INTRODUCTION TO SPECTRAL THEORY

WILLIAM CASPER

Abstract. This is an abstract.

1. Introduction
Throughout this paper, we let H denote a Hilbert space, X a measureable space
and Ω a σ-algebra of subsets of X. By an operator T on H we will always mean a
linear transformation such that the operator norm
kT k := sup{kT ψk : ψ ∈ H, kψk = 1}
is finite. An operator will be called invertible if it has an algebraic inverse T −1
which is also an operator on H.
1.1. Basic Definitions and Facts.
Definition 1.1. The spectrum of a linear operator T on H is
(1) spec(T ) = {λ ∈ C : T − λ is not invertible. }.
The approximate point spectrum of T is
(2)
aspec(T ) = {λ ∈ C : ̺(T − λ) = 0}, where ̺(T ) = inf{kT ψk : ψ ∈ H, kψk = 1}.
Theorem 1.1. aspec(T ) ⊂ spec(T ).
Theorem 1.2. If T is a normal operator, then aspec(T ) = spec(T ).
Theorem 1.3 (Transforms of Spectra).
(i) If p ∈ C[x], then
spec(p(T )) = p(spec(T )) := {p(λ) : λ ∈ spec(T )}.
(ii) If T is invertible, then
spec(T −1 ) = (spec(T ))−1 := {λ−1 : λ ∈ spec(T )}.
(iii) The spectrum of the adjoint of T satisfies
spec(T ∗ ) = (spec(T ))∗ := {λ∗ : λ ∈ spec(T )}.
Theorem 1.4. Define
(3) NT (f ) = sup{|f (λ)| : λ ∈ spec(T )}.
In general, spec(T ) is a compact subset of the complex plane and NT (x) ≤ kT k.
If T is Hermitian, then spec(T ) is a subset of R and NT (p(x)) = kp(T )k for any
polynomial p ∈ R[x].

Date: February 24, 2010.

1
2 WILLIAM CASPER

2. Spectral Measures
Definition 2.1. A spectral measure E on a measureable space (X, Ω) is a
projection-valued (idempotent, hermitian) set function
E : Ω → { projections on H}
satisfying
(i) E(X) = 1;
(ii) for any collection of pairwise disjoint sets {Ak }∞
k=1 ,
Ã∞ ! ∞
[ X
E Ak = E(Ak ).
k=1 k=1

Theorem 2.1 (Properties of Spectral measures). If E is a spectral measure, then

for all A, B ∈ Ω
(i) if A ⊂ B, then E(A) ≤ E(B);
(ii) if A ⊂ B, then E(B \ A) = E(B) − E(A);
(iii) E(A ∪ B) + E(A ∩ B) = E(A) + E(B);
(iv) E(A ∩ B) = E(A)E(B).
Theorem 2.2. A function
E : Ω → { projections on H}
is a spectral measure if and only if
(i) E(X) = 1;
(ii) for any two fixed elements ψ, φ ∈ H,the function µ : Ω → C defined by
µ(A) = hE(A)ψ, φi for all A ∈ Ω
is a complex measure. We use the notation dhE(λ)ψ, φi = dµ(λ) so that in
general
Z
1A dhE(λ)ψ, φi = hE(A)ψ, φi for all A ∈ Ω.

Theorem 2.3. If E is a spectral measure and f is an R E-measureable

R function,
then there exists a unique operator denoted by either f dE or f (λ)dE(λ) and
defined as
¿Z À Z
(4) f dEψ, φ = f (λ)dhE(λ)ψ, φi.
R
Theorem 2.4 (Properties of f dE). Given any E-measureable functions f, g and
α ∈ C,
R R
(i) R (αf )dE = α R f dE; R
(ii) (f + g)dE = f dE + gdE;
¡R ¢∗ R ∗
(iii) R f dE = ¡R f dE;¢ ¡R ¢
(iv) f gdE = f dE f dE .

R If E is a spectral measure and E(A) commutes with T for every

Theorem 2.5.
A ∈ Ω, then f dE commutes with T .
 INTRODUCTION TO SPECTRAL THEORY 3

3. Complex Spectral Measures

For the remainder of the paper, we assume that X is a locally compact Hausdorff
space and that Ω is the Borel σ-algebra on X.

Definition 3.1. A spectral measure is regular if for all A ∈ Ω,

E(A) = sup{E(C) : C ⊂ A, C is compact }.

Definition 3.2. The spectrum of a spectral measure is

spec(E) := X \{λ ∈ X : λ ∈ A, A is open, E(A) = 0}.

A spectral measure is compact if its spectrum is compact.

Theorem 3.1. If E is a regular spectral measure, then spec(E) is closed and

E(X \ spec(E)) = 0 (and therefore E(spec(E)) = 1).

Theorem 3.2. For any complex-valued, E-measureable function bounded on spec(E),

define
NE (f ) = sup{|f (λ)| : λ ∈ spec(E)}.
Then if E°Ris a compact,
° regular spectral measure and f is a continuous function on
X, then ° f dE ° = NE (f ).

Definition 3.3. A spectral measure is called complex when X = C.

Theorem 3.3. Every complex spectral measure is regular.

R
Theorem 3.4. If E is a compact, complex spectral measure and if T = λdE(λ),
then spec(T ) = spec(E).

Theorem
R 3.5. A complex spectral measure is defined completely by the operator
λdE(λ). That
R is, givenR two complex spectral measures E1 and E2 , E1 = E2 if
and only if λdE1 (λ) = λdE2 (λ).

Theorem 3.6. Let E be a complex spectral measure and T an operator.R Then T

commutes with E(A) for all A ∈ Ω if and only if T commutes with both λdE(λ)
and λ∗ dE(λ).
R

4. The Spectral Theorem

Theorem 4.1 (Spectral Theorem for Hermitian Operators). Let T be a Hermitian
operator. RThen there exists a unique compact, complex spectral measure E such
that T = λdE(λ).

Theorem 4.2 (Spectral Theorem for Normal Operators). Let T be a normal op-
erator.
R Then there exists a unique compact, complex spectral measure E such that
T = λdE(λ).

DefinitionR 4.1. For any normal operator T , we call the spectral measure E satis-
fying T = λdE(λ) the spectral measure of T .
4 WILLIAM CASPER

Appendix A. Applications of the Spectral Theorem

A.1. Weak Mixing. As a first example application of the spectral theorem, we
will use it to show that a measure-preserving transformation T is weak mixing when
the only eigenfunctions of the unitary operator UT defined by UT f (x) = f (T x) are
the constants.
Definition A.1. Let (X, Ω, µ) be a probability space and T a measure preserving
transformation (mpt) on X (µ(A) = µ(T −1 (A)) for all A ∈ Ω). Then T is called
weakly mixing if
n−1
1X
(5) lim |µ(T k (A ∩ B) − µ(A)µ(B)| → 0
n→∞ n
k=0

for all A, B ∈ Ω.
Theorem A.1. A mpt T is weakly mixing if the only measureable eigenfunctions
of UT are the constants.
Proof. We first note that T is weakly mixing if and only if
n−1
1X
(6) lim |hUTk f, gi − hf, 1ih1, gi| → 0
n→∞ n
k=0

for all f, g ∈ L2 (µ).

Let V be the closed linear subspace of the eigenfunctions of T in L2 (µ) and let
E be the spectral measure of T . Then for any λ0 ∈ spec(UT ), we have that
Z Z Z
UT E({λ0 }) = λdE(λ) 1λ0 (λ)dE(λ) = λ1{λ0 } (λ)dE(λ) = λ0 E({λ0 }),

so that in particular UT E({λ0 })f = λ0 f for every f ∈ L2 (µ). Thus E({λ0 })f ∈ V
for all f ∈ L2 (µ). If f ∈ V ⊥ , this implies that
0 = hE({λ0 })f, f i = hE({λ0 })2 f, f i = hE({λ0 })f, E({λ0 })f i,
and therefore E({λ0 })f = 0.
Now fix an f ∈ V ⊥ and g ∈ L2 (µ) and define µ to be the complex Borel measure
on the spectrum of T satisfying dµ = dhE(λ)f, gi. Then for all λ0 ∈ spec(T ), we
have that
µ({λ0 }) = hE({λ})f, gi = 0.
Setting ∆ = {(λ, ω) ∈ spec(UT ) × spec(UT ) : λ = ω}, we find
n−1 n−1 Z n−1
1X 1X 1X k
Z
|hUTk f, gi|2 = |h λk dE(λ)f, gi|2 = |λ hdE(λ)f, gi|2
n n n
k=0 k=0 k=0
n−1 n−1
1X 1X
Z Z Z
= | λk dµ(λ)|2 = λk dµ(λ) (ω)k dµ∗ (ω)
n n
k=0 k=0
n−1 n−1
1 1X
X Z Z Z
∗
= k
(λω) dµ(λ)dµ (ω) = (λω)k d(µ × µ∗ )(λ, ω)
n n
k=0 k=0
1 1 − (λω)k
Z Z
= d(µ × µ∗ )(λ, ω) + 1d(µ × µ∗ )(λ, ω).
∆c n 1 − λω ∆
 INTRODUCTION TO SPECTRAL THEORY 5

Additionally,
Z Z Z Z
1d(µ × µ∗ )(λ, ω) = 1∆ (λ, ω)dµ(λ)dµ∗ (ω) = µ({ω})dµ∗ (ω) = 0.
∆

For (λ, ω) ∈/ ∆, (1 − (λω)k )/(1 − λω) is a cyclotomic polynomial, and is therefore

bounded on the compact set spec(UT ). Thus by the bounded convergence theorem
1 1 − (λω)k
Z Z
lim d(µ × µ∗ )(λ, ω) = 0d(µ × µ∗ )(λ, ω) = 0.
n→∞ ∆c n 1 − λω ∆ c

We conclude that
n−1
1X
lim |hUTk f, gi|2 = 0.
n→∞ n
k=0
By the Cauchy-Schwartz inequality,
Ãn−1 !2 n−1
X X
k
|hUT f, gi| ≤n |hUTk f, gi|2 .
k=0 k=0
2
Dividing both sides by n , we find
à n−1 !2 n−1
1X 1X
k
|hUT f, gi| ≤ |hUTk f, gi|2 ,
n n
k=0 k=0

and therefore
n−1
1X
lim |hUTk f, gi| = 0.
n→∞ n
k=0
Now for any f, g ∈ L2 (G), if the eigenfunctions of UT are the constants, f −
hf, 1i ∈ V ⊥ . Therefore
n−1 n−1
1X 1X
lim |hUTk (f − hf, 1i), gi| = lim |hUTk f − hf, 1i, gi|
n→∞ n n→∞ n
k=0 k=0
n−1
1 X
= lim |hUTk f, gi − hf, 1ih1, gi| = 0.
n→∞ n
k=0

Therefore T is weakly mixing. ¤

A.2. Almost Periodic Functions.

Appendix B. Proofs of Theorems

B.1. Proof for Section 1.
Proof of Theorem (1.1). If λ ∈
/ spec(T ), then T − λ is invertible. Thus for any
ψ ∈ H with kψk = 1, we have that
1 = kψk = k(T − λ)−1 (T − λ)ψk ≤ k(T − λ)−1 k · k(T − λ)ψk.
Thus ̺ (as defined by Eq. (1)) satisfies ̺(T − λ) ≥ 1/k(T − λ)−1 k and so λ ∈
/
aspec(T ). This proves our theorem. ¤
Lemma B.1. An operator T is invertible if and only if its range is dense in H and
there exists a positive real number c > 0 such that kT ψk ≥ ckψk for all ψ ∈ H.
6 WILLIAM CASPER

Proof. If T is invertible, then it is a bijection and therefore the range must be H.

Moreover,
kψk = kT −1 T ψk ≤ kT −1 k · kT ψk,
so kT ψk ≥ ckψk with c = 1/kT −1 k.
Conversely, suppose the range of T is dense in H and there exists a positive
real number c > 0 such that kT ψk ≥ ckψk for all ψ ∈ H. We first show that
the range of T is H. Let {φi }∞ i=1 be a convergent sequence in the range of H
converging to φ. For every i > 0, there exists a ψi ∈ H such that T ψi = φi .
Moreover, kψi − ψj k ≤ kφi − φj k/c. It follows that the sequence {ψi }∞i=1 is Cauchy
and therefore converges to a function ψ ∈ H. Since T is continuous, φ = T ψ, and
therefore φ is in the range of T . We conclude that the range of T is closed. Since
the range of T is dense in H, the range of T must be H.
The kernel of T is trivial, since if ψ ∈ ker(T ), then kψk ≤ kT ψk/c = 0, implying
that ψ = 0. Thus T is a bijection, and all that is left to show is that the algebraic
inverse, which we call T −1 , is bounded. We have that kT −1 ψk ≤ kT T −1 ψk/c ≤
kψk/c. It follows that kT −1 k ≤ 1/c. This proves our theorem. Incidentally, this
also shows us that c = 1/kT −1 k is the ”sharpest” value for c. ¤
Proof of Theorem (1.2). By Theorem (1.1), we need only prove spec(T ) ⊂ aspec(T ).
Suppose that λ ∈ / aspec(T ). Then there exists a constant c > 0 such that k(T −
λ)ψk ≥ ckψk for all ψ ∈ H. By Lemma (B.1), we need only show that the
range of T − λ is dense in H. Since T commutes with T ∗ , T − λ commutes
with (T − λ)∗ = T ∗ − λ∗ , and it follows that k(T − λ)ψk = k(T ∗ − λ∗ )ψk for
all ψ ∈ H. If φ ∈ rangle(T − λ)⊥ , then 0 = h(T − λ)ψ, φi = hψ, (T ∗ − λ∗ )φi
for all ψ ∈ H, and therefore (T ∗ − λ∗ )φ = 0. It follows that φ = 0, since
kφk ≤ k(T − λ)φk/c = k(T ∗ − λ∗ )φk/c = 0. Thus rangle(T − λ)⊥ = {0} and
it follows that rangle(T − λ) is dense in H. This proves our theorem. ¤
Proof of Theorem (1.3). (i) Let p ∈ C[x] and λ ∈ spec(T ). Then λ is a root of
r(x) = p(x) − p(λ) and therefore there exists a polynomial q ∈ C[x] such that
q(x)(x − λ) = p(x) − p(λ). If r(T ) is invertible, then q(T ) commutes with
r−1 (T ) and
(T − λ)q(T )r(T )−1 = r(T )r(T )−1 = 1 = r(T )−1 r(T ) = r(T )−1 (T − λ)q(T )
= r(T )−1 q(T )(T − λ) = q(T )r(T )−1 (T − λ).
It follows that (T − λ) is invertible with (T − λ)−1 = q(T )r(T )−1 , which is a
contradiction. Thus r(T ) is not invertible and p(λ) ∈ spec(p(T )).
Conversely, suppose λ ∈ spec(p(T )) and let {ri }ni=1 be the roots of the
polynomial p(x) − λ. We have that p(x) − λ = (x − r1 ) . . . (x − rn ) and
therefore p(T ) − λ = (T − r1 ) . . . (T − rn ). Since p(T ) − λ is not invertible,
(T − rj ) is not invertible for some j. Whence rj ∈ spec(T ) and p(rj ) − λ = 0.
We conclude that λ ∈ p(spec(T )). This proves (i).
(ii) Note that for any λ ∈ C, we have that T −1 − λ−1 = −T −1 λ−1 (T − λ), and
it follows that T −1 − λ−1 is invertible if and only if T − λ is invertible. This
proves (ii).
(iii) If λ ∈/ spec(T ), then T − λ is invertible. It follows that (T − λ)∗ = T ∗ − λ∗
is invertible, and therefore spec(T ∗ ) ⊂ spec(T )∗ . By the same argument
with T replaced by T ∗ , spec(T ) = spec((T ∗ )∗ ) ⊂ spec(T ∗ )∗ , and therefore
spec(T )∗ ⊂ (spec(T ∗ )∗ )∗ = spec(T ∗ ). This proves our theorem.
 INTRODUCTION TO SPECTRAL THEORY 7

¤
Lemma B.2. If T is an operator such that k1 − T k < 1, then T is invertible.
Proof. Define c > 0 by c = 1 − k1 − T k. Then
kT ψk = kψ − (ψ − T ψ)k ≥ kψk − k(1 − T )ψk ≥ kψk − k(1 − T )k · kψk = ckψk.
Thus by Lemma (B.1), we need only show that the range of T is dense in H. Let
φ ∈ H and let δ = inf{kT ψ − φk : ψ ∈ H}. Suppose that δ > 0. Then for all
c
ǫ = δ 1−c , there exists ψ ∈ H such that δ ≤ kT ψ − φk < δ + ǫ. Moreover
δ ≤ kT (T ψ−φ)−(T ψ−φ)k = k(1−T )(T ψ−φ)k < (1−c)kT ψ−φk = (1−c)(δ+ǫ) ≤ δ.
That is, δ < δ, which is a contradiction. We conclude that δ = 0. Since φ ∈ H was
taken arbitrarily, this means that the range of T is dense in H. This proves our
lemma. ¤
Proof of Theorem (1.4). If λ0 ∈ / spec(T ), then T − λ0 is invertible. If λ ∈ C with
|λ − λ0 | < r := 1/k(T − λ0 )−1 k, then
k1 − (T − λ0 )−1 (T − λ)k = k(T − λ0 )−1 [(T − λ0 ) − (T − λ)]k
≤ k(T − λ0 )−1 k · |λ − λ0 | < 1.
Therefore by Lemma (B.2) (T − λ0 )−1 (T − λ) is invertible and it follows that
(T − λ) must be invertible. We conclude that the ball B(λ0 ; r) about λ0 of radius r
is contained in C \ spec(T ). It follows that C \ spec(T ) is open and therefore spec(T )
is closed. Moreover, if λ ∈ C satisfies kT k < |λ|, then k1 − (1 − T /λ)k = kT /λk < 1
and therefore 1−T /λ is invertible by Lemma(B.2). It follows that T −λ is invertible,
and therefore λ ∈ / spec(T ). Thus if λ ∈ spec(T ), then |λ| ≤ kT k necessarily. In
particular, this shows that NT (x) ≤ kT k and that spec(T ) is a closed and bounded
subset of C (and therefore compact).
Suppose T is Hermitian and λ ∈ spec(T ). Then T is normal and spec(T ) =
aspec(T ) by Theorem (1.2). Thus there exists a sequence {ψi }∞ i=1 ⊂ H such that
kψi k = 1 for all i and k(T − λ)ψi k → 0. Thus
|λ − λ∗ | = |λ − λ∗ | · kψi k2 = |h(T − λ)ψi , ψi i − h(T − λ∗ )ψi , ψi i|
= |h(T − λ)ψi , ψi i − hψi , (T − λ)ψi i|
≤ 2k(T − λ)ψi k · kψi k = 2k(T − λ)ψi k → 0.
It follows that λ is real. Moreover for any λ ∈ R, since T is Hermitian, we have the
relation
kT 2 ψ − λ2 ψk2 = hT 2 ψ − λ2 ψ, T 2 ψ − λ2 ψi
= kT 2 ψk2 + |λ|4 kψk2 − (λ2 )∗ hT 2 ψ, ψi − λ2 hψ, T 2 ψi
= kT 2 ψk2 + λ4 kψk2 − 2λ2 kT ψk2 .
Now let {ψi }∞
i=1 ⊂ H be a sequence such that kψi k = 1 for all i and kT ψi k → kT k.
Then taking λ = kT k in the above relation, we find that
k(T 2 − kT k2 )ψi k2 = kT 2 ψi − λ2 ψi k2 = kT 2 ψi k2 + λ4 kψi k2 − 2λ2 kT ψi k2
= kT 2 ψi k2 + kT k4 − 2kT k2 kT ψi k2 → 0.
Thus kT k2 ∈ spec(T 2 ), and it follows from Theorem (1.3) that either kT k ∈ spec(T )
or −kT k ∈ spec(T ). In particular, this proves NT (x) = kT k. If p(x) ∈ R[x], then
8 WILLIAM CASPER

p(T ) is Hermitian and therefore NT (p(x)) = Np(T ) (x) = kp(T )k. This proves our
theorem. ¤
B.2. Proofs for Section 2.
B.3. Proofs for Section 3.
B.4. Proofs for Section 4.
Lemma B.3 (Weierstrass Approximation Theorem). Let X be a compact subset
of R and let f be a continuous function on X. Then there exists a sequence of real
polynomials {pi }∞
i=1 such that pi → f uniformly on X.

Lemma B.4. Let L be a bounded linear functional on R[x] and let X be a compact
subset of R. Then there exists a unique Borel measure µ on X satisfying
Z
L(p) = p(λ)dµ(λ) for all p ∈ R[x].

Sketch of proof. Let Ω be the collection of all Borel subsets of X and let A ∈ Ω.
Let {pi }∞
i=1 ⊂ R[x] be a sequence of polynomials with pi → 1A uniformly on X.
Define µ(A) by
µ(A) = lim L(pi ).
i→∞
Then µ is a well-defined complex Borel measure on X. ¤
Proof of Theorem (4.1). Let ψ, φ ∈ H and define a function
L : R[x] → C
by L(p) = hp(T )ψ, φi. Then L is linear and
|L(p)| ≤ kp(T )ψk · kφk ≤ kp(T )k · kψk · kφk ≤ NT (p(x)) · kψk · kφk
and therefore L is a linear functional on R[x]. The set R[x] is a dense subset of
the collection of all continuous, real-valued functions on spec(T ), and it follows
that there exists a unique complex measure µ on X = spec(T R ) with σ-algebra Ω
consisting of all Borel subsets of spec(T ) such that L(p) = p(λ)dµ(λ) for all p ∈
C[x]. For given ψ, φ ∈ H, we denote this measure by µ(ψ,φ) . Let ψ1 , ψ2 , φ1 , φ2 ∈ H
and let α ∈ C.
Z
p(λ)dµ(ψ1 +ψ2 ,φ) (λ) = hp(T )(ψ1 + ψ2 ), φi = hp(T )ψ1 , φi + hp(T )ψ2 , φi
Z Z
= p(λ)dµ(ψ1 ,φ) (λ) + p(λ)dµ(ψ2 ,φ) (λ),

from which it follows that

µ(ψ1 +ψ2 ,φ) = µ(ψ1 ,φ) + µ(ψ2 ,φ) .
Similarly,
µ(ψ,φ1 +φ2 ) = µ(ψ,φ1 ) + µ(ψ,φ2 ) ;
µ(αψ,φ) = αµ(ψ,φ) and µ(ψ,αφ) = α∗ µ(ψ,φ) .
Lastly, for A ∈ Ω we have that
½ ¯Z ¯ ¾
1 ¯¯ ¯
|µ(ψ,φ) (A)| ≤ |µ(ψ,φ) |(X) = sup pdµ(ψ,φ) ¯ : p ∈ C[x]
¯
NT (p) ¯
= sup {|hp(T )ψ, φi|/NT (p) : p ∈ C[x]} = kψk · kφk.
 INTRODUCTION TO SPECTRAL THEORY 9

For any A ∈ Ω, we define µA (ψ, φ) = µ(ψ,φ) (A). The above properties show us that
µA is a symmetric, bilinear functional, and therefore for every A ∈ Ω, there exists a
unique Hermitian operator E(A) such that µA (ψ, φ) = hE(A)ψ, φi for all ψ, φ ∈ H.
We first show that E(A) is idempotent for all A ∈ Ω by proving the more general
result E(A ∩ B) = E(A)E(B) for all A, B ∈ Ω. Fix B ∈ Ω and let {qi }∞ i=1 ⊂ R[x]
be a fixed sequence of polynomials with qi (λ) → 1A (λ) uniformly on X. Also fix
ψ, φ ∈ H. For each i, define a measure νi by dνi (λ) = qi (λ)dµ(ψ,φ) (λ). Then for
any p ∈ C[x], q(T ) is Hermitian commutes with p(T ) and we have that
Z Z
p(λ)dνi (λ) = p(λ)qi (λ)dµ(ψ,φ) (λ) = hp(T )qi (T )ψ, φi
Z
= hp(T )ψ, qi (T )φi = p(λ)dµ(ψ,qi (T )φ) .

Let A ∈ Ω and {pi }∞ i=1 ⊂ R[x] be a sequence of polynomials with pi (λ) → 1A (λ)
uniformly on X. Then the dominated convergence theorem tells us that
Z Z
νi (A) = lim pi (λ)qi (λ)dµ(ψ,φ) (λ) = lim pi (λ)dµ(ψ,qi (T )φ) (λ)
i→∞ i→∞
Z
= 1A (λ)dµ(ψ,qi (T )φ) (λ) = µA (ψ, qi (T )φ) = hE(A)ψ, qi (T )φi
Z
= hqi (T )E(A)ψ, φi = qi (λ)dµ(E(A)ψ,φ) (λ)

The dominated convergence theorem also tells us that

Z Z
hE(A ∩ B)ψ, φi = 1A∩B (λ)dµ(ψ,φ) (λ) = 1A (λ)1B (λ)dµ(ψ,φ) (λ)
Z
= lim νi (A) = lim qi (λ)dµ(E(A)ψ,φ) (λ)
i→∞ i→∞
Z
= 1B (λ)dµ(E(A)ψ,φ) (λ) = hE(B)E(A)ψ, φi.

Since A, B ∈ Ω and ψ, φ ∈ H were arbitrary, this proves that E(A∩B) = E(A)E(B)

for all A, B ∈ Ω. Thus E is idempotent.
Lastly, we have that
Z
hE(X)ψ, φi = µX (ψ, φ) = µ(ψ,φ) (X) = 1dµ(ψ,φ) (λ) = hψ, φi,

and by Theorem (2.2) this means that E(X) is a compact, complex spectral mea-
sure. A quick calculation shows that
Z
λdµ(ψ,φ) (λ) = hT ψ, φi.

The uniqueness of the measure follows from Theorem (3.5). This proves our theo-
rem. ¤
Sketch of proof of Theorem (4.2). Let T be normal and define T1 and T2 by T1 =
1 ∗ 1 ∗
2 (T + T ) and T2 = 2i (T − T ). Then T1 and T2 are Hermitian withRT = T1 + iT2
and there exist unique spectral measures E1 , E2 such that Ti = λdEi (λ) for
i = 1, 2.
Define A = {A + iB : A, B real Borel sets } and define a projection-valued set
function E by E(A + iB) = E1 (A)E2 (B). Then A is an algebra of sets and the
10 WILLIAM CASPER

σ-algebra generated by A is Ω, the collection of all Borel subsets of C. Moreover,

E(C) = E1 (R)E2 (R) = 1 and E extends uniquely to a projection-valued measure
E on C.
¤