4.

Operators, January 21, 2010 1

Chapter 4. Operators
1 Introduction. In this chapter we show that the expression for the aver-
age value of an observable leads us to introduce a new object: the operator

A representing the observable A. This object is so central to quantum me-

chanics that no progress can be made without thoroughly understanding its
properties. We dene the adjoint and the inverse of an operator and examine
unitary and hermitian operators, which have special properties and play an
essential role in quantum mechanics.
In addition, we introduce the eigenvalue problem for operators. We show
that the pure states of an observable A are the eigenstates of the operator

A
and that the spectrum of A consists of the eigenvalues of

A.
2 Average values. If we perform an experiment that puts a system in
a state | and measure the value of an observable A, we can only know
the probability that the measurement gives a particular result. In such a
situation, it is common to describe the measurement by determining the
average value of A. It is therefore necessary to examine how such average
values are calculated in quantum mechanics. I start by reminding you how
the average of a quantity is dened in probability theory.
If we measure a quantity X, which can take the values x
1
, x
2
, x
3
, . . . with
the probabilities p
1
, p
2
,p
3
, . . . , then the average value X of X is dened by
X

i=1
x
i
p
i
(1)
If X takes continuous values x D with the probability distribution p(x)
then
X
_
xD
xp(x) dx (2)
These formulae are the ones you use to calculate the average grade in a
course.
Let us use this for quantum mechanics. If a measurement of A can give
one of the values a
1
, a
2
, . . . or D and the system is in state |, the
average value of A is
A

n=1
P

(a
n
)a
n
+
_
D
dP

() (3)
4. Operators, January 21, 2010 2
Here P

(a
n
) is the probability that a measurement of A, when the system is
in state |, yields a
n
, and P

() d is the probability that the measurement

of A yields a value between and + d. The notation A

is often used
in quantum mechanics for the average value of A when the system is in state
|. When it is clear which state the system is in, we use A.
3 The operator corresponding to an observable. We can rewrite Eq. 3 in
a dierent form. The probability P

(a
n
) is
P

(a
n
) = |a
n
| |
2
= a
n
|

a
n
| = | a
n
a
n
| (4)
Similarly,
P

() = | | (5)
Using Eqs. 4 and 5 allows us to write Eq. 3 as
A

n=1
| a
n
a
n
a
n
| +
_
D
d | | (6)
As was explained in Chapter 3, Dirac wrote this expression as
A

= ( | )
_

n=1
|a
n
a
n
a
n
| +
_
D
d||
_
( | ) (7)
and considered it to consist of the symbol

n=1
|a
n
a
n
a
n
| +
_
D
d| | (8)
sandwiched between the bra | and the ket |.

A is an operator, which
acts on an arbitrary ket | to give another ket

A| =

n=1
|a
n
a
n
a
n
| +
_
D
d| | (9)
Here a
n
| and | are complex numbers. a
n
and are real numbers,
since they are possible results of a measurement of A. |a
n
a
n
a
n
| is a ket
because multiplying the ket |a
n
by the number a
n
a
n
| gives a ket. The
sum in Eq. 9 is a ket because a sum of kets is a ket. The integral in Eq. 9
also gives a ket, because an integral is a gloried sum.
4. Operators, January 21, 2010 3
However, we might have a problem when we state that

A| is a ket,
because the sum has an innite number of terms and without establishing
convergence we are not sure that the sum is meaningful. We leave such
subtleties for the mathematicians, mainly because in almost all practical
problems we need only a nite number of terms in the sum, and convergence
is no longer an issue.
Note that we have a good reason to use the notation

A in Eq. 8: the right-
hand side of that equation contains only quantities tied to the observable A;
in particular,

A does not depend on the state | of the system on which we
perform the measurement.
If the observable is energy, and it has the spectrum E
1
, E
2
, . . . and e
[0, ], we can associate with it the operator

H =

n=1
|E
n
E
n
E
n
| +
_

0
de |e e e| (10)
For example, the pure states of the hydrogen atom are |n, , m with n =
1, 2, . . ., = 0, 1, . . . , n 1, and m = , + 1, . . . , 1, . The energy of
a bound state |E
n
is
1
E
n
=
e
4
2(4
0
)
2
h
2
1
n
2
, n = 1, 2, . . .
The energy of the dissociated electron-proton pair is
e
k
=
h
2
k
2
2
, k [0, ],
where hk is the relative momentum and is the reduced mass; the corre-
sponding ket is denoted by |k.
According to the theory we have just developed, the operator correspond-
ing to the energy of the hydrogen atom is

H =

n=1
n1

=0

m=
|n, , m
e
4
2(4
0
)
2
h
2
1
n
2
n, , m|
+
_
+

dk
x
dk
y
dk
z
|k
x
, k
y
, k
z

h
2
k
2
2
k
x
, k
y
, k
z
| (11)
1
is the reduced mass, e is the charge of the electron, and
0
is the permittivity of
vacuum
4. Operators, January 21, 2010 4
We use the notation

H because the operator corresponding to the energy is
called the Hamiltonian. Eq. 11 is a correct representation of the Hamiltonian
of the hydrogen atom, even though it looks extremely dierent from the
widely used formula:

H =
h
2
2
_

2
x
2
+

2
y
2
+

2
z
2
_

e
2
4
0
r
(12)
where r =

x
2
+y
2
+z
2
is the distance between the electron and the proton.
Exercise 1 Write a formula similar to Eq. 11 for the operator corresponding
to the angular momentum squared. The pure states of angular momentum
squared are |, m with = 0, 1, 2,. . . and m = , + 1, . . . , .
4 The pure states of an observable A are eigenstates of the operator

A and
the spectrum of A consists of all eigenvalues of

A. The result of acting with
the operator

A on a pure state |a
n
of A is particularly simple because of the
orthonormality conditions
a
n
| a
m
=
nm
(13)
and
a
n
| = 0 (14)
Indeed, we have

A|a
n
=

m=1
|a
m
a
m
a
m
| a
n
+
_
D
d| | a
n

m=1
|a
m
a
m

mn
= a
n
|a
n
(15)
and, similarly,

A| = | (16)
for in the continuous spectrum of A.
In operator theory, the equation

A| = |, (17)
4. Operators, January 21, 2010 5
where | is an unknown ket and is an unknown number, is called the
eigenvalue equation of the operator

A. The eigenvalue problem for operators
is so important to quantum mechanics that I will dedicate a separate chapter
to its mathematics.
Any ket | that satises Eq. 17 is called an eigenket or an eigenvector of

A; any that satises Eq. 17 is called an eigenvalue of

A. Eigenvalues and
eigenkets come in pairs. If

A|
1
=
1
|
1
, we say that |
1
is the eigenvector
associated with the eigenvalue
1
, or that
1
is the eigenvalue associated with
the eigenket |
1
.
The formulae in Eqs. 15 and 16 tell us that the pure states |a
n
and |
are eigenfunctions of the operator

A and that the spectrum of A consists of
the eigenvalues of

A.
People who use quantum mechanics spend most of their time solving
various eigenvalue problems. At this point, we are not able to discuss how
this is done, because we have a chicken and egg problem. We dened
operators in terms of pure states and then showed that we can calculate the
pure states by solving the eigenvalue problem for the operator; we are running
in a circle!. To make progress, we must add to our postulates dynamical laws
that allow us to obtain explicit expressions for the operators representing
observables. Once this is done, we can solve the eigenvalue problem for the
operator to obtain the eigenstates and the eigenvalues. The eigenvalues give
us the spectrum of the observable, which are the values that the observable
can have in a measurement. The eigenstates, which are the pure states of
the observable, can be used to calculate the probability that a measurement
yields a specic value from the spectrum.
5 The eigenvalue equation for f(

A). Since we dene
f(

A)

n=1
|a
n
f(a
n
) a
n
| +
_
D
d| f() | (18)
it is easy to show that
f(

A)|a
n
= f(a
n
)|a
n
(19)
and
f(

A)| = f()| (20)
Thus we see that f(

A) has the same eigenfunctions as

A and its eigenvalues
are f(a
n
), f().
4. Operators, January 21, 2010 6
6 The operators representing observables are linear. You can easily verify
that the operator

A =

n=1
|a
n
a
n
a
n
| +
_
D
d| | (21)
has the property:

A(| +|) =

A| +

A| (22)
where is a complex number and | and | are two arbitrary kets in the
space in which

A is dened. An operator for which Eq. 22 is valid is called
linear.
7 Other linear operators. I have pointed out that there is a one-to-one
correspondence between kets and elements of the linear spaces L
2
, or
2
, or
C
N
. It will turn out that a similar correspondence exists between operators
dened in the space of kets and operators dened on L
2
, or
2
, or C
N
. This
is why I give here examples of linear operators in L
2
,
2
, and C
N
.
Operators in L
2
. The set of dierentiable functions f(x) satisfying
_
+

f(x)

f(x) dx < (23)

forms the linear space L
2
. The derivative
d
dx
is an operator in this space if
f

df/ dx also satises

_
+

(x)

(x) dx < . The derivative is a linear

operator because if f(x) and g(x) belong to L
2
then
d
dx
(af(x) +g(x)) = a
df(x)
dx
+
dg(x)
dx
The operator

O dened by the expression

Of =
_
K(x y)f(y) dy (x) (24)
is an operator on functions f L
2
if the function (x) satises the condition
in Eq. 23. It is easy to see that

O is a linear operator.
Operators in C
3
. In the space of three-dimensional vectors x = {x
1
, x
2
, x
3
}
with complex components x
i
, i = 1, 2, 3, a matrix
A =

a
11
a
12
a
13
a
21
a
22
a
23
a
31
a
32
a
33

(25)
4. Operators, January 21, 2010 7
acts on x to give another vector, having the components
Ax {a
11
x
1
+a
12
x
2
+a
13
x
3
, a
21
x
1
+a
22
x
2
+a
23
x
3
, a
31
x
1
+a
32
x
2
+a
33
x
3
} (26)
It is easy to see from this denition that A(x + y) = Ax +Ay. A matrix
is therefore a linear operator in the C
3
space.
Operators in
2
. The
2
space consists of the lists
|f {f
1
, f
2
, f
3
, . . .} (27)
of complex numbers f
i
that satisfy the condition

i=1
f

i
f
i
< (28)
We can dene an operator

O in this space through
|

O|f = {
1
,
2
,
3
, . . .} (29)
where the components in the list | are dened by

j=1
O
ij
f
j
(30)

O is a legitimate operator if

i=1

i
< (31)
because the denition of the operator requires that | be an element of
2
.
You can verify easily that

O is a linear operator.
Quantum-mechanical equations can be written in dierent, equivalent
forms involving kets, or elements of L
2
, or elements of
2
, or elements of
C
N
. The operators representing observables can be dened in the space of
kets, or in L
2
, or in
2
, or in C
N
. They look very dierent but they give the
same results when used to calculate quantities that can be measured.
This ability of the theory to take dierent forms, while giving the same
results when observables are calculated, is not peculiar to quantum mechan-
ics. Classical mechanics can use Newton equation, Lagrange equation, the
4. Operators, January 21, 2010 8
Hamilton-Jacobi formulation, or a variational principle. If you use the same
forces in any one of these representations, you get the same results for
the quantities that can be measured. In statistical mechanics you can use a
canonical ensemble, a microcanonical ensemble, or a grand canonical ensem-
ble; these theories look dierent but give the same results for all observable
quantities.
8 In what follows we look at those mathematical properties of operators
that are essential to quantum mechanics. We avoid all rigor in proofs because
extensive experience with applications shows that the results are correct and
that no ambiguity arises even though we lack the precision so dear to math-
ematicians.
Exercise 2 In the space R
2
consisting of vectors of the form x = {x
1
, x
2
},
the matrix

A =
_
1 3
3 2
_
(32)
is an operator. The eigenvalue problem is

Ax = ax (33)
(a) Solve the eigenvalue problem Eq. 33.
(b) Eq. 33 has an innite number of eigenstates for each eigenvalue. The
eigenstates of

A that are of interest to physics must be pure states
of the observable A. Therefore they must be orthonormal. Pick from
among the solutions of Eq. 33 the set that satises this condition.
Exercise 3 (a) Solve the eigenvalue problem
i
df(x)
dx
= f(x) (34)
where is a real number and x [, +].
(b) Show that the operator i
d
dx
has a continuous spectrum. (That is, can
take any real value and still satisfy Eq. 34.)
4. Operators, January 21, 2010 9
(c) Denote the eigenfunction corresponding to by |. Determine the
eigenfunctions that satisfy the normalization condition
|

= (

)
Hint: use the known formula
_
+

e
i(

)x
dx = 2 (

)
Exercise 4 Solve the eigenvalue equation

d
2

dx
2
= k
2
, x [, +]
and perform the same analysis as in the previous exercise.
Exercise 5 Solve the eigenvalue equation

d
2

dx
2
= k
2

with the conditions

x [0, L]
and
(0) = (L)
Perform the same analysis as in the two previous exercises.
9 Operations with operators. An operator

C is the sum of the operators

A and

B if

C operating on any ket | is equal to

C| =

A| +

B| (35)
Since

A| is a ket and

B| is a ket, the sum

A| +

B| is also a ket.
An operator

C is the product

A

B of the operators

A and

B if

C| =

A(

B|) (36)
4. Operators, January 21, 2010 10
for any ket |. The right-hand side of Eq. 36 indicates that rst you act
with

B on | and obtain the ket |

B|, and then you act with

A on
|.
Now that we have dened the product of two operators, we can dene
how to raise an operator to a power.

A
2
is

A
2
|

A

A|
and it is calculated as follows: act with

A on | to obtain the ket |

A|
and then act with

A on | to obtain

A
2
|

A|. The meaning of

A
n
,
n 2, should now be clear, and of course

A
1
=

A. By convention,

A
0

I is
the unit operator (the do nothing operator), which is dened by

I| = |
for every ket |.
The product of two operators is not a commutative operation:

B| =

B

A| in general (37)
I am not saying that all operators must satisfy Eq. 37. Some operators do
commute, that is, they satisfy

A

B| =

B

A|
for all |.
The expression

A

B

B

A is called a commutator and is denoted by
[

A,

B]

A

B

B

A (38)
When two operators commute, their commutator is identically zero.
10 Examples. Quantum mechanics can be formulated in a variety of
spaces (a space of kets, L
2
,
2
, C
N
) and because of this I give examples of
operations with operators in these spaces.
In the space L
2
dene

Af(x)
d
dx
f(x) (39)
and

Bf(x) V (x)f(x) (40)

where V (x) is some function such that V (x)f(x) is always an element of L
2
when f(x) is.
4. Operators, January 21, 2010 11
We calculate that

A

Bf(x) =

A(V (x)f(x)) =
d
dx
(V (x)f(x))
=
dV (x)
dx
f(x) +V (x)
df(x)
dx
(41)
We can test whether

A and

B commute, by calculating

B

Af(x) =

B
_
df(x)
dx
_
= V (x)
df(x)
dx
(42)
Clearly

A

Bf(x) =

B

Af(x). We can write the commutator:
[

A,

B]|f = (

A

B

B

A)f(x) =

A

Bf(x)

B

Af(x) =
dV (x)
dx
f(x) (43)
The n-th power of the derivative operator is
_
d
dx
_
n
f(x) =
d
dx
d
dx

d
dx
f(x) =
d
n
f(x)
dx
n
(44)
Consider now the operators

A and

B on
2
dened by

A{
1
,
2
, . . .} =

j=1
A
1j

j
,

j=1
A
2j

j
, . . .

(45)

B{
1
,
2
, . . .} =

j=1
B
1j

j
,

j=1
B
2j

j
, . . .

(46)
Here | = {
1
,
2
, . . .} is an arbitrary element of
2
and the complex numbers
A
ij
and B
ij
are given (they dene the operators

A and

B). It is easy to show
that

B

A| =

i=1

j=1
B
1j
A
ji

i
,

i=1

j=1
B
2j
A
ji

i
, . . .

(47)
Exercise 6 Suppose that {|a
n
}

n=1
are pure states of an observable A. Show
the following.
1. If

P = |a
n
a
n
| then

P
2
=

P.
4. Operators, January 21, 2010 12
2. If

P = |a
n
a
n
| and

Q = |a
m
a
m
|, then

P

Q = 0 when m = n and
[

P,

Q] = 0.
3. If

A =

i=1
|a
n
a
n
a
n
|
then

A
m
=

i=1
|a
n
a
m
n
a
n
|
Exercise 7 If {|a
n
}, n = 1, 2, . . . and {|b
n
}, n = 1, 2, . . . are pure states
of two dierent observables A and B, and if

A and

B are the corresponding
operators, nd a condition that ensures that [

A,

B] = 0.
Exercise 8 Two operators

A and

B are dened by

Af =

x
f(x)
and

Bf =
_
+

e
(xy)
2
f(x) dx
Do these operators commute?
11 Functions of operators. We have shown that to each observable A,
having the spectrum a
1
, a
2
, . . . , a
n
, . . . , and D, we can associate an
operator

A =

n=1
|a
n
a
n
a
n
| +
_
D
| | d (48)
In what follows, we will need to dene a variety of functions of an operator.
If you do not pay attention, you would think that this is a trivial matter. If
the function is f(x) and you want to know f(

A), you might replace x with

A
and dene f(

A). Lets see how this works if f(x) =

x sin x and you assume

f(

A)
_

A sin

A. This seems ne, but suppose the operator

A is

x
: what
is the meaning of
_

x
or of sin

x
? This is not a trivial question.
4. Operators, January 21, 2010 13
A problem of a dierent kind appears if I want to dene functions of
two operators

A and

B. Take f(x, y) = xy
2
. If x and y are numerical
variables, then xy
2
= yxy = y
2
x. However, if

A and

B do not commute,

B
2
=

B

A

B =

B
2

A. How do I dene f(

A,

B)? Is it

A

B
2
, or

B

A

B, or

B
2

A?
Dening functions of operators is not straightforward. There is, however,
a simple solution using Eq. 48.
For

A dened by Eq. 48, we dene f(

A) to be
f(

A)

n=1
|a
n
f(a
n
) a
n
| +
_
D
| f() | d (49)
In the right-hand side, f acts on the numbers a
n
and so there is no ambi-
guity regarding the meaning of f(a
n
) and f(); the result is a well-dened
operator.
We know that |a
n
and | are eigenvectors (eigenkets) of

A, and a
n
and
are the corresponding eigenvalues. If you want to calculate f(

A), you solve
the eigenvalue problem for

A and use the result to write down the right-hand
side of Eq. 49.
This denition may seem arbitrary but it has survived because it is useful.
It is also consistent with other denitions in use. One of the most popular
applies if f(x) can be represented by a convergent power-series expansion
f(x) = f
0
+f
1
x +f
2
x
2
+ (50)
For such functions, f(

A) is dened by
f(

A) = f
0
+f
1

A+f
2

A
2
+ (51)
If we know

A and we are supposed to know it if we want to dene the
function f(

A) we can evaluate the right-hand side of Eq. 51 because it
involves only powers of

A. When we write such innite series, we must be
concerned with convergence. If the power series in Eq. 50 is convergent, this
does not mean that the power series in Eq. 51 is convergent. Nor have we
indicated how to dene the domain of values of

A for which Eq. 51 is
convergent. A physicists favorite method for dealing with convergence is to
pray that the series is meaningful. In practice they approximate f(

A) with
a nite sum f(

A) f
0
+ f
1

A + f
2

A
2
+ + f
N

A
N
. By increasing N, they
check that at some point adding a new term causes a negligible change in
4. Operators, January 21, 2010 14
f(

A). When this happens, they are satised with their approximation. This
is not always a guarantee that the series converges, but it works most of the
time.
It is very easy to show that the denition provided by Eq. 49 is consistent
with Eq. 51. Just replace f(a
n
) and f() in Eq. 48 with the expansion in
Eq. 50.
The denition in Eq. 49 allows us to construct new and interesting oper-
ators. You will see that the operator exp
_

H/k
B
T
_
is essential in statistical
mechanics and the operator exp
_
i

Ht/ h
_
gives the time evolution of the
state of a system in quantum mechanics.
12 Adjoint, Hermitian, inverse, and unitary operators: denitions. We
dened the operator

A through Eq. 48:

A =

n=1
|a
n
a
n
a
n
| +
_
D
| | d (52)
Here |a
n
are the eigenstates of

A and a
n
are its eigenvalues. Starting from
this expression we can dene new operators, which are all functions of the
operator

A.
1. The adjoint (or Hermitian conjugate) of

A is

n=1
|a
n
a

n
a
n
| +
_
D
|

| d (53)
2. An operator is called Hermitian (or self-adjoint) if all a
n
and D
are real numbers. From Eqs. 52 and 53 it follows that

A is Hermitian
if and only if

=

A (54)
3. An operator is unitary if
|a
n
| = 1 for all n and || = 1 for all D (55)
Here |z| is the absolute value of the complex number z. You can verify
that Eq. 55 is equivalent to requiring that
a
n
= e
ip
n
for n 1 and = e
ip()
for D (56)
where p
n
and p() are real numbers and i =

1.
4. Operators, January 21, 2010 15
4. The operator

A
1

n=1
|a
n

1
a
n
a
n
| +
_
D
|
1

| d (57)
is called the inverse of

A. Note that if some a
n
is zero then this expres-
sion makes no sense, and in that case we say that

A does not have an
inverse or that it is singular. If the integral in Eq. 57 is not well-dened,
the operator does not have an inverse.
You will use these denitions, and the properties that follow from them, over
and over and over. Other books use dierent, but equivalent, denitions.
The ones used here lead most eciently to useful results. The methods used
in other books are more general but we do not need this additional generality
here.
13 Notation. In what follows we derive some important properties that
follow from these denitions. Before proceeding, let us establish some nota-
tion. If |

A| then I either write
| |

A (58)
or
| |

A| (59)
Mathematicians tend to use Eq. 58 and physicists, Eq. 59.
I also use the notation
|

A | (60)
14 Some properties of adjoint operators.
Property 1. If

A

is the adjoint of

A, then for any | and | we have
|

A =

| (61)
and

A| = |

A

(62)
Here

| and

A| are the bras corresponding to the kets

A

| and

A|,
respectively. Eqs. 61 and 62 follow from the denition of the adjoint operator
and the properties of the scalar product. To simplify our lives, I will discard
4. Operators, January 21, 2010 16
the integral in the denitions of the operators. Since the integral is the limit
of a sum, what goes for the sum goes for the integral. Start with

| and
rewrite it using known relationships, to show that it is the same as |

A| :

| = |

A

(used x| y = y | x

)
= (

n=1
| a
n
a

n
a
n
| )

(used Eq. 53)

=

n=1
| a
n

a
n
a
n
|

(used (zw)

= z

)
=

n=1
| a
n
a
n
a
n
| (used x| y = y | x

)
= |

A (used Eq. 52)
This proves Eq. 61. For Eq. 62, rst note that, from the denition (Eq. 53),
it follows that
(

A

=

A (63)
Therefore
|

A

= (

A

| (used Eq. 61)

=

A| (used Eq. 63)

Many mathematicians prefer to use Eq. 61 as a denition of an adjoint
operator, because it is more general than the one used here; it is valid for
operators not covered by the denition in Eq. 52 on which the material
developed here is based.
Property 2. If

A | = |

B (64)
for any | and |, then

B is the adjoint of

A.
Write Eq. 64 as
|

A

= |

B (65)
Since this is valid for any kets | and |, it is valid for the pure states |a
n

of A (eigenstates of

A). Therefore Eq. 65 gives (using Eq. 52 and orthonor-
mality)
a
n
|

B| a
n
= a
n
|

A| a
n

= a

n
(66)
and, for m = n,
a
n
|

B| a
m
= a
n
|

A| a
m

= 0 (67)
Since the set of pure states is complete,

n
|a
n
a
n
| =

I and so we can write

B as

B =

I

B

I =

m
|a
n
a
n
|

B| a
m
a
m
| =

n
|a
n
a

n
a
n
| (68)
4. Operators, January 21, 2010 17
To obtain the last equality, I used Eqs. 66 and 67. Eq. 68 proves that

B =

A

(see Eq. 53).

We will often have to deal with products of operators, such as

A

B. What
is the adjoint of the product? That is, if we set

C =

A

B, what is

C

?
Property 3. (

A

B)

=

B

From Property 1, we know that we can write

|

C =

| (69)
for any operator

C. Let us replace

C in the left-hand side of this equation
with

C =

A

B and use Property 1 twice:
|

C = |

A(

B) =

|

B =

| (70)
This is true for any kets | and |. This means that (compare Eqs. 69 and
70)

C

=

B

, which is what we set out to prove.

15 Properties of a self-adjoint (Hermitian) operator.
Property 4. The eigenvalues of a Hermitian operator must be real numbers.
This property follows from the results in 4, where we showed that a
n
and
are the eigenvalues of

A. If

A is Hermitian, then (by the denition) a
n
and
are real.
Property 4 is very important, for the following reason. Consider an ob-
servable A with the spectrum a
n
, n = 1, 2, . . . and D. These are the
values that a measurement of A can give and they must be real numbers.
You have also learned that the spectrum of A is identical to the set of all
eigenvalues of the operator

A corresponding to the observable A (4). There-
fore all eigenvalues of

A must be real numbers. According to Property 3, this
requirement is automatically fullled if

A is Hermitian. Therefore, we expect
that all operators corresponding to observables are Hermitian.
Property 5.

A is self-adjoint (Hermitian) if and only if

A =

A

.
The proof is straightforward. By the denition of adjoint, if

A is dened
by

A =

n=1
|a
n
a
n
a
n
| (71)
4. Operators, January 21, 2010 18
then

n=1
|a
n
a

n
a
n
| (72)
Obviously if each a
n
is real (i.e.

A is Hermitian) then these two expressions
are identical.
Conversely, if

A =

A

, then for any value of n,

a
n
|

A

| a
n
= a
n
|

A| a
n
(73)
Using Eq. 72 and the fact that a
i
| a
j
=
ij
, it is easy to see that a
n
|

A

| a
n
=
a

n
. Similarly, using Eq. 71, a
n
|

A| a
n
= a
n
. These, together with Eq. 73,
imply that a

n
= a
n
and a
n
must be a real number; therefore if

A =

A

then

A is Hermitian.
Property 6. If

A is self-adjoint, then

A| = |

A.
This follows from Properties 1 and 5.
Exercise 9 Suppose that

A and

B are Hermitian. Prove the following.
(a) (

A

B)

=

B

A.
(b)

A +

B is Hermitian.
(c) [

A,

B]

=

B

A

A

B = [

A,

B]
where [

A,

B]

A

B

B

A is the commutator of

A with

B.
(d) {

A,

B}

A

B +

B

A is Hermitian.
16 Inverse operators. If

A is dened by

A =

n=1
|a
n
a
n
a
n
| (74)
then its inverse is

A
1

n=1
|a
n

1
a
n
a
n
| (75)
4. Operators, January 21, 2010 19
As noted earlier, if any a
n
in the spectrum of

A is equal to zero, then the
operator

A has no inverse (is singular). We know that each a
n
is an eigenvalue
of

A and therefore we can state:
Property 7. If one of the eigenvalues of

A is equal to zero, then

A does not
have an inverse.
Property 8. If

A and

B satisfy

B =

B

A =

I (76)
and all the eigenvalues of

A dier from zero, then

B =

A
1
.
Here

I is the unit operator. Note that the equality in Eq. 76 means that
|

A

B| = |

B

A| = | for every | and |.
Write

A =

n=1
|a
n
a
n
a
n
| (77)
I have dropped the integral since that term can be treated by analogy with
the sum. From

B

A =

I, we have, for every n,
a
n
|

B

A| a
n
= a
n
|

I | a
n
= 1 (used a
n
| a
n
= 1) (78)
But also
a
n
|

B

A| a
n
= a
n
a
n
|

B| a
n
(used

A|a
n
= a
n
|a
n
) (79)
Comparing these two values for a
n
|

B

A| a
n
, we see that
a
n
|

B| a
n
= 1/a
n
(80)
Starting again from

B

A =

I, we have
a
n
|

B

A| a
m
= a
n
|

I | a
m
= 0 (used orthogonality) (81)
for every n and m = n, and also
a
n
|

B

A| a
m
= a
m
a
n
|

B| a
n
(used

A|a
m
= a
m
|a
m
) (82)
If a
m
= 0 for all m, comparison of Eqs. 81 and 82 leads to
a
n
|

B| a
m
= 0, n = m (83)
4. Operators, January 21, 2010 20
We conclude from Eqs. 80 and 83 that when all the eigenvalues of

A are
non-zero,
a
n
|

B| a
m
=
nm
/a
m
(84)
Now write

B =

I

B

I =

m
|a
n
a
n
|

B| a
m
a
m
|
=

n

m
|a
n

nm
a
m
a
m
| (used Eq. 84)
=

n
|a
n

1
a
n
a
n
| (85)
which, from the denition of inverse operator, means that

B =

A
1
.
Inverse operators are handy when we want to solve an equation of the
form

A| = | (86)
where | is unknown and

A and | are known. Acting with

A
1
on Eq. 86
and using

A
1

A =

I and

I| = | converts Eq. 86 to
| =

A
1
| (87)
To this particular solution we can add any | having the property

A| = 0 (88)
Therefore the general solution of

A| = | is
| =

A
1
| + | (89)
since then

A| =

A(

A
1
| + |) =

A

A
1
| +

A| =

I| + 0 = |
This seems to be a very ecient way of solving Eq. 86, but the trouble is
that in many cases

A
1
is hard to calculate. Nevertheless, the inverse of an
operator is used often in quantum mechanics.
Consider the following equation in L
2
:
d
2
(x)
dx
2
= (x)
4. Operators, January 21, 2010 21
You can formally write
(x) =
_
d
2
dx
2
_
1
(x) +(x)
where (x) satises
d
2
(x)
dx
2
= 0
But what is
_
d
2
(x)
dx
2
_
1
? In this case you can calculate it (do so), but, in
general, nding the inverse of an operator is not easy.
17 Unitary operators. The operator

A dened through

A =

n=1
|a
n
a
n
a
n
| (90)
where |a
n
is an orthonormal set of kets, is called unitary if
a

n
a
n
= 1, n = 1, 2, . . . (91)
This is equivalent to requiring that
a
n
= e
i
n
, n = 1, 2, . . . (92)
where
n
is a real number.
2
Property 9. The kets |a
n
entering into Eq. 90 for a unitary operator are
the eigenkets of that operator and the numbers a
n
are the corresponding
eigenvalues. Therefore the eigenkets of a unitary operator are orthonormal
and the absolute values of the eigenvalues are equal to 1 (see Eq. 91). The
proof of these statements is through straightforward calculation.
2
We have introduced Hermitian operators through Eq. 91 based on physical arguments.
We cannot use the same arguments to lead us to the conclusion that unitary operators
have the this form. The denition give here assumes that this form is correct and then it
denes unitarity. This form implies that the kets |a
n
are orthonormal and that a
n
are
the eigenvalues and |a
n
are the eigenvectors of

A. Most texts dene the adjoint operator
A

through

| |

A and unitary operators through

A

A

=

A

A =

I. Then one
can show that an operator is unitary if and only if satises the denition given here (see
Halmos, Finite-Dimensional Vector Spaces, 80, the section on normal transformations).
4. Operators, January 21, 2010 22
Property 10.

A is unitary if and only if it has an inverse and

A
1
=

A

.
The inverse of

A =

n=1
|a
n
a
n
a
n
| is

n=1
|a
n

1
a
n
a
n
|. Since

A is uni-
tary, a
n
= exp[i
n
] for some real numbers
n
and
1
a
n
= exp[i
n
] = a

n
Therefore

A
1
=

n=1
|a
n

1
a
n
a
n
|

n=1
|a
n
a

n
a
n
| =

A

Next, assume that

A
1
=

A

. This means (using the denitions of

A
1
and

A

) that

n=1
|a
n

1
a
n
a
n
| =

n=1
|a
n
a

n
a
n
| (93)
Since a
m
| a
n
=
nm
, Eq. 93 leads to (act with Eq. 93 on |a
m
and then act
with a
m
| on the result)
1
a
m
= a

m
for each m (94)
And this means that a
m
a

m
= 1 for each m and therefore

A is unitary.
Property 11. If

A is Hermitian then the operator

U exp[i

A], where is
a real number, is unitary. Also, if

U is a unitary operator, then it can be
written as

U = exp[i

B] where

B is Hermitian.
For

A =

n=1
|a
n
a
n
a
n
|, the operator exp[i

A] is dened by
exp[i

A] =

n=1
|a
n
e
ia
n
a
n
| (95)
If

A is Hermitian then all the a
n
are real numbers and so are all a
n
; therefore
(e
ia
n
)

e
ia
n
= e
ia
n
e
ia
n
= 1 for all n and exp[i

A] is unitary.
Conversely, if an operator

U is unitary then it must have the form

U =

m=1
|u
m
e
i
m
u
m
| (96)
with
m
real.
4. Operators, January 21, 2010 23
Consider now the operator

B dened to have the spectrum
m
and the
pure states |u
m
, so that

B =

m=1
|u
m

m
u
m
| (97)

B is Hermitian because the

m
are real numbers. A function f(

B) is dened
by
f(

B) =

m=1
|u
m
f(
m
) u
m
| (98)
If f is the function f(x) = e
ix
then
f(

B) e
i

B
=

m=1
|u
m
e
i
m
u
m
| (99)
which means that e
i

B
is unitary.
Property 12. If

A is Hermitian and

U exp[i

A] then

A and

U have the
same eigenvectors (eigenkets): if

A|a
n
= a
n
|a
n
then

U|a
n
= u
n
|a
n
and
vice versa. In addition, u
n
= e
ia
n
.
The proof of this is hidden in the proof of Property 9. You can prove it by
direct calculation.
Exercise 10 Consider a function f(x) of the real variable x for which the
values f(x) are complex (exp[ix] is an example). Let

A be a Hermitian
operator with spectrum {a
n
} and eigenkets {|a
n
}. Show that if
|f(a
n
)| = 1 for all n (100)
then f(

A) is unitary. Also show the converse: if f(

A) is unitary then (100)
must be true.
Exercise 11 Show that if x varies along the real axis then y (xi)/(x+i)
varies on the unit circle in the complex plane.
4. Operators, January 21, 2010 24
Exercise 12 Show that if

A is Hermitian then

B (

Ai

I)(

A+i

I)
1
is unitary (show rst that

A + i

I has an inverse). This is called a Cayley

transform, and it is the basis of the Crank-Nicholson method for solving
dierential equations.
Property 13.

U is unitary if and only if

U|

U = | for any kets |, | (101)
We say that the scalar product is invariant under a unitary transformation.
It is easy to show that Eq. 101 holds for a unitary transformation:

U|

U = |

U

U (used Property 1)
= |

U
1

U (used Property 4)
= |
Let us prove the reverse: if Eq. 101 is valid then

U is unitary. If |u
n
are
the pure states of

U then

U =

n
|u
n
u
n
u
n
| (102)
and

I =

n
|u
n
u
n
| (103)
We start with Eq. 101:
| =

U|

U
= |

U

U (used Property 1)
= |

m

n
|u
n
u

n
u
n
| u
m
u
m
u
m
| (used Eq. 102)
= |

m

n
|u
n
u

nm
u
m
u
m
| (used u
n
| u
m
=
nm
)
= |

n
|u
n
u

n
u
n
u
n
|
This is equal to | only if

n
|u
n
u

n
u
n
u
n
| =

I
4. Operators, January 21, 2010 25
Comparing this to Eq. 103 (completeness) implies that u

n
u
n
= 1 for all n,
which means that

U is unitary.
Consider two sets of kets {|x
i
}
N
i=1
and {|y
i


U|x
i
}
N
i=1
for some operator

U. If

U is unitary and one of the sets is complete and orthonormal then the
other set is also complete and orthonormal. In addition, if both sets are
complete and orthonormal then

U is unitary. This is a corollary of Property
13.