6 The basic formalism of quantum mechanics

6.1 Spaces of functions

We will not be too precise about the space of functions to which physically
relevant wavefunctions belong in any given context. One possible example is the
set of infinitely differentiable complex-valued functions ψ(x) of one real variable x.
This forms a complex vector space which we denote by C ∞ (R). Another is the space
of normalisable wavefunctions, which again forms a complex vector space, L2 (R).
Exercise Check these are indeed vector spaces.

6.2 The inner product

We can define a natural inner product on the space of normalisable wavefunctions

by � ∞
(ψ1 , ψ2 ) = ψ1∗ (x)ψ2 (x)dx . (6.1)
−∞
Exercise: Show that this is well-defined whenever ψ1 and ψ2 are normalisable.
[Hint: look ahead to Thm. 6.6 below.]
We extend the definition of the inner product ( , ) to any ψ1 and ψ2 for which
the integral is well-defined, whether or not ψ1 or ψ2 is normalisable.

6.2.1 Properties of the inner product

1. (ψ1 , ψ2 ) = (ψ2 , ψ1 )∗

2. The inner product is anti-linear in the first entry and linear in the second:

(a1 ψ1 + a2 ψ2 , φ) = a∗1 (ψ1 , φ) + a∗2 (ψ2 , φ) ,

(ψ, a1 φ1 + a2 φ2 ) = a1 (ψ, φ1 ) + a2 (ψ, φ2 ) .

3. The inner product is positive definite on continuous wavefunctions, i.e.

(ψ, ψ) ≥ 0 and (ψ, ψ) = 0 if and only if ψ = 0 .

To see this, note that

� ∞ � ∞
∗
(ψ, ψ) = ψ (x)ψ(x)dx = |ψ(x)|2 dx .
−∞ −∞

52
6.3 Operators

Recall that we defined an operator Ô to be any linear map from a space of functions18
(for example C ∞ (R) or L2 (R)) to itself, i.e. any map such that

Ô(a1 ψ1 + a2 ψ2 ) = a1 Ôψ1 + a2 Ôψ2 (6.2)

for all complex numbers a1 , a2 and all ψ1 , ψ2 ∈ C ∞ (R).

6.3.1 Some examples of operators

�n ∂i
• Finite differential operators i=0 pi (x) ∂x i , where the pi (x) are polynomials

in x. (This class includes p̂ and x̂, defined below.)

• The translation operator

Sa : ψ(x) → ψa (x) = ψ(x − a) .

• The parity operator

P : ψ(x) → ψP (x) = ψ(−x) .

6.4 Hermitian operators

We define the hermitian conjugate A† of an operator A to be the operator such

that
(A† ψ1 , ψ2 ) = (ψ1 , Aψ2 ) (6.3)
for all normalisable wavefunctions ψ1 , ψ2 . An operator is hermitian if A = A† .
Exercise Verify the identities

(a1 A1 + a2 A2 )† = a∗1 A†1 + a∗2 A†2 ,

(AB)† = B † A† .

6.4.1 Classical states and dynamical variables

By the state of a physical system we mean its mathematical representation in
a given theory. In classical mechanics, the states are points in a 2n-dimensional
space, phase space. For example, a system of n point particles in 1D is described by
2n coordinates: (x1 , . . . , xn , p1 , . . . , pn ), where xi is the position of the ith particle
18 *Non-examinable comment: In our discussions below, we generally take the space of
functions to be the space L2 (R) of normalisable physical wavefunctions satisfying Eqn. (3.11).
Strictly speaking, a completely rigorous discussion ought to take account of subtleties which arise
from this choice. For example, the momentum operator is not actually defined on all functions
in L2 (R), but only those that are differentiable. Similarly, the position operator is not defined
on all functions in L2 (R), but only those normalisable functions ψ(x) for which xψ(x) is also
normalisable.
However, a fully rigorous treatment would go well beyond the scope of this course. We shall
follow standard practice in introductory treatments of quantum mechanics and simply assume,
whenever we have an equation in which the position, momentum or other operators act on physical
wavefunctions, that the wavefunctions are chosen such that the action of the operators is well
defined. [You should make the same assumption in tackling problem sheets or exam questions.] *

53
positions and pi its momentum. Classical dynamical variables – for instance the
energy
�n
p2i
E= + V (x1 , . . . , xn ) (6.4)
i=1
2mi
– are defined by functions on phase space. Note that the classical state of a finite
number of particles can be described by a finite number of parameters.
In principle, the classical state and (hence) the value of all classical dynamical
variables can be measured with arbitrary precision.

6.4.2 Quantum states and observables

The possible quantum mechanical states of a physical system at any given time
t are normalised wavefunctions. In particular,
� ∞ the state of a single particle in 1D is
given by a wavefunction ψ(x, t) such that −∞ |ψ(x, t)|2 dx = 1. We need infinitely
many parameters to specify a normalised wavefunction, and thus to specify the
state of a quantum system – even a single particle in 1D.

All quantum dynamical variables or observables – quantities we can measure

– are represented by hermitian operators, and vice versa. Examples of hermitian
operators defining quantum observables are the position, momentum and energy
operators

∂
p̂ = −i� (6.5)
∂x
x̂ = x (i.e. multiplication by x) (6.6)

�2 ∂ 2
Ĥ = − + V (x) (6.7)
2m ∂x2
Another example is the parity operator

P : ψ(x) → ψP (x) = ψ(−x) . (6.8)

Exercise (important!): Verify that each of the operators (6.5-6.8) is

hermitian.

54
6.5 Some theorems about hermitian operators
Theorem 6.1. The eigenvalues of a hermitian operator are real.

Proof. Let A be hermitian and ψ be a normalised eigenfunction with eigenvalue

a: Aψ = aψ. We have

(ψ, Aψ) = (ψ, aψ) = a(ψ, ψ) = a ,

(ψ, Aψ) = (A† ψ, ψ) = (Aψ, ψ)
= (aψ, ψ) = a∗ (ψ, ψ) = a∗ .
Hence a = a∗ . �

Theorem 6.2. Let A be a hermitian operator and ψ1 , ψ2 be normalised eigen-

functions with different eigenvalues a1 , a2 . Then ψ1 and ψ2 are orthogonal.

Proof. We have Aψ1 = a1 ψ1 and Aψ2 = a2 ψ2 , and by theorem 6.1 we have

that a1 , a2 are real.

a1 (ψ1 , ψ2 ) = a∗1 (ψ1 , ψ2 ) = (a1 ψ1 , ψ2 )

= (Aψ1 , ψ2 ) = (A† ψ1 , ψ2 )
= (ψ1 , Aψ2 ) = (ψ1 , a2 ψ2 )
= a2 (ψ1 , ψ2 ) .
Hence (a1 − a2 )(ψ1 , ψ2 ) = 0, and since a1 �= a2 we have (ψ1 , ψ2 ) = 0. �

Our discussion is complicated by the fact that the eigenfunctions of hermitian

operators (i.e. of quantum observables) are not necessarily all normalisable. For
example:
∂
• The momentum operator p̂ = −i� ∂x has eigenfunctions exp( �i px) with eigen-
value p.
• The position operator x̂ has eigenfunctions δ(x − q) with eigenvalue q: we
have x̂δ(x − q) = xδ(x − q) = qδ(x − q).
1 2
• The energy operator Ĥ = 2m p̂ + V (x), for a finite square well potential
V (x), has normalisable eigenfunctions (corresponding to bound states) and
unnormalisable eigenfunctions (corresponding to scattering solutions).

Theorems 6.1 and 6.2 extend to these non-normalisable eigenfunctions of p̂, x̂

and Ĥ: these are the only operators with continuous sets of eigenfunctions that we
consider in this course.
In general, a hermitian operator may have both a set of normalisable eigenfunc-
tions with discrete eigenvalues and a set of non-normalisable eigenfunctions with
continuous eigenvalues, and the theorems apply to both sets.19
19 *Non-examinable technical note: it is well beyond our scope here but, in fact, it turns out

one can find a more general notion of normalisability which covers both sets of eigenfunctions,
and more general versions of the theorems can be precisely framed in terms of this condition.
This definition includes bound states and scattering solutions to the time-independent SE, but
not solutions which blow up exponentially. A discussion can be found in, for example, Messiah,
“Quantum Mechanics”, vol. 1, chap V.9.*

55
Theorem 6.3. The discrete and continuous sets of eigenfunctions of any her-
mitian operator together form a complete orthogonal basis of the physical wavefunc-
tions, i.e. of the normalisable complex-valued functions ψ(x) of one real variable
x.
Proof. This is quite hard to prove in complete generality. We will assume it without
proof in this course. �

Note: We say the hermitian operator A has a degenerate eigenvalue λ if

it has more than one linearly independent eigenfunction with eigenvalue λ. If A
has degenerate eigenvalues, we define an orthogonal basis of its eigenfunctions by
choosing orthogonal bases for the eigenfunctions associated with each degenerate
eigenvalue λ.

Corollary 6.3.1. Let A be a hermitian operator with a discrete set of nor-

malised eigenfunctions {ψi }N
i=1 (we include N = ∞ as a possibility) and a contin-

uous set of eigenfunctions {ψα }α∈Δ , where the indexing set Δ is some sub-interval
of the real numbers. Then any wavefunction ψ can be written as

N
� �
ψ= ai ψ i + aα ψα dα , (6.9)
i=1 Δ

where ai = (ψi , ψ) and aα are complex numbers.

Comment In fact, by appropriately normalising, we can also ensure that
aα = (ψα , ψ).

Corollary 6.3.2. Let A be a hermitian operator with (only) a discrete set of

orthonormalised eigenfunctions {ψi }∞
i=1 . Then any wavefunction ψ can be written

as
∞
�
ψ= (ψi , ψ)ψi . (6.10)
i=1

Proof. Theorem 6.3 implies that

∞
�
ψ= ai ψ i (6.11)
i=1
for some complex numbers ai . Taking the inner product with ψj , and applying
theorem 6.2, we see (ψj , ψ) = aj . �

56
6.6 Quantum measurement postulates
• Postulate 1 Every quantum observable O is represented by a hermitian op-
erator Ô.
• Postulate 2 The possible outcomes of a measurement of O are the eigenvalues
of Ô.

• Postulate 3 If Ô has (only) a discrete set of normalised eigenfunctions

{ψi }∞
i=1 with corresponding distinct eigenvalues {λi }, and a measurement of

O is carried out on a particle with normalised wavefunction

∞
�
ψ= ai ψ i ,
i=1

then the probability of outcome λi is |ai |2 .

• Postulate 4 (the projection postulate) If a measurement of the observable O
is carried out on a particle with normalised wavefunction ψ(x, t) at time t and
the outcome λi is obtained, the wavefunction instantaneously after the mea-
surement becomes ψi (x). [This is sometimes referred to as the “collapse of the
wavefunction”.] The wavefunction then evolves according to the Schrödinger
equation, with initial state ψi (x) at time t, until the next measurement.

Notes

• It follows from postulates 2 and 3 that the total probability of all possible
outcomes is
� � �
|ai |2 = (ai ψi , ai ψi ) = (ai ψi , aj ψj ) = (ψ, ψ) = 1 .
i i ij
(6.12)

So the postulates are consistent: the sum of the probabilities of all possible
outcomes is 1, and so if you carry out a measurement you will certainly get
some outcome and you will only get one outcome. (We already verified this
in the case of the Born rule for position measurements.)

• It follows from postulates 3 and 4 that if an observable O is measured twice,

with infinitesimal time separating the two measurements, then if the first
outcome is λi the second will, with probability one, also be λi .

Exercise: Check this.

This has the important consequence that quantum measurements resemble

classical measurements in at least one sense: they establish a property of the
system that can be repeatedly verified. If we measure something, and then
quickly measure it again, we get the same answer. If this was not true, it
would be hard to find any good reason for the use of the term “measurement”
in quantum mechanics.

57
 • If the wavefunction ψ is an eigenfunction ψi of Ô, the measurement outcome
will be λi with probability one. For example, a stationary state obeying
Ĥψ = Eψ will always give outcome E if the energy is measured.
• But unless the wavefunction ψ is an eigenfunction of the measured observable,
the measurement outcome is not definitely predictable. In contrast to classical
mechanics, a quantum observable does not generally have a definite value on
a quantum state.
• We can extend� postulates 3 and 4 to the case when Ô has degenerate eigen-
values. If ψ = i ai ψi is measured, where {ψi }∞i=1 are orthonormalised eigen-
functions of Ô and {ψi }i∈I are a complete set of orthonormalised �
eigenfunc-
tions with the same eigenvalue λ, the probability of outcome λ is i∈I |ai |2 ,
and the state
� resulting after a measurement with outcome λ is (up to normal-
isation) i∈I ai ψi .

• The projection postulate is so called because it implies that the the post-
measurement wavefunction ψi (x, t) is obtained from the pre-measurement
wavefunction ψ(x, t) by the action of the projection operator Pi defined by
Pi : ψ → (ψi , ψ)ψi , up to normalisation. We call Pi a projection since it maps
any state onto its component in a particular linear subspace, namely the sub-
space spanned by ψi – an action analogous to, for instance, the projection of
a 3D vector (x, y, z) onto its x-component (x, 0, 0).

6.7 Expectation values

Consider a measurement of the observable A on the state ψ. If the correspond-

ing hermitian operator  has only a discrete set of normalisable eigenfunctions
{ψi }, the possible outcomes are the corresponding eigenvalues λi , and the outcome
probabilities are pi = |(ψ, ψi )|2 . The expectation value of the measured outcome, in
the standard statistical sense of the term, is thus
� �
p i λi = |(ψ, ψi )|2 λi
i i
� �
=( (ψ, ψi )ψi , λj (ψ, ψj )ψj )
i j

= (ψ, Âψ) . (6.13)

We write �Â�ψ for the expectation value of a measurement of A on the state ψ.

Thus we have
�Â�ψ = (ψ, Âψ) . (6.14)

We can similarly justify this definition of expectation value for the position op-
erator x̂ from the Born rule. Recall that the probability of obtaining a position
measurement outcome in the interval [x, x + dx] is given by |ψ(x)|2 dx. The expec-
tation value of a position measurement is thus

58
 � ∞ � ∞
2
x|ψ(x)| dx = ψ ∗ (x)xψ(x)dx = (ψ, x̂ψ) .
−∞ ∞

This encourages us to take Eqn. (6.14) as a general definition of expectation

value for any observable, whether its eigenvalues are discrete, continuous or a com-
bination. This general definition too can be similarly justified, given a careful
treatment of the eigenfunctions of general observables and their normalisation: this
is beyond our scope here, but it is important to note that the definition turns out
to agree with the statistical definition of expectation in all cases. In particular, the
definitions

�p̂�ψ = (ψ, p̂ψ) , �Ĥ�ψ = (ψ, Ĥψ) ,

often give simple ways of calculating the statistical expectation of measurements
of momentum and energy for general wavefunctions ψ, since the right hand sides
are often given by simple integrals.

Note that the expectation value is linear with respect to real scalars: i.e.

�a + bB̂�ψ = a�Â�ψ + b�B̂�ψ , (6.15)

for any hermitian operators A, B and any real numbers a, b. We restrict to a, b real
here because the interpretation of �Â�ψ as an expectation value of an observable
requires that A is hermitian, since observables are always represented by hermitian
operators. A complex multiple of a hermitian operator is not generally hermitian:
if A is hermitian then (aA)† = a∗ A† .

6.8 Commutation relations

We define the commutator of two operators by

[A, B] = AB − BA . (6.16)

It is easy to verify the identities

[A, A] = 0 ,
[A, B] = −[B, A] ,
[A, BC] = [A, B]C + B[A, C]
[AB, C] = A[B, C] + [A, C]B . (6.17)

Note that the commutator [A, B] is a sum of products of operators, and thus
itself an operator. Note also that it depends linearly on both entries.

59
 The commutator plays a crucial role in describing symmetries of quantum me-
chanical systems, as we will see when we consider angular momentum. It also
gives a way of calibrating how close two operators are to having simultaneously
determinable eigenvalues: see the following note and the later discussion of the
uncertainty principle.

Definition We say two hermitian operators A and B are simultaneously

diagonalisable if the space of normalisable wavefunctions has a complete basis of
joint eigenfunctions {ψi }, i.e. of eigenfunctions such that Aψi = ai ψi and Bψi =
bi ψi for some real numbers ai , bi .
Theorem 6.4. Two hermitian operators A and B are simultaneously diagonalisable
if and only if [A, B] = 0.

Proof. If A and B are simultaneously diagonalisable, the space of normalisable

wavefunctions has a complete basis of joint eigenfunctions {ψi }. Now for any such
eigenfunction

[A, B]ψi = ABψi − BAψi = (ai bi − bi ai )ψi = 0 . (6.18)

�
If the basis {ψi } is complete, any wavefunction ψ can be written as ψ = i c i ψi ,

and we have that

� �
[A, B]ψ = [A, B] ci ψ i = ci [A, B]ψi = 0 . (6.19)
i i
So we have the operator equation [A, B] = 0, where 0 is the zero operator, which
maps any wavefunction to the zero function.
Conversely, if [A, B] = 0 and Aψi = ai ψi , then 0 = [A, B]ψi = A(Bψi )−ai (Bψi ),
so that Bψi is also an eigenfunction of A with eigenvalue ai . Thus B maps the
eigenspace of A with eigenvalue ai to itself. If we write E for this eigenspace, and
B|E for the operator B restricted to E, then clearly B|E is a hermitian operator on
E. Thus, by theorem 6.3, we can find a basis of E in which B acts diagonally. Since
this is true for all eigenspaces of A, we can find a complete basis of simultaneous
eigenfunctions of A and B. �

6.8.1 The canonical commutation relations

∂
Recall that the position and momentum operators are p̂ = −i� ∂x and x̂ = x
(multiplication by x). We can work out their commutator by considering their
action on a general wavefunction ψ:

∂ψ
x̂p̂ψ = −ix�
∂x
∂
p̂x̂ψ = −i� (xψ)
∂x
∂ψ
= −i�ψ − ix�
∂x
so [x̂, p̂]ψ = x̂p̂ψ − p̂x̂ψ = i�ψ

60
and as this is true for all ψ we have

[x̂, p̂] = i� . (6.20)

It can be shown (though not in this course: see Part II Principles of Quantum
Mechanics) that these commutation relations essentially characterise the position
and momentum operators. That is, any pair of operators satisfying these relations
is equivalent (in a sense that can be made precise) to position and momentum.

6.9 Heisenberg’s Uncertainty Principle

We define the uncertainty Δψ A in a measurement of A on the state ψ by

(Δψ A)2 = �(A − �A�ψ )2 �ψ

= �A2 �ψ − (�A�ψ )2 . (6.21)

Note that Theorem 6.1 implies that the expectation value and the uncertainty
are always real, as we would expect if they are physically meaningful.
Exercise Verify that (Δψ A)2 is the statistical variance of the probability
distribution for the possible outcomes of the measurement of A on ψ, and Δψ A is
the distribution’s standard deviation.
Lemma 6.5. The uncertainty Δψ A ≥ 0 and Δψ A = 0 if and only if ψ is an
eigenfunction of A.

Proof. We can write

(Δψ A)2 = �(A − �A�ψ )2 �ψ

= ((A − �A�ψ )ψ, (A − �A�ψ )ψ) .
Write φ = (A − �A�ψ )ψ. We have (φ, φ) ≥ 0, with equality only if φ = 0, i.e.

Aψ = �A�ψ ψ , (6.22)

which implies that ψ is an eigenfunction of A. Conversely, if ψ is an eigenfunction

of A then (ψ , Aψ) = �A�ψ , and so Aψ = �A�ψ ψ. In other words A has eigenvalue
�A�ψ , and so Δψ A = 0. �
Theorem 6.6 (Schwarz’s inequality). If φ, ψ are any two normalisable wave-
functions, then |(φ, ψ)|2 ≤ (φ, φ)(ψ, ψ). We have equality if and only if φ = aψ for
some complex number a.
(ψ,φ)
Proof. For any a, 0 ≤ (φ − aψ, φ − aψ). If we take a = (ψ,ψ) , we see

|(ψ, φ)|2 |(ψ, φ)|2 |(ψ, φ)|2

0≤ (φ, φ) − − +
(ψ, ψ) (ψ, ψ) (ψ, ψ)
|(ψ, φ)|2
= (φ, φ) − .
(ψ, ψ)

Multiplying through by (ψ, ψ) we obtain the inequality. We have equality only if

φ − aψ = 0, for the value of a specified above. �

61
Theorem 6.7 (the generalised uncertainty relations). If A and B are any two
observables, and ψ is any state, then
1
Δψ AΔψ B ≥ |(ψ, [A, B]ψ)| . (6.23)
2
Proof. We have

(Δψ A)2 = �(A − �A�ψ )2 �ψ

= ((A − �A�ψ )ψ, (A − �A�ψ )ψ)
(Δψ B)2 = ((B − �B�ψ )ψ, (B − �B�ψ )ψ) .

So, writing A� = A − �A�ψ , B � = B − �B�ψ , we have

(Δψ A)2 (Δψ B)2 = (A� ψ, A� ψ)(B � ψ, B � ψ)

≥ |(A� ψ, B � ψ)|2 ( from Thm. 6.6)
(6.24)
� � 2 �
= |(ψ, A B ψ)| (as A is hermitian) .

Now: (1) A� B � = 12 ([A� , B � ] + {A� , B � }), where the anti-commutator {A� , B � } =

A� B � + B � A� ;
(2) [A� , B � ] = [A, B];
(3) (ψ, {A� , B � }ψ) is real, since

(ψ, {A� , B � }ψ) = (({A� , B � })† ψ, ψ) = (({A� , B � })ψ, ψ) = (ψ, {A� , B � }ψ)∗ .
(6.25)
A similar argument shows that (ψ, [A� , B � ]ψ) is imaginary.
So
1
|(ψ, A� B � ψ)|2 = (|(ψ, {A� , B � }ψ)|2 + |(ψ, [A� , B � ]ψ)|2 ) . (6.26)
4
Combining (6.24) and (6.26), we have that
1
(Δψ A)2 (Δψ B)2 ≥ |(ψ, [A, B]ψ)|2 . (6.27)
4
Taking the square root gives (6.23). �

Corollary 6.7.1. (the Heisenberg uncertainty principle for position and momen-
tum).
1
(Δψ x)(Δψ p) ≥ � (6.28)
2
Proof. Taking A = x̂ and B = p̂, we have [A, B] = i�, and the result follows from
Thm. 6.23. �
Thus, the smaller the uncertainty in position, Δψ x, the greater the minimum
possible uncertainty in momentum, Δψ p, and vice versa.

62
Lemma 6.8. If
x̂ψ = iap̂ψ (6.29)
for some real parameter a, then (Δψ x)(Δψ p) = 12 �.

Proof. If x̂ψ = iap̂ψ then we have

(ψ, {x̂, p̂}ψ) = (ψ, x̂p̂ψ) + (ψ, p̂x̂ψ)

= (x̂ψ, p̂ψ) + (p̂ψ, x̂ψ)
= (iap̂ψ, p̂ψ) + (p̂ψ, iap̂ψ)
= (−ia + ia)(p̂ψ, p̂ψ) = 0 ,

which is the condition for the first term on the RHS of (6.26) to vanish. We also
have that �x̂�ψ = ia�p̂�ψ and, since both expectations are real, this implies that
�x̂�ψ = �p̂�ψ = 0. Hence

(x̂ − �x̂�ψ )ψ = ia(p̂ − �p̂�ψ )ψ ,

which means we have equality in the Schwarz’s inequality (6.24) used to derive
(6.28). �
Lemma 6.9. The condition (6.29) holds if and only if ψ(x) = C exp(−bx2 ) for
some constants b, C.
∂
Proof. If x̂ψ = iap̂ψ for some real a, we have that xψ = a� ∂x ψ and so ψ(x) =
2 a�
C exp(−bx ) for some real b = − 2 , and because we have equality in (6.28) we know
the uncertainty is minimised. Conversely, any wavefunction of the form ψ(x) =
C exp(−bx2 ) satisfies x̂ψ = iap̂ψ for some real a. �
Note: For the wavefunction to be normalisable, we require b > 0 and C �= 0.
We can take C = |C| > 0 by multiplying ψ(x) by a phase factor (which does not
alter any physical quantity: the probabilities of outcomes for any measurement are
unaffected).
Exercise: Show that the condition (6.29) for minimum uncertainty is nec-
essary as well as sufficient. Hence show that the normalisable minimum uncertainty
states are precisely the wavefunctions defined by Gaussian functions.

6.9.1 What does the uncertainty principle tell us?

The uncertainty principle is a mathematical statement relating the uncertainties of
x and p (or more generally A and B), which are quantities defined for a given state
�
ψ. We can say, for example, that for any state ψ with Δψ x = δ, we have Δψ p ≥ 2δ .
Heisenberg originally suggested that the uncertainty principle can be understood
simply by observing that a measurement of A creates uncertainty by disturbing the
value of any observable B that does not commute with A. This is not a valid
argument! There are two problems with it.
First, according to quantum mechanics, it is not generally the case that there is a
definite fixed pre-measurement value of either A or B, which is somehow disturbed
in the process of our attempting to measure it. Unless the wavefunction is an
eigenfunction of A or B, the result of any measurement is indeterminate.
Second, the mathematical derivation of the uncertainty principle does not re-
quire us to consider measurements of A or B actually taking place. The quantity
Δψ A is mathematically defined whether or not we choose to carry out a measure-
ment of A. Even if we choose to interpret it as referring to a possible measurement,
we cannot interpret the mathematics as referring to successive measurements of A

63
and B. (Δψ A) and (Δψ B) are the standard deviations for the outcomes of mea-
surements of A and B, but these hypothetical measurements are alternative possible
measurements on the same state ψ, not actual measurements carried out one after
the other. If we measured, say, first A and then B, the first measurement would
collapse the wavefunction onto an eigenfunction of A, and the second measurement
would hence not generally be a measurement on the original state ψ.

6.10 Ehrenfest’s theorem

Theorem 6.10 (Ehrenfest’s theorem). The expectation value �A�ψ of an op-

erator A in the state ψ evolves by
d i ∂A
�A�ψ = �[Ĥ, A]�ψ + � �ψ . (6.30)
dt � ∂t

Proof. We have
�
d�A�ψ d ∞ ∗
= ψ Aψdx
dt dt −∞
� ∞
∂ψ ∗ ∂A ∂ψ
= ( Aψ + ψ ∗ ψ + ψ ∗ A )dx
−∞ ∂t ∂t ∂t
� ∞
∂A i
= � �ψ + ((Ĥψ)∗ Aψ − ψ ∗ A(Ĥψ))dx
∂t � −∞
�
∂A i ∞ ∗
= � �ψ + (ψ ĤAψ − ψ ∗ A(Ĥψ))dx
∂t � −∞
i ∂A
= �[Ĥ, A]�ψ + � �ψ . (6.31)
� ∂t
�

6.10.1 Applications of Ehrenfest’s theorem

p̂2
For Ĥ = 2m + V (x), we have

[Ĥ, p̂] = [V (x), p̂]

∂
= [V (x), −i� ]
∂x
dV
= i� (6.32)
dx
p̂2
[Ĥ, x̂] = [ , x̂]
2m
1 −i�
= 2[p̂, x̂]p̂ = p̂ (6.33)
2m m
[Ĥ, Ĥ] = 0 . (6.34)

64
 ∂ Ĥ
Since none of these operators is explicitly time-dependent, we have that ∂t =
∂ x̂ ∂ p̂
∂t = ∂t = 0 and so the � ∂A
∂t �ψ term on the RHS of (6.30) vanishes in each case,

giving

d dV
�p̂�ψ = −� �ψ ,
dt dx
d 1
�x̂�ψ = �p̂�ψ ,
dt m
d
�Ĥ�ψ = 0 . (6.35)
dt
d 1
These are quantum versions of the classical laws dt x= m p (which follows from
d dV d
p = mv), dt p = − dx (which follows from F = ma), and dt E = 0 (conservation of
total energy).

* Ehrenfest’s theorem and the harmonic oscillator. *

65
6.11 *The harmonic oscillator revisited

By considering commutation relations, we can give a much nicer and more illumi-
nating derivation of the energy spectrum of the harmonic oscillator. This derivation
forms part of the material for the Part II Principles of Quantum Mechanics course.
It is non-examinable material, in the sense that if you are asked to derive the energy
spectrum (without any method being stipulated) then the derivation given earlier is
a perfectly adequate answer. However, the derivation below is simpler and slicker,
and of course it also may be used in this context.
Recall that the harmonic oscillator hamiltonian is
p̂2 1
Ĥ = + mω 2 x̂2
2m 2
1 iω
= (p̂ + imω x̂)(p̂ − imω x̂) + [p̂, x̂]
2m 2
1 �ω
= (p̂ + imω x̂)(p̂ − imω x̂) + (6.36)
2m 2

Define the operator a = √ 1 (p̂ − imω x̂). Since p̂ and x̂ are hermitian, we have
2m
†
a = √ 1 (p̂ + imω x̂), and
2m

1
Ĥ = a† a + �ω . (6.37)
2

We have the following commutation relations:

1
[a, a† ] = (−imω)2[x̂, p̂] = �ω , (6.38)
2m
[Ĥ, a] = [a† a, a] = [a† , a]a = −�ωa , (6.39)
† † † † † †
[Ĥ, a ] = [a a, a ] = a [a, a ] = �ωa . (6.40)

Suppose now that ψ is a harmonic oscillator eigenfunction of energy E:

Ĥψ = Eψ .

We then have

Ĥaψ = [Ĥ, a]ψ + aĤψ = (E − �ω)aψ (6.41)

Ĥa† ψ = [Ĥ, a† ]ψ + a† Ĥψ = (E + �ω)a† ψ , (6.42)

so that aψ and a† ψ are eigenfunctions of energy (E − �ω) and (E + �ω).

We can use this to prove by induction that an ψ and (a† )n ψ are eigenfunctions
of energy (E − n�ω) and (E + n�ω). For example,

Ĥan ψ = Ĥa(an−1 ψ) = (En−1 − �ω)an ψ , (6.43)

where Er is the energy eigenvalue of ar ψ. Since E0 = E, it follows by induction

that En = (E − n�ω).

66
 In particular, if it were true that an ψ �= 0 for all n, there would be eigenfunctions
of arbitrarily low energy, and so there would be no ground state.
However, given any physical wavefunction ψ, we have that
� ∞
−�2 d2 ψ 1
�Ĥ�ψ = ψ∗ ( + mω 2 x2 ψ)dx
−∞ 2m dx2 2
≥ 0,

since both terms are non-negative. (Important note: this argument can obviously
be generalised to show that, if we have any potential V such that V (x) ≥ 0 for all
x, then �Ĥ�ψ ≥ 0 for all states ψ.)

So there cannot be negative energy eigenfunctions. Thus there must be a lowest

energy (i.e. ground state) eigenfunction ψ0 such that
1
0 = aψ0 = √ (p̂ − imω x̂)ψ0 , (6.44)
2m
which implies
dψ0
−i� = imωxψ0 (6.45)
dx
and hence
mωx2
ψ0 (x) = C exp(− ), (6.46)
2�
which is indeed the ground state wavefunction we previously obtained. (This time,
though, we derived it much more simply, by introducing the operators a, a† and
considering their commutation relations.)

Since Ĥ = a† a+ �ω
2 and aψ0 = 0, we have Ĥψ0 =
�ω
2 ψ0 , giving us the previously
�ω
obtained value of 2 for the ground state energy. We have also obtained a closed
form expression (6.46) for the ground state and hence for the excited states,

† n 1 n mωx2
(a ) ψ0 = C( √ (p̂ + imωx̂)) exp(− ), (6.47)
2m 2�
and we see immediately that their energies are (n + 12 )�ω, as previously obtained
by a less direct argument.
We can similarly show that there cannot be eigenfunctions ψ � with energies
taking values other than (n + 12 )�ω). If there were, then am ψ � cannot vanish for
any m, since ψ0 is the unique wavefunction annihilated by a. So there would be
negative energy eigenfunctions, which contradicts the result shown above.
With a little more thought, we can similarly also show that the eigenspaces
must all be non-degenerate: i.e. there is (up to scalar multiplication) only one
eigenfunction of each energy.
The derivation of the harmonic oscillator spectrum in this subsection illustrates
an important general feature: symmetries or regularities in a quantum mechanical
spectrum (such as the regular spacing of the harmonic oscillator energy levels)
suggest the existence of a set of operators whose commutation relations define the
symmetry or explain the regularity (in this case, the operators Ĥ, a and a† ). *

67