a

r
X
i
v
:
1
3
0
6
.
6
6
5
3
v
1


[
m
a
t
h
.
F
A
]


2
7

J
u
n

2
0
1
3
NON-NEGATIVE SPECTRAL MEASURES AND
REPRESENTATIONS OF C

-ALGEBRAS
ALJA

Z ZALAR
Abstract. Regular normalized W-valued spectral measures on a compact
Hausdor space X are in one-to-one correspondence with unital -represen-
tations : C(X, C) W, where W stands for a von Neumann algebra. In
this paper we show that for every compact Hausdor space X and every von
Neumann algebras W
1
, W
2
there is a one-to-one correspondence between unital
-representations : C(X, W
1
) W
2
and special B(W
1
, W
2
)-valued measures
on X that we call non-negative spectral measures. Such measures are special
cases of non-negative measures that we introduced in our previous paper [3]
in connection with moment problems for operator polynomials.
1. Introduction
A -representation of a C

-algebra / is an algebra homomorphism : / W

such that (a

) = (a)

for every a A, where W is a von Neumann alge-

bra. Our main result is the following theorem on -representations of the form
: C(X, W
1
) W
2
, where X is a compact Hausdor space and W
1
, W
2
are von
Neumann algebras. It is a generalization of the usual situation : C(X, C) W
(see Theorem 2 below).
Theorem 1. Let X, W
1
, W
2
be as above and : C(X, W
1
) W
2
a linear map.
Let Bor(X) be a Borel -algebra on X. The following statements are equivalent.
(1) : C(X, W
1
) W
2
is a unital -representation.
(2) There exists a unique regular normalized non-negative spectral measure M :
Bor(X) B(W
1
, W
2
) such that
(F) =
_
X
F dM
for every F C(X, W
1
).
A set function
M: Bor(X) B(W
1
, W
2
)
is a non-negative spectral measure if for every hermitian projection P W
1
the set
function
M
P
: Bor(X) W
2
, M
P
() := M()(P)
is a spectral measure such that
M
P
(
1
)M
Q
(
2
) = M
PQ
(
1

2
)
holds for all hermitian projections P, Q W
1
and all sets
1
,
2
Bor(X).
Date: July 1, 2013.
2010 Mathematics Subject Classication. 28B05, 46G10, 46L05, 46L10, 46L51, 47A67.
Key words and phrases. -representations, C

-algebras, operator-valued measures.

1
2 ALJA

Z ZALAR
Remark 1. (1) Spectral measures and their adaptations are well-studied in
the representation theory (e.g. [2], [6], [7], [10]). We introduced non-
negative measures in [3], where we studied moment problems in the case
of operator polynomials (see Section 3 below). Non-negative spectral mea-
sures are their special cases (see Section 7).
(2) Note that since every von Neumann algebra is a dual of a Banach space,
the existence of a representing measure in Theorem 1 is already covered as
a special case of [8, Theorem 3.3.]. The interesting part of Theorem 1 is a
concrete description of the representing measure in this special case and a
one-to-one correspondence between -representations and measures.
The paper is structured in the following way. In Section 2 we introduce some
terminology and state a well-known representation theorem for abelian C

-algebras.
In Section 3 we present the complex version of the measure and integration theory
from [3] in a more systematic way. Section 4 provides a characterisation of non-
negative measures (see Theorem 4). In Section 5 we extend the integration theory to
a Banach space which in particular constists of all bounded measurable W
1
-valued
functions and obtain a slight extension of [3, Proposition 2]; see Theorem 5. In
Section 6 we show how our measures are connected with the measures from [8] (see
Proposition 4). In Section 7 we introduce non-negative spectral measures. Section
8 provides a characterisation of non-negative spectral measures (see Theorem 6),
which is then used in Section 9 to prove Theorem 1 (see Theorem 7 and Corollary
2).
2. Preliminaries
Spectral measure is a positive operator-valued measure with an additional prop-
erty that it maps into the set of hermitian projections; see [1, Denition 2]. Spectral
measure F is normalized if F(X) = Id
H
, where (X, o, 1) is a measure space, i.e. X
is a set, o a -algebra of X and 1 a Hilbert space. For a locally compact space
X and a Borel -algebra of X, F is regular if the measures F
g,h
: Bor(X) C,
F
g,h
() := F()g, h are regular for all g, h 1. It is well-known that every
normal operator can be represented as an integral with respect to a unique regular
normalized spectral measure. A -representation of a C

-algebra / is an algebra
homomorphism : / W such that (a

) = (a)

for every a A, where W is a

von Neumann algebra. Spectral measures are interesting also due to the following
result; see [4, p. 259] and note that B(1) can be replaced by W by [9, Theorem
2.7.4].
Theorem 2. Let X be a compact Hausdor space, W a von Neumann algebra and
: C(X, C) W a linear map. Let Bor(X) be a Borel -algebra on X. The
following statements are equivalent.
(1) : C(X, C) W is a unital -representation.
(2) There exists a unique regular normalized spectral measure F : Bor(X) W
such that (f) =
_
X
f dF for every f C(X, C).
Remark 2. (1) The assumptions that X is a compact Hausdor space and
a linear map of the form : C(X, C) W can be replaced by the
assumptions that X is a locally compact Hausdor space and a linear
map of the form : C
0
(X, C) W, where C
0
(X, C) denotes the space of
functions vanishing at innity. By compactifying X with one point to X

NON-NEGATIVE SPECTRAL MEASURES AND REPRESENTATIONS OF C

-ALGEBRAS 3
and using Theorem 2, -representations of the form : C
0
(X, C) W are
in one-to-one correspondence with the regular normalized spectral measures
on X

mapping into W. However, this result is also covered by [8, Theorem

4.1.].
(2) The Baire -algebra is a -algebra generated by all compact subsets of X,
which are G

sets, i.e. a countable intersection of open sets. [1, Theorem

19] is the same result as Theorem 2, where the Borel -algebra is replaced
by the Baire -algebra. In general one has to be cautious when working
with Baire or Borel -algebras. For -compact and metrizable spaces they
coincide, but for general topological spaces this is not the case. The reason
for Borel -algebra being appropriate in Theorem 2 is the following lemma
(see [4, Proposition V.4.1]) and working with nets instead of sequences.
Lemma 1. The ball in C(X) is a dense subset of the ball in C(X)

equipped with a weak

-topology.
3. Non-negative measures
For Banach spaces }, : we denote by B(}, :) the Banach space of all bounded
linear operators from } to :. In the case } = : we write B(}) for B(}, }). Let
W be a von Neumann algebra. By W
h
, W
+
we denote the subsets of W of all
hermitian operators and all positive operators respectively.
Let X be a set, o a -algebra on X and W
1
, W
2
von Neumann algebras. A set
function
m: o B(W
1
, W
2
)
is a non-negative measure if for every A (W
1
)
+
the set function
m
A
: o W
2
, m
A
() := m()(A),
is a positive operator-valued measure. A quadruple (X, o, W
1
, W
2
) is a measure
space and a pentuple (X, o, W
1
, W
2
, m) a space with a measure m.
Let (X, o, W
1
, W
2
, m) be a space with a measure m. A o-measurable complex
function f : X C is m-integrable, if it is m
A
-integrable for every A (W
1
)
+
.
The set of all m-integrable functions is denoted by J(m).
Remark 3. Given a positive operator-valued measure E : o B(/), where /
is a Hilbert space, a o-measurable function f : X C is called E-integrable, if
there exists a constant K
f
R such that
_
X
[f[ dE
k
K
f
|k|
2
for every k /.
See [3, Remark 4] for details, where the real polarization identity is replaced by the
complex one.
The set J(m) is a complex vector space and it consists of at least all bounded o-
measurable complex functions. In particular, for o = Bor(X) we have C
c
(X, C)
J(m).
The following convergence theorem will be frequently used in the sequel.
Theorem 3. Let f
n

nN
be an increasing sequence of positive E-integrable func-
tions that pointwise converges to a o-measurable function f. If there exists B
B(/) such that
_
X
f
n
dE _ B, then f is E-integrable and lim
n
_
X
f
n
dE =
_
X
f dE
in the strong operator topology.
4 ALJA

Z ZALAR
Proof. Since by the usual convergence theorem we have
_
X
f dE
k
= lim
n
_
X
f
n
dE
k

Bk, k for every k /, f is E-integrable (take K
f
= |B| in Remark 3). Then
proceed as in the proof of [1, Theorem 11(iii)].
Given A W
1
we write Re(A) :=
1
2
(A+A

) W
1
and Im(A) :=
i
2
(A

A) W
1
for its the real and imaginary part, while for A (W
1
)
h
we write A
+
and A

for
its positive and negative part (A
+
, A

W by [5, Proposition ]).

For each m-integrable function f and each operator A W
1
we dene
_
X
f dm
A
as
_
X
f dm
Re(A)+

_
X
f dm
Re(A)
+i
_
X
f dm
Im(A)+
i
_
X
f dm
Im(A)
.
Let J(m)
C
W
1
be an algebraic tensor product of J(m) and W
1
over C. We dene
the map
B : J(m) W
1
W
2
, B(f, A) =
_
X
f dm
A
.
Let J(m)
+
be the set of all functions f J(m), such that f(x) 0 for every x X.
Proposition 1. The map B is bilinear.
Proof. It suces to consider f, g J(m)
+
, A, B (W
1
)
h
. Equality B(f +
g, A) = B(f, A) +B(g, A) easily follows by denitions. Equality B(f, A+B) =
B(f, A) + B(f, B) is equivalent to the equality of C :=
_
X
f dm
(A+B)+
+
_
X
f dm
A
+
_
X
f dm
B
and D :=
_
X
f dm
(A+B)
+
_
X
f dm
A+
+
_
X
f dm
B+
.
There is an increasing sequence s
k

kN
of simple functions s
k
J(m)
+
such
that lim
k
s
k
= f. By Theorem 3, C = lim
k
_
X
s
k
dm
(A+B)+
+
_
X
s
k
dm
A
+
_
X
s
k
dm
B
= lim
k
_
X
s
k
dm
(A+B)++A+B
=
_
X
f dm
(A+B)++A+B
. Similarly
D =
_
X
f dm
(A+B)+A++B+
. By (A+B)
+
+A

+B

= (A+B)

+A
+
+B
+
it
follows C = D, which concludes the proof.
By the universal property of the tensor product the bilinear map B can be
extended to the linear map

B : J(m)
C
W
1
W
2
,

B
_
F :=
n

i=1
f
i
A
i
_
=
n

i=1
_
X
f
i
dm
Ai
=:
_
X
F dm.
We call F J(m) W
1
positive if F(x) _ 0 for every x X and write F _ 0.
In the following proposition we list some properties of the integral with respect to
m.
Proposition 2. Let (X, o, W
1
, W
2
, m) be a space with a measure m. For all F, G
J(m) W
1
, all operators A W
1
, all complex numbers C and all sets o
the following equalities hold.
(1)
_
X
(F +G) dm =
_
X
F dm+
_
X
G dm,
(2)
_
X
F dm =
_
X
F dm.
(3)
_
X
(

A) dm = m
A
().
NON-NEGATIVE SPECTRAL MEASURES AND REPRESENTATIONS OF C

-ALGEBRAS 5
If F J(m) W
1
satises F _ 0, then
(4)
_
X
F dm _ 0.
Proof. (1), (2) follow by construction of the map

B. It suces to prove (3) for A
(W
1
)
+
. Since m
A
is a positive operator-valued measure it follows
_
X
(

A) dm =
_
X

dm
A
= m
A
(). It remains to prove (4). F J(m) W
1
, F _ 0 can be
expressed as

n1
i=1
r
i
B
i

n2
j=1
s
j
C
j
, where r
i
B
i
, s
j
C
j
J(m)
+
(W
1
)
+
,
n
1
, n
2
N. For every N we dene the set X

:=
_
i
r
1
i
[0, ]
_ _
i
s
1
j
[0, ]
_
.
The sequence X

is increasing and X =
N
X

. For every i, j there are posi-

tive simple functions t
1
i
, t
2
j
such that
_
_
t
1
i

r
i
_
_

1
2n1Bi
,
_
_
_t
2
j

s
j
_
_
_

1
2n2Cj
, where

is a characteristic function of X

. For every N we dene

G

(x) :=
_

n1
i=1
t
1
i
B
i

n2
j=1
t
2
j
C
j
_
. Therefore |

F G

|
1

. Together
with

F _ 0 it follows that G

_
1

Id
Id
W

1
. Each G

is of the form

k
D
k
,
where
k
o,
k

k

= for k ,= k

,
k

k
= X and D
k
_
1

Id
W

1
. It
follows
_
X
G

dm =
_
X
(

k
D
k
) dm =

k
m
D
k
(
k
) _

k
m

IdH
(
k
)
=
1

k
m
Id
H
(
k
) =
1

m
Id
H
(
k

k
) =
1

m
Id
H
(X)
Since for every i, j, the functions r
i
, s
j
are positive and the sequence X

increases,
the sequences t
1
i
, t
2
j
can be chosen such that they increase (w.r.t. ). By Theo-
rem 3, lim

_
X
t
1
i
dm
Bi
=
_
X
r
i
dm
Bi
, lim

_
X
t
2
j
dm
Cj
=
_
X
s
j
dm
Cj
. It follows
_
X
F dm = lim

_
X
G

dm _ lim

m
Id
H
(X)
_
= 0. This proves
_
X
F dm _ 0,
which is (4).
4. Characterisation of non-negative measures
Let 1 be a Hilbert space. [1, Theorem 2] characterizes positive operator-valued
measures via families
h

hH
of nite positive measures. We would like to have
an analoguous characterisation in the case of non-negative measures.
Let (X, o, W
1
, W
2
) be a measure space (see Section 3). The following theorem
provides a characterization of non-negative measures via families E
A

A(W1)+
of
positive operator-valued measures.
Theorem 4. Let (X, o, W
1
, W
2
) be a measure space and E
A

A(W1)+
a family of
positive operator-valued measures E
A
: o W
2
.
There exists a unique non-negative measure m such that
m
A
= E
A
for all operators A (W
1
)
+
i the following conditions hold.
(5) E
A+B
() = E
A
() +E
B
(),
(6) E
A
() = E
A
(),
6 ALJA

Z ZALAR
for all operators A, B (W
1
)
+
, all real numbers R
+
, and all sets o, and
for each set o there exists a constant k

R
>0
such that
(7) |E
A
()| k

|A|
for all operators A (W
1
)
+
.
Every family E
A

A(W1)+
which satises the conditions above is called a com-
patible family of positive operator-valued measures.
Proof. The nontrivial direction is the if part. We have to show that the set function
m: o B(W
1
, W
2
),
m()(A) :=
_
E
Re(A)+
() E
Re(A)
()
_
+i
_
E
Im(A)+
() E
Im(A)
()
_
is well-dened, which means that for each o the map m

: W
1
W
2
,
m

(A) := m()(A) is linear and bounded. By the usual decompositions of C

and A W
1
to the real and imaginary part it suces to prove R-linearity and
boundedness of m

over (W
1
)
h
.
Additivity of m

. For A, B (W
1
)
h
the equality m

(A+B) = m

(A)+m
S
(B)
is equivalent to
E
(A+B)+
() E
(A+B)
() =
_
E
A+
() E
A
()
_
+
_
E
B+
() E
B
_
(),
which is further equivalent to
E
(A+B)+
() +E
A
() +E
B
() = E
(AB)+
() +E
A+
() +E
B+
().
By (5) this is equivalent to E
(A+B)++A+B
() = E
(A+B)+A++B+
(), which is
true due to (A +B)
+
+A

+B

= (A +B)

+A
+
+B
+
.
Homogeneity of m

. To prove m

(A) = m

(A) for A (W
1
)
h
, R it
suces to consider A (W
1
)
+
by A = A
+
A

and additivity of m

. For
0 this is (6). For < 0 we have
m

(A) := E
(A)+
() E
(A)
() = E
||A
() =
..
by (6)
[[ E
A
() = m

(A).
Boundedness of m

. For A (W
1
)
h
we have
_
_
m

(A)
_
_
=
_
_
E
A+
() E
A
()
_
_

_
_
E
A+
()
_
_
+
_
_
E
A
()
_
_

..
by (7)
k

(|A
+
| +|A

|)
..
A+,AA
2 |A| k

5. Extension of integration to J(M) W

1
Assume the notation from Section 3. Let J(M) W
1
be a norm completion of
J(M) W
1
. For every F J(M) W
1
we dene
(8)
_
X
F dm := lim
i
_
X
F
i
dm,
where F
i

i
is any sequence of elements from J(M) W
1
converging to F in the
supremum norm.
The denition is well-dened by the following proposition.
Proposition 3. The integral
_
X
F dm exists and is independent of the choice of
the sequence F
i

i
.
NON-NEGATIVE SPECTRAL MEASURES AND REPRESENTATIONS OF C

-ALGEBRAS 7
Proof. Since F
i

i
is a Cauchy sequence, for each > 0 there exists n

N such that
|F
m
F
n
|

< for every m, n n

. By (4),
_
X
Id
H
dm _
_
X
(F
n
F
m
) dm _

_
X
Id
H
dm. Hence
_
_
_
X
(F
n
F
m
) dm
_
_

_
_
_
X
( Id
H
) dm
_
_
= |m
Id
H
(X)| .
Therefore
_
X
F
i
dm is a Cauchy sequence and hence convergent.
Independence of
_
X
F dm of the sequence is proved similarly. Namely, for the
sequences F
i
, G
i
converging to F, the sequence F
i
G
i
converges to 0 and by the
above argument
_
X
(F
i
G
i
) dm is a convergent sequence with the limit 0.
For a locally compact Hausdor space X and a Banach space } over F R, C,
let C
c
(X, }), C
0
(X, }) be the vector spaces of }-valued functions with compact
support and }-valued functions which vanish at innity respectively.
Let (X, Bor(X), W
1
, W
2
, m) be a space with a measure m. Since C
c
(X, F) } is
dense in C
0
(X, }) endowed with the supremum norm (see [12, Proposition 44.2.]),
we have
C
0
(X, }) J(M) W
1
.
Theorem 5 is a version of the Riesz representation theorem and slightly extends
[3, Proposition 2] from the case of a positive bounded linear map L : C(X)
B(1) B(/) to the case of a positive bounded linear map L : C
0
(X, W
1
) W
2
.
We also replace the assumptions X is a locally compact and -compact metrizable
space by the weaker assumptions X is a locally compact Hausdor space.
Theorem 5. Let X be a locally compact Hausdor space, W
1
and W
2
von Neumann
algebras and L : C
0
(X, W
1
) W
2
a positive bounded linear map. Then there
exists a unique regular non-negative measure m : Bor(X) B(W
1
, W
2
) such that
L(F) =
_
X
F dm holds for all F C
0
(X, W
1
).
Proof. Step 1 - [3, Proposition 2] remains true for a locally compact Hausdor space
X: Since C
c
(X, C) is dense in C
0
(X, C), T in [1, Theorem 19] can be uniquelly
extended to the bounded map on C
0
(X, C). Now the construction of the measure
E in the proof of [1, Theorem 19] remains the same, just that we use the version of
Riesz theorem for C
0
(X, C) (see [4, C.17. Theorem.]), use Lemma 1 and work with
nets as in the proof of [4, IX.1.14. Theorem.]. Applying this to [3, Proposition 2]
yields step 1.
Step 2 - [3, Proposition 2] remains true with B(1), B(/) replaced by W
1
, W
2
:
This follows trivially.
Step 3 - Replacing C
c
(X) W
1
by C
0
(X, W
1
): By Step 2, L[
Cc(X,C)W1
has
a unique non-negative representing measure m : Bor(X) B(W
1
, W
2
). For F
C
0
(X, W
1
) there exists a sequence F
i
C
c
(X, C) W
1
such that lim
i
F
i
= F and
by the continuity of L we have L(F) = lim
i
L(F
i
). By denition also
_
X
F dm =
lim
i
_
X
F
i
dm. Hence L(F) =
_
X
F dm.
Remark 4. Theorem 5 is true with the same proof also when the underlying Hilbert
spaces 1, / are real and linear maps are of the form L : C
0
(X, (W
1
)
h
) (W
2
)
h
.
In this case the measure m is of the form m : Bor(X) B((W
1
)
h
, (W
2
)
h
).
6. Connection with measures and integration from [8]
Let (X, o, W
1
, W
2
) be a measure space (see Section 3). From now on up to the
end of this section we will assume that X is a locally compact Hausdor space X,
and o = Bor(X) a -algebra of Borel sets. Recall (see [8]) that the semivariation
of m is the map m : Bor(X) [0, ] dened by m(X) = sup|

n
i=1
m
Ai
(
i
)|,
8 ALJA

Z ZALAR
where supremum is over all nite collections of disjoint sets
1
,
2
, . . . ,
n
such
that X =
n
i=1

i
and all A
1
, A
2
, . . . , A
n
W

1
have norm at most 1. For every
T (W
2
)

and every A W
1
the map T, m
A
() : Bor(X) C is dened
by T, m
A
() := T(m
A
()). Assume that m
A
are regular measures for every
A (W
1
)
+
(i.e. for every h, k /, m
A
()h, k is regular). The next proposition
shows that in this case our measures have the properties of measures obtained in
[8, Theorem 3.3].
Proposition 4. Assume X, o and m
A
are as above. We claim that m is a nitely
additive measure with a nite semivariation (i.e. m(X) < ) such that for every
T (W
2
)

and every A W
1
the maps T, m
A
() are regular countably additive
measures with a bounded variation.
Proof. Firstly we prove m(X) < .
m(X) = sup|
n

i=1
m
Ai
(
i
)| = sup|
n

i=1
(m
Re(Ai)
(
i
) i m
Im(Ai)
(
i
))|
= sup|
n

i=1
((m
Re(Ai)+
m
Re(Ai)
)(
i
) +i (m
Im(Ai)+
m
Im(Ai)
(
i
)))|
4 sup|
n

i=1
m
Id
(
i
)| = 4|m
Id
(X)| < ,
where supremum is over all nite collections of disjoint sets
1
,
2
, . . . ,
n
, such
that X =
n
i=1

i
and all A
1
, A
2
, . . . , A
n
W

1
have norm at most 1. Note that
the rst inequality in the last line follows by |Re(A
i
)

|, |Im(A
i
)

| 1 and hence
m
Re(Ai)
(
i
), m
Im(Ai)
(
i
) _ m
Id
(
i
) for every i = 1, 2, . . . , n.
We have to prove that the measures
(9) T, m
A
() : Bor(X) C, T, m
A
() ,
where T (W
2
)

and A W
1
, are countably additive, regular and have a nite
variation. It suces to take A (W
1
)
+
.
Let us rst prove the countable additivity. Take ,
i
Bor(X) for i N,
such that =
n
i=1

i
and
i
are mutually disjoint sets. To prove T, m
A
() =

i=1
T, m
A
(
i
) it suces to show, that m
A
() =

i=1
m
A
(
i
) in the ultra-
strong topology of W
2
. Since the ultra-strong and strong topologies coincide on
bounded subsets of W
2
, this imediatelly follows by denition of m
A
.
By the countable additivity, it follows that the sum

i=1
T, m
A
(
i
) is abso-
lutely convergent and hence the variation of (9) is nite.
To prove the inner regularity, let us take an open set U. By [5, I.4. Theorem
8.(i)] every T (W
2
)

can be decomposed as T := T
+
T

, where T
+
, T

(W
2
)

are positive linear forms, i.e. T(B) 0 for every B (W

2
)
+
. Therefore we can
assume T is a positive linear form. Hence T, m
A
() is a positive measure and it
suces to prove, that for every > 0 there exists a compact set K

U, such
that T, m
A
(U K

) < . Since T is ultra-strongly continuous by [5, I.3. Theorem

1.(ii)], there is a sequence h
j

j=1
, where h
j
/ and

i=1
|h
j
|
2
< , such that
[T, m
A
(U) B[ < for every B satisfying

j=1
|(m
A
(U) B)h
j
|
2
< 1. There
exists N N, such that

i=N+1
|h
j
|
2
<
1
2mA(X)
2
. By the inner regularity of the
measures m
A
, for every j = 1, 2, . . . , N there exists a compact set K
j
U, such
NON-NEGATIVE SPECTRAL MEASURES AND REPRESENTATIONS OF C

-ALGEBRAS 9
that
m
A
(U K
j
)h
j
, h
j
<
1
2
2
N
2
|m
A
(X)|
3
|h
j
|
2
.
Therefore by the generalized Cauchy-Schwarz inequality we get |m
A
(U K
j
)h
j
|
2

1
2N
. Hence for K

:=
N
j=1
K
j
it follows

N
j=1
|m
A
(U K

)h
j
|
2

1
2
. Therefore

j=1
|m
A
(U K

)h
j
|
2
<
1
2
+|m
A
(U K

)|
2

j=N+1
|h
j
|
2

1
2
+|m
A
(X)|
2
1
2 |m
A
(X)|
2
1.
It follows that T, m
A
(U K

) < . Since > 0 was arbitrary, this proves the

inner regularity.
The outer regularity is proved analoguously.
By Proposition 4, m has a nite semivariation. Now we will compare our inte-
gration with the integration with respect to the measure of nite semivariation from
[8, Section 3]. Let B stand for the simple Borel measurable functions on X. By
(3), Theorem 3 and (8), it easily follows that the integrations coincide on B W
1
.
But B W
1
consists just of bounded W
1
-valued function, while not all elements
from J(M) W
1
are necessarily bounded. Hence our integration theory extends
the integration theory, when m is regarded as a nitely additive measure with a
nite semivariation.
Now we will comment on the connection between Theorem 5 and [8, Theo-
rem 3.3.]. Since every von Neumann algebra is a dual of a Banach space, L :
C
0
(X, W
1
) W
2
from Theorem 5 has a representing measure given by [8, Theo-
rem 3.3.]. By Proposition 4 and uniqueness of measures from Theorem 5 and [8,
Theorem 3.3.], it follows that both representing measures coincide. Hence, we de-
rived another proof for the special case of [8, Theorem 3.3.] and obtained a concrete
description of measures given by [8, Theorem 3.3.] for this special case.
7. Non-negative spectral measures
Notation remains as in Section 3. In addition we denote by (W
1
)
p
the set of
all hermitian projections in W
1
. Let (X, o, W
1
, W
2
) be a measure space. A non-
negative measure
M: o B(W
1
, W
2
)
is a non-negative spectral measure if for every P (W
1
)
p
the set function
M
P
: o W
2
, M
P
() := M()(P),
is a spectral measure and if the equality
M
P
(
1
)M
Q
(
2
) = M
PQ
(
1

2
)
holds for all hermitian projections P, Q (W
1
)
p
and all sets
1
,
2
o. A pentu-
ple (X, o, W
1
, W
2
, M) is a space with a measure M.
We will show in the next section (see Corollary 1) that this denition is the same
as the one from the introduction.
10 ALJA

Z ZALAR
By the well-known result, e.g. [11, Theorem 5.1], every hermitian operator A
(W
1
)
h
B(1), where 1 is a Hilbert space, has a unique spectral measure E :
Bor([a, b]) B(1) such that A =
_
[a,b]
dE(), where (A) [a, b]. From A
(W
1
)
h
it follows by [5, I.Proposition 2] that E() W
1
for every Bor([a, b]).
By [11, p. 63-64], there exists a sequence S

(A) of Riemann sums of the form

(10) S

(A) =
n

k=1

k,
(E(
k,
) E(
k1,
)),
where the family E() [ R is the resolution of identity (see [11, Denition
4.1]) corresponding to the spectral measure E, S
+1
(A) is a renement of S

(A)
and |AS

(A)|
1

.
Recall, that the set J(M) of all M-integrable functions is a complex vector space
and it consists of at least all bounded o-measurable complex functions.
Proposition 5.
(11)
_
X
FG dM =
__
X
F dM
___
X
G dM
_
holds for all F, G J(M) W
1
.
Proof. Step 1: We will rst consider the case F, G J(M) W
1
. By the linearity
it suces to consider F = f A, G = g B for f, g J(M)
+
, A, B (W
1
)
+
and
prove
(12)
_
X
(fg AB) dM = (
_
X
(f A) dM)(
_
X
(g B) dM)
Let us rst show (12) holds for every A, B (W
1
)
p
. There are increasing sequences
s
k

k
, t
k

k
of simple functions such that lim
k
s
k
= f, lim
k
t
k
= g. By Theo-
rem (3), lim
k
_
X
(s
k
t
k
PQ) dM =
_
X
(fg PQ) dM, lim
k
_
X
(s
k
P) dM =
_
X
(f P) dM and lim
k
_
X
(t
k
Q) dM =
_
X
(g Q) dM, where all the limits
are in the strong operator topology. (For the rst equality we have also used the
decomposition of PQ into four positive parts and applied convergence theorem to
each of them.) By denition of M and linearity of integration equality (12) is true
for all simple functions s
k
, t
k
above; hence it is true for f, g J(M)
+
.
Let now A, B (W
1
)
+
be arbitrary. By (4) and
1

Id _ A S

(A) _ 0,
1

Id _
B S

(B) _ 0, where S

(A), S

(B) are as in (10), we get

|
_
X
(f (A S

(A))) dM|
1

|
_
X
(f Id) dM|,
|
_
X
(g (B S

(B))) dM|
1

|
_
X
(g Id) dM|.
Therefore
_
X
(f A) dM = lim

_
X
(f S

(A)) dM and
_
X
(gB) dM = lim

_
X
(g
S

(B)) dM. Since (12) holds for projections, then it follows by linearity that
_
X
(fg S

(A)S

(B)) dM = (
_
X
(f S

(A)) dM)(
_
X
(g S

(B)) dM).
It remains to prove that
(13)
_
X
(fg AB) dM = lim

_
X
(fg S

(A)S

(B)) dM.
NON-NEGATIVE SPECTRAL MEASURES AND REPRESENTATIONS OF C

-ALGEBRAS 11
We denote C

:= AB S

(A)S

(B) and

= |C

|. By the usual decomposition of

C

into the linear combination of positive elements we conclude

|
_
X
(fg C

) dM| 4

|
_
X
(fg Id) dM|.
Here we used |Re(C

| , |Im(C

| |C

| and (4). Since lim

= 0, (13)
follows.
Step 2: Now we consider the general case F, G J(M) W
1
. By denition
(see (8)),
_
X
F dM = lim
i
_
X
F
i
dM,
_
X
G dM = lim
i
_
X
G
i
dM where F
i
, G
i

C(X, C) W
1
are any sequences converging to F and G in the supremum norm.
Therefore F
i
G
i
converges to FG in the supremum norm and we have
_
X
FG dM =
lim
i
_
X
F
i
G
i
dM. By Step 1, equality (11) holds for every pair F
i
, G
i
; hence also
for the pair F, G.
Remark 5. In the sequel we will need the result from Proposition 5 just for
F, G C
0
(X, W
1
) on a locally compact Hausdor space X and a Borel -algebra
Bor(X). Since elements from C
0
(X, W
1
) are bounded, we need to prove the validity
of Proposition 5 just for bounded functions from J(M) W
1
. But this is simple.
Let us write it down.
Proof. Take bounded functions F, G from J(M) W
1
. There are sequences S
k

k
,
T
k

k
of simple functions from J(M) W
1
such that lim
k
|F S
k
|

= 0 and
lim
k
|GT
k
|

= 0. Therefore also lim

k
|FGS
k
T
k
|

= 0. It follows
_
X
F dM =
lim
k
_
X
S
k
dM,
_
X
G dM = lim
k
_
X
T
k
dM and
_
X
FG dM = lim
k
_
X
S
k
T
k
dM.
By denition of M, (
_
X
S
k
dM)(
_
X
T
k
dM) =
_
X
S
k
T
k
dM. Therefore (11)
holds.
8. Characterisation of non-negative spectral measures
Theorem 6 is a characterisation of non-negative spectral measures via families
F
P

P(W1)p
of spectral measures. This characterisation will be used to prove our
main result (see Theorem 7 and Corollary 2) in the next section.
Theorem 6. Let (X, o, W
1
, W
2
) be a measure space, F
P

P(W1)p
a family of
spectral measures F
P
: o W
2
.
There is a unique non-negative spectral measure M such that
M
P
= F
P
for all hermitian projections P (W
1
)
p
i the following conditions hold.
(14)
n

i=1

i
F
Pi
() =
m

j=1

j
F
Qj
(),
for all operators P
i
, Q
j
(W
1
)
p
, all real numbers
i
,
j
R, and all sets o
such that

n
i=1

i
P
i
=

m
j=1

j
Q
j
, for each set o there exists a constant
k

R
>0
such that
(15) |F
P
()| k

for all operators P (W

1
)
p
, and
M
P
(
1
)M
Q
(
2
) = M
PQ
(
1

2
)
holds for all hermitian projections P, Q (W
1
)
p
and all sets
1
,
2
o.
12 ALJA

Z ZALAR
Every family F
P

P(W1)p
which satises the conditions above is called a com-
patible family of spectral measures.
Proof. The nontrivial direction is the if part. For every A (W
1
)
+
we dene the
set functions M
A
: o W
2
in two steps:
(i) If A has a nite spectral decomposition

n
k=1

k
P
k
, where P
k
are mutually
orthogonal hermitian projections, then M
A
() :=

n
k=1

k
F
P
k
().
(ii) If A does not have a nite spectral decomposition, then for S

(A) as in (10)
M
A
() := lim

M
S

(A)
in the norm topology.
We have to show that M
A

A(W1)+
is a well-dened family of positive operator-
valued measures, which satises the conditions (5), (6), (7) of Theorem 4.
Existence and uniqueness of M
A
(): For A as in (i) this is clear. It remains
to prove for A as in (ii). We notice that S
1
(A) S
2
(A) =:

k
Q
k
has a nite
spectral decomposition and |S
1
(A) S
2
(A)|
2

for
1
,
2
. Hence
|M
S

1
(A)
() M
S

2
(A)
()| =
..
by (14)
|

k
M
Q
k
()|

..
MQ
k
()0,
|
k
|
2

k
M
Q
k
()| =
..
by (14)
2

|M

k
Q
k
()|
..
by (15)
for

k
Q
k
2

.
Therefore M
S

(A)
() is a Cauchy sequence in W
2
and hence convergent. Uniqueness
is proved in an analoguous manner. It remains to prove that M
A
is a positive
operator-valued measure. For A (W
1
)
+
all constants
k,
in (10) can be chosen
such that
k,l
0 and by M
A
() =
..
(14)
lim

k=1

k,
(M
E(
k,
)E(
k1,
)
()), it
follows M
A
() (W
2
)
+
. Let =

j=1

j
, where ,
j
o and
j
are mutually
disjoint. M
A
_

j=1

j
_
=
..
SOT

j=1
M
A
(
j
) follows by
M
S

(A)
_

j=1

j
_
=
n

k=1

k,
_
M
E(
k,
)E(
k1,
)
_

j=1

j
__
=
..
SOT
k

i=1

j=1

k,
_
M
E(
k,
)E(
k1,
)
(
j
)
_
=

j=1
n

k=1

k,
_
M
E(
k,
)E(
k1,
)
(
j
)
_
=

j=1
M
S

(A)
(
j
),
where the rst and forth equality follow by (14), the second by M
E(
k,
)E(
k1,
)
being spectral measures and the third holds since all the operators in the sum are
positive.
Condition (5). Take A, B (W
1
)
+
. For A, B as in (i) it follows by (14). If not
both A, B are as in (i), then we have to prove
(16) lim

_
M
S

(A+B)
() M
S

(A)
() M
S

(B)
()
_
= 0
in the norm topology, where S

(A) (resp. S

(B), S

(A + B)) is taken constant if

A (resp. B, A + B) is as in (i). Dene T

:= S

(A + B) S

(A) S

(B) and
notice |T

|
3

. Further on, for every > 0 there exists N N such that for
NON-NEGATIVE SPECTRAL MEASURES AND REPRESENTATIONS OF C

-ALGEBRAS 13
j N, |T

S
j
(T

)| and
_
_
M
T

() M
Sj(T

)
()
_
_
. As for the existence
and uniqueness we estimate
_
_
M
Sj(T

)
()
_
_

_
3

+
_
k

and hence |M
T

()|
+
_
3

+
_
k

. Since > 0 was arbitrary, we conclude |M

T

()|
3k

. Hence
lim

= 0 which proves (16).

Condition (6). Analoguous to the proof of the additivity.
Condition (7). Take A (W
1
)
+
. Each S

(A) =:

i
P
i
has a nite spectral
decomposition. By |A S

(A)|
1

and P
i
P
j
for i ,= j we have |A|
1

i=1

i
P
i
| |A| +
1

and hence max

i
[
i
[
_
|A|
1

, |A| +
1

_
. It follows
|M
A
()| = | lim

M
S

(A)
()| = lim

|
k

i=1

i
M
P
i
()|

..
MP
i
()0
lim

(max
i=1
[
i
[|
k

i=1
M
P
i
()|) lim

((|A| +
1

)|
k

i=1
M
P
i
()|)

..
by (14)
lim

((|A| +
1

)|M

k
i=1
P
i
()|)
..
by (15)
lim

(|A| +
1

) k

= |A|k

As a corollary we obtain the following equivalent deniton of a non-negative

spectral measure.
Corollary 1. A set function M: o B(W
1
, W
2
) is a non-negative spectral mea-
sure if for every hermitian projection P W
1
the set functions M
P
are spectral
measures and M
P
(
1
)M
Q
(
2
) = M
PQ
(
1

2
) holds for all hermitian projec-
tions P, Q W
1
and all sets
1
,
2
o.
9. Integral representations of representations : C(X, W
1
) W
2
Let (X, Bor(X), W
1
, W
2
, M) be a space with a measure M, where X is a compact
Hausdor space and Bor(X) a Borel -algebra. We call M regular if the measures
M
P
are regular for every P (W
1
)
p
. M is normalized if M(Id
W1
) = Id
W2
holds
for all hermitian projections P, Q (W
1
)
p
and all sets
1
,
2
Bor(X).
The main result of this article is the following.
Theorem 7. Let X, W
1
, W
2
be as above and
: C(X, C) W
1
W
2
a bounded linear map. The following statements are equivalent.
(1) : C(X, C) W
1
W
2
is a unital algebra homomorphism such that
(F

) = (F)

for every F C(X, C) W

1
.
(2) There exists a unique regular normalized non-negative spectral measure M :
Bor(X) B(W
1
, W
2
) such that
(F) =
_
X
F dM
for every F C(X, C) W
1
.
14 ALJA

Z ZALAR
Proof. Direction (2) (1). Multiplicativity follows by Proposition 5. To show
_
X
F

dM =
__
X
F dM
_

it suces by linearity to consider F = f A, f J(M)

+
,
A (W
1
)
+
. Since M
A
is a positive operator-valued measure, the result follows.
Direction (1) (2). Since is an algebra homomorphism such that (F

) =
(F)

for every F C(X, C)W

1
, the maps
P
: C(X, C) W
2
,
P
(f) := (fP)
are -representations for every P (W
1
)
p
. By Theorem 2, there exist unique
spectral measures F
P
: Bor(X) W
2
such that
P
(f) =
_
X
f dF
P
for every
f C(X, C). The idea is to show that F
P

P(W1)p
is a compatible family of
spectral measures and apply Theorem 6 to obtain a non-negative spectral measure
M representing .
The condition (14) follows from the linearity of and equalities (P) =
_
X
f dF
P
.
Namely, pick a Borel set Bor(X). The function

is a bounded Borel function.

By Lemma 1, there is a net f
k
B such that
_
X
f
k
d () for every measure
C(X)

. Therefore for all g, h 1

(
n

i=1

i
_
X
f
k
d(F
Pi
)
g,h
)
n

i=1

i
_
X

d(F
Pi
)
g,h
=
n

i=1

i
(F
Pi
)
g,h
().
and analoguously (

m
j=1

j
_
X
f
k
d(F
Qj
)
g,h
)

m
j=1

j
(F
Qj
)
g,h
(). It follows
that

n
i=1

i
F
Pi
() =

m
j=1

j
F
Qj
().
The condition (15) follows by the continuity of . Namely,
|F
P
(X)| = |
_
X
1 dF
P
| = |
P
(1)| = |(1 P)| || |1 P|

= || .
By the nite additivity of F
P
, it follows |F
P
()| || for every o.
Now we will show that M represents . By the linearity and continuity of
and
_
it suces to consider F = f P C(X) (W
1
)
p
. By the construction of
measures F
P
, we have (f P) =
_
X
f dF
P
=
_
X
(f P) dM.
The uniqueness, regularity and normalization condition of M follow from the
uniqueness and regularity of each F
P
and the unitality of .
Multiplicativity of M is proved as the condition (14) above. Namely, pick

1
,
2
Bor(X). There exists a net f
k
B, such that
_
X
f
k
g d
_
X

1
g d for every C(X)

and every bounded Borel function g. Therefore

_
X
f
k
g dM
PQ

_
X

1
g dM
PQ
and
_
X
f
k
dM
P

_
X

1
dM
P
in the weak
operator topology. Hence for g C(X) (all the limits are in WOT)
_
X

1
g dM
PQ
= lim
_
X
f
k
g dM
PQ
= lim(f
k
g PQ)
= lim(f
k
P) (g Q) = lim(
_
X
f
k
dM
P
) (
_
X
g dM
Q
)
= (
_
X

1
dM
P
) (
_
X
g dM
Q
).
There exists a net g
k
B, such that
_
X

1
g
k
d
_
X

1

2
d and
_
X
g
k
d
_
X

2
d for every C(X)

. Hence for g C(X) (all the limits

are in WOT)
_
X

1

2
dM
PQ
= lim
_
X

1
g
k
dM
PQ
= lim(
_
X

1
dM
P
) (
_
X
g
k
dM
Q
)
= (
_
X

1
dM
P
) (
_
X

2
dM
Q
).
NON-NEGATIVE SPECTRAL MEASURES AND REPRESENTATIONS OF C

-ALGEBRAS 15
It follows that M
PQ
(
1

2
) = M
P
(
1
)M
Q
(
2
).
Corollary 2 generalizes Theorem 2 to the case C(X, W
1
).
Corollary 2. Let X be a compact Hausdor space, W
1
, W
2
von Neumann algebras
and
: C(X, W
1
) W
2
a linear map. The following statements are equivalent.
(1) : C(X, W
1
) W
2
is a unital -representation.
(2) There exists a unique regular normalized non-negative spectral measure M :
Bor(X) B(W
1
, W
2
) such that
(F) =
_
X
F dM
for every F C(X, W
1
).
Proof. Direction (2) (1) follows as in the proof of Theorem 7. (Note that
_
X
F

dM = (
_
X
F dM)

follows from
_
X
F

i
dM = (
_
X
F
i
dM)

, where F
i

C(X, C)W
1
is a Cauchy sequence converging to F in the norm topology.) For the
direction (1) (2) we rst notice that is bounded by [4, 4.8. Theorem., p. 247].
Then we apply Theorem 7 to [
C(X,C)W1
to obtain the unique representing mea-
sure M for [
C(X,C)W1
. Now we proceed as in the proof of Theorem 5 to show
that it also represents .
Remark 6. As in Remark 2.(1) there is a corresponding version of Corollary 2 for a
locally compact X and linear maps : C
0
(X, W
1
) W
2
. Namely, -representations
: C
0
(X, W
1
) W
2
are in one-to-one correspondence with the regular normalized
non-negative spectral measures F on X

mapping into W
2
.
Acknowledgement. I would like to thank to my advisor Jaka Cimpric for many
useful discussions while studying the material above and suggesting improvements
of the manuscript. I am also grateful to Bojan Magajna for the discussions con-
cerning von Neumann algebras, vector measures and Remark 2.(1).
References
[1] S. K. Berberian, Notes on Spectral Theory, D. van Nostrand Company, Princeton, 1966.
[2] M.A. Bastos, C.A. Fernandes, Y.I. Karlovich, Spectral measures in C

-algebras of singular
integral operators with shifts, J. Funct. Anal. 242 (2007) 86126.
[3] J. Cimpric, A. Zalar, Moment problems for operator polynomials, J. Math. Anal. Appl. 401
(2013) 307316.
[4] J. Conway, A Course in Functional Analysis, Springer-Verlag, New York, 1990.
[5] J. Dixmier, Von Neumann Algebras, North-Holland Publishing Co., Amsterdam-New York-
Oxford, 1981.
[6] P. Masani, The normality of time invariant, subordinate operators in a Hilbert space,
Bull. Amer. Math. Soc. 71 (1965) 546550.
[7] P. Masani, M. Rosenberg, When is an Operator the Integral of a Given Spectral Measure,
J. Funct. Anal. 21 (1976) 88121.
[8] S. K. Mitter, S. K. Young, Integration with Respect to Operator-Valued Measures with Ap-
plications to Quantum Estimation Theory, Ann. Mat. Pura Apl. 137 (1984) 139.
[9] J. Peterson, Notes on von Neumann algebras, http://www.math.vanderbilt.edu/
~
peters10/teaching/spring2013/vonNeumannAl
9.4.2013.
[10] M. Rosenberg, Operators as Spectral Integrals of Operator-Valued Functions from the Study
of Multivariate Stationary Stochastic Processes, J. Multivariate. Anal. 4 (1974) 166209.
16 ALJA

Z ZALAR
[11] K. Schm udgen, Unbounded Self-Adjoint Operators on Hilbert Spaces, Springer, Netherlands,
2012.
[12] F. Treves, Topological Vector Spaces, Distributions and Kernels, Academic Press, New York,
1967.
Aljaz Zalar, University of Ljubljana, Faculty of Math. and Phys., Dept. of Math.
E-mail address: aljaz.zalar@student.fmf.uni-lj.si