can be decomposed into three
disjoint subsets:
the discrete (or point) spectrum:

d
(A) := { C | ker(AI) = 0};
the continuous spectrum:

c
(A) := { C | ker(AI) = 0, R(AI) = X, (AI)
1
unbounded};
the residual (or essential ) spectrum:

r
(A) := (A)\(
d
(A)
c
(A)) = { C | ker(AI) = 0, R(AI) = X}.
Not so much can be said in general about
d
(A),
c
(A),
r
(A). If H is
nite dimensional, then (A) =
p
(A), for all A B(H) (see Exercise 3). In
the general setting, if A B(H) is compact, then (A)
p
(A) {0} (but

c
(A) and
r
(A) might be non-empty). If A is normal (that is, AA

= A

A),
then ker A = ker A

, and, since ker A

= R(A)

, we see that A is injective

i it has dense image; this yields that
r
(A) = . We give a few simple
examples:
(i) For the right-shift operator s
r
: l
2
l
2
dened in (1) we have:

d
(s
r
) = ,
c
(s
r
) = { | || = 1},
r
(s
r
) = { | || < 1}.
On the other hand, for the left-shift operator s
l
: l
2
l
2
dened in (4), have
that

d
(s
l
) = { | || < 1, }
c
(s
l
) = { | || = 1},
r
(s
l
) = .
(Note that s
l
= s

r
.)
(ii) A : l
2
l
2
such that
A(x) = (x
n
/n)
nN
.
17
We have seen in (3) that
d
(A) = {1/n | n N}. Now A has dense image,
but if B is an inverse for A then Bx = (nx
n
)
nN
and B is not bounded.
Conclude that 0
c
(A) and therefore, also from (3), that

d
= {1/n | n N},
c
(A) = {0},
r
(A) = .
(Note that A is compact and self-adjoint.)
(iii) If A B(H) is induced by a sequence (
n
) as in (3), we always have
that A is normal (A

Ae
n
= AA

e
n
= ||
2
), hence
r
(A) = and we have
then

d
= {
n
| n N},
c
(A) = {
n
| n N}\{
n
| n N}.
(iv) A : l
2
l
2
such that
A(x) = (0, x
1
, x
2
/2, ..., x
n
/n, ...).
Then
d
(A) =
c
(A) = ,
r
(A) = {0}.
(v) A : l
2
l
2
such that
A(x) = (x
2
, x
3
/2, ..., x
n+1
/n, ...).
Then
d
(A) ={0},
c
(A) =
r
(A) = .
18
Lecture 6 (6-11-03)
1. It is clear that C(X, Y ) is an (algebraic) ideal of C(X), for any Y X,
and C(X, Y ) is closed if Y is closed.
Conversely, let I be a closed ideal of C(X) and Y be its zero-set, that is,
Y := {x X : f(x) = 0, for all f I}.
We have clearly that I C(X, Y ). To prove the converse, let g C(X, Y )
and > 0 be arbitrary. For any x / Y , there is f
x
I and a neighbourhood
U
x
of x such that such that f
x
(y) = 0, for y U
x
. Let
h
x
(y) := f
x
(y)
g(x)
f
x
(x)
, y X.
We have then that h
x
I, and, for y U
x
, / Y ,
|g(y) h
x
(y)| = |f
x
(y)|

g(y)
f
x
(y)

g(x)
f
x
(x)

f
x

g(y)
f(y)

g(x)
f(x)

<
on a suciently small neighbourhood of x, which we denote again by U
x
,
since g/f is continuous. For x Y , take a small neighbourhood U
x
Y ;
then g(y) f(y) = 0, for all f I, y U
x
. Since X is compact, can take
U
x
i
, i = 1, ..., p such that X
i=1,...,p
U
x
i
. Let u
i
C(X) be such that
0 u
i
1, supp(u
i
) U
i
and

i=1,..,p
u
i
= 1, and dene
f :=
p

i=1
u
i
h
x
i
I,
where, for x
i
Y , we put h
x
i
= h, for some xed h I. We have that
f I, since I is an ideal, and moreover
|g(x) f(x)| =

i=1
u
i
(x)(g(x) h
i
(x))

i=1
u
i
(x) |g(x) h
i
(x)|
g f = sup
xX
|g(x) f(x)|
p

i=1
sup
xUx
i
|g(x) h
i
(x)| p .
We conclude then that g I = I, since I is closed, and therefore that
C(X, Y ) I.
19
We sketch two alternative proofs of this last result.
1 Dene I
XY
:= {f
|XY
| f I}, and show that I
XY
C
0
(X Y )
satises the conditions of the locally compact version of the theorem of
Stone-Weierstrass to get that I
XY
is dense in C
0
(X Y ). Now every
g C(X, Y ) denes by restriction a function g
|XY
C
0
(X Y ), and
hence for any > 0 there is f I such that sup
xXY
|f(x) g(x)| < .
Since g(x) = f(x) = 0, for x Y , we have in fact f g < and g I = I,
that is, C(X, Y ) I.
2 This proof also uses the fact that I is closed, but now in that in this
case we have (I

= I. Dene
I

= {F (C(X))

| F(f) = 0, for all f I}

(I

= {f C(X) | F(f) = 0, for all F I

}.
Identifying (C(X))

with the space of all nite regular Borel signed measures

on X, denoted here by M(X), one has then that
I

= { M(X) |
_
X
df = 0, for all f I}.
One then checks that if I

then supp() X Y (one sees that for

all f I, |f|
2
= 0 on supp()). Now for g C(X, Y ), g = 0 on Y and
therefore
_
X
dg = 0 for all I

, that is, g (I

= I, and therefore
C(X, Y ) I.
Note The class of maximal ideals will then be given by
{I
x
:= C(X, {x}) : x X}
and is in one-to-one correspondence with X (in fact, I
x
= ker(
x
), where

x
(C(X)) is the evaluation functional,
x
(f) = f(x)).
2. First note that C
0
(X)

= {f C(

X) : f() = 0}. Indeed, the map
C
0
(X) {f C(

X) : f() = 0}, f

f such that

f(x) = f(x), x X and

f() = 0
is easily seen to be well-dened and bijective, and moreover f =

f.
20
Now let I be an ideal of C
0
(X); then, under the isomorphism above, I
is an ideal of C(

X): just note that for any

f C(

X),

f

f() C
0
(X)
and therefore, for g I,
g

f = g(

f

f()) +

f()g I.
On the other hand, if J is an ideal in C(

X), we have also that J {f
C(

X) : f() = 0} is an ideal in C
0
(X). We conclude that the ideals of
C
0
(X) are in one-to-one correspondence with J {f C(

X) : f() = 0},
where J is an ideal in C(

X).
1
From Ex.1, we then have that the ideals of
C
0
(X) are given by
{f C
0
(X) |

f C(

X, Y )}
with Y

X closed such that Y . Since Y
0
X is closed i Y
0
{} is
closed in

X, we have nally that every ideal of C
0
(X) is of the form
C
0
(X, Y
0
) := {f C
0
(X) | f(x) = 0, x Y
0
}
with Y
0
X closed.
3.(a) Have that J

=
1
(0) is closed, since is continuous, and for any
a A, j J

,
(aj) = (a)(j) = 0
which shows that J

is an ideal of A. To check that it is maximal, let a A

be arbitrary, and write
a = x +(a)1, with x := a (a)1 ker = J

.
There is then a vector space isomorphism A

= J

C, that is, J

has codi-
mension 1 and is therefore maximal.
(b) Assume J
w
1
= J
w
2
. Then for any a A, a w
2
(a)1 J
w
1
, since
w
2
(a w
2
(a)1) = 0. But
w
1
(a w
2
(a)1) = w
1
(a) w
2
(a)w
2
(1) = w
1
(a) w
2
(a) = 0
and hence w
1
(a) = w
2
(a) for all a A.
4. Recall that

A is dened as the C

-algebra AC with
(a +)(b +) := (ab +a +b +)
1
This is just a particular instance of a general fact: if A is a non-unital C

-algebra and

A is its unitization, then the ideals of A are given by J A, with J an ideal of

A; see Ex.4.
21
a + := (a) +I
L(A)
where : A L(A) is dened as (a)b = ab and I is the identity in L(A).
In Ex.2, we have considered the isometric homomorphism C
0
(X) C(

X),
f

f such that

f(x) = f(x), for x X and

f() = 0. Dene now
:

A C(

X) by
(f +) :=

f +;
it is immediate to check that is an algebra homomorphism. If (f +) =
(g + ) then (f + )() = = (g + )() = and also

f = g,
which gives f = g; we conclude that is one-to-one. On the other hand,
let h C(

X). Then f := h h() C
0
(X) and, with := h(),
(f + ) =

f + = h. Hence is a bijection. Since any (algebraic)
isomorphism between C

-algebras is isometric, we have nally that is a

homeomorphism. (In this case, one can check directly that
(f +) =

f +

= max {sup
xX
|f(x) +|, |}
= sup
g=1
sup
xX
|(f(x) +)g(x)|
= f +1

A
.)
5.(a) Clearly, is linear, since is, and since (1) = 1, is not identically
zero. We need to check that it is multiplicative:
((a +1)(b +1)) = (ab +a +b +1)
= (a)(b) +(a) +(b) +
= (a +1) (b +1).
(b) We have that

is clearly linear and, since

(1) = 1,

is not
identically zero. Now, computations as above give that

((a +1)(b +1)) = =

(a +1)

(b +1).
(c) Let (

A). Since is linear and (1) = 1, we have
(a +1) = (a + 0) +(1) =
A
(a) +.
It is immediate that the restriction
A
is multiplicative in A and therefore ei-
ther
A
(A), and is given by (a), or
A
= 0, and =

dened in (b).
22
(d) First note that the map : (A) (

A), , with as in (a),
is a continuous injection. Indeed, if {
i
}
iI
is a convergent net in (A),

i
w, then by denition
i
(a) (a), for all a A and therefore

i
(a +1) =
i
(a) + (a) + = (a +1).
Now we check that can be extended to a homeomorphism

:

(A) (

A),

() :=

,
where

is as dened in (b). It is clear that

is a one-to-one map and
surjectivity follows from (c). If we prove that

is continuous in

(A),
then since

(A) is compact and (

A) is Hausdor, we have that
1
is
continuous as well, and the proof is concluded. Let then U

(

A) be a
neighbourhood of

, which we write as
U

= { (

A) | |(a +1)

(a +1)| < }.
We have then
U

= { (

A) | | (a)| < } {

}
and therefore
1
(U

) = {} { (A) | |(a)| < }. Now (A)

1
(U

) = { (A) | |(a)| } is closed and hence, by the Banach-

Alaoglu theorem, it is compact. We conclude that
1
(U

) is indeed a
neighbourhood of in

(A) and is continuous.
23
Lecture 7 (13-11-03)
1. Let C be a category according to Denition 6.1 and let (f, g), (g, h) C
2
.
Then s(f) = t(g) and s(g) = t(h), and, from 2 in Denition 6.1, we have
s(fg) = s(g) = t(h),
t(gh) = t(g) = s(f). (11)
Therefore, (fg, h) C
2
, (f, gh) C
2
and, from 3 in Denition 6.1, (fg)h =
f(gh). Conversely, assume that whenever (f, g) C
2
and (g, h) C
2
we
have that (fg)h, f(gh) are dened and equal. Let (f, g) C
2
; Only need
to show that s(fg) = s(g) and t(fg) = t(f). Take h such that (g, h) C
2
(for instance, h = i(s(g))); then we have s(g) = t(h) and by assumption
(fg, h) C
2
. Hence
s(fg) = t(h) = s(g).
A similar argument shows that t(fg) = t(f), and the two denitions are
indeed equivalent.
2. From 1 in Denition 6.1, we have that s(i(x)) = t(i(x)) = x, for all
objects x; hence s i and t i are bijections. Therefore, s, t are surjective
and i is injective.
3. Let C be a category as in Denition 6.1; we will show that the arrows
only denition holds. Let C
1
be the class of arrows, and C
2
C
1
C
1
be the subclass of composable arrows, m : C
2
C
1
given by composition,
m(f, g) = fg.
We rst prove (a). If fg and gh are dened, then s(f) = t(g) and
s(g) = t(h). Since s(fg) = s(g) = t(h), have that (fg)h is dened. On
the other hand, if (fg)h is dened, then in particular fg is dened and
s(f) = t(g) = t(gh) and hence f(gh) is dened. We can see in this way that
f(gh) is dened i (fg)h is dened and in this case clearly fg and gh are
dened. We then have that (i) (ii) (iii). Using 3. in Denition 6.1,
we now have that (i) (iv) and therefore (a) holds ((iv) (iii) is trivial).
As for (b), let s(f) := i(s(f)) C
1
, for f C
1
; then (f, s(f)) C
2
.
If ( s(f), h) C
2
, then s(( s(f)) = s(i(s(f))) = s(f) = t(h) and there-
fore s(f)h = i(s(f))h = i(t(h))h = h; similarly, if (g, s(f)) C
2
, then
s(g) = t(i(s(f)) = s(f) and g s(f) = gi(s(f)) = gi(s(g)) = g. Conclude that
s(f) is the required identity, and we can dene in the same way an identity

t(f) with (

t(f), f) C
2
.
24
Conversely, assume that C
1
is a category according to the arrows only
denition. Dene the class of objects
C
0
:= {e C
1
| e is an identity}
and let i : C
0
C
1
be the inclusion map. By assumption, we know that
to each f C
1
one can associate an identity t(f) (respectively, s(f)) such
that (t(f), f) (respectively, (f, s(f)) is dened. We show that t(f) (and
similarly s(f)) is the unique identity with this property and that will yield
the denition of the target (and source) map. Let f C
1
and e, e

be
identities such that (e, f), (e

, f) C
2
. Then ef = f and therefore e

(ef) is
dened; by (a), we have that e

e is dened and
e is an identity e

e = e

is an identity e

e = e.
We have therefore well-dened source and target maps
s : C
1
C
0
, f s(f)
t : C
1
C
0
, f t(f).
We now show that C
2
= {(f, g) C
1
C
1
| s(f) = t(g)} (where C
2
is
given by the arrows only denition). Let f, g C
1
such that s(f) = t(g);
then by denition of t(g), (s(f), g) C
2
, and since (f, s(f)) C
2
, have from
(a) that (fs(f), g) = (f, g) C
2
. Conversely, let (f, g) C
2
; it suces to
show that s(f)g is dened, since in this case, because s(f) is an identity, we
have s(f) = t(g). But fs(f) = f, and hence (fs(f))g is dened; by (a), we
have s(f)g dened as well.
Finally we check that the axioms in Denition 6.1 are satised.
1. Want to show that for x C
0
, s(i(x)) = t(i(x)) = x, that is, that for
any identity e, s(e) = t(e) = e. It suces to show that ee is dened: in this
case, since e is an identity, have one one hand that e = s(e) and similarly,
e = t(e). Now (t(e), e) C
2
and, since t(e)e = e, we have (t(e)e, e) C
2
;
from (a), conclude that (e, e) C
2
as well, that is, ee is dened.
1
2. Let (f, g) C
2
; to check that t(fg) = t(f) it suces to show that
t(fg)f is dened, again since t(fg) is an identity. But t(fg)(fg) is dened,
1
Moreover, it is clear that if f C1 is such that f = s(f) = t(f) then f is an identity;
we have then that C0 = {f C1 | s(f) = t(f) = f}. Also, note that we have shown in
particular that s and t are surjective, that is, C0 = Im(s) = Im(t). One could then have
started with any of these denitions for C0 and proceed from there.
25
and hence, by (a), t(fg)f is dened. Similarly, we have gs(fg) dened and
thus s(fg) = s(g).
3. If (f, g) C
2
and (g, h) C
2
, then by (a) we have (fg)h = f(gh).
4. For any f C
1
, i(t(f)) f = t(f)f = f, since i is the inclusion and
t(f) is an identity for f. Similarly, f i(s(f)) = f.
4. Let C be a category as in Denition 6.1. We show that the objects
denition holds. Let C
0
be the class of objects and for each (x, y) C
0
C
0
let
(x, y) := {f C
1
| s(f) = y, t(f) = x}.
Then, if f (x, y), g (y, z), have s(f) = t(g), that is (f, g) C
2
and we
dene
m : (x, y) (y, z) (x, z), (f, g) fg.
Now we verify (a),(b)(c).
(a) If f (x, y) (x

, y

) then s(f) = y = y

, t(f) = x = x

.
(b) For each x C
0
, have t(i(x)) = s(i(x)) = x; hence i(x) (x, x) and
for f (x, y), g (y, x), i(x)f = i(t(f))f = f and gi(x) = gi(s(g)) = g.
(c) This is just a restatement of 3. in Denition 6.1.
Conversely, let C be a category according to the objects denition.
Take C
0
for the class of objects and dene
C
1
:=
_
x,yC
0
(x, y)
s, t : C
1
C
0
such that if f (x, y) then s(f) := y, t(f) := x (note that s, t are well-
dened because of (a)). Then, from (b), we have a map
i : C
0
C
1
, x i(x)
and, dening C
2
:= {(f, g) C
1
C
1
| s(f) = t(g)} we have that (f, g) C
2
i f (x, y), g (y, z), for x = t(f), y = s(f) = t(g), z = s(g), that is, i
fg is dened. We now only need to check that the axioms in Denition 6.1
are satised.
1. For x C
0
, i(x) (x, x); hence s(i(x)) = t(i(x)) = x.
2. Let (f, g) C
2
. Then, as above, f (x, y), g (y, z) and therefore
fg (x, z); this gives s(fg) = z = s(g), t(fg) = x = t(f).
3. If (f, g), (g, h) C
2
, then f (x, y), g (y, z), h (z, w), and by (c),
(fg)h = f(gh).
26
4. For f C
1
, have that f (t(f), s(f)) and from (b), i(t(f))f = f,
fi(s(f)) = f.
5. From (b), and since s
G
is surjective, we have that
0
is determined by

1
, in that, for all x G
0
, we have

0
(x) =
0
(s
G
(i
G
(x)) = s
h

1
(i
G
(x)).
Now assume that
1
is a functor according to the arrows denition.
Dene

0
: G
0
H
0
,
0
:= s
H

1
i
G
.
We check that (a),(b),(c) hold.
(a) By denition, i
H

0
= i
H
s
H

1
i
G
. For any x G
0
,

1
i
G
(x) i
H
(s
H
(
1
i
G
(x))) =
1
i
G
(x);
on the other hand, by (a),
1
i
G
(x) is an identity
1
, since i
G
(x) is, hence,
noting that
1
i
G
(x) i
H

0
(x) is dened, have

1
i
G
(x) i
H

0
(x) = i
H

0
(x).
Conclude that i
H

0
(x) =
1
i
G
(x).
(b) For f G
1
, have that f = f i
G
(s
G
(f)) and, from (b),

1
(f) =
1
(f i
G
(s
G
(f))) =
1
(f)
1
(i
G
(s
G
(f)))
hence s
H

1
(f) = s
H

1
i
G
s
G
(f) =
0
s
G
(f).
Similarly, we have that t
H

1
(f) = t
H

1
i
G
t
G
(f); from (a), we
know that
1
i
G
s
G
(f) is an identity (since i
G
s
G
(f) is) and therefore its
source and target coincide, that is, we have t
H

1
(f) = s
H

1
i
G
t
G
(f) =

0
t
G
(f).
(c) This is a restatement of (b) in the arrows denition.
We nally show that the usual denition yields the arrows denition.
Noting that an identity e is of the form i
G
(x) for some x G
0
, have from (a)
that
1
(e) is also an identity in H. On the other hand, if (f, g) G
2
, then
(
1
(f),
1
(g)) H
2
by (b), and
1
(fg) =
1
(f)
1
(g), by (c). Therefore
the arrows denition holds.
1
Recall that an identity in a category G is an element of the form iG(x), for some
x G0. This is coherent with the denition of identity given in Exercise 3: it is clear
that any iG(x) satises those conditions and on the other hand if e is an identity and ef
is dened, by unicity, we have e = iG(tG(f)), and any identity is of the form iG(sG(f)),
iG(tG(f)).
27
Lecture 8 (20-11-03)
1. Let I : V ec V ec be the identity functor and : V ec V ec be
dened on objects by V V

, where V

denotes the second dual of the

vector space V , and on arrows by

where

(w) = w

, with

(f) = f , for w V

, f W

, : V W.
It is clear that if i
V
: V V denotes the identity, then (i
V
)

= i
V
,
and that is compatible with the source and target maps. It is also easy
to check that if , are composable arrows then ( )

, and
we have then that is a (covariant) functor. We now show that

= I.
For each object V , dene
V
: V V

by
V
(v) := v, where v(f) :=
f(v), for v V , f V

. We have that
V
is injective, since V

separates
the points of V , and therefore also surjective, since dim(V ) = dim(V

) <
. Hence
V
is an isomorphism for any V . We check that the collection
{
V
}
V V ec
0 denes a natural transformation between I and , that is, that
for any arrow : V W the following diagram commutes:
V

V
V

W
W

.
Indeed, for any v V , f W

, we have
(
W
)(v)(f) =

(v)(f) = f (v),
(

V
)(v)(f) = (
V
(v)

)(f) =
V
(v)(f ) = f (v).
We conclude that

= I.
2. Assume that G and H are equivalent categories, that is, that there exist
functors : G H and : H G such that

= id
H
,

= id
G
. We
want to show that is essentially surjective and fully faithful.
For each v H
0
, we have (v)

= v and hence, taking u := (v), we see

that is essentially surjective. To show that it is fully faithful, note that
to each u G
0
, we have isomorphisms
u
: u
0

0
(u) such that for any
given arrow , the following diagram commutes
u
u

0

0
(u)

_
1

(
)
u

0
(u

).
28
The arrow is then completely determined by
1
(), in that we have =

1
u

(
1

1
())
u
, and hence is injective. (The same argument for shows
that is injective as well.) Now let :
0
(u)
0
(u

) be an arrow and
dene : x y be :=
1
u

1
()
u
. Again using the commutativity of the
above diagram, we have that
=
1
u

(
1

1
())
u
=
1
u

1
()
u
,
that is, we have

1
() =
1
()
1
() = ,
since is injective, and we conclude then that is also surjective.
Conversely, let : G H be an essentially surjective, fully faithful
functor. We want to show that G, H are equivalent, that is, we want to
dene a functor : H G such that

= id
H
,

= id
G
.
Note In order to use the axiom of choice, we have to assume extra condi-
tions on the categories G and H (since the axiom of choice is not dened on
the universe of classes).
For each v H
0
let G
v
:= {u G
0
|
0
(u)

= v}. Since is essentially
surjective, G
v
= . Using the axiom of choice, we have that there is a map

0
: H
0
G
0
such that
0
(v) G
v
, that is, such that

0
(v)

= v, v H
0
;
this denes the functor on objects.
To dene on arrows, pick for each v H
0
an isomorphism
v
: v

=

0
(v). Each arrow (v, v

)
H
denes an arrow
1
v

v
(
0
(u),
0
(u

)),
where u =
0
(v), u

=
0
(v

). Since
1
: (u, u

)
G
(
0
(u),
0
(u

))
H
is
bijective
1
, there is a uniquely dened arrow (u, u

) such that

1
() =
1
v

v
.
We dene
1
() := . Note that with these denitions we have immediately
that the diagram
v
v

0

0
(v)

_
1

(
)=()
v

0
(v

)
1
Note that
1
: G
1
H
1
is not necessarily bijective, and in fact, in this case, we would
have that
0
is surjective not just essentially surjective.
29
commutes, for any objects v, v

and any arrow , that is, once we prove that

is indeed a functor, we have that

= id
H
. Now, it is immediate that
is compatible with the source and target maps; to check compatibility with
the inclusion maps, note that, from the denition of a category,
v
i
H
(v) =

v
= i
H
(
0

0
(v))
v
and the following diagram is commutative
v
v

0

0
(v)
i
H
(v)

_i
H
(
0

0
(v))=
1
(i
G
(
0
(v))
v
v

0
(v)
where we used the fact that is a functor; from the denition of
1
, we
have that
1
(i
H
(v)) = i
G
(
0
(v)). A similar argument shows that
1
() =

1
()
1
(), for all compatible arrows , , and is therefore a functor. To
nish the proof, we need only show that

= id
G
. Now for each arrow
(u, u

)
G
, we know that the diagram

0
(u)

0
(u)

0

0
(
0
(u))

1
()

_
1

1
(
1
())

0
(u

0
(u

1
(
1
(u

))
commutes. Since
1
: (u,
0

0
(u))
G
(
0
(u),
0
(
0

0
(u)))
H
is injective,
there is a unique arrow
u
(u,
0

0
(u))
G
such that
1
(
u
) =

0
(u)
, and
u
is an isomorphism (because
u
is). Since is a functor, the commutativity
of the above diagram yields

1
(
1

1
()
u
) =
1
(
u
)
and by the injectivity of
1
: (u,
0

0
(u

))
G
(
0
(u),
0
(
0

0
(u

)))
H
, we
have that the diagram
u
u

0

0
(u)

_
1

1
()
u

0
(u

)
also commutes, which proves nally that

= id
G
.
3. For the rst diagram, let a A, (B). If
A
(a) := a C((A)),
where a is the Gelfand transform of a, then,
(
B
)(a)() =
B
((a))() =

(a)() = (a)
30
(

A
(a))() =
A
(a)( ) = a( ) = (a).
For the second diagram, let x X, f C(Y ). If
X
(x) (C(X)) is the
evaluation functional, we have,
(
Y
)(x)(f) =
Y
((x))(f) = f (x)
(

X
)(x)(f) = (
X
(x)

)(f) =
X
(x)(f ) = f (x).
4. Let A, B be non-unital C

-algebras and : A B be a non-degenerate

homomorphism. If we denote by

the extension of to the unitization

A, we know from the equivalence of the categories CH and CC

1
A that the
diagram

A
C((

A))

B
C((

A))
is commutative, where

A
,

B
denote the Gelfand transform and

is the
usual map induced by

. Now, since is non-degenerate,

: (B) (A)
is proper and therefore induces a map

: C
0
((A)) C
0
((B)). It is
easy to check that under the isomorphism C((

A))

=

C
0
((A))
1
one
obtains a unital map

C
0
((A))

C
0
((B)), which coincides with the map

), the extension of

to the unitization. We then obtain the following

commutative diagram

A
C((

A))

=

C
0
((A))

_

(

B
C((

B))

=

C
0
((B))
which, since the maps on the top and bottom lines are (unital) isomorphisms,
restricts to
A

A
C
0
((A))

B
C
0
((A))
1
given in Lecture 6, Ex.3
31
with
A
,
B
the Gelfand transform on A, B (again isomorphisms).
Now let X, Y be locally compact spaces and : X Y a proper map.
Again from the equivalence of the categories CH and CC
1
A, we have a
commutative diagram

X
(C(

X))

Y
(C(

Y ))
where

is the extension of to the one-point compactication (which is
well-dened since is proper),

is the usual map induced by

and

X
,
which is known to be an isomorphism. Since is proper, it induces a non-
degenerate morphism

: C
0
(X) C
0
(Y ) and this induces a proper map

: (C
0
(X)) (C
0
(Y )). Now, under the isomorphism (C(

X))

=

(C
0
(X)), we get a map

(C
0
(X))

(C
0
(Y )) which is easily seen to
restrict to

, that is, we have the following commutative diagram

X
(C(

X))

=

(C
0
(X))

_

(

Y
(C(

Y ))

=

(C
0
(Y ))
which, noting that

X
() =

, restricts to
X

X
(C
0
(X))

Y
(C
0
(Y ))
with
X
,
Y
the evaluation maps (of course also isomorphisms).
32
Lecture 9 (27-11-03)
1. Recall that a subspace H
0
H is called invariant for a representation
if (A)H
0
H
0
. In this case, (a)
|H
0
, a A, denes a representation of A
on H
0
, a subrepresentation of , denoted by
H
0
(A).
For each H, H

:= (A) H is an invariant subspace and the

representation is cyclic for the subrepresentation
H

(A), since
H

(A) =
(A)
|H

yields
H

(A) = (A). Picking an arbitrary H we can then

write
H = H

, (A) =
H

.
We want repeat the argument for H

so that eventually we are able to write

(A) as a sum of cyclic representations, that is, we want to show that there
is a maximal family {
i
} such that (A)
i
(A)
j
. For this, we use
Zorns lemma. Let
F := {F H | (A)
i
(A)
j
,
i
,
j
F, i = j}.
This is a set partially ordered by inclusion, and any chain {F
i
} F (that is,
any totally ordered subset) has an upper bound, namely, F
i
. Therefore F
has a maximal element F = {
i
}. Let H
i
:= (A)
i
. If H\ H
i
, then
(, (a)
i
) = 0, for all a A, i I, and hence ((b), (a)
i
) = 0, for all
a, b A, i I, which would contradict the maximality of F. Conclude that
H = H
i
and, denoting by
i
(A) the (cyclic) subrepresentation induced by
H
i
, we have
(A) =
i
(A),
and the proof is concluded.
2. Need to show that if v V is such that (v, v) = 0 then (v, w) = 0, for all
w V . But, from the Cauchy-Schwarz inequality
|(v, w)|
2
(v, v)(w, w) = 0.
Now let be a state in a C

-algebra A and N

= {a A| (a

a) = 0}.
Want to show that N

is a closed left-ideal in A. Since is linear, N

is a
linear space. Dene a semi-denite positive form on A by
(a, b) := (a

b).
We check that it is hermitian; for this it suces to show that (a

) = (a),
since (b, a) = ((a

b)

). Now if a is self-adjoint, we have from the continuous

33
functional calculus that aI a 0
1
and hence (a1 a) 0. Writing
(a) =
1
2
((a1 +a) (a a)),
we have then that (a) R, that is, (a) = (a), for a selfadjoint. For
arbitrary a A, the result follows if we write a = a
1
+ ia
2
with a
1
, a
2
selfadjoint.
Now, from above, we have then that N

= {a A| (b

a) = 0, for all b
A}. Now let b A, a N

, then
((ba)

(ba)) = (a

ba) = ((b

ba)

a) = 0.
Therefore ba N

and N

is a left-ideal. Since both

2
and the map a a

are continuous, N

is closed.
3. Since is cyclic for and

is cyclic for

, that is, since

(A)

=
H

,
(
A) = H, it suces to show that we have an unitary bounded oper-
ator
U
0
:

(A)

(
A)
such that (a)U
0
= U
0

(a), for a A, since in this case U

0
can be
extended to an unitary map U : H

H. Now, for any a A,

(a)
2
= ((a), (a))
= (, (a

a))
=

(a

a)
= (

(a

a)

)
=

(a)

2
.
Hence there is a well-dened map U
0
such that
U
0
(

(a)

) := (a)
and U
0
is an isometry and therefore it is unitary (it preserves the inner
product). Now for any a, b A,
(a)U
0
(

(b)

) = (a)(b)
= U
0
(

(ab)

)
= U
0
(

(a)

(b)

)
1
See also Lecture 11.
2
See Exercise 1, Lecture 11.
34
and therefore (a)U
0
= U
0

(a).
4. From Exercise 3, if
1
(A) and
2
(A) are cyclic representations, then both

1
(A) and
2
(A) are unitarily equivalent to the GNS-representation and
hence unitarily equivalent: if there are unitaries U
1
, U
2
such that

1
(a)U
1
= U
1

(a),
2
(a)U
2
= U
2

(a)
then
1
(a)U
1
U
1
2
= U
1
U
1
2

2
(a), and U
1
U
1
2
is unitary.
35
Lecture 10 (4-12-03)
1. Assume that

iI
f(i) is nite and let J := {i I | f(i) = 0}. We want
to show that J is countable. First note that for any > 0, the set F

:=
{i I | f(i) > } is nite; otherwise, for any n N, one can nd S
n
F

such that (S) = n and in that case

iSn
f(i) > n which contradicts the
fact that

iI
f(i) < . Now we clearly have
J =
_
nN
F
1/n
and therefore, being a countable union of nite sets, J is countable.
2. Let (x
n
) be a Cauchy sequence in H
u
, that is, for each n N, x
n
=
{x

n
}
S(A)
with x

n
H

and we are assuming that for any > 0 there

is N such that for n, m > N, x
n
x
m
< , that is, for any nite set
F S(A),

F
x

n
x

2
<
2
. (12)
In particular, for each xed S(A), (x

n
)
nN
is a Cauchy sequence in H

and therefore it converges. Let x

:= limx

n
and x := {x

}
S(A)
. Letting
m in (12), we have that there is N such that for n > N and for any
nite set F,

F
x

n
x

2
<
2
.
We conclude that x
n
x H
u
, and therefore that x H
u
, and moreover
x
n
x
2
= sup
FS(A), nite

F
x

n
x

2
<
2
which shows that H
u
is complete.
3. We rst show that a = a

; this is straightforward from

(v, av) (v, a

v) = (v, av) (av, v) = (v, av) (v, av) = 2Im(v, av) = 0

since (v, av) R. Now let (a). Since a is selfadjoint, we know that
R, so that a I is selfadjoint as well. We rst show that a selfadjoint
operator b is invertible i
inf{bx | x = 1} > 0. (13)
36
It is clear from (13) that b is injective. Since ker(b)

= ker(b

= Im(b), we
have also that Im(b) is dense. If bx
n
y, then (bx
n
) is a Cauchy sequence
in H and (13) yields that (x
n
) is also a Cauchy sequence in H, therefore
convergent, x
n
x; since b is bounded, bx
n
bx and hence Im(b) is closed.
We conclude then that b is a bijective operator and (13) gives that b
1
is
bounded, that is, b is invertible.
Now, since a I is selfadjoint, we therefore have that (a) i
inf{(a I)x | x = 1} = 0
that is, if there exists a sequence (x
n
) such that x
n
= 1 and
(a I)x
n
0.
In this case, (x
n
, (a I)x
n
) 0 as well, and since (x
n
, x
n
) = and
(x
n
, ax
n
) 0, we conclude that 0, which nishes our proof.
Notes 1- One can check directly that, for < 0,
(aI)x
2
= ((a I)x, (a I)x) = ax
2
+
2
x
2
2(x, ax)
2
x
2
(we have used the fact that (x, ax) 0), and therefore a I is invertible.
2 - We can give a proof of this result without using the fact (a) R not-
ing that the invertibility criterion given above for selfadjoint operators also
holds for normal operators (such that b

b = bb

), since in this case one has

always ker(b) = ker(b

). Note that if a is self-adjoint, then aI is normal,

for any C.
4. To characterize S(M
2
(C)), we rst identify (M
2
(C))

. To do this, we
dene a Hilbert space structure on M
2
(C) and use Riesz Representation
theorem. For a, b M
2
(C) dene
(a, b) := Tr(a

b).
(Recall that for a = (a
ij
), Tr(a) =

a
ii
.) Since Tr is linear and Tr(a

) =
Tr(a), one easily checks that ( , ) is an hermitian bilinear form. Also, if
a is self-adjoint, one has Tr(a) =

i
, where
i
are the eigenvalues of a
1
Therefore,
(a, a) = Tr(a

a) =

i
0,
1
Since, in this case, one has a = u
1
du where u is unitary and d is the matrix repre-
sentation of a on a basis of eigenvectors; hence Tr(a) = Tr(u
1
du) = Tr(u
1
ud) = Tr(d).
37
since a

a is selfadjoint and positive, and if (a, a) = 0 then

i
= 0, i = 1, 2
and hence a

a = 0 which gives a = 0. We conclude that ( , ) is a an inner

product and that M
2
(C) is indeed an Hilbert space.
1
Now by Riesz Representation theorem, every (M
2
(C))

is given by
(a) = Tr(a

a)
for some (unique) matrix a

M
2
(C). To characterize the state space
S(M
2
(C)) (M
2
(C))

, we show the following:

(i) (I) = 1 i Tr(a

) = 1;
(ii) (a

a) 0, for all a A i a

is positive in M
2
(C).
Only (ii) needs proof. Let a

M
2
(C) be positive; then, for any a A,
we have that a

a is also positive and therefore Tr(a

a) 0, that is,
w(a

a) 0. Conversely, we show that a

= a

and that (a

) R
+
. To see
that a

is selfadjoint, note that for selfadjoint a, we have Tr(a

a) = Tr(a

a)
and writing an arbitrary a A as a = b+ic with b, c selfadjoint, we see that
Tr(a

a) = Tr(a

a) for all a A; by unicity, we have a

= a

. Replacing,
if necessary, a

by d

, the diagonal matrix representing a

on a basis of
eigenvectors, one can see that there are positive matrices b
1
, b
2
such that
Tr(a

b
i
) = Tr(d

b
i
) =
i
.
where
1
,
2
are the eigenvalues of a

. Hence, since is positive, (b

i
) =

i
0 and we have that a

is positive.
We conclude that S(M
2
(C))

= {a M
2
(C) | a 0, Tr(a) = 1}. (Note
that this isomorphism is induced by (M
2
(C))

= M
2
(C) and is then obviously
a homeomorphism.) One can easily check that a M
2
(C) is such that a 0
and Tr(a) = 1 i
a =
_
1 +x y +iz
y iz 1 x
_
for some x, y, z R. Now this matrix is positive (that is, semi-denite
positive) i 1 + x 0 and det(a) 0, that is, i x
2
+ y
2
+ z
2
1 and we
conclude that S(M
2
(C))

= B
3
= {(x, y, z) R
3
| x
2
+y
2
+z
2
1}.
1
Since M2(C) is nite dimensional, all norms are equivalent and therefore it is complete
with respect to the norm induced by the dened inner product.
38
Lecture 11 (11-12-03)
1. (i) Let : A C be a positive functional (A not necessarily unital). We
rst show that it suces to prove boundedness on A
+
. Given an arbitrary
a A, one can write a = b +ic, where b, c are self-adjoint; every self-adjoint
element b can be written as b = b
+
b

, where b
+
, b

are positive. Hence

we can write a = b
+
b

+i(c
+
c

), with b
+
, b

, c
+
, c

positive elements,
and it is clear that u a, for u = b

, u = c

. We have then
|(a)| |(b
+
)| +|(b

)| +|(c
+
)| +|(c

)|.
If is bounded on A
+
, that is, if there exists C > 0 such that for all positive
u A, |(u)| Cu, we conclude
|(a)| 4Ca
and is bounded on A.
We now prove that is bounded by contradiction. Assume then that
is not bounded. Then, for any n N we can nd a
n
, with a
n
= 1, such
that
|(a
n
)| > n
3
and by the argument above, we can assume that a
n
is positive. The series

nN
an
n
2
obviously converges to some a A. The sequence s
n
:=

n
k=1
an
n
2
is such that s
n
0 and s
n+1
s
n
. By the little lemma we prove below, one
has that a = lims
n
is positive and a s
n
for all n N. Since is positive,
(a) (s
n
)
_
a
n
n
2
_
n, for all n N,
which is absurd, because then (a) = . Therefore is bounded.
We still need to prove the following:
Lemma: Let (a
n
) be a sequence on a C

-algebra A such that a

n
a,
a
n
0, and a
n+1
a
n
, for all n N. Then a 0 and, for any N N,
a a
N
.
Proof : Can assume that A is unital
1
. It is clear that a is self-adjoint, since
the involution is continuous. From
(a) = {a() | (C

(a, 1))},
1
Since an element in a non-unital algebra is positive i it is positive in the unitization;
recall that on a non-unital C

-algebra A, A(a) is dened as

A
(a + 0).
39
one easily checks that a 0 i a() 0 for all . Now we know that
the Gelfand transform is continuous, therefore a
n
a; but since a
n
0,
a
n
() 0 and hence a() 0, that is, a 0.
To show that, for any xed N N, a a
N
0, just apply the result
just proved to the sequence b
n
:= a
n+N
a
N
.
(ii) We show that a functional : A C, A unital, is positive i it is
continuous and = (1). Assume then that is positive; we know from
(i) that it is bounded. Since A is unital, one can give a dierent proof
that will yield . Let rst a be self-adjoint. Then from the continuous
functional calculus on C

(a, 1), one has that a1 a a1, that is,

a + a||1 0 (just take f(t) := t + a, and use the fact that f 0 to
conclude that f(a) 0). Therefore, since is positive,
|(a)| (1)a.
For arbitrary a A, from |w(a

b)|
2
(a

a)(b

b), we have now

|(a)|
2
(1)(a

a) (1)
2
a

a = (1)
2
a
2

and therefore (1). It is trivial that (1) and hence =

(1).
Conversely, let be a bounded functional such that = (1) and
a A be positive. Again from the continuous functional calculus, and since
(a) R
+
, we have that there is an s R
+
such that
(1 sa) = {1 st | t (a)} [0, 1].
Since 1 sa = r(1 sa), the spectral radius of 1 sa, we then have
1 sa 1. But then
|(1) s(a)| = |(1 sa)| (1)1 sa (1).
Hence (1) s(a) (1) and therefore (a) 0. We conclude that is
positive.
(iii) If A is unital, then is a state if it is positive and (1) = 1. From (ii),
is positive i = (1), and hence is a state i = 1.
On a non-unital A, is a state if is positive in

A, the unitization of
A, where (a +) := (a) +. It is clear that we have in this case
= sup
aA,a=1
|(a)| sup
a+

A,a+=1
|| (a +)| = .
40
From (ii), since is positive, = (0 + 1) = 1 and therefore 1.
To show that = 1, we consider an approximate identity in A, that is,
a net {I

}
I
of self-adjoint elements, with I

= 1, such that, for all

a A, I

a a 0 and aI

a 0. Then, since is bounded,

(I

a a) = ((I

) 1) (a) 0, for any a A, that is,

(I

) 1.
Therefore, for any > 0, there is I such that (I

) > 1. We conclude
that 1.
(iv) On a non-unital C

-algebra A, we have by denition that is a state

i is positive in

A, the unitization of A, where (a +) := (a) +. We
have then (0 + 1) = 1 and hence is a state. (It is clear that is the
unique extension of to

A.)
(v) By denition, is a state on A if its extension to

A is positive. There-
fore the restriction of a state (hence a positive functional) on

A to A is a
state.
2. Let a A be self-adjoint and X := (C

(a, 1)). We have then the

diagram
C((a))

C

(a, 1)
a

_
C(X) C(X),
where the maps involved are as follows:
a

is induced by the Gelfand transform of a regarded as a map a :

X (a). It was proved in class that a is a homeomorphism, and therefore
a

is an isomorphism.
is the Gelfand transform on C

(a, 1), which is an isomorphism by

the Gelfand-Neumark theorem (since C

(a, 1) is commutative).
is given by the continuous functional calculus, (f) := f(a), f
C((a)). We know by the denition of f(a) that f(a = f

; to show
that is an isomorphism one therefore only need to show that it is surjec-
tive. Recall that C

(a, 1) A is dened as the closure in A of the set of all

polynomials in a. It is clear that p(a) = (p) where p(t) is any polynomial
on (a). If b = limp
n
(a), for some sequence of polynomials p
n
, then (p
n
(a))
is a Cauchy sequence in A and therefore also (p
n
) is a Cauchy sequence in
C((a)), since is isometric. Hence there is f C((a)) such that p
n
f
41
and then we have also p
n
(a) f(a), that is, b = (f). Conclude that is
surjective, and therefore an isomorphism.
We now nally check that the diagram indeed commutes. Need to show
that for any f C((a)),

f(a) = a

(f),
where, as usual,

f(a), a denote the Gelfand transforms of f(a), a. For any
X,

f(a)() = (f(a))
a

(f)() = (f a)() = f((a)).

If f(a) is given by a polynomial in a, we have (f(a)) = f((a)), since is
multiplicative (and linear). The boundedness of then yields this equality
for general f.
42