2 Observables

In this chapter we are going to set up the formalism to describe observables in

Quantum Mechanics. This is an essential part of the formulation of the theory, as
it deals with the description of the outcome of experiments. Beyond any theoretical
sophistication, a physical theory is first and foremost a description of natural phe-
nomena; therefore it requires a very precise framework that allows the observer to
relate the outcome of experiments to theoretical predictions. As we will see, this is
particularly true for quantum phenomena. The necessary formalism is very different
from the intuitive one used in Classical Mechanics. A subtle point is that the state
of the system is not an observable by itself. As seen in the previoius chapter, the
state of the system is specified by a complex vector. Given a quantum system in a
state that is described by a ket, we can perform measurements in that particular
state; the results of these measurements are what we call observables. Therefore we
need a formalism that allows us to determine

• the possible outcomes of a measurement;

• what can be predicted about these outcomes, assuming that we know the complex
vector that describes the state of the system.

In order to achieve these goals we will need to set up a new mathematical structure,
based on operators that act on the state vectors. In this chapter you will learn the
exact relation between states, operators and the outcome of experiments. This is a
crucial step in building a framework that can yield predictions for the outcome of
experiments, i.e. to connect the theory to the experiment. This chapter is focussed
on the general formalism and a few initial examples. More observables, like e.g. the
angular momentum, will be introduced in detail in following chapters.

2.1 Introducing Observables

We define an observable O to be the outcome of an experiment that is measured

by a given apparatus. The description of observables needs to be specified very
precisely. This precise definition is encoded in a series of postulates that we are
going to introduce in this chapter and discuss in detail. The overall picture can be
summarised as follows.
For a quantum system in a state described by some complex vector |ψi. The
i
ii Observables

outcome of a measurement of the observable O is a stochastic variable. The theory

can predict

1. the probability distribution function (pdf), pO , that describes the possible results
of the measurement and their respective frequencies;
2. the state of the system after the measurement.

We should note in passing that the knowledge of the pdf is equivalent to the
knowledge of all its moments. Remember that the result of a measurement needs
to be treated as a random variable and depending on the context we may focus
either on its pdf or on its moments, i.e. we will want to compute its mean value,
its variance, etc.

Mathematical aside: moments of probability distributions

Let us consider a random real variable r, characterised by a probability distribution
function p(r). The probability of finding r in the interval I = [a, b] is
Z b
P (r ∈ I) = dx p(x) . (2.1)
a

(Note that x is a dummy variable of integration, we could use any name to identify
the integration variable. If this does not sound familiar, think about it...)
The n-th moment of the distribution p is defined as:
Z
µn = dr p(r) rn . (2.2)

For a properly normalised probability distribution function:

µ0 = 1 . (2.3)

The mean value of the random variable r is:

Z
µ1 = hri = dr p(r) r . (2.4)

The variance of the variable r is:

Var[r] = µ2 − (µ1 )2 . (2.5)

Exercise 2.1.1 Compute the mean value and the variance of a Gaussian variable
r, i.e. a random variable with probability distribution function:
" #
2
1 (x − x0 )
p(r) = √ exp − (2.6)
2πσ 2 2σ 2

Note that the Gaussian distribution is completely determined by the knowl-

edge of its first two moments.
iii Introducing Observables

2.1.1 A First Example: Position Operator

As discussed in the previous chapter, the modulus squared of the wave function
|ψ(x)|2 describes the probability density of finding the system in the state |ψi at
x. So the outcome of a measurement of the position of the system, x, is a real
stochastic variable and its pdf is |ψ(x)|2 . The mean value of the position of the
system is therefore
Z
hxi = dx |ψ(x)|2 x . (2.7)

The crucial point here is the realization that the mean value of the position of
the system in the state |ψi, hxi, can be written as the expectation value of a
suitably-defined operator X̂ in the state |ψi, i.e. the matrix element of the operator
X̂ between |ψi and |ψi.
The Hermitian operator X̂ associated to the position of the system is such that
the states |xi introduced in the previous chapter are eigenstates of X̂ to the eigen-
value x – pay attention to the notation here! Using the fact that |xi are eigenstates
of X̂, we have
 ∗
hx|X̂|ψi = hψ|X̂|xi (2.8)
∗
= x (hψ|xi) (2.9)
= xhx|ψi ; (2.10)

with a slight abuse of notation, we can think of the operator X̂ acting on the wave
function as

X̂ψ(x) = xψ(x) . (2.11)

It is important to realise that Eq. (2.11) does not mean that ψ(x) is an eigenfunction
of X̂ with eigenvalue x. The eigenvalue needs to be a constant that multiplies the
state vector. The mean value in Eq. (2.7) can then be written as:
Z
hX̂i = hψ|X̂|ψi = dx ψ(x)∗ xψ(x) = hxi . (2.12)

As shown explicitly in the equation above, the expectation value in Eq. (2.12)
depends on the state ψ. When there is no ambiguity, we shall omit the explicit ψ
dependence, and write simply hX̂i, but remember that this is the mean value in a
given state.
The mean value of the outcome of a measurement, hxi is therefore expressed as
the expectation value of an operator acting in the space of physical states. We will
now generalize this idea to a generic observable and discuss in detail the physical
meaning of the mean value in the equation above.
iv Observables

2.1.2 Generic Observables

We define an observable to be a quantity that can be measured in an experiment.

Experiments here are considered as black boxes: the system is prepared in some
quantum state |ψi, then the experiment is performed and yields a value for the
observable O. We will not dwell upon the details of the experimental settings, but
we will assume that the experiment is reproducible, i.e. that is possible to repeatedly
prepare the system in the same state and perform the same measurement. Following
the example discussed in Sect. 2.1.1 for the position of a quantum system, we are
going to associate an operator to each observable. This is the second postulate
of Quantum Mechanics.

Postulate 1 In Quantum Mechanics each observable O is associated to a linear

operator Ô acting in the Hilbert space of physical states:

Ô :H → H
|ψi 7→ |ψ 0 i = Ô|ψi . (2.13)

This is a weird statement! The postulate associates an operator to each observ-

able. However be aware that the measurement is NOT associated with acting with
the operator on the state. By analogy with the example of the position operator,
we expect that the mean value of the observable O in a quantum state can be
computed as the expectation value of the operator Ô in that state:

hOi = hψ|Ô|ψi , (2.14)

The link between operators and the outcome of experiments needs to be explained
in detail and this will be the focus of this chapter.

Example 2.1 The operator associated to the energy of the system is the Hamil-
tonian. The Hamiltonian of the system may be written as Ĥ = T̂ + V̂ , where the
kinetic and potential energy operators are defined by:

P̂ 2
T̂ = , V̂ = V (X̂) . (2.15)
2m
The operator V (X̂), which is a function of the position operator X̂, acts as:

V (X̂)ψ(x) = V (x)ψ(x) . (2.16)

Any other operator that is a function of X̂ can be defined in the same way. We
shall see later how to define the operator P̂ , associated to the momentum of the
system, and more complicated operators. Ĥ acts on state vectors and returns state
vectors.
v Observing Observables

2.2 Observing Observables

Let us now explain in more detail the process of measurement in Quantum Mechan-
ics and in particular what we mean when we say that the result of an experiment
is a stochastic variable. If the observable O 
is measured several times under identi-
cal conditions, the results is a set of values O(1) , O(2) , . . . , O(n) , where the suffix
labels each measurement. This is a characteristic feature of Quantum Mechanics:
the results of several independent measurements performed on a system prepared
always in the same state |ψi are different; they are distributed according to some
pdf that can be predicted by the theoretical framework. Hence the mean value of
the observables in the given state can be computed by taking the average of the
outcomes of the measurements, in the limit where the number of measurements
becomes large. We see here the first explicit connection between the operator Ô
and experimental results.
The expectation value of the operator Ô defined in Eq. (2.14) is equal to the
Pn
average k=1 O(k) /n in the limit where the number of measurements n → ∞.
So far our examples have focused on the mean value of an observable, i.e. the
first moment of its probability distribution. Moving beyond the first moment of the
pdf, the theoretical framework of Quantum Mechanics allows us to predict:

1. the possible outcomes of the measurements above;

2. the probability of obtaining each of these possible outcomes.

It is important to realise that all questions in Quantum Mechanics need to be

formulated in terms of possible outcomes of an experiment and their respective
probabilities. These are the only quantities that are predicted by the theory. Trying
to shortcut this procedure leads to well-known paradoxes.

Eigenvalues and Eigenstates

Our next step in building the theoretical framework is to specify the relation be-
tween the outcomes of measurements and some properties of the operator Ô. This
is encapsulated in the third postulate of Quantum Mechanics .

Postulate 2 The possible outcomes of experiments, Ok , are the eigenvalues of

the operator Ô, i.e. the solutions of the eigenvalue equation:

Ô|ψk i = Ok |ψk i , (2.17)

where |ψk i is the eigenstate corresponding to the eigenvalue Ok . Note that an

eigenstate is always associated to a specific eigenvalue.

Mathematical aside: eigenvalues and degeneracies

vi Observables

An eigenvalue Ok of an operator Ô is called g-fold degenerate if there are exactly

g linearly independent eigenvectors corresponding to the same eigenvalue:
(n) (n) (n)
∃ |uk i, such that Ô|uk i = Ok |uk i, for n = 1, . . . , g . (2.18)

Note that any linear combination

g
(n)
X
cn |uk i, with cn ∈ C , (2.19)
n=1

is also an eigenstate of Ô with the same eigenvalue Ok . The space of eigenstates

corresponding to the eigenvalue Ok is g-dimensional vector space.

The eigenfunction represents the state in which the measurement of O yields the
value Ok with probability 1. To check this statement, we can compute the variance
of O in the state |ψk i:

Vark [O] = hO2 i − hOi2 (2.20)

2 2
= hψk |Ô |ψk i − hψk |Ô|ψk i =
2
= Ok2 hψk |ψk i − (Ok hψk |ψk i) = 0 , (2.21)

where we have used the fact that the eigenfunctions are normalized to one.
It is important in the equations above to distinguish the outcome of the i-th
measurement, denoted O(i) , from the set of possible outcomes Ok . The latter are a
property of the operator Ô and do not depend on the state of the system.

2.3 Hermitian Operators

We stated above that every observable is represented by an operator; ultimately we

want to establish a correspondence

observable ⇐⇒ operator
total energy ⇐⇒ Ĥ
position ⇐⇒ X̂
momentum ⇐⇒ P̂
.. ..
. ⇐⇒ .
Before we look into the details of associating observables with operators, we need
to establish some properties of these operators. The main constraint comes from
the fact that observables take real values only. Therefore we must require that
the operators that represent observables have only real eigenvalues, since we
want to identify the eigenvalues with the possible results of measurements. We can
guarantee this if we only use Hermitian operators to represent observables.
vii Hermitian Operators

2.3.1 Hermitian Conjugate

Let us define first the Hermitian conjugate Ô† of an operator Ô.

Let |ψi and |φi be arbitrary states in H, then
 ∗
hφ|Ô† |ψi = hψ|Ô|φi (2.22)

In order to compute the matrix element of Ô† between |φi and |ψi, we need to
compute the matrix element of Ô swapping the states and then taking the complex
conjugate.

Mathematical aside: a comment on Hermitian conjugation

Compare Eq. (2.22) with the more familiar expression from linear algebra:
† ∗
Oij = Oji . (2.23)

The quantities in Eq. (2.22) are the matrix elements of the operator Ô, just like
Oij are the matrix elements of a usual matrix O. The quantum states |φi and |ψi
are the “indices” that label the matrix elements. Using this identification, many
equations that we encounter in Quantum Mechanics become rather familiar.

d
Example 2.2 The operator Ô = is defined by specifying its action on the
dx
wave function of a quantum system:
d
Ôψ(x) = ψ(x) . (2.24)
dx
Following our prescriptions, the matrix element of Ô between two states |ψi and
|φi is given by
Z
d
hψ|Ô|φi = dx ψ(x)∗ φ(x) ; (2.25)
dx
then we can integrate by parts to obtain
Z ∞ Z ∞
d ∞ d
ψ(x)∗ φ(x) dx = [φ(x)ψ(x)∗ ]−∞ − φ(x) ψ(x)∗ dx (2.26)
−∞ dx −∞ dx
We can discard the constant term on the right hand side, since physically acceptable
wave functions vanish at x = ±∞, and if we then take the complex conjugate of
the resulting equation we obtain
Z ∞ ∗ Z ∞
∗ d d
ψ(x) φ(x) dx = − φ(x)∗ ψ(x) dx
−∞ dx −∞ dx
Z ∞  †
∗ d
≡ φ(x) ψ(x) dx
−∞ dx
viii Observables

from the definition of Hermitian conjugate. Thus we can make the identification
 †
d d
=− .
dx dx

2.3.2 Hermitian Operators

We can now define a Hermitian operator. An operator Ô acting on a Hilbert

space H is called Hermitian if

Ô† ≡ Ô . (2.27)

d
Example 2.3 Eq. (2.27) is clearly not true for all operators; is NOT Hermitian
dx
since we have just shown that
 †
d d
=− ,
dx dx

d
whereas the operator −i~ IS Hermitian; the proof is straightforward and is left
dx
as an exercise to the reader.

Strictly speaking physical observables are associated to self-adjoint operators. In

order to be able to define self-adjoint operators we first need to define the domain,
D(Ô) of an operator Ô. The domain D(Ô) is the subspace of H upon which the
operator acts, i.e.

Ô|ψi ∈ H =⇒ |ψi ∈ D(Ô) . (2.28)

A self-adjoint operator is a Hermitian operator such that

D(Ô† ) = D(Ô) . (2.29)

While the Hermitian operators that we encounter in Quantum Mechanics are mostly
automatically self-adjoint, this pedantic distinction becomes relevant in some phys-
ically meaningful problems. Typically problems with non-trivial boundary condi-
tions require some care to extend an Hermitian operator so that it is actually
self-adjoint.
ix Properties of Hermitian Operators

2.4 Properties of Hermitian Operators

Hermitian operators obey properties that are important for building the logical
framework of Quantum Mechanics.
1. Hermitian operators have real eigenvalues. The eigenvalue equation is:
Ô|ψk i = Ok |ψk (x)i , k = 1, . . . .
|ψk i are the eigenfunctions of Ô, Ok are the eigenvalues. Then we have:
Ô = Ô† =⇒ Ok ∈ R
2. The eigenstates of a Hermitian operator that belong to different eigenvalues are
orthogonal.
3. If Ô is a Hermitian operator acting on a vector space H, there exists an orthog-
onal basis of H made of eigenvectors of Ô. In other words, every vector |ψi can
be expanded as:
X
|ψi = ck |ψk i , (2.30)
k

where ck are complex coefficients, computed by taking the projection of |ψi onto
the states |ψk i:
ck = hψk |ψi .
As you can see from the equation above, for each state |ψi there is a set of
coefficients ck , they are the coordinates of the function |ψi in the basis {|ψk i , k =
1, . . .}. Do not confuse the coefficients ck with the eigenvalues Ok ! The latter are
a characteristic of the operator Ô and have nothing to do with the state ψ.
4. The commutator of two Hermitian operators is anti-Hermitian. While this prop-
erty is trivial to prove, it is useful to keep in mind as a way to check your results.
Every time you compute a commutator of Hermitian operators, you have a san-
ity check of your answer.

Mathematical aside: proof of the properties

We shall now prove the first two properties above. The proofs are useful examples of
manipulations involving operators acting on wave functions. Familiarity with these
kind of manipulations is essential for solving problems in Quantum Mechanics.
1. Hermitian operators have real eigenvalues.
Proof. Suppose Ô is a Hermitian operator so that Ô† = Ô, and let Ô have an
eigenvalue Ok , with corresponding eigenfunction |ψk i:
Ô |ψk i = Ok |ψk i
Then
hψk |Ô|ψk i = Ok hψk |ψk i = Ok ,
x Observables

where in the last equality we have used the fact that the ket |ψk i is properly
normalised to one.
If we take the complex conjugate of this equation, we obtain
 ∗
hψk |Ô|ψk i = Ok∗

but if we make use of the definition of the Hermitian conjugate, we can rewrite the
left-hand side of this equation in terms of Ô† and use the fact that Ô† = Ô by
hypothesis:
 ∗
hψk |Ô|ψk i = hψk |Ô† |ψk i = hψk |Ô|ψk i .

The right-hand side is now just the matrix element that appears in the first equation
and is equal to Ok , so we have proved that
Ok∗ = Ok
thus showing that the eigenvalue Ok is real as stated.
2. The eigenfunctions of a Hermitian operator which belong to different eigenval-
ues are orthogonal.
Proof. Suppose that
Ô |ψ1 i = O1 |ψ1 i and (2.31)
Ô |ψ2 i = O2 |ψ2 i with O1 6= O2 . (2.32)
From Eq. (2.31) we have
hψ2 |Ô|ψ1 i = O1 hψ2 |ψ1 i , (2.33)
whereas from Eq. (2.32)
hψ1 |Ô|ψ2 i = O2 hψ1 |ψ2 i . (2.34)
Taking the complex conjugate of Eq. (2.34) yields on the left hand side
 ∗
hψ1 |Ô|ψ2 i ≡ hψ2 |Ô† |ψ1 i = hψ2 |Ô|ψ1 i , (2.35)

whereas the right hand side gives

O2∗ hψ2 |ψ1 i = O2 hψ2 |ψ1 i , (2.36)
using the fact that O2 = O2∗ .
Comparing with Eq. (2.33) we see that
O2 hψ2 |ψ1 i = O1 hψ2 |ψ1 i , (2.37)
which we can rearrange to yield the result
(O2 − O1 )hψ2 |ψ1 i = 0 . (2.38)
Given that O2 6= O1 by hypothesis, this implies that
hψ2 |ψ1 i = 0 (2.39)
xi Spectral Decomposition

which is the desired result.

2.5 Spectral Decomposition

Using the fact that we can define a basis in the space of physical states made
of eigenstates of the operator Ô as shown in Eq. (2.30), we can formulate the
rule to compute the probability of finding any given eigenvalue when performing a
measurement. This is the fourth postulate of Quantum Mechanics.

Postulate 3 Given the decomposition in Eq. (2.30), the probability of finding

the non-degenerate eigenvalue Ok when measuring O in the state described by the
normalised vector |ψi is given by:
2
Pk = |ck | . (2.40)

Clearly the sum of probabilities should be properly normalized and therefore:

X X 2
Pk = |ck | = 1 . (2.41)
k k

This is simply a confirmation of the fact that the state vector |ψi is normalised to
one.

Example 2.4 Note that the concept of superposition of states is very different
from anything we have encountered in Classical Mechanics. Consider two quantum
states |Ai and |Bi, such that the measurement of an observable O yields the result
a with probability 1 when the system is in the state |Ai, and the result b with
probability 1 when the system is in the state |Bi. The superposition principle
states that the state vector:

|Ci = cA |Ai + cB |Bi , (2.42)

where cA and cB are complex numbers such that |cA |2 + |cB |2 6= 0, describes a
possible physical state of the system. According to the postulates that we formulated
so far, the measurement of the observable O in the state |Ci can only yield the value
a or b, with respective probabilities:
|cA |2 |cB |2
pa = , pb = . (2.43)
|cA + |cB |2
|2 |cA + |cB |2
|2
No other results are possible for the measured value of O in the state |Ci.

Example 2.5 Let us consider again the discretized one-dimensional system in

xii Observables

Fig. ??; we can have a state |1i where the particle is localized e.g. at site 1, and
a state |2i where the particle is localized at site 2. Measuring the position of the
particle in state |1i yields x = 1 with probability 1. Likewise we obtain x = 2 with
probability 1 for a particle described by the state vector |2i.
The state √12 |1i + √12 |2i is an admissible quantum state. Measuring the position
of the particle in this latter state, the outcome will be x = 1, or x = 2 with 50%
probability. No other value is allowed in this state.

Degenerate Eigenvalues
In the case where the eigenvalue Ok is g-fold degenerate the probability of Ok
being the result of a measurement of the observable O in the state |ψi is obtained
by summing over the contributions from the whole subspace spun by the degenerate
eigenvectors. Following the notation introduced in Eq. (2.18), the probability is
g
2
(n)
X
Pk = huk |ψi . (2.44)
n=1

2.6 Collapse of the State Vector

Given the state vector |ψi, the fourth postulate of Quantum Mechanics allows
us to compute the probability of any given outcome for the measurement of an
observable. However, after the measurement has been performed, the outcome of
the experiment is known with probability 1, and therefore we must conclude that
the state of the system has changed. This is an important aspect of measurement in
Quantum Mechanics: the measurement must change the state of the system. This
is summarised in the fifth postulate of Quantum Mechanics.

Postulate 4 Immediately after a measurement that gave the result Ok , where

Ok is a non-degenerate eigenvalue, the system is in the state |ψk i, the eigenvector
of Ô associated to the eigenvalue Ok . The state vector has been projected onto the
eigenstate by the process of performing the measurement.

We will discuss more precisely what we mean by immediately after when we

discuss the time evolution of quantum systems in the next chapter.
We can express the same concept using the projection operator introduced in
Eq. (??); immediately after a measurement yielding the value Ok the state of the
system is transformed according to:
1
|ψi 7→ q P̂k |ψi , (2.45)
hψ|P̂k |ψi
where we have explicitly normalized the projected vector to have unit norm. This is
xiii Compatible Observables

sometimes referred to as the collapse of the state vector. Clearly this is an idealized
description of a much more complicated process where a ’classical’ instrument and
the quantum system interact. While a fully satisfactory description of quantum
measurement is a subtle issue, the fifth postulate gives a practical recipe, which
underlies the numerous successful predictions of Quantum Mechanics.
After a measurement that yielded the value Ok , the wave function of the system
coincides with the eigenfunction |ψk i. Then, as discussed below Eq. (2.17), if we
perform immediately another measurement of O we will find the same value Ok
with probability 1.
Conversely, if the wave function does not coincide with one of the eigenfunctions,
then the observable O does not have a given value in the state |ψi. We can only
compute the probability for each eigenvalue to be the outcome of the experiment.
Clearly these phenomena do not have a classical analogue. The description of
a physical system in Quantum Mechanics is radically different from the classical
one. You need to practice in order to get familiar with the quantum mechanical
framework.

Degenerate Eigenvalues
In the case of degenerate eigenvalues the state vector before the measurement can
be expanded as
gk
(n)
XX
|ψi = ckn |uk i , (2.46)
k n=1

where the index k runs over all eigenvalues, and for each eigenvalue we have a sum
over n running over the number of degenerate eigenvectors that correspond to the
k-th eigenvalue.
After the measurement the state of the system is still given by Eq. (2.45), but
the projector P̂k needs to project on the subspace of degenerate eigenstates, i.e.
gk
(n) (n)
X
P̂k = |uk ihuk | . (2.47)
n=1

Exercise 2.6.1 Show that the projected (and normalized) state according to
Eq. (2.47) is
gk
1 X (n)
qP ckn |uk i . (2.48)
gk 2
n=1 |ckn | n=1

2.7 Compatible Observables

Suppose A and B are observables and we perform the following sequence of mea-
surements in succession on a single system:
xiv Observables

1. measure A 2. measure B 3. remeasure A

Then if and only if the result of 3 is certain to be the same as the result of 1, we
say that A and B are compatible observables.
In general, this will not be the case: according to our previous discussion about
the collapse of state vector, the measurement of B will project the state of the
system onto an eigenstate of B, and therefore “spoil” the result of 1. Let us analyse
this statement in a little more detail, following the ideas that we have introduced
so far about observables and measurements. Suppose that A and B are represented
by operators  and B̂ respectively, with eigensystems:

 |ui i = Ai |ui i ,
B̂ |vi i = Bi |vi i .

For simplicity we are going to consider the case of non-degenerate eigenvalues –

taking into account degeneracies adds a little complexity for no intellectual gain.
Measurement 1 must return one of the eigenvalues of the operator Â, Aj say, col-
lapsing the system into the state |uj i. Measurement 2 yields an eigenvalue of B̂,
Bk , forcing the system into the state |vk i, so that measurement 3 is made with the
system in the state |vk i. The only way that 3 is certain to yield the result Aj as
obtained in 1 is if |vk i ≡ |uj i. For this to be true in all circumstances it must be
the case that each eigenvector |vk i of B̂ is identical with some eigenvector |uj i of
Â. If there is no degeneracy this implies a one-to-one correspondence between the
eigenvectors of  and the eigenvectors of B̂. We say that  and B̂ have a com-
mon eigenbasis. These properties are summarized in the so-called compatibility
theorem.

Theorem 2.1 Given two observables, A and B, represented by Hermitian oper-

ators  and B̂, then any one of the following three statements implies the other
two:

1. A and B are compatible observables;

2. Â and B̂ have a common eigenbasis;
3. the operators  and B̂ commute: [Â, B̂] = 0

Example Proof. Let us show, for instance, that 3 ⇒ 2. We have

 |ui i = Ai |ui i
B̂ |vi i = Bi |vi i

so that for any eigenvector of Â

ÂB̂ |ui i = B̂ Â |ui i by virtue of 3

= B̂ Ai |ui i
= Ai B̂ |ui i
xv Complete Sets of Commuting Observables

Thus B̂ |ui i is an eigenvector of  belonging to the eigenvalue Ai . If we assume

that the eigenvalues are non-degenerate, then B̂ |ui i must be some multiple of |ui i:

B̂ |ui i = ρ |ui i say (2.49)

This just says that |ui i is an eigenstate of B̂ belonging to the eigenvalue ρ, and we
must have that, for some j,

ρ = Bj and |ui i = |vj i (2.50)

Thus any eigenstate of the set {|ui i} coincides with some member of the set {|vj i}.
The correspondence has to be one-to-one because both sets are orthonormal; if we
assume that two states in one set coincide with a single state in the other set, we
are led to a contradiction that two orthogonal vectors are identical to the same
vector. By simply relabelling all the vectors in one set we can always ensure that

|u1 i = |v1 i, |u2 i = |v2 i, |u3 i = |v3 i, . . . etc , (2.51)

and this is the common eigenbasis. A more general proof, in the case where the
eigenvalues are degenerate is left as an exercise.

2.8 Complete Sets of Commuting Observables

Consider an observable A, and a basis made of eigenstates of Â, {|u1 i, |u2 i, . . .}.
If all the eigenvalues are non-degenerate, each eigenvalue identifies uniquely one
eigenstate. Hence we can label the eigenstates by their eigenvalue; if

Â|un i = an |un i , (2.52)

then we can rename:

|un i ≡ |an i . (2.53)

In this case, the observable A constitutes by itself a complete set of commut-

ing observables (CSCO), i.e. the eigenvalues of  are sufficient to identify the
eigenvectors that form a basis of the space of physical states.
However this is no longer true if some of the eigenvalues are degenerate, since
in this case there are several eigenstates corresponding to the same degenerate
eigenvalue. In order to distinguish these eigenstates, we can use the eigenvalues of
a second observable B, which commutes with A. According to the compatibility
theorem, we can find a basis of common eigenstates of  and B̂. If each pair of
eigenvalues {an , bp } identifies uniquely one vector of the basis, then the set {A, B}
is a CSCO. If this is not the case, then there must be at least one pair {an , bp } for
which there exists more than one eigenvector with these eigenvalues, i.e. there exist
xvi Observables

at least two vectors |w1 i and |w2 i, such that:

Â|w1 i = an |w1 i, B̂|w1 i = bp |w1 i , (2.54)
Â|w2 i = an |w2 i, B̂|w2 i = bp |w2 i . (2.55)
(2.56)
In this case specifying the values of an and bp is not sufficient to identify uniquely
one eigenvector, since any linear combination of |w1 i and |w2 i is also a simultaneous
eigenvector of  and B̂ with the same eigenvalues. Iterating the above procedure,
we add to our set of observables one more quantity C, which commutes with both
A and B, and we choose a basis made of simultaneous eigenvalues of the three
operators Â, B̂, Ĉ. If each eigenstate in the basis is uniquely identified by the set
of eigenvalues {an , bp , cq }, then {A, B, C} is a CSCO. If not, we need to add one
more observable to our set, and so on.
A set of observables A, B, C, . . . is called a CSCO if:
i. all the observables commute by pairs;
ii. specifying the eigenvalues of all the operators in the CSCO identifies a unique
common eigenvector.
Given a CSCO, we can choose a basis for the space of states made of common
eigenvectors of the operators associated to the observables. Each eigenvector is
uniquely identified by the values of the eigenvalues to which it corresponds.

Â|an , bp , cq , . . .i = an |an , bp , cq , . . .i ,
B̂|an , bp , cq , . . .i = bp |an , bp , cq , . . .i ,
Ĉ|an , bp , cq , . . .i = cq |an , bp , cq , . . .i ,
.... (2.57)
Given a CSCO, we can expand any generic wave function in the basis of common
eigenstates labeled by the eigenvalues of the observables:
X
|ψi = ψn,p,q |an , bp , cq i ,
n,p,q
2
The modulus square of the coefficients, |ψn,p,q | , yields the probability of finding
simultaneously the values an , bp , cq if we measure A, B, C in the state |ψi.

2.9 Continuous Spectrum

Until now we have discussed a number of examples where operators have a discrete
spectrum, i.e. where the eigenvalues are numbered by some integer index k. 1 How-
1 In the case of a discrete spectrum, the total number of eigenvalues may well be infinite, however
the eigenvalues are labeled by integer numbers.
xvii Continuous Spectrum

ever there are operators that have a continuous spectrum, like e.g. the energy and
the momentum of an unbound state, or the position operator X̂. In order to deal
with these cases, we need to generalize the formalism that we have introduced so far.
As you will see below, the modifications are minimal, and rather straightforward.

2.9.1 Eigenvalue Equation

Let us denote by fˆ the Hermitian operator associated to an observable with a

continuous spectrum, the eigenvalue equation takes the form:

fˆ|f i = f |f i . (2.58)

Note that in Eq. (2.58) we have denoted the eigenstate simply by |f i. The set
of eigenvectors is a complete set, and we can write the completeness relation by
substituting the sum over the eigenvalues with an integral, viz.
Z
df |f ihf | = 1 . (2.59)

Once again, using the completeness relation, a generic state can be expanded as a
superposition of eigenstates:
Z
|ψi = df ψ(f )|f i ; (2.60)

you should compare this expression with its analogue Eq. (2.30) in the case of
discrete eigenvalues. Here we are implicitly assuming that the integral over vectors
in H can be properly defined and yields another element of the same vector space
H. The coefficient in the expansion Eq. (2.60) are obtained by taking the scalar
product:
ψ(f ) = hf |ψi . (2.61)

Example 2.6 For a quantum system on a line, we can use the completeness re-
lation
Z
dx |xihx| = 1 (2.62)

and rewrite Eq. (2.61) as

Z Z
hf |ψi = dx hf |xihx|ψi = dx f (x)∗ ψ(x) . (2.63)

The probabilistic interpretation of the state vector, i.e. the fourth postulate of
Quantum Mechanics, can be generalized to the case of a continuum spectrum. The
xviii Observables

probability of finding a result between f and f + df when measuring the observable

f is given by:
2
|ψ(f )| df .
Thus we derive the normalization condition:
Z
2
df |ψ(f )| = 1 . (2.64)

The integral in Eq. (2.60) defines a normalizable state if and only if the function
ψ(f ) is square-integrable.

2.9.2 Orthonormality

Following the steps performed to obtain Eq. (2.39), we have:

hf 0 |fˆ|f i = f hf 0 |f i . (2.65)

Taking the complex conjugate of Eq. (2.65), and using the fact that fˆ is Hermitian,
yields:
 ∗
hf |fˆ|f 0 i = hf 0 |fˆ† |f i , (2.66)

= hf 0 |fˆ|f i . (2.67)
Combining the results above:
(f − f 0 ) hf 0 |f i = 0 , (2.68)
and therefore, if f 6= f 0 ,
hf 0 |f i = 0 . (2.69)
Now comes a subtle point. The norm of the state |ψi is given by:
Z
hψ|ψi = df df 0 ψ(f )∗ ψ(f 0 )hf |f 0 i (2.70)
Z Z
2
= df |ψ(f )| df 0 hf |f 0 i , (2.71)

where we used the orthogonality result Eq. (2.69). Now the integral over df 0 in
Eq. (2.71) vanishes for any finite value of hf |f i. Therefore we need to impose that
hf |f i is infinite and normalized so that
Z
df 0 hf 0 |f i = 1 . (2.72)

The delta function introduced by Dirac satisfies precisely this condition, and we
can require:
hf 0 |f i = δ(f − f 0 ) . (2.73)
The eigenstates of the continuum spectrum have infinite norm, and therefore can-
not be considered as genuine physical states. However they provide a useful basis
xix Continuous Spectrum

for expanding normalizable state vectors that correspond to physical states, as in

Eq. (2.60).

2.9.3 Spectral Decomposition

For a generic operator Ô, the eigenvalue spectrum can be made of discrete eigen-
values On and continuous values f . In this case we need to consider both the
eigenfunctions of the discrete spectrum and those of the continuous spectrum to
have a basis to expand quantum states in:
X Z
|ψi = hn|ψi |ni + df hf |ψi |f i . (2.74)
n

Eq. (2.74) summarizes the completeness relation for both discrete and continuous
spectra.
Taking the scalar product with eigenstates of the position eigenstates |xi
X Z
ψ(x) = ψn un (x) + df ψ(f )uf (x) , (2.75)
n

where

ψn = hn|ψi, un (x) = hx|ni , (2.76)

ψ(f ) = hf |ψi, uf (x) = hx|f i . (2.77)

Example 2.7 Gaussian wave functions

For a quantum system on a line, the position operator X̂ has a continuous spectrum.
The eigenvalue equation
X̂|xi = x|xi , (2.78)

has a solution for each value of x. The eigenvectors are normalized to the Dirac
delta:
hx|x0 i = δ (x − x0 ) , (2.79)

and a generic vector can be expanded as

Z Z
|ψi = dx |xihx|ψi = dx ψ(x)|xi . (2.80)

Consider now the state defined by the wave function

" #
2
i (x − x0 )
ψ(x) = hx|ψi = C exp p0 x − . (2.81)
~ 2ξ 2

The state has a finite norm:

√
Z
2 2
dx |ψ(x)| = |C| πξ, (2.82)
xx Observables

which yields the normalization constant:

1
C= √ . (2.83)
π 1/4 ξ

Having properly normalized the state vector, we can compute the mean value of x,
i.e. the average of multiple measurements of the position of the system prepared in
the state |ψi, specified by Eq. (2.81):
Z
2
hxi = dx |ψ(x)| x = x0 . (2.84)

The variance of the same ensemble of measurements is

Z
2 2
h∆x i = h(x − x0 ) i = dx |ψ(x)| (x − x0 ) = ξ 2 .
2 2
(2.85)

In the last two equations, the quantities on the left-hand side are obtained by
preparing the system several times in the same state |ψi, performing measurements
and taking averages. The quantities on the right-hand side, i.e. x0 and ξ are pa-
rameters that define the state of the system. By performing measurements, and
looking at the moments of the observables, we can determine these parameters, i.e.
we learn about the state of the system.

2.10 Momentum Operator

The momentum operator is defined as a differential operator acting on the wave

function:
d
P̂ ψ(x) = hx|P̂ |ψi = −i~ ψ(x) . (2.86)
dx

This can be seen as a realization of de Broglie’s duality hypothesis; according to

wave-particle duality to a particle with momentum p we can associate a wave with
wavelength h/p. A wave with a fixed wavelength is a plane wave, described by the
function:

ψp (x) = C exp [ipx/~] . (2.87)

When we act with the operator P̂ defined in Eq. (2.86), we see that ψp (x) is an
eigenstate of P̂ with eigenvalue p, and therefore we can associate the plane wave to
a state with given momentum p.
Note that there are no restrictions on the possible values of p. The spectrum of
xxi Momentum Operator

P̂ is therefore a continuous spectrum. The normalization of the eigenstates is

Z
hp |pi = dx hp0 |xihx|pi
0

Z  
i
= |C|2 dx exp (p − p0 )x
~
= |C|2 2π~δ(p − p0 ) . (2.88)
√
Following the convention we set in 2.9.2, we can choose C = 1/ 2π~, and hence:
1
ψp (x) = hx|pi = √ exp [ipx/~] . (2.89)
2π~

Example 2.8 We have seen previously that the action of the position operator
X̂ is:
X̂ψ(x) = xψ(x) , (2.90)
i.e. the wave function is simply multiplied by the value of x. Consider the case
d
Ô1 = X̂, Ô2 = dx . Then:
 
d
Ô1 Ô2 ψ(x) = X̂ ψ(x) (2.91)
dx
d
= x ψ(x) , (2.92)
dx
while
 
Ô2 Ô1 ψ(x) = Ô2 X̂ψ(x) (2.93)
= Ô2 (xψ(x)) (2.94)
d
= (xψ(x)) (2.95)
dx
d
= ψ(x) + x ψ(x) . (2.96)
dx
Putting the two results together, we obtain for this particular choice of Ô1 and Ô2 :
h i
Ô1 , Ô2 ψ(x) = −ψ(x) , (2.97)

i.e.
h i
Ô1 , Ô2 = −1 . (2.98)

From the example above we deduce the fundamental canonical commutation

relation:
h i
X̂, P̂ = i~ . (2.99)
xxii Observables

2.10.1 Momentum as Generator of Translations

Consider the wave function of a one-dimensional system, and translate the coordi-
nate x by some amount a. Expanding at first order in a,
d
ψ(x + a) = ψ(x) + a ψ(x) + O(a2 ) . (2.100)
dx
We want to describe the change in the wave function by the action of a Hermitian
operator acting on it. As discussed above the differential operator d/dx is anti-
Hermitian; we can introduce an Hermitian operator by rewriting Eq. (2.100) as

δa ψ(x) = ψ(x + a) − ψ(x)

 
i d
= a −i~ ψ(x) + O(a2 )
~ dx
!
P̂
= ia ψ(x) + O(a2 ) . (2.101)
~

The quantity in the bracket on the right-hand side of Eq. 2.101 is a Hermitian
operator. It can be easily checked that it has dimensions of inverse length, and
actually coincides with the momentum operator that we defined above in order to
implement de Broglie’s idea of duality.
If we only consider infinitesimal transformations and neglect the non-linear de-
pendence on the transformation parameter, we can write

δa ψ(x) = iaT̂ ψ(x) . (2.102)

The operator T̂ is called the generator of translations. The momentum operator

coincides with the generator of translations up to the factor 1/~, which takes care
of dimensions.

2.10.2 Position and Momentum Representations

So far we have adopted a heuristic approach in order to define the position and
momentum operators, based on physical considerations. It is worthwhile to have a
more formal introduction, which follows closely the ideas of vectors and bases that
we have developed in this course.
The state of the quantum system is described by a vector |ψi. This vector can
be expanded in a basis made of eigenstates of the position operator {|xi},
Z
|ψi = dx |xihx|ψi , (2.103)

and the vector is fully specified by specifying its coordinates in that basis:

ψ(x) = hx|ψi . (2.104)

This is called position representation of the state.

xxiii Momentum Operator

The same vector can be expanded in eigenstates of momentum,

Z
|ψi = dp |pihp|ψi . (2.105)

In this case the vector is fully described by the wave function in the so-called
momentum representation;

ψ̃(p) = hp|ψi . (2.106)

Using the completeness relations allows us to write

Z
ψ̃(p) = dxhp|xihx|ψi (2.107)
Z
dx −i px
= √ e ~ ψ(x) , (2.108)
2π~
which shows that the two representations are related by Fourier transforms.
The vector |ψi has finite norm, while the eigenstates of both position and mo-
mentum are states with an infinite norm and do not represent physical states.
Nonetheless they are useful as a basis to expand the state vector and define a pdf
for observing given values of x and p in the state |ψi.

2.10.3 Wave Packets

Physical states, i.e. state vectors with a finite norm, can be constructed as linear
superpositions of momentum eigenstates. These states are called wave packets:
Z
|W i = dp g(p) |pi , g(p) ∈ C . (2.109)

For the wave packet to be normalizable, we require

Z
hW |W i = dp0 dp g(p0 )∗ g(p) hp0 |pi
Z
2
= dp |g(p)| < ∞ . (2.110)

If we choose the function g(p) such that |g(p)|2 is peaked around some value p0 ,
then the probability of measuring a given value for the momentum of the system in
the state |W i has a maximum at p0 and rapidly goes to zero as we move away from
p0 . While the wave packet is not an eigenstate of momentum, we can think of it as a
physical realization of a state with momentum approximately equal to p0 . Making
the distribution more peaked around its maximum yields a more “collimated” beam.
In this section we are going to focus on Gaussian wave packets, normalized to
one, viz.
 
1 1 2
g(p) = exp − (p − p0 ) . (2.111)
(2πσ 2 )
1/4 4σ 2
xxiv Observables

Exercise 2.10.1 Check that the wave packet is properly normalized, hW |W i = 1.

Compute the expectation values
Z
hW |P̂ |W i = dp0 dp g(p0 )∗ g(p) hp0 |P̂ |pi
Z
2
= dp |g(p)| p = p0 ,

hW |P̂ 2 |W i = p20 + σ 2 .
Using the results of the exercise above, we can determine the uncertainty on the
momentum
q
∆p = hW |P̂ 2 |W i − hW |P̂ |W i2
= σ. (2.112)
Starting from Eq. 2.109, we can readily compute
Z
hp|W i = dp0 g(p0 )hp|p0 i = g(p) , (2.113)

which is sometimes referred to as the wave function in momentum space.

A slightly lengthier computation yields the wave function in position space
Z  
1 1 2 1 h px i
hx|W i = dp exp − (p − p 0 ) exp i (2.114)
(2πσ 2 )
1/4 4σ 2 (2π~)
1/2 ~
 2 1/4
σ 2 x2
 
2σ p0 x
= exp − + i . (2.115)
π~2 ~2 ~
Exercise 2.10.2 Compute the integral yielding the wave function in Eq. 2.115.
Check that the modulus square of the wave function is properly normalized
to one, as required by the Parsefal identity.
Using the explicit expression for the wave function in position space yields
hW |X̂|W i = 0 ,
~2
hW |X̂ 2 |W i = , (2.116)
4σ 2
and therefore the uncertainty on the position of the system in the state |W i,
~
∆x = . (2.117)
2σ

2.11 The Uncertainty Principle

In Classical Mechanics the state of a particle in a one-dimensional world is com-

pletely determined by the value of its position x(t) and momentum p(t), i.e. by its
trajectory.
xxv The Uncertainty Principle

The situation is radically different in Quantum Mechanics. The probabilistic in-

terpretation of the wave function implies that we can at best obtain the probability
density for a particle to be at a given position x at time t. As a consequence the
concept of classical trajectory used in Newtonian mechanics does not make sense in
Quantum Mechanics. The position and momentum of the particle can be defined,
but their values cannot be measured simultaneously with arbitrary precision. As
expected, when the scales in the problem are much larger than the Planck constant
h, the classical results are recovered.
These two features are summarized in the so-called uncertainty relations, first
derived by Heisenberg. The uncertainty relations state that the standard deviations
∆x and ∆p of the measured position and momentum in a given quantum state
satisfy
~
∆x · ∆p ≥ . (2.118)
2
The derivation of this inequality is left as an exercise for the reader, see Problem 3.12
below. It is clear from Eq. (2.118) that if the position of the particle is known exactly,
i.e. if the state is an eigenstate of position, then the knowledge of its momentum is
completely lost, in the sense that the standard deviation of the measured momentum
is divergent. In general the product of the two uncertainties has to be greater than
~/2. Note that the Gaussian wave packet discussed in Section 2.10.3 saturates the
bounds exactly.
It is important to appreciate that Heisenberg’s inequalities reflect a physical
limitation. The outcomes of measurements are stochastic variables, the state of
the quantum system determines the expecation value and the standard deviation
of these stochastic variables. A better experimental apparatus would not allow a
higher precision to be obtained. The uncertainty is a quantum-mechanical property
of the system.
The uncertainty principle also encodes the idea that in Quantum Mechanics the
measurement of a quantity interferes with the state of the system. If we measure
exactly the position of the particle, then we lose all knowledge of its momentum
and viceversa. You should contrast this with the situation in Classical Mechanics,
where we can assume that measurements do not perturb the state of the system.

Summary

• The basic concept introduced in this chapter is that observables are the results
of experiments. While this may sound trivial, it is a very improtant state-
ment. The only knowledge we have of physical system are the outcomes of
experiments.[2.1]
• In Quantum Mechanics, observables are associated to Hermitian operators
acting on the space of physical states. Every time you want to consider an
xxvi Observables

observable, make sure that you know which operator is associated to it.
[2.1.2]
• The measurement of an observable can only yield one of the eigenvalues of
the operator associated to the observable, independently of the state of the
system. [2.2]
• Properties of the spectrum of Hermitian operators. These are extremely im-
portant in order to be able to manipulate operators. [2.3]
• The results of an experiment must be considered as a stochastic variable. The
state of the system in which the experiment is performed determines the
probability of the possible outcomes of measurements. [2.5]
• Measurements in a quantum-mechanical system change the state of the sys-
tem. This phenomenon is sometimes called collapse of the wave function.
[2.6]
• If the operators associated to two observbles commute, then the observables
are called compatible. Compatible observables have a common set of eigen-
vectors. [2.7]
• Some operators have a continuous spectrum, and we have introduced the tools
to deal with those. States of the continuum spectrum are not normalizable.
[2.9]
• Definition of the momentum operator, interpretation of the momentum op-
erator as the generator of translations. Construction of normalizable wave
packets. [2.10]
• The uncertainty principle is a mathematical consequence of the canonical
commutation relations between position and momentum. It is a property
of quantum-mechanical systems that does not depend on the quality of the
expeerimental apparatus. [2.11]

Problems

2.1 When acting on the wave function, the position operator, X̂, corresponds
simply to multiplication by x:

X̂ψ(x) = xψ(x) .

Use the definition of Hermitian conjugation to show that X̂ is Hermitian and

hence that the potential energy operator V̂ ≡ V (x) is also Hermitian.
xxvii Problems

2.2 Prove the following relations:

(fˆ† )† = fˆ ,
(fˆĝ)† = ĝ † fˆ† ,
h i h i h i
fˆ, ĝ ĥ = ĝ fˆ, ĥ + fˆ, ĝ ĥ ,
h i h i h i
fˆĝ, ĥ = fˆ ĝ, ĥ + fˆ, ĥ ĝ .
h i
If fˆ, ĝ are Hermitian, show that fˆĝ + ĝ fˆ, and i fˆ, ĝ are also Hermitian.
Show that for any operator Â,

hA† Ai ≥ 0 ,

for any state.

2.3 The observables A and B are represented by operators  and B̂ with eigen-
functions {ui (x)} and {vi (x)} respectively, such that
√
v1 (x) = { 3 u1 (x) + u2 (x)}/2
√
v2 (x) = {u1 (x) − 3 u2 (x)}/2
vn (x) = un (x), n ≥ 3.

Verify that these relations are consistent with orthonormality of both bases.
A certain system is subjected to three successive measurements:
(i) a measurement of A
(ii) a measurement of B
(iii) another measurement of A
Show that if measurement (i) yields any of the values A3 , A4 , . . . then (iii)
gives the same result but that if (i) yields the value A1 there is a probability
of 85 that (iii) will yield A1 and a probability of 38 that it will yield A2 . What
may be said about the compatibility of A and B?
2.4 Consider a two-state system. We denote the two orthonormal states by |1i,
and |2i. In the general case, the Hamiltonian of the system can be written as
a 2 × 2 matrix, where the elements of the matrix are given by:

Hij = hi|Ĥ|ji .

Let us consider the Hamiltonian:

 
E0 −η
Ĥ = , E0 real .
−η E0

1. Write the action of Ĥ on the states |1i and |2i.

2. Show that η has to be real.
3. Compute the eigenvalues of Ĥ, and the normalized eigenvectors.
xxviii Observables

2.5 A two-state quantum system describes a qubit in quantum computing. We

shall assume that we have specified a basis in the space of physical states, and
consider a qubit whose Hamiltonian is described by the matrix:
 
1 0
Ĥ = E0 .
0 −1
Two observables A and B are associated respectively to the Hermitian oper-
ators A and B that are represented by the matrices:
   √ 
0 −i 2 − 2i
 = , B̂ = √ .
i 0 2i 1

1. Find the eigenvalues and eigenvectors for Â, and B̂.

2. Are  and B̂ compatible? Do they commute with the Hamiltonian?
3. Suppose that an observation of  has resulted in A = 1, what would be
the results for B̂, and what would be the respective probabilities?
4. What would be the probability of finding A = 1 if a second measurement
is made immediately after the first one?
5. What is the probability of finding A = 1 if a measurement of A is made
immediately after a measurement of B that yielded the larger eigenvalue
of B?
2.6 Let us consider three operators fˆ, ĝ, and ĥ. Show that:
h h ii h h ii h h ii
fˆ, ĝ, ĥ + ĝ, ĥ, fˆ + ĥ, fˆ, ĝ = 0 .

This result is known as the Jacobi identity.

2.7 Given the wave function
−α2 x2
 π −1/4  
ψ(x) = exp ,
α2 2
p
calculate hxn i and ∆x ≡ hx2 i − hxi2 .
Now calculate the momentum space wave function associated with ψ(x):
Z
dx ipx/~
ψ̃(p) = √ e ψ(x) .
2π~
p
Using ψ̃(p), calculate hpn i and ∆p ≡ hp2 i − hpi2 .
With the above results, what do you find for ∆x∆p?
2.8 Let us consider two Hermitian operators Â, B̂, such that:
h i
Â, B̂ = 0 ,

and let us denote |ψk i the eigenfunctions of B̂:

B̂|ψk i = Bk |ψk i .

1. Show that Â|ψk i is an eigenstate of B̂ with eigenvalue Bk .

xxix Problems

In general there will be a finite number of eigenstates corresponding to the

same eigenvalue Bk . We denote these orthonormal eigenstates by
n o
(n)
|ψk i, n = 1, 2, . . . , gk .

2. Using the result in (a), deduce that

gk
(n) (m)
X
Â|ψk i = Anm |ψk i ,
m=1

where Anm are complex numbers.

3. Show that Anm = A∗mn , i.e. that A is an gk × gk Hermitian matrix.
Any finite-dimensional Hermitian matrix can be diagonalized by a unitary
transformation U :
 
A1 0 ... 0
 0 A2 ... 0 
U † AU =  . .
 
.. .. ..
 .. . . . 
0 0 ... Agk
(n)
We can choose the following set of orthonormal linear combinations of |ψk i,
(n) P † (m)
for a given k: |φk i = m Unm |ψk i.
(n) (n)
4. Show that Â|φk i = An |φk i .
We have therefore found a basis of simultaneous eigenstates of  and B̂:
(n) (n)
Â|φk i = An |φk i ,
(n) (n)
B̂|φk i = Bk |φk i .

By repeating the same argument for all eigenvalues Bk , we can explicitly

construct a basis of simultaneous eigenvalues of  and B̂. The two observables
are called compatible.
2.9 The action of the momentum operator P̂ on the wave function of a one-
dimensional system is given by
d
P̂ ψ(x) = hx|P̂ |ψi = −i~ ψ(x) .
dx
Compute:

P̂ X̂ n ψ(x)
X̂ n P̂ ψ(x) .

Deduce that
h i
P̂ , X̂ n = −i~nX̂ n−1 .
xxx Observables

The operator V̂ = V (X̂) is defined via the Taylor expansion of the function
V
X 1
V̂ = V (X̂) = V (k) (0)X̂ k
k!
k

(k)
where V (0) are the coefficients of the expansion, i.e. they are numbers
computed by evaluating the k-th derivative of the function V at x = 0. Show
that
h i d
P̂ , V̂ = −i~ V (X̂) .
dx
2.10 Denoting by |pi the eigenstates of momentum with eigenvalue p, the momen-
tum space wave function can be defined as:

ψ̃(p) = hp|ψi
Z
= dx hp|xi hx|ψi
Z
= dx e−ipx/~ ψ(x) .

A generic operator Ô acts on ψ(p) according:

Z
Ôψ̃(p) = dxe−ipx/~ Ôψ(x) .

Find the action of the momentum operator P̂ , and the position operator X̂ on
ψ̃(p). Check that this new representation of X̂ and P̂ satisfies the canonical
commutation relation.
2.11 Hellmann-Feynman theorem. Let us consider a system where the potential
depends on an external parameter g, V̂ = V (X̂, g). Show that for the energy
eigenvalues we have:
∂En ∂V
= hψn | |ψn i .
∂g ∂g
Remember that the eigenfunctions also depend on g!
This result, know as the Helmann-Feynman theorem, will be discussed in
detail later in the book, in the chapter about methods beyond perturbation
theory.
2.12 Generalized uncertainty principle
If ∆A and ∆B denote the uncertainties in the observables A and B respec-
tively in the state Ψ(x, t) then the generalised uncertainty relation states that
1
∆A ∆B ≥ 2 ||h[Â, B̂]i|| .

In order to prove this relation consider the operators:

X̂ = Â − hÂi
Ŷ = B̂ − hB̂i .
xxxi Problems

The uncertainties in the observables are given by:

(∆A)2 = hX̂ 2 i ,
(∆B)2 = hŶ 2 i .
For any real number λ, we can construct the state:
|φi = X̂|ψi + iλŶ |ψi .
The norm of the state |φi is positive by definition:
||φ||2 = hφ|φi .
Use this fact to prove the generalized uncertainty relation.