Quantum Mechanics
:
A graduate level course
Richard Fitzpatrick
Associate Professor of Physics
The University of Texas at Austin
Contents
1 Introduction 5
1.1 Major sources . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5
2 Fundamental concepts 6
2.1 The breakdown of classical physics . . . . . . . . . . . . . . . . . . 6
2.2 The polarization of photons . . . . . . . . . . . . . . . . . . . . . . 7
2.3 The fundamental principles of quantum mechanics . . . . . . . . . 9
2.4 Ket space . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10
2.5 Bra space . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14
2.6 Operators . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17
2.7 The outer product . . . . . . . . . . . . . . . . . . . . . . . . . . . 19
2.8 Eigenvalues and eigenvectors . . . . . . . . . . . . . . . . . . . . . 20
2.9 Observables . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21
2.10 Measurements . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 24
2.11 Expectation values . . . . . . . . . . . . . . . . . . . . . . . . . . . 25
2.12 Degeneracy . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 26
2.13 Compatible observables . . . . . . . . . . . . . . . . . . . . . . . . 27
2.14 The uncertainty relation . . . . . . . . . . . . . . . . . . . . . . . . 28
2.15 Continuous spectra . . . . . . . . . . . . . . . . . . . . . . . . . . . 31
3 Position and momentum 33
3.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 33
3.2 Poisson brackets . . . . . . . . . . . . . . . . . . . . . . . . . . . . 33
3.3 Wavefunctions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 37
3.4 Schr¨ odinger’s representation  I . . . . . . . . . . . . . . . . . . . . 39
3.5 Schr¨ odinger’s representation  II . . . . . . . . . . . . . . . . . . . 43
3.6 The momentum representation . . . . . . . . . . . . . . . . . . . . 46
3.7 The uncertainty relation . . . . . . . . . . . . . . . . . . . . . . . . 48
3.8 Displacement operators . . . . . . . . . . . . . . . . . . . . . . . . 50
4 Quantum dynamics 55
4.1 Schr¨ odinger’s equations of motion . . . . . . . . . . . . . . . . . . 55
4.2 Heisenberg’s equations of motion . . . . . . . . . . . . . . . . . . . 59
2
4.3 Ehrenfest’s theorem . . . . . . . . . . . . . . . . . . . . . . . . . . 61
4.4 Schr¨ odinger’s waveequation . . . . . . . . . . . . . . . . . . . . . 65
5 Angular momentum 71
5.1 Orbital angular momentum . . . . . . . . . . . . . . . . . . . . . . 71
5.2 Eigenvalues of angular momentum . . . . . . . . . . . . . . . . . . 74
5.3 Rotation operators . . . . . . . . . . . . . . . . . . . . . . . . . . . 78
5.4 Eigenfunctions of orbital angular momentum . . . . . . . . . . . . 81
5.5 Motion in a central ﬁeld . . . . . . . . . . . . . . . . . . . . . . . . 84
5.6 Energy levels of the hydrogen atom . . . . . . . . . . . . . . . . . . 86
5.7 Spin angular momentum . . . . . . . . . . . . . . . . . . . . . . . 89
5.8 Wavefunction of a spin onehalf particle . . . . . . . . . . . . . . . 91
5.9 Rotation operators in spin space . . . . . . . . . . . . . . . . . . . 93
5.10 Magnetic moments . . . . . . . . . . . . . . . . . . . . . . . . . . . 96
5.11 Spin precession . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 97
5.12 Pauli twocomponent formalism . . . . . . . . . . . . . . . . . . . . 99
5.13 Spin greater than onehalf systems . . . . . . . . . . . . . . . . . . 105
5.14 Addition of angular momentum . . . . . . . . . . . . . . . . . . . . 110
6 Approximation methods 120
6.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 120
6.2 The twostate system . . . . . . . . . . . . . . . . . . . . . . . . . . 120
6.3 Nondegenerate perturbation theory . . . . . . . . . . . . . . . . . 122
6.4 The quadratic Stark effect . . . . . . . . . . . . . . . . . . . . . . . 124
6.5 Degenerate perturbation theory . . . . . . . . . . . . . . . . . . . . 129
6.6 The linear Stark effect . . . . . . . . . . . . . . . . . . . . . . . . . 132
6.7 Fine structure . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 134
6.8 The Zeeman effect . . . . . . . . . . . . . . . . . . . . . . . . . . . 140
6.9 Timedependent perturbation theory . . . . . . . . . . . . . . . . . 144
6.10 The twostate system . . . . . . . . . . . . . . . . . . . . . . . . . . 146
6.11 Spin magnetic resonance . . . . . . . . . . . . . . . . . . . . . . . 149
6.12 The Dyson series . . . . . . . . . . . . . . . . . . . . . . . . . . . . 150
6.13 Constant perturbations . . . . . . . . . . . . . . . . . . . . . . . . . 154
6.14 Harmonic perturbations . . . . . . . . . . . . . . . . . . . . . . . . 158
6.15 Absorption and stimulated emission of radiation . . . . . . . . . . 159
3
6.16 The electric dipole approximation . . . . . . . . . . . . . . . . . . . 162
6.17 Energyshifts and decaywidths . . . . . . . . . . . . . . . . . . . . 165
7 Scattering theory 170
7.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 170
7.2 The LipmannSchwinger equation . . . . . . . . . . . . . . . . . . 170
7.3 The Born approximation . . . . . . . . . . . . . . . . . . . . . . . . 175
7.4 Partial waves . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 178
7.5 The optical theorem . . . . . . . . . . . . . . . . . . . . . . . . . . 181
7.6 Determination of phaseshifts . . . . . . . . . . . . . . . . . . . . . 182
7.7 Hard sphere scattering . . . . . . . . . . . . . . . . . . . . . . . . . 184
7.8 Low energy scattering . . . . . . . . . . . . . . . . . . . . . . . . . 186
7.9 Resonances . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 188
4
1 INTRODUCTION
1 Introduction
1.1 Major sources
The textbooks which I have consulted most frequently while developing course
material are:
The principles of quantum mechanics, P.A.M. Dirac, 4th Edition (revised), (Ox
ford University Press, Oxford, UK, 1958).
The Feynman lectures on physics, R.P. Feynman, R.B. Leighton, and M. Sands,
Volume III (AddisonWesley, Reading MA, 1965).
Quantum mechanics, E. Merzbacher, 2nd Edition (John Wiley & Sons, New York
NY, 1970).
Modern quantum mechanics, J.J. Sakurai, (Benjamin/Cummings, Menlo Park
CA, 1985).
5
2 FUNDAMENTAL CONCEPTS
2 Fundamental concepts
2.1 The breakdown of classical physics
The necessity for a departure from classical mechanics is clearly demonstrated
by:
1. The anomalous stability of atoms and molecules: According to classical physics,
an electron orbiting a nucleus should lose energy by emission of synchrotron
radiation, and gradually spiral in towards the nucleus. Experimentally, this
is not observed to happen.
2. The anomalously low speciﬁc heats of atoms and molecules: According to the
equipartition theorem of classical physics, each degree of freedom of an
atomic or molecular system should contribute R/2 to its molar speciﬁc heat,
where R is the ideal gas constant. In fact, only the translational and some
rotational degrees of freedom seem to contribute. The vibrational degrees
of freedom appear to make no contribution at all (except at high temper
atures). Incidentally, this fundamental problem with classical physics was
known and appreciated in the middle of the nineteenth century. Stories that
physicists at the start of the twentieth century thought that classical physics
explained everything, and that there was nothing left to discover, are largely
apocryphal (see Feynman, Vol. I, Cha. 40).
3. The ultraviolet catastrophe: According to classical physics, the energy density
of an electromagnetic ﬁeld in vacuum is inﬁnite due to a divergence of en
ergy carried by short wavelength modes. Experimentally, there is no such
divergence, and the total energy density is ﬁnite.
4. Waveparticle duality: Classical physics can deal with waves or particles. How
ever, various experiments (e.g., light interference, the photoelectric effect,
electron diffraction) show quite clearly that waves sometimes act as if they
were streams of particles, and streams of particles sometimes act as if they
were waves. This is completely inexplicable within the framework of classi
cal physics.
6
2.2 The polarization of photons 2 FUNDAMENTAL CONCEPTS
2.2 The polarization of photons
It is known experimentally that when plane polarized light is used to eject photo
electrons there is a preferred direction of emission of the electrons. Clearly, the
polarization properties of light, which are more usually associated with its wave
like behaviour, also extend to its particlelike behaviour. In particular, a polariza
tion can be ascribed to each individual photon in a beam of light.
Consider the following wellknown experiment. A beam of plane polarized
light is passed through a polaroid ﬁlm, which has the property that it is only
transparent to light whose plane of polarization lies perpendicular to its optic
axis. Classical electromagnetic wave theory tells us that if the beam is polarized
perpendicular to the optic axis then all of the light is transmitted, if the beam is
polarized parallel to the optic axis then none of the light is transmitted, and if the
light is polarized at an angle α to the axis then a fraction sin
2
α of the beam is
transmitted. Let us try to account for these observations at the individual photon
level.
A beam of light which is plane polarized in a certain direction is made up of a
stream of photons which are each plane polarized in that direction. This picture
leads to no difﬁculty if the plane of polarization lies parallel or perpendicular
to the optic axis of the polaroid. In the former case, none of the photons are
transmitted, and, in the latter case, all of the photons are transmitted. But, what
happens in the case of an obliquely polarized incident beam?
The above question is not very precise. Let us reformulate it as a question
relating to the result of some experiment which we could perform. Suppose that
we were to ﬁre a single photon at a polaroid ﬁlm, and then look to see whether
or not it emerges from the other side. The possible results of the experiment are
that either a whole photon, whose energy is equal to the energy of the incident
photon, is observed, or no photon is observed. Any photon which is transmitted
though the ﬁlm must be polarized perpendicular to the optic axis. Furthermore,
it is impossible to imagine (in physics) ﬁnding part of a photon on the other side
of the ﬁlm. If we repeat the experiment a great number of times then, on average,
a fraction sin
2
α of the photons are transmitted through the ﬁlm, and a fraction
7
2.2 The polarization of photons 2 FUNDAMENTAL CONCEPTS
cos
2
α are absorbed. Thus, we conclude that a photon has a probability sin
2
α of
being transmitted as a photon polarized in the plane perpendicular to the optic
axis, and a probability cos
2
α of being absorbed. These values for the probabilities
lead to the correct classical limit for a beamcontaining a large number of photons.
Note that we have only been able to preserve the individuality of photons,
in all cases, by abandoning the determinacy of classical theory, and adopting a
fundamentally probabilistic approach. We have no way of knowing whether an
individual obliquely polarized photon is going to be absorbed by or transmitted
through a polaroid ﬁlm. We only know the probability of each event occurring.
This is a fairly sweeping statement, but recall that the state of a photon is fully
speciﬁed once its energy, direction of propagation, and polarization are known.
If we imagine performing experiments using monochromatic light, normally in
cident on a polaroid ﬁlm, with a particular oblique polarization, then the state of
each individual photon in the beam is completely speciﬁed, and there is nothing
left over to uniquely determine whether the photon is transmitted or absorbed by
the ﬁlm.
The above discussion about the results of an experiment with a single obliquely
polarized photon incident on a polaroid ﬁlm answers all that can be legitimately
asked about what happens to the photon when it reaches the ﬁlm. Questions as
to what decides whether the photon is transmitted or not, or how it changes its
direction of polarization, are illegitimate, since they do not relate to the outcome
of a possible experiment. Nevertheless, some further description is needed in
order to allow the results of this experiment to be correlated with the results of
other experiments which can be performed using photons.
The further description provided by quantum mechanics is as follows. It is
supposed that a photon polarized obliquely to the optic axis can be regarded as
being partly in a state of polarization parallel to the axis, and partly in a state of
polarization perpendicular to the axis. In other words, the oblique polarization
state is some sort of superposition of two states of parallel and perpendicular
polarization. Since there is nothing special about the orientation of the optic
axis in our experiment, we must conclude that any state of polarization can be
regarded as a superposition of two mutually perpendicular states of polarization.
8
2.3 The fundamental principles of quantum mechanics 2 FUNDAMENTAL CONCEPTS
When we make the photon encounter a polaroid ﬁlm, we are subjecting it
to an observation. In fact, we are observing whether it is polarized parallel or
perpendicular to the optic axis. The effect of making this observation is to force
the photon entirely into a state of parallel or perpendicular polarization. In other
words, the photon has to jump suddenly from being partly in each of these two
states to being entirely in one or the other of them. Which of the two states it will
jump into cannot be predicted, but is governed by probability laws. If the photon
jumps into a state of parallel polarization then it is absorbed. Otherwise, it is
transmitted. Note that, in this example, the introduction of indeterminacy into
the problem is clearly connected with the act of observation. In other words, the
indeterminacy is related to the inevitable disturbance of the system associated
with the act of observation.
2.3 The fundamental principles of quantum mechanics
There is nothing special about the transmission and absorption of photons through
a polaroid ﬁlm. Exactly the same conclusions as those outlined above are ob
tained by studying other simple experiments, such as the interference of photons
(see Dirac, Sect. I.3), and the SternGerlach experiment (see Sakurai, Cha. 1;
Feynman, Cha. 5). The study of these simple experiments leads us to formulate
the following fundamental principles of quantum mechanics:
1. Dirac’s razor: Quantum mechanics can only answer questions regarding the
outcome of possible experiments. Any other questions lie beyond the realms
of physics.
2. The principle of superposition of states: Any microscopic system (i.e., an atom,
molecule, or particle) in a given state can be regarded as being partly in
each of two or more other states. In other words, any state can be regarded
as a superposition of two or more other states. Such superpositions can be
performed in an inﬁnite number of different ways.
3. The principle of indeterminacy: An observation made on a microscopic system
causes it to jump into one or more particular states (which are related to
9
2.4 Ket space 2 FUNDAMENTAL CONCEPTS
the type of observation). It is impossible to predict into which ﬁnal state
a particular system will jump, however the probability of a given system
jumping into a given ﬁnal state can be predicted.
The ﬁrst of these principles was formulated by quantum physicists (such as Dirac)
in the 1920s to fend off awkward questions such as “How can a system suddenly
jump from one state into another?”, or “How does a system decide which state to
jump into?”. As we shall see, the second principle is the basis for the mathemat
ical formulation of quantum mechanics. The ﬁnal principle is still rather vague.
We need to extend it so that we can predict which possible states a system can
jump into after a particular type of observation, as well as the probability of the
system making a particular jump.
2.4 Ket space
Consider a microscopic system composed of particles or bodies with speciﬁc prop
erties (mass, moment of inertia, etc.) interacting according to speciﬁc laws of
force. There will be various possible motions of the particles or bodies consistent
with the laws of force. Let us term each such motion a state of the system. Accord
ing to the principle of superposition of states, any given state can be regarded as
a superposition of two or more other states. Thus, states must be related to math
ematical quantities of a kind which can be added together to give other quantities
of the same kind. The most obvious examples of such quantities are vectors.
Let us consider a particular microscopic system in a particular state, which we
label A: e.g., a photon with a particular energy, momentum, and polarization.
We can represent this state as a particular vector, which we also label A, residing
in some vector space, where the other elements of the space represent all of the
other possible states of the system. Such a space is called a ket space (after Dirac).
The state vector A is conventionally written
A). (2.1)
Suppose that state A is, in fact, the superposition of two different states, B and
10
2.4 Ket space 2 FUNDAMENTAL CONCEPTS
C. This interrelation is represented in ket space by writing
A) = B) + C), (2.2)
where B) is the vector relating to the state B, etc. For instance, state B) might
represent a photon propagating in the zdirection, and plane polarized in the x
direction, and state C) might represent a similar photon plane polarized in the
ydirection. In this case, the sum of these two states represents a photon whose
plane of polarization makes an angle of 45
◦
with both the x and ydirections (by
analogy with classical physics). This latter state is represented by B) + C) in ket
space.
Suppose that we want to construct a state whose plane of polarization makes
an arbitrary angle α with the xdirection. We can do this via a suitably weighted
superposition of states B and C. By analogy with classical physics, we require
cos α of state B, and sinα of state C. This new state is represented by
cos αB) + sinαC) (2.3)
in ket space. Note that we cannot form a new state by superposing a state with
itself. For instance, a photon polarized in the ydirection superposed with another
photon polarized in the ydirection (with the same energy and momentum) gives
the same photon. This implies that the ket vector
c
1
A) +c
2
A) = (c
1
+c
2
)A) (2.4)
corresponds to the same state that A) does. Thus, ket vectors differ from con
ventional vectors in that their magnitudes, or lengths, are physically irrelevant.
All the states of the system are in one to one correspondence with all the possi
ble directions of vectors in the ket space, no distinction being made between the
directions of the ket vectors A) and −A). There is, however, one caveat to the
above statements. If c
1
+ c
2
= 0 then the superposition process yields nothing at
all: i.e., no state. The absence of a state is represented by the null vector 0) in
ket space. The null vector has the fairly obvious property that
A) + 0) = A), (2.5)
for any vector A). The fact that ket vectors pointing in the same direction repre
sent the same state relates ultimately to the quantization of matter: i.e., the fact
11
2.4 Ket space 2 FUNDAMENTAL CONCEPTS
that it comes in irreducible packets called photons, electrons, atoms, etc. If we ob
serve a microscopic system then we either see a state (i.e., a photon, or an atom,
or a molecule, etc.) or we see nothing—we can never see a fraction or a multiple
of a state. In classical physics, if we observe a wave then the amplitude of the
wave can take any value between zero and inﬁnity. Thus, if we were to represent
a classical wave by a vector, then the magnitude, or length, of the vector would
correspond to the amplitude of the wave, and the direction would correspond to
the frequency and wavelength, so that two vectors of different lengths pointing
in the same direction would represent different wave states.
We have seen, in Eq. (2.3), that any plane polarized state of a photon can
be represented as a linear superposition of two orthogonal polarization states
in which the weights are real numbers. Suppose that we want to construct a
circularly polarized photon state. Well, we know from classical physics that a cir
cularly polarized wave is a superposition of two waves of equal amplitude, plane
polarized in orthogonal directions, which are in phase quadrature. This suggests
that a circularly polarized photon is the superposition of a photon polarized in
the xdirection (state B) and a photon polarized in the ydirection (state C), with
equal weights given to the two states, but with the proviso that state C is 90
◦
out of phase with state B. By analogy with classical physics, we can use complex
numbers to simultaneously represent the weighting and relative phase in a linear
superposition. Thus, a circularly polarized photon is represented by
B) + i C) (2.6)
in ket space. A general elliptically polarized photon is represented by
c
1
B) +c
2
C), (2.7)
where c
1
and c
2
are complex numbers. We conclude that a ket space must be
a complex vector space if it is to properly represent the mutual interrelations
between the possible states of a microscopic system.
Suppose that the ket R) is expressible linearly in terms of the kets A) and B),
so that
R) = c
1
A) +c
2
B). (2.8)
12
2.4 Ket space 2 FUNDAMENTAL CONCEPTS
We say that R) is dependent on A) and B). It follows that the state R can be
regarded as a linear superposition of the states A and B. So, we can also say that
state R is dependent on states A and B. In fact, any ket vector (or state) which
is expressible linearly in terms of certain others is said to be dependent on them.
Likewise, a set of ket vectors (or states) are termed independent if none of them
are expressible linearly in terms of the others.
The dimensionality of a conventional vector space is deﬁned as the number
of independent vectors contained in the space. Likewise, the dimensionality of
a ket space is equivalent to the number of independent ket vectors it contains.
Thus, the ket space which represents the possible polarization states of a photon
propagating in the zdirection is twodimensional (the two independent vectors
correspond to photons plane polarized in the x and ydirections, respectively).
Some microscopic systems have a ﬁnite number of independent states (e.g., the
spin states of an electron in a magnetic ﬁeld). If there are N independent states,
then the possible states of the system are represented as an Ndimensional ket
space. Some microscopic systems have a denumerably inﬁnite number of inde
pendent states (e.g., a particle in an inﬁnitely deep, onedimensional potential
well). The possible states of such a system are represented as a ket space whose
dimensions are denumerably inﬁnite. Such a space can be treated in more or less
the same manner as a ﬁnitedimensional space. Unfortunately, some microscopic
systems have a nondenumerably inﬁnite number of independent states (e.g., a
free particle). The possible states of such a system are represented as a ket space
whose dimensions are nondenumerably inﬁnite. This type of space requires a
slightly different treatment to spaces of ﬁnite, or denumerably inﬁnite, dimen
sions.
In conclusion, the states of a general microscopic system can be represented as
a complex vector space of (possibly) inﬁnite dimensions. Such a space is termed
a Hilbert space by mathematicians.
13
2.5 Bra space 2 FUNDAMENTAL CONCEPTS
2.5 Bra space
A snack machine inputs coins plus some code entered on a key pad, and (hope
fully) outputs a snack. It also does so in a deterministic manner: i.e., the same
money plus the same code produces the same snack (or the same error message)
time after time. Note that the input and output of the machine have completely
different natures. We can imagine building a rather abstract snack machine which
inputs ket vectors and outputs complex numbers in a deterministic fashion. Math
ematicians call such a machine a functional. Imagine a general functional, labeled
F, acting on a general ket vector, labeled A, and spitting out a general complex
number φ
A
. This process is represented mathematically by writing
¸F(A)) = φ
A
. (2.9)
Let us narrow our focus to those functionals which preserve the linear dependen
cies of the ket vectors upon which they operate. Not surprisingly, such functionals
are termed linear functionals. A general linear functional, labeled F, satisﬁes
¸F(A) + B)) = ¸F(A)) +¸F(B)), (2.10)
where A) and B) are any two kets in a given ket space.
Consider an Ndimensional ket space [i.e., a ﬁnitedimensional, or denumer
ably inﬁnite dimensional (i.e., N → ∞), space]. Let the i) (where i runs from 1
to N) represent N independent ket vectors in this space. A general ket vector can
be written
1
A) =
N
i=1
α
i
i), (2.11)
where the α
i
are an arbitrary set of complex numbers. The only way the func
tional F can satisfy Eq. (2.10) for all vectors in the ket space is if
¸F(A)) =
N
i=1
f
i
α
i
, (2.12)
1
Actually, this is only strictly true for ﬁnitedimensional spaces. Only a special subset of denumerably inﬁnite
dimensional spaces have this property (i.e., they are complete), but since a ket space must be complete if it is to
represent the states of a microscopic system, we need only consider this special subset.
14
2.5 Bra space 2 FUNDAMENTAL CONCEPTS
where the f
i
are a set of complex numbers relating to the functional.
Let us deﬁne N basis functionals ¸i which satisfy
¸i(j)) = δ
ij
. (2.13)
It follows from the previous three equations that
¸F =
N
i=1
f
i
¸i. (2.14)
But, this implies that the set of all possible linear functionals acting on an N
dimensional ket space is itself an Ndimensional vector space. This type of vector
space is called a bra space (after Dirac), and its constituent vectors (which are
actually functionals of the ket space) are called bra vectors. Note that bra vectors
are quite different in nature to ket vectors (hence, these vectors are written in
mirror image notation, ¸  and  ), so that they can never be confused). Bra
space is an example of what mathematicians call a dual vector space (i.e., it is
dual to the original ket space). There is a one to one correspondence between
the elements of the ket space and those of the related bra space. So, for every
element A of the ket space, there is a corresponding element, which it is also
convenient to label A, in the bra space. That is,
A)
DC
←→¸A, (2.15)
where DC stands for dual correspondence.
There are an inﬁnite number of ways of setting up the correspondence between
vectors in a ket space and those in the related bra space. However, only one
of these has any physical signiﬁcance. For a general ket vector A, speciﬁed by
Eq. (2.11), the corresponding bra vector is written
¸A =
N
i=1
α
∗
i
¸i, (2.16)
where the α
∗
i
are the complex conjugates of the α
i
. ¸A is termed the dual vector
to A). It follows, from the above, that the dual to c¸A is c
∗
A), where c is a
complex number. More generally,
c
1
A) +c
2
B)
DC
←→c
∗
1
¸A +c
∗
2
¸B. (2.17)
15
2.5 Bra space 2 FUNDAMENTAL CONCEPTS
Recall that a bra vector is a functional which acts on a general ket vector, and
spits out a complex number. Consider the functional which is dual to the ket
vector
B) =
N
i=1
β
i
i) (2.18)
acting on the ket vector A). This operation is denoted ¸B(A)). Note, however,
that we can omit the round brackets without causing any ambiguity, so the oper
ation can also be written ¸BA). This expression can be further simpliﬁed to give
¸BA). According to Eqs. (2.11), (2.12), (2.16), and (2.18),
¸BA) =
N
i=1
β
∗
i
α
i
. (2.19)
Mathematicians term ¸BA) the inner product of a bra and a ket.
2
An inner prod
uct is (almost) analogous to a scalar product between a covariant and contravari
ant vector in some curvilinear space. It is easily demonstrated that
¸BA) = ¸AB)
∗
. (2.20)
Consider the special case where B) →A). It follows from Eqs. (2.12) and (2.20)
that ¸AA) is a real number, and that
¸AA) ≥ 0. (2.21)
The equality sign only holds if A) is the null ket [i.e., if all of the α
i
are zero in
Eq. (2.11)]. This property of bra and ket vectors is essential for the probabilistic
interpretation of quantum mechanics, as will become apparent later.
Two kets A) and B) are said to be orthogonal if
¸AB) = 0, (2.22)
which also implies that ¸BA) = 0.
Given a ket A) which is not the null ket, we can deﬁne a normalized ket 
˜
A),
where

˜
A) =
_
_
1
_
¸AA)
_
_
A), (2.23)
2
We can now appreciate the elegance of Dirac’s notation. The combination of a bra and a ket yields a “bra(c)ket”
(which is just a number).
16
2.6 Operators 2 FUNDAMENTAL CONCEPTS
with the property
¸
˜
A
˜
A) = 1. (2.24)
Here,
_
¸AA) is known as the norm or “length” of A), and is analogous to the
length, or magnitude, of a conventional vector. Since A) and cA) represent
the same physical state, it makes sense to require that all kets corresponding to
physical states have unit norms.
It is possible to deﬁne a dual bra space for a ket space of nondenumerably
inﬁnite dimensions in much the same manner as that described above. The main
differences are that summations over discrete labels become integrations over
continuous labels, Kronecker deltafunctions become Dirac deltafunctions, com
pleteness must be assumed (it cannot be proved), and the normalization conven
tion is somewhat different. More of this later.
2.6 Operators
We have seen that a functional is a machine which inputs a ket vector and spits
out a complex number. Consider a somewhat different machine which inputs a
ket vector and spits out another ket vector in a deterministic fashion. Mathemati
cians call such a machine an operator. We are only interested in operators which
preserve the linear dependencies of the ket vectors upon which they act. Such
operators are termed linear operators. Consider an operator labeled X. Suppose
that when this operator acts on a general ket vector A) it spits out a new ket
vector which is denoted XA). Operator X is linear provided that
X(A) + B)) = XA) +XB), (2.25)
for all ket vectors A) and B), and
X(cA)) = cXA), (2.26)
for all complex numbers c. Operators X and Y are said to be equal if
XA) = YA) (2.27)
17
2.6 Operators 2 FUNDAMENTAL CONCEPTS
for all kets in the ket space in question. Operator X is termed the null operator if
XA) = 0) (2.28)
for all ket vectors in the space. Operators can be added together. Such addition
is deﬁned to obey a commutative and associate algebra:
X +Y = Y +X, (2.29)
X + (Y +Z) = (X +Y) +Z. (2.30)
Operators can also be multiplied. The multiplication is associative:
X(YA)) = (XY)A) = XYA), (2.31)
X(Y Z) = (XY)Z = XY Z. (2.32)
However, in general, it is noncommutative:
XY ,= Y X. (2.33)
So far, we have only considered linear operators acting on ket vectors. We can
also give a meaning to their operating on bra vectors. Consider the inner product
of a general bra ¸B with the ket XA). This product is a number which depends
linearly on A). Thus, it may be considered to be the inner product of A) with
some bra. This bra depends linearly on ¸B, so we may look on it as the result of
some linear operator applied to ¸B. This operator is uniquely determined by the
original operator X, so we might as well call it the same operator acting on B). A
suitable notation to use for the resulting bra when X operates on ¸B is ¸BX. The
equation which deﬁnes this vector is
(¸BX)A) = ¸B(XA)) (2.34)
for any A) and ¸B. The triple product of ¸B, X, and A) can be written ¸BXA)
without ambiguity, provided we adopt the convention that the bra vector always
goes on the left, the operator in the middle, and the ket vector on the right.
Consider the dual bra to XA). This bra depends antilinearly on A) and must
therefore depend linearly on ¸A. Thus, it may be regarded as the result of some
18
2.7 The outer product 2 FUNDAMENTAL CONCEPTS
linear operator applied to ¸A. This operator is termed the adjoint of X, and is
denoted X
†
. Thus,
XA)
DC
←→¸AX
†
. (2.35)
It is readily demonstrated that
¸BX
†
A) = ¸AXB)
∗
, (2.36)
plus
(XY)
†
= Y
†
X
†
. (2.37)
It is also easily seen that the adjoint of the adjoint of a linear operator is equiva
lent to the original operator. A Hermitian operator ξ has the special property that
it is its own adjoint: i.e.,
ξ = ξ
†
. (2.38)
2.7 The outer product
So far we have formed the following products: ¸BA), XA), ¸AX, XY, ¸BXA).
Are there any other products we are allowed to form? How about
B)¸A ? (2.39)
This clearly depends linearly on the ket A) and the bra B). Suppose that we
rightmultiply the above product by the general ket C). We obtain
B)¸AC) = ¸AC)B), (2.40)
since ¸AC) is just a number. Thus, B)¸A acting on a general ket C) yields
another ket. Clearly, the product B)¸A is a linear operator. This operator also
acts on bras, as is easily demonstrated by leftmultiplying the expression (2.39)
by a general bra ¸C. It is also easily demonstrated that
(B)¸A)
†
= A)¸B. (2.41)
Mathematicians term the operator B)¸A the outer product of B) and ¸A. The
outer product should not be confused with the inner product, ¸AB), which is just
a number.
19
2.8 Eigenvalues and eigenvectors 2 FUNDAMENTAL CONCEPTS
2.8 Eigenvalues and eigenvectors
In general, the ket XA) is not a constant multiple of A). However, there are
some special kets known as the eigenkets of operator X. These are denoted
x
), x
), x
) . . . , (2.42)
and have the property
Xx
) = x
x
), Xx
) = x
x
) . . . , (2.43)
where x
, x
, . . . are numbers called eigenvalues. Clearly, applying X to one of its
eigenkets yields the same eigenket multiplied by the associated eigenvalue.
Consider the eigenkets and eigenvalues of a Hermitian operator ξ. These are
denoted
ξξ
) = ξ
ξ
), (2.44)
where ξ
) is the eigenket associated with the eigenvalue ξ
. Three important
results are readily deduced:
(i) The eigenvalues are all real numbers, and the eigenkets corresponding to
different eigenvalues are orthogonal. Since ξ is Hermitian, the dual equation to
Eq. (2.44) (for the eigenvalue ξ
) reads
¸ξ
ξ = ξ
∗
¸ξ
. (2.45)
If we leftmultiply Eq. (2.44) by ¸ξ
, rightmultiply the above equation by ξ
),
and take the difference, we obtain
(ξ
−ξ
∗
)¸ξ
ξ
) = 0. (2.46)
Suppose that the eigenvalues ξ
and ξ
are the same. It follows from the above
that
ξ
= ξ
∗
, (2.47)
where we have used the fact that ξ
) is not the null ket. This proves that the
eigenvalues are real numbers. Suppose that the eigenvalues ξ
and ξ
are differ
ent. It follows that
¸ξ
ξ
) = 0, (2.48)
20
2.9 Observables 2 FUNDAMENTAL CONCEPTS
which demonstrates that eigenkets corresponding to different eigenvalues are
orthogonal.
(ii) The eigenvalues associated with eigenkets are the same as the eigenvalues
associated with eigenbras. An eigenbra of ξ corresponding to an eigenvalue ξ
is
deﬁned
¸ξ
ξ = ¸ξ
ξ
. (2.49)
(iii) The dual of any eigenket is an eigenbra belonging to the same eigenvalue,
and conversely.
2.9 Observables
We have developed a mathematical formalism which comprises three types of
objects—bras, kets, and linear operators. We have already seen that kets can be
used to represent the possible states of a microscopic system. However, there is
a one to one correspondence between the elements of a ket space and its dual
bra space, so we must conclude that bras could just as well be used to repre
sent the states of a microscopic system. What about the dynamical variables of
the system (e.g., its position, momentum, energy, spin, etc.)? How can these be
represented in our formalism? Well, the only objects we have left over are oper
ators. We, therefore, assume that the dynamical variables of a microscopic system
are represented as linear operators acting on the bras and kets which correspond to
the various possible states of the system. Note that the operators have to be linear,
otherwise they would, in general, spit out bras/kets pointing in different direc
tions when fed bras/kets pointing in the same direction but differing in length.
Since the lengths of bras and kets have no physical signiﬁcance, it is reasonable
to suppose that nonlinear operators are also without physical signiﬁcance.
We have seen that if we observe the polarization state of a photon, by placing
a polaroid ﬁlm in its path, the result is to cause the photon to jump into a state
of polarization parallel or perpendicular to the optic axis of the ﬁlm. The former
state is absorbed, and the latter state is transmitted (which is how we tell them
apart). In general, we cannot predict into which state a given photon will jump
21
2.9 Observables 2 FUNDAMENTAL CONCEPTS
(except in a statistical sense). However, we do know that if the photon is initially
polarized parallel to the optic axis then it will deﬁnitely be absorbed, and if it is
initially polarized perpendicular to the axis then it will deﬁnitely be transmitted.
We also known that after passing though the ﬁlm a photon must be in a state of
polarization perpendicular to the optic axis (otherwise it would not have been
transmitted). We can make a second observation of the polarization state of
such a photon by placing an identical polaroid ﬁlm (with the same orientation of
the optic axis) immediately behind the ﬁrst ﬁlm. It is clear that the photon will
deﬁnitely be transmitted through the second ﬁlm.
There is nothing special about the polarization states of a photon. So, more
generally, we can say that when a dynamical variable of a microscopic system
is measured the system is caused to jump into one of a number of independent
states (note that the perpendicular and parallel polarization states of our photon
are linearly independent). In general, each of these ﬁnal states is associated with
a different result of the measurement: i.e., a different value of the dynamical
variable. Note that the result of the measurement must be a real number (there
are no measurement machines which output complex numbers). Finally, if an
observation is made, and the system is found to be a one particular ﬁnal state,
with one particular value for the dynamical variable, then a second observation,
made immediately after the ﬁrst one, will deﬁnitely ﬁnd the system in the same
state, and yield the same value for the dynamical variable.
How can we represent all of these facts in our mathematical formalism? Well,
by a fairly nonobvious leap of intuition, we are going to assert that a measure
ment of a dynamical variable corresponding to an operator X in ket space causes
the system to jump into a state corresponding to one of the eigenkets of X. Not
surprisingly, such a state is termed an eigenstate. Furthermore, the result of the
measurement is the eigenvalue associated with the eigenket into which the system
jumps. The fact that the result of the measurement must be a real number implies
that dynamical variables can only be represented by Hermitian operators (since only
Hermitian operators are guaranteed to have real eigenvalues). The fact that the
eigenkets of a Hermitian operator corresponding to different eigenvalues (i.e., dif
ferent results of the measurement) are orthogonal is in accordance with our ear
lier requirement that the states into which the system jumps should be mutually
22
2.9 Observables 2 FUNDAMENTAL CONCEPTS
independent. We can conclude that the result of a measurement of a dynamical
variable represented by a Hermitian operator ξ must be one of the eigenvalues of
ξ. Conversely, every eigenvalue of ξ is a possible result of a measurement made
on the corresponding dynamical variable. This gives us the physical signiﬁcance
of the eigenvalues. (From now on, the distinction between a state and its rep
resentative ket vector, and a dynamical variable and its representative operator,
will be dropped, for the sake of simplicity.)
It is reasonable to suppose that if a certain dynamical variable ξ is measured
with the system in a particular state, then the states into which the system may
jump on account of the measurement are such that the original state is dependent
on them. This fairly innocuous statement has two very important corollaries.
First, immediately after an observation whose result is a particular eigenvalue ξ
,
the system is left in the associated eigenstate. However, this eigenstate is orthog
onal to (i.e., independent of) any other eigenstate corresponding to a different
eigenvalue. It follows that a second measurement made immediately after the
ﬁrst one must leave the system in an eigenstate corresponding to the eigenvalue
ξ
. In other words, the second measurement is bound to give the same result as
the ﬁrst. Furthermore, if the system is in an eigenstate of ξ, corresponding to an
eigenvalue ξ
, then a measurement of ξ is bound to give the result ξ
. This follows
because the system cannot jump into an eigenstate corresponding to a different
eigenvalue of ξ, since such a state is not dependent on the original state. Second,
it stands to reason that a measurement of ξ must always yield some result. It fol
lows that no matter what the initial state of the system, it must always be able to
jump into one of the eigenstates of ξ. In other words, a general ket must always
be dependent on the eigenkets of ξ. This can only be the case if the eigenkets
form a complete set (i.e., they span ket space). Thus, in order for a Hermitian oper
ator ξ to be observable its eigenkets must form a complete set. A Hermitian operator
which satisﬁes this condition is termed an observable. Conversely, any observable
quantity must be a Hermitian operator with a complete set of eigenstates.
23
2.10 Measurements 2 FUNDAMENTAL CONCEPTS
2.10 Measurements
We have seen that a measurement of some observable ξ of a microscopic system
causes the system to jump into one of the eigenstates of ξ. The result of the
measurement is the associated eigenvalue (or some function of this quantity). It
is impossible to determine into which eigenstate a given system will jump, but it is
possible to predict the probability of such a transition. So, what is the probability
that a system in some initial state A) makes a transition to an eigenstate ξ
) of an
observable ξ, as a result of a measurement made on the system? Let us start with
the simplest case. If the system is initially in an eigenstate ξ
) then the transition
probability to a eigenstate ξ
) corresponding to a different eigenvalue is zero,
and the transition probability to the same eigenstate ξ
) is unity. It is convenient
to normalize our eigenkets such that they all have unit norms. It follows from the
orthogonality property of the eigenkets that
¸ξ
ξ
) = δ
ξ
ξ
, (2.50)
where δ
ξ
ξ
is unity if ξ
= ξ
, and zero otherwise. For the moment, we are
assuming that the eigenvalues of ξ are all different.
Note that the probability of a transition from an initial eigenstate ξ
) to a ﬁ
nal eigenstate ξ
) is the same as the value of the inner product ¸ξ
ξ
). Can we
use this correspondence to obtain a general rule for calculating transition prob
abilities? Well, suppose that the system is initially in a state A) which is not an
eigenstate of ξ. Can we identify the transition probability to a ﬁnal eigenstate
ξ
) with the inner product ¸Aξ
)? The straight answer is “no”, since ¸Aξ
) is, in
general, a complex number, and complex probabilities do not make much sense.
Let us try again. How about if we identify the transition probability with the mod
ulus squared of the inner product, ¸Aξ
)
2
? This quantity is deﬁnitely a positive
number (so it could be a probability). This guess also gives the right answer for
the transition probabilities between eigenstates. In fact, it is the correct guess.
Since the eigenstates of an observable ξ form a complete set, we can express
any given state A) as a linear combination of them. It is easily demonstrated that
A) =
ξ
ξ
)¸ξ
A), (2.51)
24
2.11 Expectation values 2 FUNDAMENTAL CONCEPTS
¸A =
ξ
¸Aξ
)¸ξ
, (2.52)
¸AA) =
ξ
¸Aξ
)¸ξ
A) =
ξ
¸Aξ
)
2
, (2.53)
where the summation is over all the different eigenvalues of ξ, and use has been
made of Eq. (2.20), and the fact that the eigenstates are mutually orthogonal.
Note that all of the above results follow from the extremely useful (and easily
proved) result
ξ
ξ
)¸ξ
 = 1, (2.54)
where 1 denotes the identity operator. The relative probability of a transition to
an eigenstate ξ
), which is equivalent to the relative probability of a measure
ment of ξ yielding the result ξ
, is
P(ξ
) ∝ ¸Aξ
)
2
. (2.55)
The absolute probability is clearly
P(ξ
) =
¸Aξ
)
2
ξ
¸Aξ
)
2
=
¸Aξ
)
2
¸AA)
. (2.56)
If the ket A) is normalized such that its norm is unity, then this probability simply
reduces to
P(ξ
) = ¸Aξ
)
2
. (2.57)
2.11 Expectation values
Consider an ensemble of microscopic systems prepared in the same initial state
A). Suppose a measurement of the observable ξ is made on each system. We
know that each measurement yields the value ξ
with probability P(ξ
). What is
the mean value of the measurement? This quantity, which is generally referred
to as the expectation value of ξ, is given by
¸ξ) =
ξ
ξ
P(ξ
) =
ξ
ξ
¸Aξ
)
2
25
2.12 Degeneracy 2 FUNDAMENTAL CONCEPTS
=
ξ
ξ
¸Aξ
)¸ξ
A) =
ξ
¸Aξξ
)¸ξ
A), (2.58)
which reduces to
¸ξ) = ¸AξA) (2.59)
with the aid of Eq. (2.54).
Consider the identity operator, 1. All states are eigenstates of this operator
with the eigenvalue unity. Thus, the expectation value of this operator is always
unity: i.e.,
¸A1A) = ¸AA) = 1, (2.60)
for all A). Note that it is only possible to normalize a given ket A) such that
Eq. (2.60) is satisﬁed because of the more general property (2.21) of the norm.
This property depends on the particular correspondence (2.16), that we adopted
earlier, between the elements of a ket space and those of its dual bra space.
2.12 Degeneracy
Suppose that two different eigenstates ξ
a
) and ξ
b
) of ξ correspond to the same
eigenvalue ξ
. These are termed degenerate eigenstates. Degenerate eigenstates
are necessarily orthogonal to any eigenstates corresponding to different eigen
values, but, in general, they are not orthogonal to each other (i.e., the proof of
orthogonality given in Sect. 2.8 does not work in this case). This is unfortunate,
since much of the previous formalism depends crucially on the mutual orthogo
nality of the different eigenstates of an observable. Note, however, that any linear
combination of ξ
a
) and ξ
b
) is also an eigenstate corresponding to the eigenvalue
ξ
. It follows that we can always construct two mutually orthogonal degenerate
eigenstates. For instance,
ξ
1
) = ξ
a
), (2.61)
ξ
2
) =
ξ
b
) −¸ξ
a
ξ
b
)ξ
a
)
1 − ¸ξ
a
ξ
b
)
2
. (2.62)
26
2.13 Compatible observables 2 FUNDAMENTAL CONCEPTS
This result is easily generalized to the case of more than two degenerate eigen
states. We conclude that it is always possible to construct a complete set of mu
tually orthogonal eigenstates for any given observable.
2.13 Compatible observables
Suppose that we wish to simultaneously measure two observables, ξ and η, of
a microscopic system? Let us assume that we possess an apparatus which is ca
pable of measuring ξ, and another which can measure η. For instance, the two
observables in question might be the projection in the x and zdirections of the
spin angular momentum of a spin onehalf particle. These could be measured us
ing appropriate SternGerlach apparatuses (see Sakurai, Sect. 1.1). Suppose that
we make a measurement of ξ, and the system is consequently thrown into one
of the eigenstates of ξ, ξ
), with eigenvalue ξ
. What happens if we now make
a measurement of η? Well, suppose that the eigenstate ξ
) is also an eigenstate
of η, with eigenvalue η
. In this case, a measurement of η will deﬁnitely give the
result η
. A second measurement of ξ will deﬁnitely give the result ξ
, and so on.
In this sense, we can say that the observables ξ and η simultaneously have the
values ξ
and η
, respectively. Clearly, if all eigenstates of ξ are also eigenstates
of η then it is always possible to make a simultaneous measurement of ξ and η.
Such observables are termed compatible.
Suppose, however, that the eigenstates of ξ are not eigenstates of η. Is it
still possible to measure both observables simultaneously? Let us again make an
observation of ξ which throws the system into an eigenstate ξ
), with eigenvalue
ξ
. We can now make a second observation to determine η. This will throw
the system into one of the (many) eigenstates of η which depend on ξ
). In
principle, each of these eigenstates is associated with a different result of the
measurement. Suppose that the system is thrown into an eigenstate η
), with
the eigenvalue η
. Another measurement of ξ will throw the system into one
of the (many) eigenstates of ξ which depend on η
). Each eigenstate is again
associated with a different possible result of the measurement. It is clear that if
the observables ξ and η do not possess simultaneous eigenstates then if the value
27
2.14 The uncertainty relation 2 FUNDAMENTAL CONCEPTS
of ξ is known (i.e., the system is in an eigenstate of ξ) then the value of η is
uncertain (i.e., the system is not in an eigenstate of η), and vice versa. We say
that the two observables are incompatible.
We have seen that the condition for two observables ξ and η to be simultane
ously measurable is that they should possess simultaneous eigenstates (i.e., every
eigenstate of ξ should also be an eigenstate of η). Suppose that this is the case.
Let a general eigenstate of ξ, with eigenvalue ξ
, also be an eigenstate of η, with
eigenvalue η
. It is convenient to denote this simultaneous eigenstate ξ
η
). We
have
ξξ
η
) = ξ
ξ
η
), (2.63)
ηξ
η
) = η
ξ
η
). (2.64)
We can leftmultiply the ﬁrst equation by η, and the second equation by ξ, and
then take the difference. The result is
(ξ η −ηξ)ξ
η
) = 0) (2.65)
for each simultaneous eigenstate. Recall that the eigenstates of an observable
must form a complete set. It follows that the simultaneous eigenstates of two
observables must also form a complete set. Thus, the above equation implies that
(ξ η −ηξ)A) = 0), (2.66)
where A) is a general ket. The only way that this can be true is if
ξ η = ηξ. (2.67)
Thus, the condition for two observables ξ and η to be simultaneously measurable is
that they should commute.
2.14 The uncertainty relation
We have seen that if ξ and η are two noncommuting observables, then a deter
mination of the value of ξ leaves the value of η uncertain, and vice versa. It is
28
2.14 The uncertainty relation 2 FUNDAMENTAL CONCEPTS
possible to quantify this uncertainty. For a general observable ξ, we can deﬁne a
Hermitian operator
∆ξ = ξ −¸ξ), (2.68)
where the expectation value is taken over the particular physical state under con
sideration. It is obvious that the expectation value of ∆ξ is zero. The expectation
value of (∆ξ)
2
≡ ∆ξ ∆ξ is termed the variance of ξ, and is, in general, nonzero.
In fact, it is easily demonstrated that
¸(∆ξ)
2
) = ¸ξ
2
) −¸ξ)
2
. (2.69)
The variance of ξ is a measure of the uncertainty in the value of ξ for the particu
lar state in question (i.e., it is a measure of the width of the distribution of likely
values of ξ about the expectation value). If the variance is zero then there is no
uncertainty, and a measurement of ξ is bound to give the expectation value, ¸ξ).
Consider the Schwarz inequality
¸AA)¸BB) ≥ ¸AB)
2
, (2.70)
which is analogous to
a
2
 b
2
≥ a b
2
(2.71)
in Euclidian space. This inequality can be proved by noting that
(¸A +c
∗
¸B)(A) +cB)) ≥ 0, (2.72)
where c is any complex number. If c takes the special value −¸BA)/¸BB) then
the above inequality reduces to
¸AA)¸BB) − ¸AB)
2
≥ 0, (2.73)
which is the same as the Schwarz inequality.
Let us substitute
A) = ∆ξ ), (2.74)
B) = ∆η ), (2.75)
29
2.14 The uncertainty relation 2 FUNDAMENTAL CONCEPTS
into the Schwarz inequality, where the blank ket  ) stands for any general ket.
We ﬁnd
¸(∆ξ)
2
)¸(∆η)
2
) ≥ ¸∆ξ ∆η)
2
, (2.76)
where use has been made of the fact that ∆ξ and ∆η are Hermitian operators.
Note that
∆ξ ∆η =
1
2
[∆ξ, ∆η] +
1
2
{∆ξ, ∆η} , (2.77)
where the commutator, [∆ξ, ∆η], and the anticommutator, {∆ξ, ∆η}, are deﬁned
[∆ξ, ∆η] ≡ ∆ξ ∆η −∆η∆ξ, (2.78)
{∆ξ, ∆η} ≡ ∆ξ ∆η +∆η∆ξ. (2.79)
The commutator is clearly antiHermitian,
([∆ξ, ∆η])
†
= (∆ξ ∆η −∆η∆ξ)
†
= ∆η∆ξ −∆ξ ∆η = − [∆ξ, ∆η] , (2.80)
whereas the anticommutator is obviously Hermitian. Now, it is easily demon
strated that the expectation value of a Hermitian operator is a real number,
whereas the expectation value of an antiHermitian operator is a pure imaginary
number. It is clear that the right hand side of
¸∆ξ ∆η) =
1
2
¸[∆ξ, ∆η]) +
1
2
¸{∆ξ, ∆η}), (2.81)
consists of the sum of a purely real and a purely imaginary number. Taking the
modulus squared of both sides gives
¸∆ξ ∆η)
2
=
1
4
¸[ξ, η])
2
+
1
4
¸{∆ξ, ∆η})
2
, (2.82)
where use has been made of ¸∆ξ) = 0, etc. The ﬁnal term in the above expression
is positive deﬁnite, so we can write
¸(∆ξ)
2
)¸(∆η)
2
) ≥
1
4
¸[ξ, η])
2
, (2.83)
where use has been made of Eq. (2.76). The above expression is termed the
uncertainty relation. According to this relation, an exact knowledge of the value
of ξ implies no knowledge whatsoever of the value of η, and vice versa. The one
exception to this rule is when ξ and η commute, in which case exact knowledge
of ξ does not necessarily imply no knowledge of η.
30
2.15 Continuous spectra 2 FUNDAMENTAL CONCEPTS
2.15 Continuous spectra
Up to now, we have studiously avoided dealing with observables possessing eigen
values which lie in a continuous range, rather than having discrete values. The
reason for this is because continuous eigenvalues imply a ket space of nonde
numerably inﬁnite dimension. Unfortunately, continuous eigenvalues are un
avoidable in quantum mechanics. In fact, the most important observables of all,
namely position and momentum, generally have continuous eigenvalues. Fortu
nately, many of the results we obtained previously for a ﬁnitedimensional ket
space with discrete eigenvalues can be generalized to ket spaces of nondenumer
ably inﬁnite dimensions.
Suppose that ξ is an observable with continuous eigenvalues. We can still
write the eigenvalue equation as
ξξ
) = ξ
ξ
). (2.84)
But, ξ
can now take a continuous range of values. Let us assume, for the sake of
simplicity, that ξ
can take any value. The orthogonality condition (2.50) gener
alizes to
¸ξ
ξ
) = δ(ξ
−ξ
), (2.85)
where δ(x) denotes the famous Dirac deltafunction. Note that there are clearly a
nondenumerably inﬁnite number of mutually orthogonal eigenstates of ξ. Hence,
the dimensionality of ket space is nondenumerably inﬁnite. Note, also, that eigen
states corresponding to a continuous range of eigenvalues cannot be normalized
so that they have unit norms. In fact, these eigenstates have inﬁnite norms: i.e.,
they are inﬁnitely long. This is the major difference between eigenstates in a
ﬁnitedimensional and an inﬁnitedimensional ket space. The extremely useful
relation (2.54) generalizes to
_
dξ
ξ
)¸ξ
 = 1. (2.86)
Note that a summation over discrete eigenvalues goes over into an integral over
a continuous range of eigenvalues. The eigenstates ξ
) must form a complete set
if ξ is to be an observable. It follows that any general ket can be expanded in
31
2.15 Continuous spectra 2 FUNDAMENTAL CONCEPTS
terms of the ξ
). In fact, the expansions (2.51)–(2.53) generalize to
A) =
_
dξ
ξ
)¸ξ
A), (2.87)
¸A =
_
dξ
¸Aξ
)¸ξ
, (2.88)
¸AA) =
_
dξ
¸Aξ
)¸ξ
A) =
_
dξ
¸Aξ
)
2
. (2.89)
These results also follow simply from Eq. (2.86). We have seen that it is not possi
ble to normalize the eigenstates ξ
) such that they have unit norms. Fortunately,
this convenient normalization is still possible for a general state vector. In fact,
according to Eq. (2.89), the normalization condition can be written
¸AA) =
_
dξ
¸Aξ
)
2
= 1. (2.90)
We have now studied observables whose eigenvalues can take a discrete num
ber of values as well as those whose eigenvalues can take any value. There are
number of other cases we could look at. For instance, observables whose eigen
values can only take a ﬁnite range of values, or observables whose eigenvalues
take on a ﬁnite range of values plus a set of discrete values. Both of these cases
can be dealt with using a fairly straightforward generalization of the previous
analysis (see Dirac, Cha. II and III).
32
3 POSITION AND MOMENTUM
3 Position and momentum
3.1 Introduction
So far, we have considered general dynamical variables represented by general
linear operators acting in ket space. However, in classical mechanics the most
important dynamical variables are those involving position and momentum. Let
us investigate the role of such variables in quantum mechanics.
In classical mechanics, the position q and momentum p of some component
of a dynamical system are represented as real numbers which, by deﬁnition, com
mute. In quantum mechanics, these quantities are represented as noncommuting
linear Hermitian operators acting in a ket space which represents all of the pos
sible states of the system. Our ﬁrst task is to discover a quantum mechanical
replacement for the classical result qp − pq = 0. Do the position and momen
tum operators commute? If not, what is the value of qp −pq?
3.2 Poisson brackets
Consider a dynamic system whose state at a particular time t is fully speciﬁed
by N independent classical coordinates q
i
(where i runs from 1 to N). Associ
ated with each generalized coordinate q
i
is a classical canonical momentum p
i
.
For instance, a Cartesian coordinate has an associated linear momentum, an an
gular coordinate has an associated angular momentum, etc. As is wellknown,
the behaviour of a classical system can be speciﬁed in terms of Lagrangian or
Hamiltonian dynamics. For instance, in Hamiltonian dynamics,
dq
i
dt
=
∂H
∂p
i
, (3.1)
dp
i
dt
= −
∂H
∂q
i
, (3.2)
where the function H(q
i
, p
i
, t) is the energy of the system at time t expressed
in terms of the classical coordinates and canonical momenta. This function is
33
3.2 Poisson brackets 3 POSITION AND MOMENTUM
usually referred to as the Hamiltonian of the system.
We are interested in ﬁnding some construct of classical dynamics which con
sists of products of dynamical variables. If such a construct exists we hope to
generalize it somehow to obtain a rule describing how dynamical variables com
mute with one another in quantum mechanics. There is, indeed, one wellknown
construct in classical dynamics which involves products of dynamical variables.
The Poisson bracket of two dynamical variables u and v is deﬁned
[u, v] =
N
i=1
_
∂u
∂q
i
∂v
∂p
i
−
∂u
∂p
i
∂v
∂q
i
_
, (3.3)
where u and v are regarded as functions of the coordinates and momenta q
i
and
p
i
. It is easily demonstrated that
[q
i
, q
j
] = 0, (3.4)
[p
i
, p
j
] = 0, (3.5)
[q
i
, p
j
] = δ
ij
. (3.6)
The time evolution of a dynamical variable can also be written in terms of a
Poisson bracket by noting that
du
dt
=
N
i=1
_
∂u
∂q
i
dq
i
dt
+
∂u
∂p
i
dp
i
dt
_
=
N
i=1
_
∂u
∂q
i
∂H
∂p
i
−
∂u
∂p
i
∂H
∂q
i
_
= [u, H], (3.7)
where use has been made of Hamilton’s equations.
Can we construct a quantum mechanical Poisson bracket in which u and v are
noncommuting operators, instead of functions? Well, the main properties of the
classical Poisson bracket are as follows:
[u, v] = −[v, u], (3.8)
34
3.2 Poisson brackets 3 POSITION AND MOMENTUM
[u, c] = 0, (3.9)
[u
1
+u
2
, v] = [u
1
, v] + [u
2
, v], (3.10)
[u, v
1
+v
2
] = [u, v
1
] + [u, v
2
] (3.11)
[u
1
u
2
, v] = [u
1
, v]u
2
+u
1
[u
2
, v], (3.12)
[u, v
1
v
2
] = [u, v
1
]v
2
+v
1
[u, v
2
], (3.13)
and
[u, [v, w]] + [v, [w, u]] + [w, [u, v]] = 0. (3.14)
The last relation is known as the Jacobi identity. In the above, u, v, w, etc.,
represent dynamical variables, and c represents a number. Can we ﬁnd some
combination of noncommuting operators u and v, etc., which satisﬁes all of the
above relations?
Well, we can evaluate the Poisson bracket [u
1
u
2
, v
1
v
2
] in two different ways,
since we can use either of the formulae (3.12) or (3.13) ﬁrst. Thus,
[u
1
u
2
, v
1
v
2
] = [u
1
, v
1
v
2
]u
2
+u
1
[u
2
, v
1
v
2
] (3.15)
= {[u
1
, v
1
]v
2
+v
1
[u
1
, v
2
]} u
2
+u
1
{[u
2
, v
1
]v
2
+v
1
[u
2
, v
2
]}
= [u
1
, v
1
]v
2
u
2
+v
1
[u
1
, v
2
]u
2
+u
1
[u
2
, v
1
]v
2
+u
1
v
1
[u
2
, v
2
],
and
[u
1
u
2
, v
1
v
2
] = [u
1
u
2
, v
1
]v
2
+v
1
[u
1
u
2
, v
2
] (3.16)
= [u
1
, v
1
]u
2
v
2
+u
1
[u
2
, v
1
]v
2
+v
1
[u
1
, v
2
]u
2
+v
1
u
1
[u
2
, v
2
].
Note that the order of the various factors has been preserved, since they now
represent noncommuting operators. Equating the above two results yields
[u
1
, v
1
](u
2
v
2
−v
2
u
2
) = (u
1
v
1
−v
1
u
1
)[u
2
, v
2
]. (3.17)
Since this relation must hold for u
1
and v
1
quite independent of u
2
and v
2
, it
follows that
u
1
v
1
−v
1
u
1
= i ¯h[u
1
, v
1
], (3.18)
u
2
v
2
−v
2
u
2
= i ¯h[u
2
, v
2
], (3.19)
35
3.2 Poisson brackets 3 POSITION AND MOMENTUM
where ¯h does not depend on u
1
, v
1
, u
2
, v
2
, and also commutes with (u
1
v
1
−v
1
u
1
).
Since u
1
, etc., are quite general operators, it follows that ¯h is just a number. We
want the quantum mechanical Poisson bracket of two Hermitian operators to
be an Hermitian operator itself, since the classical Poisson bracket of two real
dynamical variables is real. This requirement is satisﬁed if ¯h is a real number.
Thus, the quantum mechanical Poisson bracket of two dynamical variables u and
v is given by
[u, v] =
uv −v u
i ¯h
, (3.20)
where ¯h is a new universal constant of nature. Quantum mechanics agrees with
experiments provided that ¯h takes the value h/2π, where
h = 6.6261 10
−34
J s (3.21)
is Planck’s constant. Somewhat confusingly, the notation [u, v] is convention
ally reserved for the commutator uv − v u in quantum mechanics. We will use
[u, v]
quantum
to denote the quantum Poisson bracket. Thus,
[u, v]
quantum
=
[u, v]
i ¯h
. (3.22)
It is easily demonstrated that the quantum mechanical Poisson bracket, as deﬁned
above, satisﬁes all of the relations (3.8)–(3.14).
The strong analogy we have found between the classical Poisson bracket,
deﬁned in Eq. (3.3), and the quantum mechanical Poisson bracket, deﬁned in
Eq. (3.22), leads us to make the assumption that the quantum mechanical bracket
has the same value as the corresponding classical bracket, at least for the simplest
cases. In other words, we are assuming that Eqs. (3.4)–(3.6) hold for quantum
mechanical as well as classical Poisson brackets. This argument yields the funda
mental commutation relations
[q
i
, q
j
] = 0, (3.23)
[p
i
, p
j
] = 0, (3.24)
[q
i
, p
j
] = i ¯hδ
ij
. (3.25)
These results provide us with the basis for calculating commutation relations be
tween general dynamical variables. For instance, if two dynamical variables, ξ
36
3.3 Wavefunctions 3 POSITION AND MOMENTUM
and η, can both be written as a power series in the q
i
and p
i
, then repeated
application of Eqs. (3.8)–(3.13) allows [ξ, η] to be expressed in terms of the fun
damental commutation relations (3.23)–(3.25).
Equations (3.23)–(3.25) provide the foundation for the analogy between quan
tum mechanics and classical mechanics. Note that the classical result (that every
thing commutes) is obtained in the limit ¯h → 0. Thus, classical mechanics can be
regarded as the limiting case of quantum mechanics when ¯h goes to zero. In classi
cal mechanics, each pair of generalized coordinate and its conjugate momentum,
q
i
and p
i
, correspond to a different classical degree of freedom of the system. It is
clear from Eqs. (3.23)–(3.25) that in quantum mechanics the dynamical variables
corresponding to different degrees of freedom all commute. It is only those variables
corresponding to the same degree of freedom which may fail to commute.
3.3 Wavefunctions
Consider a simple system with one classical degree of freedom, which corre
sponds to the Cartesian coordinate x. Suppose that x is free to take any value
(e.g., x could be the position of a free particle). The classical dynamical vari
able x is represented in quantum mechanics as a linear Hermitian operator which
is also called x. Moreover, the operator x possesses eigenvalues x
lying in the
continuous range −∞ < x
< +∞ (since the eigenvalues correspond to all the
possible results of a measurement of x). We can span ket space using the suit
ably normalized eigenkets of x. An eigenket corresponding to the eigenvalue x
is denoted x
). Moreover, [see Eq. (2.85)]
¸x
x
) = δ(x
−x
). (3.26)
The eigenkets satisfy the extremely useful relation [see Eq. (2.86)]
_
+∞
−∞
dx
x
)¸x
 = 1. (3.27)
This formula expresses the fact that the eigenkets are complete, mutually orthog
onal, and suitably normalized.
37
3.3 Wavefunctions 3 POSITION AND MOMENTUM
A state ket A) (which represents a general state A of the system) can be
expressed as a linear superposition of the eigenkets of the position operator using
Eq. (3.27). Thus,
A) =
_
+∞
−∞
dx
¸x
A)x
) (3.28)
The quantity ¸x
A) is a complex function of the position eigenvalue x
. We can
write
¸x
A) = ψ
A
(x
). (3.29)
Here, ψ
A
(x
) is the famous wavefunction of quantum mechanics. Note that state
A is completely speciﬁed by its wavefunction ψ
A
(x
) [since the wavefunction
can be used to reconstruct the state ket A) using Eq. (3.28)]. It is clear that
the wavefunction of state A is simply the collection of the weights of the cor
responding state ket A), when it is expanded in terms of the eigenkets of the
position operator. Recall, from Sect. 2.10, that the probability of a measurement
of a dynamical variable ξ yielding the result ξ
when the system is in state A is
given by ¸ξ
A)
2
, assuming that the eigenvalues of ξ are discrete. This result is
easily generalized to dynamical variables possessing continuous eigenvalues. In
fact, the probability of a measurement of x yielding a result lying in the range
x
to x
+ dx
when the system is in a state A) is ¸x
A)
2
dx
. In other words,
the probability of a measurement of position yielding a result in the range x
to
x
+dx
when the wavefunction of the system is ψ
A
(x
) is
P(x
, dx
) = ψ
A
(x
)
2
dx
. (3.30)
This formula is only valid if the state ket A) is properly normalized: i.e., if
¸AA) = 1. The corresponding normalization for the wavefunction is
_
+∞
−∞
ψ
A
(x
)
2
dx
= 1. (3.31)
Consider a second state B represented by a state ket B) and a wavefunction
ψ
B
(x
). The inner product ¸BA) can be written
¸BA) =
_
+∞
−∞
dx
¸Bx
)¸x
A) =
_
+∞
−∞
ψ
∗
B
(x
) ψ
A
(x
) dx
, (3.32)
38
3.4 Schr¨ odinger’s representation  I 3 POSITION AND MOMENTUM
where use has been made of Eqs. (3.27) and (3.29). Thus, the inner product of
two states is related to the overlap integral of their wavefunctions.
Consider a general function f(x) of the observable x [e.g., f(x) = x
2
]. If B) =
f(x)A) then it follows that
ψ
B
(x
) = ¸x
f(x)
_
+∞
−∞
dx
ψ
A
(x
)x
)
=
_
+∞
−∞
dx
f(x
) ψ
A
(x
)¸x
x
), (3.33)
giving
ψ
B
(x
) = f(x
) ψ
A
(x
), (3.34)
where use has been made of Eq. (3.26). Here, f(x
) is the same function of the
position eigenvalue x
that f(x) is of the position operator x: i.e., if f(x) = x
2
then
f(x
) = x
2
. It follows, from the above result, that a general state ket A) can be
written
A) = ψ
A
(x)), (3.35)
where ψ
A
(x) is the same function of the operator x that the wavefunction ψ
A
(x
)
is of the position eigenvalue x
, and the ket ) has the wavefunction ψ(x
) = 1.
The ket ) is termed the standard ket. The dual of the standard ket is termed the
standard bra, and is denoted ¸. It is easily seen that
¸ψ
∗
A
(x)
DC
←→ψ
A
(x)). (3.36)
Note, ﬁnally, that ψ
A
(x)) is often shortened to ψ
A
), leaving the dependence on
the position operator x tacitly understood.
3.4 Schr¨ odinger’s representation  I
Consider the simple system described in the previous section. A general state ket
can be written ψ(x)), where ψ(x) is a general function of the position operator x,
and ψ(x
) is the associated wavefunction. Consider the ket whose wavefunction
39
3.4 Schr¨ odinger’s representation  I 3 POSITION AND MOMENTUM
is dψ(x
)/dx
. This ket is denoted dψ/dx). The new ket is clearly a linear func
tion of the original ket, so we can think of it as the result of some linear operator
acting on ψ). Let us denote this operator d/dx. It follows that
d
dx
ψ) =
dψ
dx
). (3.37)
Any linear operator which acts on ket vectors can also act on bra vectors.
Consider d/dx acting on a general bra ¸φ(x). According to Eq. (2.34), the bra
¸φd/dx satisﬁes
_
¸φ
d
dx
_
ψ) = ¸φ
_
d
dx
ψ)
_
. (3.38)
Making use of Eqs. (3.27) and (3.29), we can write
_
+∞
−∞
¸φ
d
dx
x
) dx
ψ(x
) =
_
+∞
−∞
φ(x
) dx
dψ(x
)
dx
. (3.39)
The righthand side can be transformed via integration by parts to give
_
+∞
−∞
¸φ
d
dx
x
) dx
ψ(x
) = −
_
+∞
−∞
dφ(x
)
dx
dx
ψ(x
), (3.40)
assuming that the contributions from the limits of integration vanish. It follows
that
¸φ
d
dx
x
) = −
dφ(x
)
dx
, (3.41)
which implies
¸φ
d
dx
= −¸
dφ
dx
. (3.42)
The neglect of contributions from the limits of integration in Eq. (3.40) is rea
sonable because physical wavefunctions are squareintegrable [see Eq. (3.31)].
Note that
d
dx
ψ) =
dψ
dx
)
DC
←→¸
dψ
∗
dx
= −¸ψ
∗
d
dx
, (3.43)
where use has been made of Eq. (3.42). It follows, by comparison with Eqs. (2.35)
and (3.36), that
_
d
dx
_
†
= −
d
dx
. (3.44)
40
3.4 Schr¨ odinger’s representation  I 3 POSITION AND MOMENTUM
Thus, d/dx is an antiHermitian operator.
Let us evaluate the commutation relation between the operators x and d/dx.
We have
d
dx
x ψ) =
d(x ψ)
dx
) = x
d
dx
ψ) +ψ). (3.45)
Since this holds for any ket ψ), it follows that
d
dx
x −x
d
dx
= 1. (3.46)
Let p be the momentumconjugate to x (for the simple systemunder consideration
p is a straightforward linear momentum). According to Eq. (3.25), x and p
satisfy the commutation relation
x p −px = i ¯h. (3.47)
It can be seen, by comparison with Eq. (3.46), that the Hermitian operator
−i ¯hd/dx satisﬁes the same commutation relation with x that p does. The most
general conclusion which may be drawn from a comparison of Eqs. (3.46) and
(3.47) is that
p = −i ¯h
d
dx
+f(x), (3.48)
since (as is easily demonstrated) a general function f(x) of the position operator
automatically commutes with x.
We have chosen to normalize the eigenkets and eigenbras of the position oper
ator so that they satisfy the normalization condition (3.26). However, this choice
of normalization does not uniquely determine the eigenkets and eigenbras. Sup
pose that we transform to a new set of eigenbras which are related to the old set
via
¸x

new
= e
i γ
¸x

old
, (3.49)
where γ
≡ γ(x
) is a real function of x
. This transformation amounts to a
rearrangement of the relative phases of the eigenbras. The new normalization
condition is
¸x
x
)
new
= ¸x
e
i γ
e
−i γ
x
)
old
= e
i (γ
−γ
)
¸x
x
)
old
= e
i (γ
−γ
)
δ(x
−x
) = δ(x
−x
). (3.50)
41
3.4 Schr¨ odinger’s representation  I 3 POSITION AND MOMENTUM
Thus, the new eigenbras satisfy the same normalization condition as the old
eigenbras.
By deﬁnition, the standard ket ) satisﬁes ¸x
) = 1. It follows from Eq. (3.49)
that the new standard ket is related to the old standard ket via
)
new
= e
−i γ
)
old
, (3.51)
where γ ≡ γ(x) is a real function of the position operator x. The dual of the
above equation yields the transformation rule for the standard bra,
¸
new
= ¸
old
e
i γ
. (3.52)
The transformation rule for a general operator A follows from Eqs. (3.51) and
(3.52), plus the requirement that the triple product ¸A) remain invariant (this
must be the case, otherwise the probability of a measurement yielding a certain
result would depend on the choice of eigenbras). Thus,
A
new
= e
−i γ
A
old
e
i γ
. (3.53)
Of course, if A commutes with x then A is invariant under the transformation. In
fact, d/dx is the only operator (we know of) which does not commute with x, so
Eq. (3.53) yields
_
d
dx
_
new
= e
−i γ
d
dx
e
i γ
=
d
dx
+ i
dγ
dx
, (3.54)
where the subscript “old” is taken as read. It follows, from Eq. (3.48), that the
momentum operator p can be written
p = −i ¯h
_
d
dx
_
new
− ¯h
dγ
dx
+f(x). (3.55)
Thus, the special choice
¯hγ(x) =
_
x
f(x) dx (3.56)
yields
p = −i ¯h
_
d
dx
_
new
. (3.57)
42
3.5 Schr¨ odinger’s representation  II 3 POSITION AND MOMENTUM
Equation (3.56) ﬁxes γ to within an arbitrary additive constant: i.e., the special
eigenkets and eigenbras for which Eq. (3.57) is true are determined to within an
arbitrary common phasefactor.
In conclusion, it is possible to ﬁnd a set of basis eigenkets and eigenbras of
the position operator x which satisfy the normalization condition (3.26), and for
which the momentum conjugate to x can be represented as the operator
p = −i ¯h
d
dx
. (3.58)
A general state ket is written ψ(x)), where the standard ket ) satisﬁes ¸x
) =
1, and where ψ(x
) = ¸x
ψ(x)) is the wavefunction. This scheme of things is
known as Schr¨ odinger’s representation, and is the basis of wave mechanics.
3.5 Schr¨ odinger’s representation  II
In the preceding sections, we have developed Schr¨ odinger’s representation for
the case of a single operator x corresponding to a classical Cartesian coordinate.
However, this scheme can easily be extended. Consider a system with N general
ized coordinates, q
1
q
N
, which can all be simultaneously measured. These are
represented as N commuting operators, q
1
q
N
, each with a continuous range
of eigenvalues, q
1
q
N
. Ket space is conveniently spanned by the simultaneous
eigenkets of q
1
q
N
, which are denoted q
1
q
N
). These eigenkets must form
a complete set, otherwise the q
1
q
N
would not be simultaneously observable.
The orthogonality condition for the eigenkets [i.e., the generalization of Eq. (3.26)]
is
¸q
1
q
N
q
1
q
N
) = δ(q
1
−q
1
) δ(q
2
−q
2
) δ(q
N
−q
N
). (3.59)
The completeness condition [i.e., the generalization of Eq. (3.27)] is
_
+∞
−∞
_
+∞
−∞
dq
1
dq
N
q
1
q
N
)¸q
1
q
N
 = 1. (3.60)
The standard ket ) is deﬁned such that
¸q
1
q
N
) = 1. (3.61)
43
3.5 Schr¨ odinger’s representation  II 3 POSITION AND MOMENTUM
The standard bra ¸ is the dual of the standard ket. A general state ket is written
ψ(q
1
q
N
)). (3.62)
The associated wavefunction is
ψ(q
1
q
N
) = ¸q
1
q
N
ψ). (3.63)
Likewise, a general state bra is written
¸φ(q
1
q
N
), (3.64)
where
φ(q
1
q
N
) = ¸φq
1
q
N
). (3.65)
The probability of an observation of the system ﬁnding the ﬁrst coordinate in the
range q
1
to q
1
+dq
1
, the second coordinate in the range q
2
to q
2
+dq
2
, etc., is
P(q
1
q
N
; dq
1
dq
N
) = ψ(q
1
q
N
)
2
dq
1
dq
N
. (3.66)
Finally, the normalization condition for a physical wavefunction is
_
+∞
−∞
_
+∞
−∞
ψ(q
1
q
N
)
2
dq
1
dq
N
= 1. (3.67)
The N linear operators ∂/∂q
i
(where i runs from 1 to N) are deﬁned
∂
∂q
i
ψ) =
∂ψ
∂q
i
). (3.68)
These linear operators can also act on bras (provided the associated wavefunctions
are square integrable) in accordance with [see Eq. (3.42)]
¸φ
∂
∂q
i
= −¸
∂φ
∂q
i
. (3.69)
Corresponding to Eq. (3.46), we can derive the commutation relations
∂
∂q
i
q
j
−q
j
∂
∂q
i
= δ
ij
. (3.70)
44
3.5 Schr¨ odinger’s representation  II 3 POSITION AND MOMENTUM
It is also clear that
∂
∂q
i
∂
∂q
j
ψ) =
∂
2
ψ
∂q
i
∂q
j
) =
∂
∂q
j
∂
∂q
i
ψ), (3.71)
showing that
∂
∂q
i
∂
∂q
j
=
∂
∂q
j
∂
∂q
i
. (3.72)
It can be seen, by comparison with Eqs. (3.23)–(3.25), that the linear oper
ators −i ¯h∂/∂q
i
satisfy the same commutation relations with the q’s and with
each other that the p’s do. The most general conclusion we can draw from this
coincidence of commutation relations is (see Dirac)
p
i
= −i ¯h
∂
∂q
i
+
∂F(q
1
q
N
)
∂q
i
. (3.73)
However, the function F can be transformed away via a suitable readjustment of
the phases of the basis eigenkets (see Sect. 3.4, and Dirac). Thus, we can always
construct a set of simultaneous eigenkets of q
1
q
N
for which
p
i
= −i ¯h
∂
∂q
i
. (3.74)
This is the generalized Schr¨ odinger representation.
It follows from Eqs. (3.61), (3.68), and (3.74) that
p
i
) = 0. (3.75)
Thus, the standard ket in Schr¨ odinger’s representation is a simultaneous eigenket
of all the momentum operators belonging to the eigenvalue zero. Note that
¸q
1
q
N

∂
∂q
i
ψ) = ¸q
1
q
N

∂ψ
∂q
i
) =
∂ψ(q
1
q
N
)
∂q
i
=
∂
∂q
i
¸q
1
q
N
ψ). (3.76)
Hence,
¸q
1
q
N

∂
∂q
i
=
∂
∂q
i
¸q
1
q
N
, (3.77)
so that
¸q
1
q
N
p
i
= −i ¯h
∂
∂q
i
¸q
1
q
N
. (3.78)
45
3.6 The momentum representation 3 POSITION AND MOMENTUM
The dual of the above equation gives
p
i
q
1
q
N
) = i ¯h
∂
∂q
i
q
1
q
N
). (3.79)
3.6 The momentum representation
Consider a system with one degree of freedom, describable in terms of a coordi
nate x and its conjugate momentum p, both of which have a continuous range
of eigenvalues. We have seen that it is possible to represent the system in terms
of the eigenkets of x. This is termed Schr¨ odinger’s representation. However, it is
also possible to represent the system in terms of the eigenkets of p.
Consider the eigenkets of p which belong to the eigenvalues p
. These are
denoted p
). The orthogonality relation for the momentum eigenkets is
¸p
p
) = δ(p
−p
), (3.80)
and the corresponding completeness relation is
_
+∞
−∞
dp
p
)¸p
 = 1. (3.81)
A general state ket can be written
φ(p)) (3.82)
where the standard ket ) satisﬁes
¸p
) = 1. (3.83)
Note that the standard ket in this representation is quite different to that in
Schr¨ odinger’s representation. The momentum space wavefunction φ(p
) sat
isﬁes
φ(p
) = ¸p
φ). (3.84)
The probability that a measurement of the momentum yields a result lying in the
range p
to p
+dp
is given by
P(p
, dp
) = φ(p
)
2
dp
. (3.85)
46
3.6 The momentum representation 3 POSITION AND MOMENTUM
Finally, the normalization condition for a physical momentumspace wavefunction
is
_
+∞
−∞
φ(p
)
2
dp
= 1. (3.86)
The fundamental commutation relations (3.23)–(3.25) exhibit a particular
symmetry between coordinates and their conjugate momenta. If all the coor
dinates are transformed into their conjugate momenta, and vice versa, and i is
then replaced by −i, the commutation relations are unchanged. It follows from
this symmetry that we can always choose the eigenkets of p in such a manner
that the coordinate x can be represented as (see Sect. 3.4)
x = i ¯h
d
dp
. (3.87)
This is termed the momentum representation.
The above result is easily generalized to a system with more than one degree
of freedom. Suppose the system is speciﬁed by N coordinates, q
1
q
N
, and
N conjugate momenta, p
1
p
N
. Then, in the momentum representation, the
coordinates can be written as
q
i
= i ¯h
∂
∂p
i
. (3.88)
We also have
q
i
) = 0, (3.89)
and
¸p
1
p
N
q
i
= i ¯h
∂
∂p
i
¸p
1
p
N
. (3.90)
The momentum representation is less useful than Schr¨ odinger’s representa
tion for a very simple reason. The energy operator (i.e., the Hamiltonian) of
most simple systems takes the form of a sum of quadratic terms in the momenta
(i.e., the kinetic energy) plus a complicated function of the coordinates (i.e., the
potential energy). In Schr¨ odinger’s representation, the eigenvalue problem for
the energy translates into a secondorder differential equation in the coordinates,
with a complicated potential function. In the momentum representation, the
47
3.7 The uncertainty relation 3 POSITION AND MOMENTUM
problem transforms into a highorder differential equation in the momenta, with
a quadratic potential. With the mathematical tools at our disposal, we are far bet
ter able to solve the former type of problem than the latter. Hence, Schr¨ odinger’s
representation is generally more useful than the momentum representation.
3.7 The uncertainty relation
How is a momentum space wavefunction related to the corresponding coordi
nate space wavefunction? To answer this question, let us consider the represen
tative ¸x
p
) of the momentum eigenkets p
) in Schr¨ odinger’s representation for
a system with a single degree of freedom. This representative satisﬁes
p
¸x
p
) = ¸x
pp
) = −i ¯h
d
dx
¸x
p
), (3.91)
where use has been made of Eq. (3.78) (for the case of a system with one degree
of freedom). The solution of the above differential equation is
¸x
p
) = c
exp(i p
x
/¯h), (3.92)
where c
= c
(p
). It is easily demonstrated that
¸p
p
) =
_
+∞
−∞
¸p
x
) dx
¸x
p
) = c
∗
c
_
∞
−∞
exp[−i (p
−p
) x
/¯h] dx
. (3.93)
The wellknown mathematical result
_
+∞
−∞
exp(i ax) dx = 2πδ(a), (3.94)
yields
¸p
p
) = c

2
hδ(p
−p
). (3.95)
This is consistent with Eq. (3.80), provided that c
= h
−1/2
. Thus,
¸x
p
) = h
−1/2
exp(i p
x
/¯h). (3.96)
48
3.7 The uncertainty relation 3 POSITION AND MOMENTUM
Consider a general state ket A) whose coordinate wavefunction is ψ(x
), and
whose momentum wavefunction is Ψ(p
). In other words,
ψ(x
) = ¸x
A), (3.97)
Ψ(p
) = ¸p
A). (3.98)
It is easily demonstrated that
ψ(x
) =
_
+∞
−∞
dp
¸x
p
)¸p
A)
=
1
h
1/2
_
+∞
−∞
Ψ(p
) exp(i p
x
/¯h) dp
(3.99)
and
Ψ(p
) =
_
+∞
−∞
dx
¸p
x
)¸x
A)
=
1
h
1/2
_
+∞
−∞
ψ(x
) exp(−i p
x
/¯h) dx
, (3.100)
where use has been made of Eqs. (3.27), (3.81), (3.94), and (3.96). Clearly, the
momentum space wavefunction is the Fourier transform of the coordinate space
wavefunction.
Consider a state whose coordinate space wavefunction is a wavepacket. In
other words, the wavefunction only has nonnegligible amplitude in some spa
tially localized region of extent ∆x. As is wellknow, the Fourier transform of a
wavepacket ﬁlls up a wavenumber band of approximate extent δk ∼ 1/∆x. Note
that in Eq. (3.99) the role of the wavenumber k is played by the quantity p
/¯h. It
follows that the momentum space wavefunction corresponding to a wavepacket
in coordinate space extends over a range of momenta ∆p ∼ ¯h/∆x. Clearly, a mea
surement of x is almost certain to give a result lying in a range of width ∆x.
Likewise, measurement of p is almost certain to yield a result lying in a range of
width ∆p. The product of these two uncertainties is
∆x ∆p ∼ ¯h. (3.101)
This result is called Heisenberg’s uncertainty principle.
49
3.8 Displacement operators 3 POSITION AND MOMENTUM
Actually, it is possible to write Heisenberg’s uncertainty principle more exactly
by making use of Eq. (2.83) and the commutation relation (3.47). We obtain
¸(∆x)
2
)¸(∆p)
2
) ≥
¯h
2
4
(3.102)
for any general state. It is easily demonstrated that the minimum uncertainty
states, for which the equality sign holds in the above relation, correspond to
Gaussian wavepackets in both coordinate and momentum space.
3.8 Displacement operators
Consider a system with one degree of freedom corresponding to the Cartesian
coordinate x. Suppose that we displace this system some distance along the x
axis. We could imagine that the system is on wheels, and we just give it a little
push. The ﬁnal state of the system is completely determined by its initial state,
together with the direction and magnitude of the displacement. Note that the
type of displacement we are considering is one in which everything to do with the
system is displaced. So, if the system is subject to an external potential, then the
potential must be displaced.
The situation is not so clear with state kets. The ﬁnal state of the system
only determines the direction of the displaced state ket. Even if we adopt the
convention that all state kets have unit norms, the ﬁnal ket is still not completely
determined, since it can be multiplied by a constant phasefactor. However, we
know that the superposition relations between states remain invariant under the
displacement. This follows because the superposition relations have a physical
signiﬁcance which is unaffected by a displacement of the system. Thus, if
R) = A) + B) (3.103)
in the undisplaced system, and the displacement causes ket R) to transform to
ket Rd), etc., then in the displaced system we have
Rd) = Ad) + Bd). (3.104)
50
3.8 Displacement operators 3 POSITION AND MOMENTUM
Incidentally, this determines the displaced kets to within a single arbitrary phase
factor to be multiplied into all of them. The displaced kets cannot be multiplied
by individual phasefactors, because this would wreck the superposition relations.
Since Eq. (3.104) holds in the displaced system whenever Eq. (3.103) holds in
the undisplaced system, it follows that the displaced ket Rd) must be the result
of some linear operator acting on the undisplaced ket R). In other words,
Rd) = DR), (3.105)
where D an operator which depends only on the nature of the displacement. The
arbitrary phasefactor by which all displaced kets may be multiplied results in D
being undetermined to an arbitrary multiplicative constant of modulus unity.
We now adopt the ansatz that any combination of bras, kets, and dynamical
variables which possesses a physical signiﬁcance is invariant under a displace
ment of the system. The normalization condition
¸AA) = 1 (3.106)
for a state ket A) certainly has a physical signiﬁcance. Thus, we must have
¸AdAd) = 1. (3.107)
Now, Ad) = DA) and ¸Ad = ¸AD
†
, so
¸AD
†
DA) = 1. (3.108)
Since this must hold for any state ket A), it follows that
D
†
D = 1. (3.109)
Hence, the displacement operator is unitary. Note that the above relation implies
that
A) = D
†
Ad). (3.110)
The equation
v A) = B), (3.111)
51
3.8 Displacement operators 3 POSITION AND MOMENTUM
where the operator v represents a dynamical variable, has some physical signiﬁ
cance. Thus, we require that
v
d
Ad) = Bd), (3.112)
where v
d
is the displaced operator. It follows that
v
d
Ad) = DB) = Dv A) = Dv D
†
Ad). (3.113)
Since this is true for any ket Ad), we have
v
d
= Dv D
†
. (3.114)
Note that the arbitrary numerical factor in D does not affect either of the results
(3.109) and (3.114).
Suppose, now, that the system is displaced an inﬁnitesimal distance δx along
the xaxis. We expect that the displaced ket Ad) should approach the undisplaced
ket A) in the limit as δx →0. Thus, we expect the limit
lim
δx→0
Ad) − A)
δx
= lim
δx→0
D−1
δx
A) (3.115)
to exist. Let
d
x
= lim
δx→0
D−1
δx
, (3.116)
where d
x
is denoted the displacement operator along the xaxis. The fact that D
can be replaced by Dexp(i γ), where γ is a real phaseangle, implies that d
x
can
be replaced by
lim
δx→0
Dexp(i γ) −1
δx
= lim
δx→0
D−1 + i γ
δx
= d
x
+ i a
x
, (3.117)
where a
x
is the limit of γ/δx. We have assumed, as seems reasonable, that γ tends
to zero as δx → 0. It is clear that the displacement operator is undetermined to
an arbitrary imaginary additive constant.
For small δx, we have
D = 1 +δx d
x
. (3.118)
52
3.8 Displacement operators 3 POSITION AND MOMENTUM
It follows from Eq. (3.109) that
(1 +δx d
†
x
)(1 +δx d
x
) = 1. (3.119)
Neglecting order (δx)
2
, we obtain
d
†
x
+d
x
= 0. (3.120)
Thus, the displacement operator is antiHermitian. Substituting into Eq. (3.114),
and again neglecting order (δx)
2
, we ﬁnd that
v
d
= (1 +δx d
x
) v (1 −δx d
x
) = v +δx (d
x
v −v d
x
), (3.121)
which implies
lim
δx→0
v
d
−v
δx
= d
x
v −v d
x
. (3.122)
Let us consider a speciﬁc example. Suppose that a state has a wavefunction
ψ(x
). If the system is displaced a distance δx along the xaxis then the new
wavefunction is ψ(x
− δx) (i.e., the same shape shifted in the xdirection by a
distance δx). Actually, the new wavefunction can be multiplied by an arbitrary
number of modulus unity. It can be seen that the new wavefunction is obtained
from the old wavefunction according to the prescription x
→x
−δx. Thus,
x
d
= x −δx. (3.123)
A comparison with Eq. (3.122), using x = v, yields
d
x
x −x d
x
= −1. (3.124)
It follows that i ¯hd
x
obeys the same commutation relation with x that p
x
, the
momentum conjugate to x, does [see Eq. (3.25)]. The most general conclusion
we can draw from this observation is that
p
x
= i ¯hd
x
+f(x), (3.125)
where f is Hermitian (since p
x
is Hermitian). However, the fact that d
x
is unde
termined to an arbitrary additive imaginary constant (which could be a function
of x) enables us to transform the function f out of the above equation, leaving
p
x
= i ¯hd
x
. (3.126)
53
3.8 Displacement operators 3 POSITION AND MOMENTUM
Thus, the displacement operator in the xdirection is proportional to the momen
tum conjugate to x. We say that p
x
is the generator of translations along the
xaxis.
A ﬁnite translation along the xaxis can be constructed from a series of very
many inﬁnitesimal translations. Thus, the operator D(∆x) which translates the
system a distance ∆x along the xaxis is written
D(∆x) = lim
N→∞
_
1 − i
∆x
N
p
x
¯h
_
N
, (3.127)
where use has been made of Eqs. (3.118) and (3.126). It follows that
D(∆x) = exp(−i p
x
∆x/¯h) . (3.128)
The unitary nature of the operator is now clearly apparent.
We can also construct displacement operators which translate the system along
the y and zaxes. Note that a displacement a distance ∆x along the xaxis com
mutes with a displacement a distance ∆y along the yaxis. In other words, if the
system is moved ∆x along the xaxis, and then ∆y along the yaxis then it ends
up in the same state as if it were moved ∆y along the yaxis, and then ∆x along
the xaxis. The fact that translations in independent directions commute is clearly
associated with the fact that the conjugate momentum operators associated with
these directions also commute [see Eqs. (3.24) and (3.128)].
54
4 QUANTUM DYNAMICS
4 Quantum dynamics
4.1 Schr¨ odinger’s equations of motion
Up to now, we have only considered systems at one particular instant of time. Let
us now investigate how quantum mechanical systems evolve with time.
Consider a system in a state A which evolves in time. At time t the state of the
system is represented by the ket At). The label A is needed to distinguish the
ket from any other ket (Bt), say) which is evolving in time. The label t is needed
to distinguish the different states of the system at different times.
The ﬁnal state of the system at time t is completely determined by its initial
state at time t
0
plus the time interval t − t
0
(assuming that the system is left
undisturbed during this time interval). However, the ﬁnal state only determines
the direction of the ﬁnal state ket. Even if we adopt the convention that all state
kets have unit norms, the ﬁnal ket is still not completely determined, since it
can be multiplied by an arbitrary phasefactor. However, we expect that if a
superposition relation holds for certain states at time t
0
then the same relation
should hold between the corresponding timeevolved states at time t, assuming
that the system is left undisturbed between times t
0
and t. In other words, if
Rt
0
) = At
0
) + Bt
0
) (4.1)
for any three kets, then we should have
Rt) = At) + Bt). (4.2)
This rule determines the timeevolved kets to within a single arbitrary phase
factor to be multiplied into all of them. The evolved kets cannot be multiplied by
individual phasefactors, since this would invalidate the superposition relation at
later times.
According to Eqs. (4.1) and (4.2), the ﬁnal ket Rt) depends linearly on the
initial ket Rt
0
). Thus, the ﬁnal ket can be regarded as the result of some linear
operator acting on the initial ket: , i.e.,
Rt) = TRt
0
), (4.3)
55
4.1 Schr¨ odinger’s equations of motion 4 QUANTUM DYNAMICS
where T is a linear operator which depends only on the times t and t
0
. The
arbitrary phasefactor by which all time evolved kets may be multiplied results
in T(t, t
0
) being undetermined to an arbitrary multiplicative constant of modulus
unity.
Since we have adopted a convention in which the norm of any state ket is
unity, it make sense to deﬁne the time evolution operator T in such a manner
that it preserves the length of any ket upon which it acts (i.e., if a ket is prop
erly normalized at time t then it will remain normalized at all subsequent times
t > t
0
). This is always possible, since the length of a ket possesses no physical
signiﬁcance. Thus, we require that
¸At
0
At
0
) = ¸AtAt) (4.4)
for any ket A, which immediately yields
T
†
T = 1. (4.5)
Hence, the time evolution operator T is a unitary operator.
Up to now, the time evolution operator T looks very much like the spatial
displacement operator D introduced in the previous section. However, there are
some important differences between time evolution and spatial displacement.
In general, we do expect the expectation value of some observable ξ to evolve
with time, even if the system is left in a state of undisturbed motion (after all,
time evolution has no meaning unless something observable changes with time).
The triple product ¸AξA) can evolve either because the ket A) evolves and the
operator ξ stays constant, the ket A) stays constant and the operator ξ evolves,
or both the ket A) and the operator ξ evolve. Since we are already committed to
evolving state kets, according to Eq. (4.3), let us assume that the time evolution
operator T can be chosen in such a manner that the operators representing the
dynamical variables of the system do not evolve in time (unless they contain some
speciﬁc time dependence).
We expect, from physical continuity, that as t → t
0
then At) → At
0
) for any
ket A. Thus, the limit
lim
t→t
0
At) − At
0
)
t −t
0
= lim
t→t
0
T −1
t −t
0
At
0
) (4.6)
56
4.1 Schr¨ odinger’s equations of motion 4 QUANTUM DYNAMICS
should exist. Note that this limit is simply the derivative of At
0
) with respect to
t
0
. Let
τ(t
0
) = lim
t→t
0
T(t, t
0
) −1
t −t
0
. (4.7)
It is easily demonstrated from Eq. (4.5) that τ is antiHermitian: i.e.,
τ
†
+τ = 0. (4.8)
The fact that T can be replaced by T exp(i γ) (where γ is real) implies that τ is
undetermined to an arbitrary imaginary additive constant (see previous section).
Let us deﬁne the Hermitian operator H(t
0
) = i ¯hτ. This operator is undetermined
to an arbitrary real additive constant. It follows from Eqs. (4.6) and (4.7) that
i ¯h
dAt
0
)
dt
0
= i ¯h lim
t→t
0
At) − At
0
)
t −t
0
= i ¯hτ(t
0
)At
0
) = H(t
0
)At
0
). (4.9)
When written for general t this equation becomes
i ¯h
dAt)
dt
= H(t)At). (4.10)
Equation (4.10) gives the general law for the time evolution of a state ket in
a scheme in which the operators representing the dynamical variables remain
ﬁxed. This equation is denoted Schr¨ odinger’s equation of motion. It involves a
Hermitian operator H(t) which is, presumably, a characteristic of the dynamical
system under investigation.
We saw, in the previous section, that if the operator D(x, x
0
) displaces the
system along the xaxis from x
0
to x then
p
x
= i ¯h lim
x→x
0
D(x, x
0
) −1
x −x
0
, (4.11)
where p
x
is the operator representing the momentum conjugate to x. We now
have that if the operator T(t, t
0
) evolves the system in time from t
0
to t then
H(t
0
) = i ¯h lim
t→t
0
T(t, t
0
) −1
t −t
0
. (4.12)
57
4.1 Schr¨ odinger’s equations of motion 4 QUANTUM DYNAMICS
Thus, the dynamical variable corresponding to the operator H stands to time t as
the momentum p
x
stands to the coordinate x. By analogy with classical physics,
this suggests that H(t) is the operator representing the total energy of the system.
(Recall that, in classical physics, if the equations of motion of a system are in
variant under an xdisplacement of the system then this implies that the system
conserves momentum in the xdirection. Likewise, if the equations of motion are
invariant under a temporal displacement then this implies that the system con
serves energy.) The operator H(t) is usually called the Hamiltonian of the system.
The fact that the Hamiltonian is undetermined to an arbitrary real additive con
stant is related to the wellknown phenomenon that energy is undetermined to
an arbitrary additive constant in physics (i.e., the zero of potential energy is not
welldeﬁned).
Substituting At) = TAt
0
) into Eq. (4.10) yields
i ¯h
dT
dt
At
0
) = H(t) TAt
0
). (4.13)
Since this must hold for any initial state At
0
) we conclude that
i ¯h
dT
dt
= H(t) T. (4.14)
This equation can be integrated to give
T(t, t
0
) = exp
_
_
−i
_
t
t
0
H(t
) dt
/¯h
_
_
, (4.15)
where use has been made of Eqs. (4.5) and (4.6). (Here, we assume that Hamil
tonian operators evaluated at different times commute with one another). It is
now clear how the fact that H is undetermined to an arbitrary real additive con
stant leaves T undetermined to a phasefactor. Note that, in the above analysis,
time is not an operator (we cannot observe time, as such), it is just a parameter
(or, more accurately, a continuous label). Since we are only dealing with non
relativistic quantum mechanics, the fact that position is an operator, but time is
only a label, need not worry us unduly. In relativistic quantum mechanics, time
and space coordinates are treated on the same footing by relegating position from
being an operator to being just a label.
58
4.2 Heisenberg’s equations of motion 4 QUANTUM DYNAMICS
4.2 Heisenberg’s equations of motion
We have seen that in Schr¨ odinger’s scheme the dynamical variables of the system
remain ﬁxed during a period of undisturbed motion, whereas the state kets evolve
according to Eq. (4.10). However, this is not the only way in which to represent
the time evolution of the system.
Suppose that a general state ket A is subject to the transformation
A
t
) = T
†
(t, t
0
)A). (4.16)
This is a timedependent transformation, since the operator T(t, t
0
) obviously
depends on time. The subscript t is used to remind us that the transformation is
timedependent. The time evolution of the transformed state ket is given by
A
t
t) = T
†
(t, t
0
)At) = T
†
(t, t
0
) T(t, t
0
)At
0
) = A
t
t
0
), (4.17)
where use has been made of Eqs. (4.3), (4.5), and the fact that T(t
0
, t
0
) = 1.
Clearly, the transformed state ket does not evolve in time. Thus, the transforma
tion (4.16) has the effect of bringing all kets representing states of undisturbed
motion of the system to rest.
The transformation must also be applied to bras. The dual of Eq. (4.16) yields
¸A
t
 = ¸AT. (4.18)
The transformation rule for a general observable v is obtained from the require
ment that the expectation value ¸AvA) should remain invariant. It is easily seen
that
v
t
= T
†
v T. (4.19)
Thus, a dynamical variable, which corresponds to a ﬁxed linear operator in Schr¨ o
dinger’s scheme, corresponds to a moving linear operator in this new scheme. It
is clear that the transformation (4.16) leads us to a scenario in which the state
of the system is represented by a ﬁxed vector, and the dynamical variables are
represented by moving linear operators. This is termed the Heisenberg picture, as
opposed to the Schr¨ odinger picture, which is outlined in Sect. 4.1.
59
4.2 Heisenberg’s equations of motion 4 QUANTUM DYNAMICS
Consider a dynamical variable v corresponding to a ﬁxed linear operator in the
Schr¨ odinger picture. According to Eq. (4.19), we can write
T v
t
= v T. (4.20)
Differentiation with respect to time yields
dT
dt
v
t
+T
dv
t
dt
= v
dT
dt
. (4.21)
With the help of Eq. (4.14), this reduces to
HT v
t
+ i ¯hT
dv
t
dt
= v HT, (4.22)
or
i ¯h
dv
t
dt
= T
†
v HT −T
†
HT v
t
= v
t
H
t
−H
t
v
t
, (4.23)
where
H
t
= T
†
HT. (4.24)
Equation (4.23) can be written
i ¯h
dv
t
dt
= [v
t
, H
t
]. (4.25)
Equation (4.25) shows how the dynamical variables of the system evolve in
the Heisenberg picture. It is denoted Heisenberg’s equation of motion. Note that
the timevarying dynamical variables in the Heisenberg picture are usually called
Heisenberg dynamical variables to distinguish them from Schr¨ odinger dynamical
variables (i.e., the corresponding variables in the Schr¨ odinger picture), which do
not evolve in time.
According to Eq. (3.22), the Heisenberg equation of motion can be written
dv
t
dt
= [v
t
, H
t
]
quantum
, (4.26)
where [ ]
quantum
denotes the quantumPoisson bracket. Let us compare this equa
tion with the classical time evolution equation for a general dynamical variable
v, which can be written in the form [see Eq. (3.7)]
dv
dt
= [v, H]
classical
. (4.27)
60
4.3 Ehrenfest’s theorem 4 QUANTUM DYNAMICS
Here, [ ]
classical
is the classical Poisson bracket, and H denotes the classical
Hamiltonian. The strong resemblance between Eqs. (4.26) and (4.27) provides
us with further justiﬁcation for our identiﬁcation of the linear operator H with
the energy of the system in quantum mechanics.
Note that if the Hamiltonian does not explicitly depend on time (i.e., the sys
tem is not subject to some timedependent external force) then Eq. (4.15) yields
T(t, t
0
) = exp [−i H(t −t
0
)/¯h] . (4.28)
This operator manifestly commutes with H, so
H
t
= T
†
HT = H. (4.29)
Furthermore, Eq. (4.25) gives
i ¯h
dH
dt
= [H, H] = 0. (4.30)
Thus, if the energy of the system has no explicit timedependence then it is rep
resented by the same nontimevarying operator H in both the Schr¨ odinger and
Heisenberg pictures.
Suppose that v is an observable which commutes with the Hamiltonian (and,
hence, with the time evolution operator T). It follows from Eq. (4.19) that v
t
= v.
Heisenberg’s equation of motion yields
i ¯h
dv
dt
= [v, H] = 0. (4.31)
Thus, any observable which commutes with the Hamiltonian is a constant of the mo
tion (hence, it is represented by the same ﬁxed operator in both the Schr¨ odinger
and Heisenberg pictures). Only those observables which do not commute with
the Hamiltonian evolve in time in the Heisenberg picture.
4.3 Ehrenfest’s theorem
We have now derived all of the basic elements of quantum mechanics. The only
thing which is lacking is some rule to determine the form of the quantum mechan
ical Hamiltonian. For a physical system which possess a classical analogue, we
61
4.3 Ehrenfest’s theorem 4 QUANTUM DYNAMICS
generally assume that the Hamiltonian has the same form as in classical physics
(i.e., we replace the classical coordinates and conjugate momenta by the corre
sponding quantum mechanical operators). This scheme guarantees that quantum
mechanics yields the correct classical equations of motion in the classical limit.
Whenever an ambiguity arises because of noncommuting observables, this can
usually be resolved by requiring the Hamiltonian H to be an Hermitian oper
ator. For instance, we would write the quantum mechanical analogue of the
classical product x p, appearing in the Hamiltonian, as the Hermitian product
(1/2)(x p + px). When the system in question has no classical analogue then we
are reduced to guessing a form for H which reproduces the observed behaviour
of the system.
Consider a threedimensional systemcharacterized by three independent Carte
sian position coordinates x
i
(where i runs from 1 to 3), with three corresponding
conjugate momenta p
i
. These are represented by three commuting position op
erators x
i
, and three commuting momentum operators p
i
, respectively. The com
mutation relations satisﬁed by the position and momentum operators are [see
Eq. (3.25)]
[x
i
, p
j
] = i ¯hδ
ij
. (4.32)
It is helpful to denote (x
1
, x
2
, x
3
) as x and (p
1
, p
2
, p
3
) as p. The following useful
formulae,
[x
i
, F(p)] = i ¯h
∂F
∂p
i
, (4.33)
[p
i
, G(x)] = −i ¯h
∂G
∂x
i
, (4.34)
where F and G are functions which can be expanded as power series, are easily
proved using the fundamental commutation relations Eq. (4.32).
Let us now consider the threedimensional motion of a free particle of mass m
in the Heisenberg picture. The Hamiltonian is assumed to have the same form as
in classical physics:
H =
p
2
2 m
=
1
2 m
3
i=1
p
2
i
. (4.35)
62
4.3 Ehrenfest’s theorem 4 QUANTUM DYNAMICS
In the following, all dynamical variables are assumed to be Heisenberg dynamical
variables, although we will omit the subscript t for the sake of clarity. The time
evolution of the momentum operator p
i
follows from Heisenberg’s equation of
motion (4.25). We ﬁnd that
dp
i
dt
=
1
i ¯h
[p
i
, H] = 0, (4.36)
since p
i
automatically commutes with any function of the momentum operators.
Thus, for a free particle the momentum operators are constants of the motion,
which means that p
i
(t) = p
i
(0) at all times t (for i is 1 to 3). The time evolution
of the position operator x
i
is given by
dx
i
dt
=
1
i ¯h
[x
i
, H] =
1
i ¯h
1
2 m
i ¯h
∂
∂p
i
_
_
_
3
j=1
p
2
j
_
_
_ =
p
i
m
=
p
i
(0)
m
, (4.37)
where use has been made of Eq. (4.33). It follows that
x
i
(t) = x
i
(0) +
_
_
p
i
(0)
m
_
_
t, (4.38)
which is analogous to the equation of motion of a classical free particle. Note
that even though
[x
i
(0), x
j
(0)] = 0, (4.39)
where the position operators are evaluated at equal times, the x
i
do not commute
when evaluated at different times. For instance,
[x
i
(t), x
i
(0)] =
_
_
p
i
(0) t
m
, x
i
(0)
_
_
=
−i ¯ht
m
. (4.40)
Combining the above commutation relation with the uncertainty relation (2.83)
yields
¸(∆x
i
)
2
)
t
¸(∆x
i
)
2
)
t=0
≥
¯h
2
t
2
4 m
2
. (4.41)
This result implies that even if a particle is welllocalized at t = 0, its position
becomes progressively more uncertain with time. This conclusion can also be
obtained by studying the propagation of wavepackets in wave mechanics.
63
4.3 Ehrenfest’s theorem 4 QUANTUM DYNAMICS
Let us now add a potential V(x) to our free particle Hamiltonian:
H =
p
2
2 m
+V(x). (4.42)
Here, V is some function of the x
i
operators. Heisenberg’s equation of motion
gives
dp
i
dt
=
1
i ¯h
[p
i
, V(x)] = −
∂V(x)
∂x
i
, (4.43)
where use has been made of Eq. (4.34). On the other hand, the result
dx
i
dt
=
p
i
m
(4.44)
still holds, because the x
i
all commute with the new term V(x) in the Hamilto
nian. We can use the Heisenberg equation of motion a second time to deduce
that
d
2
x
i
dt
2
=
1
i ¯h
_
dx
i
dt
, H
_
=
1
i ¯h
_
p
i
m
, H
_
=
1
m
dp
i
dt
= −
1
m
∂V(x)
∂x
i
. (4.45)
In vectorial form, this equation becomes
m
d
2
x
dt
2
=
dp
dt
= −∇V(x). (4.46)
This is the quantum mechanical equivalent of Newton’s second law of motion.
Taking the expectation values of both sides with respect to a Heisenberg state ket
that does not move with time, we obtain
m
d
2
¸x)
dt
2
=
d¸p)
dt
= −¸∇V(x)). (4.47)
This is known as Ehrenfest’s theorem. When written in terms of expectation
values, this result is independent of whether we are using the Heisenberg or
Schr¨ odinger picture. In contrast, the operator equation (4.46) only holds if x and
p are understood to be Heisenberg dynamical variables. Note that Eq. (4.47) has
no dependence on ¯h. In fact, it guarantees to us that the centre of a wavepacket
always moves like a classical particle.
64
4.4 Schr¨ odinger’s waveequation 4 QUANTUM DYNAMICS
4.4 Schr¨ odinger’s waveequation
Let us nowconsider the motion of a particle in three dimensions in the Schr¨ odinger
picture. The ﬁxed dynamical variables of the system are the position operators
x ≡ (x
1
, x
2
, x
3
), and the momentum operators p ≡ (p
1
, p
2
, p
3
). The state of the
system is represented as some time evolving ket At).
Let x
) represent a simultaneous eigenket of the position operators belonging
to the eigenvalues x
≡ (x
1
, x
2
, x
3
). Note that, since the position operators are
ﬁxed in the Schr¨ odinger picture, we do not expect the x
) to evolve in time. The
wavefunction of the system at time t is deﬁned
ψ(x
, t) = ¸x
At). (4.48)
The Hamiltonian of the system is taken to be
H =
p
2
2 m
+V(x). (4.49)
Schr¨ odinger’s equation of motion (4.10) yields
i ¯h
∂¸x
At)
∂t
= ¸x
HAt), (4.50)
where use has been made of the time independence of the x
). We adopt Schr¨ od
inger’s representation in which the momentum conjugate to the position operator
x
i
is written [see Eq. (3.74)]
p
i
= −i ¯h
∂
∂x
i
. (4.51)
Thus,
_
x
¸
¸
¸
¸
¸
¸
p
2
2 m
¸
¸
¸
¸
¸
¸
At
_
= −
_
_
¯h
2
2 m
_
_
∇
2
¸x
At), (4.52)
where use has been made of Eq. (3.78). Here, ∇
≡ (∂/∂x
, ∂/∂y
, ∂/∂z
) denotes
the gradient operator written in terms of the position eigenvalues. We can also
write
¸x
V(x) = V(x
)¸x
, (4.53)
65
4.4 Schr¨ odinger’s waveequation 4 QUANTUM DYNAMICS
where V(x
) is a scalar function of the position eigenvalues. Combining Eqs. (4.49),
(4.50), (4.52), and (4.53), we obtain
i ¯h
∂¸x
At)
∂t
= −
_
_
¯h
2
2 m
_
_
∇
2
¸x
At) +V(x
)¸x
At), (4.54)
which can also be written
i ¯h
∂ψ(x
, t)
∂t
= −
_
_
¯h
2
2 m
_
_
∇
2
ψ(x
, t) +V(x
) ψ(x
, t). (4.55)
This is Schr¨ odinger’s famous waveequation, and is the basis of wave mechanics.
Note, however, that the waveequation is just one of many possible representa
tions of quantum mechanics. It just happens to give a type of equation which we
know how to solve. In deriving the waveequation, we have chosen to represent
the system in terms of the eigenkets of the position operators, instead of those
of the momentum operators. We have also ﬁxed the relative phases of the x
)
according to Schr¨ odinger’s representation, so that Eq. (4.51) is valid. Finally, we
have chosen to work in the Schr¨ odinger picture, in which state kets evolve and
dynamical variables are ﬁxed, instead of the Heisenberg picture, in which the
opposite is true.
Suppose that the ket At) is an eigenket of the Hamiltonian belonging to the
eigenvalue H
:
HAt) = H
At). (4.56)
Schr¨ odinger’s equation of motion (4.10) yields
i ¯h
dAt)
dt
= H
At). (4.57)
This can be integrated to give
At) = exp[−i H
(t −t
0
)/¯h]At
0
). (4.58)
Note that At) only differs from At
0
) by a phasefactor. The direction of the
vector remains ﬁxed in ket space. This suggests that if the system is initially in an
eigenstate of the Hamiltonian then it remains in this state for ever, as long as the
66
4.4 Schr¨ odinger’s waveequation 4 QUANTUM DYNAMICS
system is undisturbed. Such a state is called a stationary state. The wavefunction
of a stationary state satisﬁes
ψ(x
, t) = ψ(x
, t
0
) exp[−i H
(t −t
0
)/¯h]. (4.59)
Substituting the above relation into Schr¨ odinger’s wave equation (4.55), we
obtain
−
_
_
¯h
2
2 m
_
_
∇
2
ψ
0
(x
) + (V(x
) −E) ψ
0
(x
) = 0, (4.60)
where ψ
0
(x
) ≡ ψ(x
, t
0
), and E = H
is the energy of the system. This is
Schr¨ odinger’s timeindependent waveequation. A bound state solution of the
above equation, in which the particle is conﬁned within a ﬁnite region of space,
satisﬁes the boundary condition
ψ
0
(x
) →0 as x
 →∞. (4.61)
Such a solution is only possible if
E < lim
x
→∞
V(x
). (4.62)
Since it is conventional to set the potential at inﬁnity equal to zero, the above
relation implies that bound states are equivalent to negative energy states. The
boundary condition (4.61) is sufﬁcient to uniquely specify the solution of Eq. (4.60).
The quantity ρ(x
, t), deﬁned by
ρ(x
, t) = ψ(x
, t)
2
, (4.63)
is termed the probability density. Recall, from Eq. (3.30), that the probability of
observing the particle in some volume element d
3
x
around position x
is propor
tional to ρ(x
, t) d
3
x
. The probability is equal to ρ(x
, t) d
3
x
if the wavefunction
is properly normalized, so that
_
ρ(x
, t) d
3
x
= 1. (4.64)
Schr¨ odinger’s timedependent waveequation, (4.55), can easily be written in
the form of a conservation equation for the probability density:
∂ρ
∂t
+∇
j = 0. (4.65)
67
4.4 Schr¨ odinger’s waveequation 4 QUANTUM DYNAMICS
The probability current j takes the form
j(x
, t) = −
_
i ¯h
2 m
_
[ψ
∗
∇
ψ− (∇
ψ
∗
) ψ] =
_
¯h
m
_
Im(ψ
∗
∇
ψ). (4.66)
We can integrate Eq. (4.65) over all space, using the divergence theorem, and the
boundary condition ρ →0 as x
 →∞, to obtain
∂
∂t
_
ρ(x
, t) d
3
x
= 0. (4.67)
Thus, Schr¨ odinger’s waveequation conserves probability. In particular, if the
wavefunction starts off properly normalized, according to Eq. (4.64), then it
remains properly normalized at all subsequent times. It is easily demonstrated
that
_
j(x
, t) d
3
x
=
¸p)
t
m
, (4.68)
where ¸p)
t
denotes the expectation value of the momentum evaluated at time t.
Clearly, the probability current is indirectly related to the particle momentum.
In deriving Eq. (4.65) we have, naturally, assumed that the potential V(x
) is
real. Suppose, however, that the potential has an imaginary component. In this
case, Eq. (4.65) generalizes to
∂ρ
∂t
+∇
j =
2 Im(V)
¯h
ρ, (4.69)
giving
∂
∂t
_
ρ(x
, t) d
3
x
=
2
¯h
Im
_
V(x
) ρ(x
, t) d
3
x
. (4.70)
Thus, if Im(V) < 0 then the total probability of observing the particle anywhere
in space decreases monotonically with time. Thus, an imaginary potential can
be used to account for the disappearance of a particle. Such a potential is often
employed to model nuclear reactions in which incident particles can be absorbed
by nuclei.
The wavefunction can always be written in the form
ψ(x
, t) =
_
ρ(x
, t) exp
_
_
i S(x
, t)
¯h
_
_
, (4.71)
68
4.4 Schr¨ odinger’s waveequation 4 QUANTUM DYNAMICS
where ρ and S are both real functions. The interpretation of ρ as a probability
density has already been given. What is the interpretation of S? Note that
ψ
∗
∇
ψ =
√
ρ ∇
(
√
ρ) +
_
i
¯h
_
ρ∇
S. (4.72)
It follows from Eq. (4.66) that
j =
ρ ∇
S
m
. (4.73)
Thus, the gradient of the phase of the wavefunction determines the direction of
the probability current. In particular, the probability current is locally normal to
the contours of the phasefunction S.
Let us substitute Eq. (4.71) into Schr¨ odinger’s timedependent waveequation.
We obtain
−
1
2 m
_
¯h
2
∇
2
√
ρ +2i ¯h∇
(
√
ρ)∇
S −
√
ρ ∇
S
2
+ i ¯h
√
ρ ∇
2
S
_
+
√
ρ V
=
_
i ¯h
∂
√
ρ
∂t
−
√
ρ
∂S
∂t
_
. (4.74)
Let us treat ¯h as a small quantity. To lowest order, Eq. (4.74) yields
−
∂S(x
, t)
∂t
=
1
2 m
∇
S(x
, t)
2
+V(x
, t) = H(x
, ∇
S, t), (4.75)
where H(x, p, t) is the Hamiltonian operator. The above equation is known as the
HamiltonJacobi equation, and is one of the many forms in which we can write
the equations of classical mechanics. In classical mechanics, S is the action (i.e.,
the pathintegral of the Lagrangian). Thus, in the limit ¯h → 0, wave mechanics
reduces to classical mechanics. It is a good approximation to neglect the terms
involving ¯h in Eq. (4.74) provided that
¯h∇
2
S ¸∇
S
2
. (4.76)
Note that, according to Eq. (4.71),
¯ λ =
¯h
∇
S
, (4.77)
69
4.4 Schr¨ odinger’s waveequation 4 QUANTUM DYNAMICS
where ¯ λ is the de Broglie wavelength divided by 2π. The inequality (4.76) is
equivalent to
∇
¯ λ ¸1. (4.78)
In other words, quantum mechanics reduces to classical mechanics whenever
the de Broglie wavelength is small compared to the characteristic distance over
which things (other than the quantum phase) vary. This distance is usually set by
the variation scalelength of the potential.
70
5 ANGULAR MOMENTUM
5 Angular momentum
5.1 Orbital angular momentum
Consider a particle described by the Cartesian coordinates (x, y, z) ≡ r and their
conjugate momenta (p
x
, p
y
, p
z
) ≡ p. The classical deﬁnition of the orbital angular
momentum of such a particle about the origin is L = r p, giving
L
x
= yp
z
−z p
y
, (5.1)
L
y
= z p
x
−x p
z
, (5.2)
L
z
= x p
y
−yp
x
. (5.3)
Let us assume that the operators (L
x
, L
y
, L
z
) ≡ L which represent the compo
nents of orbital angular momentum in quantum mechanics can be deﬁned in an
analogous manner to the corresponding components of classical angular momen
tum. In other words, we are going to assume that the above equations specify
the angular momentum operators in terms of the position and linear momentum
operators. Note that L
x
, L
y
, and L
z
are Hermitian, so they represent things which
can, in principle, be measured. Note, also, that there is no ambiguity regard
ing the order in which operators appear in products on the righthand sides of
Eqs. (5.1)–(5.3), since all of the products consist of operators which commute.
The fundamental commutation relations satisﬁed by the position and linear
momentum operators are [see Eqs. (3.23)–(3.25)]
[x
i
, x
j
] = 0, (5.4)
[p
i
, p
j
] = 0, (5.5)
[x
i
, p
j
] = i ¯hδ
ij
, (5.6)
where i and j stand for either x, y, or z. Consider the commutator of the operators
L
x
and L
z
:
[L
x
, L
y
] = [(yp
z
−z p
y
), (z p
x
−x p
z
)] = y[p
z
, z] p
x
+x p
y
[z, p
z
]
= i ¯h(−yp
x
+x p
y
) = i ¯hL
z
. (5.7)
71
5.1 Orbital angular momentum 5 ANGULAR MOMENTUM
The cyclic permutations of the above result yield the fundamental commutation
relations satisﬁed by the components of an angular momentum:
[L
x
, L
y
] = i ¯hL
z
, (5.8)
[L
y
, L
z
] = i ¯hL
x
, (5.9)
[L
z
, L
x
] = i ¯hL
y
. (5.10)
These can be summed up more succinctly by writing
L L = i ¯hL. (5.11)
The three commutation relations (5.8)–(5.10) are the foundation for the whole
theory of angular momentum in quantum mechanics. Whenever we encounter
three operators having these commutation relations, we know that the dynamical
variables which they represent have identical properties to those of the compo
nents of an angular momentum (which we are about to derive). In fact, we shall
assume that any three operators which satisfy the commutation relations (5.8)–
(5.10) represent the components of an angular momentum.
Suppose that there are N particles in the system, with angular momentum
vectors L
i
(where i runs from 1 to N). Each of these vectors satisﬁes Eq. (5.11),
so that
L
i
L
i
= i ¯hL
i
. (5.12)
However, we expect the angular momentum operators belonging to different par
ticles to commute, since they represent different degrees of freedom of the sys
tem. So, we can write
L
i
L
j
+L
j
L
i
= 0, (5.13)
for i ,= j. Consider the total angular momentum of the system, L =
N
i=1
L
i
. It is
clear from Eqs. (5.12) and (5.13) that
L L =
N
i=1
L
i
N
j=1
L
j
=
N
i=1
L
i
L
i
+
1
2
N
i,j=1
(L
i
L
j
+L
j
L
i
)
= i ¯h
N
i=1
L
i
= i ¯hL. (5.14)
72
5.1 Orbital angular momentum 5 ANGULAR MOMENTUM
Thus, the sum of two or more angular momentum vectors satisﬁes the same
commutation relation as a primitive angular momentum vector. In particular,
the total angular momentum of the system satisﬁes the commutation relation
(5.11).
The immediate conclusion which can be drawn from the commutation rela
tions (5.8)–(5.10) is that the three components of an angular momentum vector
cannot be speciﬁed (or measured) simultaneously. In fact, once we have speci
ﬁed one component, the values of other two components become uncertain. It is
conventional to specify the zcomponent, L
z
.
Consider the magnitude squared of the angular momentum vector, L
2
≡ L
2
x
+
L
2
y
+L
2
z
. The commutator of L
2
and L
z
is written
[L
2
, L
z
] = [L
2
x
, L
z
] + [L
2
y
, L
z
] + [L
2
z
, L
z
]. (5.15)
It is easily demonstrated that
[L
2
x
, L
z
] = −i ¯h(L
x
L
y
+L
y
L
x
), (5.16)
[L
2
y
, L
z
] = +i ¯h(L
x
L
y
+L
y
L
x
), (5.17)
[L
2
z
, L
z
] = 0, (5.18)
so
[L
2
, L
z
] = 0. (5.19)
Since there is nothing special about the zaxis, we conclude that L
2
also commutes
with L
x
and L
y
. It is clear from Eqs. (5.8)–(5.10) and (5.19) that the best we can
do in quantum mechanics is to specify the magnitude of an angular momentum
vector along with one of its components (by convention, the zcomponent).
It is convenient to deﬁne the shift operators L
+
and L
−
:
L
+
= L
x
+ i L
y
, (5.20)
L
−
= L
x
− i L
y
. (5.21)
Note that
[L
+
, L
z
] = −¯hL
+
, (5.22)
73
5.2 Eigenvalues of angular momentum 5 ANGULAR MOMENTUM
[L
−
, L
z
] = +¯hL
−
, (5.23)
[L
+
, L
−
] = 2 ¯hL
z
. (5.24)
Note, also, that both shift operators commute with L
2
.
5.2 Eigenvalues of angular momentum
Suppose that the simultaneous eigenkets of L
2
and L
z
are completely speciﬁed by
two quantum numbers, l and m. These kets are denoted l, m). The quantum
number m is deﬁned by
L
z
l, m) = m¯hl, m). (5.25)
Thus, m is the eigenvalue of L
z
divided by ¯h. It is possible to write such an
equation because ¯h has the dimensions of angular momentum. Note that m is a
real number, since L
z
is an Hermitian operator.
We can write
L
2
l, m) = f(l, m) ¯h
2
l, m), (5.26)
without loss of generality, where f(l, m) is some real dimensionless function of l
and m. Later on, we will show that f(l, m) = l (l +1). Now,
¸l, mL
2
−L
2
z
l, m) = ¸l, mf(l, m) ¯h
2
−m
2
¯h
2
l, m) = [f(l, m) −m
2
]¯h
2
, (5.27)
assuming that the l, m) have unit norms. However,
¸l, mL
2
−L
2
z
l, m) = ¸l, mL
2
x
+L
2
y
l, m)
= ¸l, mL
2
x
l, m) +¸l, mL
2
y
l, m). (5.28)
It is easily demonstrated that
¸Aξ
2
A) ≥ 0, (5.29)
where A) is a general ket, and ξ is an Hermitian operator. The proof follows
from the observation that
¸Aξ
2
A) = ¸Aξ
†
ξA) = ¸BB), (5.30)
74
5.2 Eigenvalues of angular momentum 5 ANGULAR MOMENTUM
where B) = ξA), plus the fact that ¸BB) ≥ 0 for a general ket B) [see Eq. (2.21)].
It follows from Eqs. (5.27)–(5.29) that
m
2
≤ f(l, m). (5.31)
Consider the effect of the shift operator L
+
on the eigenket l, m). It is easily
demonstrated that
L
2
(L
+
l, m)) = ¯h
2
f(l, m) (L
+
l, m)), (5.32)
where use has been made of Eq. (5.26), plus the fact that L
2
and L
z
commute. It
follows that the ket L
+
l, m) has the same eigenvalue of L
2
as the ket l, m). Thus,
the shift operator L
+
does not affect the magnitude of the angular momentum of
any eigenket it acts upon. Note that
L
z
L
+
l, m) = (L
+
L
z
+ [L
z
, L
+
])l, m) = (L
+
L
z
+ ¯hL
+
)l, m)
= (m+1) ¯hL
+
l, m), (5.33)
where use has been made of Eq. (5.22). The above equation implies that L
+
l, m)
is proportional to l, m+1). We can write
L
+
l, m) = c
+
l,m
¯hl, m+1), (5.34)
where c
+
l,m
is a number. It is clear that when the operator L
+
acts on a simulta
neous eigenstate of L
2
and L
z
, the eigenvalue of L
2
remains unchanged, but the
eigenvalue of L
z
is increased by ¯h. For this reason, L
+
is called a raising operator.
Using similar arguments to those given above, it is possible to demonstrate
that
L
−
l, m) = c
−
l,m
¯hl, m−1). (5.35)
Hence, L
−
is called a lowering operator.
The shift operators step the value of m up and down by unity each time they
operate on one of the simultaneous eigenkets of L
2
and L
z
. It would appear, at
ﬁrst sight, that any value of m can be obtained by applying the shift operators a
sufﬁcient number of times. However, according to Eq. (5.31), there is a deﬁnite
upper bound to the values that m
2
can take. This bound is determined by the
75
5.2 Eigenvalues of angular momentum 5 ANGULAR MOMENTUM
eigenvalue of L
2
[see Eq. (5.26)]. It follows that there is a maximum and a
minimum possible value which m can take. Suppose that we attempt to raise
the value of m above its maximum value m
max
. Since there is no state with
m > m
max
, we must have
L
+
l, m
max
) = 0). (5.36)
This implies that
L
−
L
+
l, m
max
) = 0). (5.37)
However,
L
−
L
+
= L
2
x
+L
2
y
+ i [L
x
, L
y
] = L
2
−L
2
z
− ¯hL
z
, (5.38)
so Eq. (5.37) yields
(L
2
−L
2
z
− ¯hL
z
)l, m
max
) = 0). (5.39)
The above equation can be rearranged to give
L
2
l, m
max
) = (L
2
z
+ ¯hL
z
)l, m
max
) = m
max
(m
max
+1) ¯h
2
l, m
max
). (5.40)
Comparison of this equation with Eq. (5.26) yields the result
f(l, m
max
) = m
max
(m
max
+1). (5.41)
But, when L
−
operates on n, m
max
) it generates n, m
max
− 1), n, m
max
− 2), etc.
Since the lowering operator does not change the eigenvalue of L
2
, all of these
states must correspond to the same value of f, namely m
max
(m
max
+1). Thus,
L
2
l, m) = m
max
(m
max
+1) ¯h
2
l, m). (5.42)
At this stage, we can give the unknown quantum number l the value m
max
, with
out loss of generality. We can also write the above equation in the form
L
2
l, m) = l (l +1) ¯h
2
l, m). (5.43)
It is easily seen that
L
−
L
+
l, m) = (L
2
−L
2
z
− ¯hL
z
)l, m) = ¯h
2
[l (l +1) −m(m+1)]l, m). (5.44)
Thus,
¸l, mL
−
L
+
l, m) = ¯h
2
[l (l +1) −m(m+1)]. (5.45)
76
5.2 Eigenvalues of angular momentum 5 ANGULAR MOMENTUM
However, we also know that
¸l, mL
−
L
+
l, m) = ¸l, mL
−
¯hc
+
l,m
l, m+1) = ¯h
2
c
+
l,m
c
−
l,m+1
, (5.46)
where use has been made of Eqs. (5.34) and (5.35). It follows that
c
+
l,m
c
−
l,m+1
= [l (l +1) −m(m+1)]. (5.47)
Consider the following:
¸l, mL
−
l, m+1) = ¸l, mL
x
l, m+1) − i ¸l, mL
y
l, m+1)
= ¸l, m+1L
x
l, m)
∗
− i ¸l, m+1L
y
l, m)
∗
= (¸l, m+1L
x
l, m) + i ¸l, m+1L
y
l, m))
∗
= ¸l, m+1L
+
l, m)
∗
, (5.48)
where use has been made of the fact that L
x
and L
y
are Hermitian. The above
equation reduces to
c
−
l,m+1
= (c
+
l,m
)
∗
(5.49)
with the aid of Eqs. (5.34) and (5.35).
Equations (5.47) and (5.49) can be combined to give
c
+
l,m

2
= [l (l +1) −m(m+1)]. (5.50)
The solution of the above equation is
c
+
l,m
=
_
l (l +1) −m(m+1). (5.51)
Note that c
+
l,m
is undetermined to an arbitrary phasefactor [i.e., we can replace
c
+
l,m
, given above, by c
+
l,m
exp(i γ), where γ is real, and we still satisfy Eq. (5.50)].
We have made the arbitrary, but convenient, choice that c
+
l,m
is real and posi
tive. This is equivalent to choosing the relative phases of the eigenkets l, m).
According to Eq. (5.49),
c
−
l,m
= (c
+
l,m−1
)
∗
=
_
l (l +1) −m(m−1). (5.52)
We have already seen that the inequality (5.31) implies that there is a maxi
mum and a minimum possible value of m. The maximum value of m is denoted
77
5.3 Rotation operators 5 ANGULAR MOMENTUM
l. What is the minimum value? Suppose that we try to lower the value of m
below its minimum value m
min
. Since there is no state with m < m
min
, we must
have
L
−
l, m
min
) = 0. (5.53)
According to Eq. (5.35), this implies that
c
−
l,m
min
= 0. (5.54)
It can be seen from Eq. (5.52) that m
min
= −l. We conclude that m can take a
“ladder” of discrete values, each rung differing from its immediate neighbours by
unity. The top rung is l, and the bottom rung is −l. There are only two possible
choices for l. Either it is an integer (e.g., l = 2, which allows m to take the values
−2, −1, 0, 1, 2), or it is a halfinteger (e.g., l = 3/2, which allows m to take the
values −3/2, −1/2, 1/2, 3/2). We will prove in the next section that an orbital
angular momentum can only take integer values of l.
In summary, using just the fundamental commutation relations (5.8)–(5.10),
plus the fact that L
x
, L
y
, and L
z
are Hermitian operators, we have shown that the
eigenvalues of L
2
≡ L
2
x
+L
2
y
+L
2
z
can be written l (l +1) ¯h
2
, where l is an integer,
or a halfinteger. We have also demonstrated that the eigenvalues of L
z
can only
take the values m¯h, where m lies in the range −l, −l + 1, l − 1, l. Let l, m)
denote a properly normalized simultaneous eigenket of L
2
and L
z
, belonging to
the eigenvalues l (l +1) ¯h
2
and m¯h, respectively. We have shown that
L
+
l, m) =
_
l (l +1) −m(m+1) ¯hl, m+1) (5.55)
L
−
l, m) =
_
l (l +1) −m(m−1) ¯hl, m−1), (5.56)
where L
±
= L
x
±i L
y
are the socalled shift operators.
5.3 Rotation operators
Consider a particle described by the spherical polar coordinates (r, θ, ϕ). The
classical momentum conjugate to the azimuthal angle ϕ is the zcomponent of
angular momentum, L
z
. According to Sect. 3.5, in quantum mechanics we can
78
5.3 Rotation operators 5 ANGULAR MOMENTUM
always adopt Schr¨ odinger’s representation, for which ket space is spanned by the
simultaneous eigenkets of the position operators r, θ, and φ, and L
z
takes the
form
L
z
= −i ¯h
∂
∂ϕ
. (5.57)
We can do this because there is nothing in Sect. 3.5 which speciﬁes that we
have to use Cartesian coordinates—the representation (3.74) works for any well
deﬁned set of coordinates.
Consider an operator R(∆ϕ) which rotates the system an angle ∆ϕabout the z
axis. This operator is very similar to the operator D(∆x), introduced in Sect. 3.8,
which translates the system a distance ∆x along the x axis. We were able to
demonstrate in Sect. 3.8 that
p
x
= i ¯h lim
δx→0
D(δx) −1
δx
, (5.58)
where p
x
is the linear momentum conjugate to x. There is nothing in our deriva
tion of this result which speciﬁes that x has to be a Cartesian coordinate. Thus,
the result should apply just as well to an angular coordinate. We conclude that
L
z
= i ¯h lim
δϕ→0
R(δϕ) −1
δϕ
. (5.59)
According to Eq. (5.59), we can write
R(δϕ) = 1 − i L
z
δϕ/¯h (5.60)
in the limit δϕ → 0. In other words, the angular momentum operator L
z
can be
used to rotate the system about the zaxis by an inﬁnitesimal amount. We say
that L
z
is the generator of rotations about the zaxis. The above equation implies
that
R(∆ϕ) = lim
N→∞
_
1 − i
∆ϕ
N
L
z
¯h
_
N
, (5.61)
which reduces to
R(∆ϕ) = exp(−i L
z
∆ϕ/¯h). (5.62)
79
5.3 Rotation operators 5 ANGULAR MOMENTUM
Note that R(∆ϕ) has all of the properties we would expect of a rotation operator
R(0) = 1, (5.63)
R(∆ϕ) R(−∆ϕ) = 1, (5.64)
R(∆ϕ
1
) R(∆ϕ
2
) = R(∆ϕ
1
+∆ϕ
2
). (5.65)
Suppose that the system is in a simultaneous eigenstate of L
2
and L
z
. As before,
this state is represented by the eigenket l, m), where the eigenvalue of L
2
is
l (l + 1) ¯h
2
, and the eigenvalue of L
z
is m¯h. We expect the wavefunction to
remain unaltered if we rotate the system 2π degrees about the zaxis. Thus,
R(2π)l, m) = exp(−i L
z
2π/¯h)l, m) = exp(−i 2 πm)l, m) = l, m). (5.66)
We conclude that m must be an integer. This implies, from the previous section,
that l must also be an integer. Thus, orbital angular momentum can only take on
integer values of the quantum numbers l and m.
Consider the action of the rotation operator R(∆ϕ) on an eigenstate possessing
zero angular momentum about the zaxis (i.e., an m = 0 state). We have
R(∆ϕ)l, 0) = exp(0)l, 0) = l, 0). (5.67)
Thus, the eigenstate is invariant to rotations about the zaxis. Clearly, its wave
function must be symmetric about the zaxis.
There is nothing special about the zaxis, so we can write
R
x
(∆ϕ
x
) = exp(−i L
x
∆ϕ
x
/¯h), (5.68)
R
y
(∆ϕ
y
) = exp(−i L
y
∆ϕ
y
/¯h), (5.69)
R
z
(∆ϕ
y
) = exp(−i L
z
∆ϕ
z
/¯h), (5.70)
by analogy with Eq. (5.62). Here, R
x
(∆ϕ
x
) denotes an operator which rotates the
system by an angle ∆ϕ
x
about the xaxis, etc. Suppose that the system is in an
eigenstate of zero overall orbital angular momentum (i.e., an l = 0 state). We
know that the system is also in an eigenstate of zero orbital angular momentum
about any particular axis. This follows because l = 0 implies m = 0, according
80
5.4 Eigenfunctions of orbital angular momentum 5 ANGULAR MOMENTUM
to the previous section, and we can choose the zaxis to point in any direction.
Thus,
R
x
(∆ϕ
x
)0, 0) = exp(0)0, 0) = 0, 0), (5.71)
R
y
(∆ϕ
y
)0, 0) = exp(0)0, 0) = 0, 0), (5.72)
R
z
(∆ϕ
z
)0, 0) = exp(0)0, 0) = 0, 0). (5.73)
Clearly, a zero angular momentum state is invariant to rotations about any axis.
Such a state must possess a spherically symmetric wavefunction.
Note that a rotation about the xaxis does not commute with a rotation about
the yaxis. In other words, if the system is rotated an angle ∆ϕ
x
about the xaxis,
and then ∆ϕ
y
about the yaxis, it ends up in a different state to that obtained
by rotating an angle ∆ϕ
y
about the yaxis, and then ∆ϕ
x
about the xaxis. In
quantum mechanics, this implies that R
y
(∆ϕ
y
) R
x
(∆ϕ
x
) ,= R
x
(∆ϕ
x
) R
y
(∆ϕ
y
), or
L
y
L
x
,= L
x
L
y
, [see Eqs. (5.68)–(5.70)]. Thus, the noncommuting nature of the
angular momentum operators is a direct consequence of the fact that rotations
do not commute.
5.4 Eigenfunctions of orbital angular momentum
In Cartesian coordinates, the three components of orbital angular momentum can
be written
L
x
= −i ¯h
_
y
∂
∂z
−z
∂
∂y
_
, (5.74)
L
y
= −i ¯h
_
z
∂
∂x
−x
∂
∂z
_
, (5.75)
L
z
= −i ¯h
_
x
∂
∂y
−y
∂
∂x
_
, (5.76)
using the Schr¨ odinger representation. Transforming to standard spherical polar
coordinates,
x = r sinθ cos ϕ, (5.77)
81
5.4 Eigenfunctions of orbital angular momentum 5 ANGULAR MOMENTUM
y = r sinθ sinϕ, (5.78)
z = r cos θ, (5.79)
we obtain
L
x
= i ¯h
_
sinϕ
∂
∂θ
+ cot θcos ϕ
∂
∂ϕ
_
(5.80)
L
y
= −i ¯h
_
cos ϕ
∂
∂θ
− cot θsinϕ
∂
∂ϕ
_
(5.81)
L
z
= −i ¯h
∂
∂ϕ
. (5.82)
Note that Eq. (5.82) accords with Eq. (5.57). The shift operators L
±
= L
x
± i L
y
become
L
±
= ±¯h exp(±i ϕ)
_
∂
∂θ
±i cot θ
∂
∂ϕ
_
. (5.83)
Now,
L
2
= L
2
x
+L
2
y
+L
2
z
= L
2
z
+ (L
+
L
−
+L
−
L
+
)/2, (5.84)
so
L
2
= −¯h
2
_
_
1
sinθ
∂
∂θ
sinθ
∂
∂θ
+
1
sin
2
θ
∂
2
∂ϕ
2
_
_
. (5.85)
The eigenvalue problem for L
2
takes the form
L
2
ψ = λ ¯h
2
ψ, (5.86)
where ψ(r, θ, ϕ) is the wavefunction, and λ is a number. Let us write
ψ(r, θ, ϕ) = R(r) Y(θ, ϕ). (5.87)
Equation (5.86) reduces to
_
_
1
sinθ
∂
∂θ
sinθ
∂
∂θ
+
1
sin
2
θ
∂
2
∂ϕ
2
_
_
Y +λ Y = 0, (5.88)
where use has been made of Eq. (5.85). As is wellknown, square integrable
solutions to this equation only exist when λ takes the values l (l + 1), where l is
an integer. These solutions are known as spherical harmonics, and can be written
Y
m
l
(θ, ϕ) =
¸
¸
¸
¸
_
2 l +1
4π
(l −m)!
(l +m)!
(−1)
m
e
i mϕ
P
m
l
(cos ϕ), (5.89)
82
5.4 Eigenfunctions of orbital angular momentum 5 ANGULAR MOMENTUM
where m is a positive integer lying in the range 0 ≤ m ≤ l. Here, P
m
l
(ξ) is an
associated Legendre function satisfying the equation
d
dξ
_
(1 −ξ
2
)
dP
m
l
dξ
_
−
m
2
1 −ξ
2
P
m
l
+l (l +1) P
m
l
= 0. (5.90)
We deﬁne
Y
−m
l
= (−1)
m
(Y
m
l
)
∗
, (5.91)
which allows m to take the negative values −l ≤ m < 0. The spherical harmonics
are orthogonal functions, and are properly normalized with respect to integration
over the entire solid angle:
_
π
0
_
2π
0
Y
m∗
l
(θ, ϕ) Y
m
l
(θ, ϕ) sinθdθdϕ = δ
ll
δ
mm
. (5.92)
The spherical harmonics also form a complete set for representing general func
tions of θ and ϕ.
By deﬁnition,
L
2
Y
m
l
= l (l +1) ¯h
2
Y
m
l
, (5.93)
where l is an integer. It follows from Eqs. (5.82) and (5.89) that
L
z
Y
m
l
= m¯hY
m
l
, (5.94)
where m is an integer lying in the range −l ≤ m ≤ l. Thus, the wavefunction
ψ(r, θ, ϕ) = R(r) Y
m
l
(θ, φ), where R is a general function, has all of the expected
features of the wavefunction of a simultaneous eigenstate of L
2
and L
z
belonging
to the quantum numbers l and m. The wellknown formula
dP
m
l
dξ
=
1
√
1 −ξ
2
P
m+1
l
−
mξ
1 −ξ
2
P
m
l
= −
(l +m)(l −m+1)
√
1 −ξ
2
P
m−1
l
+
mξ
1 −ξ
2
P
m
l
(5.95)
can be combined with Eqs. (5.83) and (5.89) to give
L
+
Y
m
l
=
_
l (l +1) −m(m+1) ¯hY
m+1
l
, (5.96)
L
−
Y
m
l
=
_
l (l +1) −m(m−1) ¯hY
m−1
l
. (5.97)
83
5.5 Motion in a central ﬁeld 5 ANGULAR MOMENTUM
These equations are equivalent to Eqs. (5.55)–(5.56). Note that a spherical har
monic wavefunction is symmetric about the zaxis (i.e., independent of ϕ) when
ever m = 0, and is spherically symmetric whenever l = 0 (since Y
0
0
= 1/
√
4π).
In summary, by solving directly for the eigenfunctions of L
2
and L
z
in Schr¨ od
inger’s representation, we have been able to reproduce all of the results of Sect. 5.2.
Nevertheless, the results of Sect. 5.2 are more general than those obtained in this
section, because they still apply when the quantum number l takes on halfinteger
values.
5.5 Motion in a central ﬁeld
Consider a particle of mass M moving in a spherically symmetric potential. The
Hamiltonian takes the form
H =
p
2
2 M
+V(r). (5.98)
Adopting Schr¨ odinger’s representation, we can write p = −(i/¯h)∇. Hence,
H = −
¯h
2
2 M
∇
2
+V(r). (5.99)
When written in spherical polar coordinates, the above equation becomes
H = −
¯h
2
2 M
_
_
1
r
2
∂
∂r
r
2
∂
∂r
+
1
r
2
sinθ
∂
∂θ
sinθ
∂
∂θ
+
1
r
2
sin
2
θ
∂
2
∂ϕ
2
_
_
+V(r). (5.100)
Comparing this equation with Eq. (5.85), we ﬁnd that
H =
¯h
2
2 M
_
_
−
1
r
2
∂
∂r
r
2
∂
∂r
+
L
2
¯h
2
r
2
_
_
+V(r). (5.101)
Now, we know that the three components of angular momentum commute
with L
2
(see Sect. 5.1). We also know, from Eqs. (5.80)–(5.82), that L
x
, L
y
,
and L
z
take the form of partial derivative operators of the angular coordinates,
84
5.5 Motion in a central ﬁeld 5 ANGULAR MOMENTUM
when written in terms of spherical polar coordinates using Schr¨ odinger’s repre
sentation. It follows from Eq. (5.101) that all three components of the angular
momentum commute with the Hamiltonian:
[L, H] = 0. (5.102)
It is also easily seen that L
2
commutes with the Hamiltonian:
[L
2
, H] = 0. (5.103)
According to Sect. 4.2, the previous two equations ensure that the angular mo
mentum L and its magnitude squared L
2
are both constants of the motion. This
is as expected for a spherically symmetric potential.
Consider the energy eigenvalue problem
Hψ = Eψ, (5.104)
where E is a number. Since L
2
and L
z
commute with each other and the Hamil
tonian, it is always possible to represent the state of the system in terms of the
simultaneous eigenstates of L
2
, L
z
, and H. But, we already know that the most
general form for the wavefunction of a simultaneous eigenstate of L
2
and L
z
is
(see previous section)
ψ(r, θ, ϕ) = R(r) Y
m
l
(θ, ϕ). (5.105)
Substituting Eq. (5.105) into Eq. (5.101), and making use of Eq. (5.93), we ob
tain
_
_
¯h
2
2 M
_
_
−
1
r
2
d
dr
r
2
d
dr
+
l (l +1)
r
2
_
_
+V(r) −E
_
_
R = 0. (5.106)
This is a SturmLiouville equation for the function R(r). We know, from the gen
eral properties of this type of equation, that if R(r) is required to be wellbehaved
at r = 0 and as r → ∞ then solutions only exist for a discrete set of values of E.
These are the energy eigenvalues. In general, the energy eigenvalues depend on
the quantum number l, but are independent of the quantum number m.
85
5.6 Energy levels of the hydrogen atom 5 ANGULAR MOMENTUM
5.6 Energy levels of the hydrogen atom
Consider a hydrogen atom, for which the potential takes the speciﬁc form
V(r) = −
e
2
4π
0
r
. (5.107)
The radial eigenfunction R(r) satisﬁes Eq. (5.106), which can be written
_
_
¯h
2
2 µ
_
_
−
1
r
2
d
dr
r
2
d
dr
+
l (l +1)
r
2
_
_
−
e
2
4π
0
r
−E
_
_
R = 0. (5.108)
Here, µ = m
e
m
p
/(m
e
+m
p
) is the reduced mass, which takes into account the fact
that the electron (of mass m
e
) and the proton (of mass m
p
) both rotate about a
common centre, which is equivalent to a particle of mass µ rotating about a ﬁxed
point. Let us write the product r R(r) as the function P(r). The above equation
transforms to
d
2
P
dr
2
−
2 µ
¯h
2
_
_
l (l +1)¯h
2
2 µr
2
−
e
2
4π
0
r
−E
_
_
P = 0, (5.109)
which is the onedimensional Schr¨ odinger equation for a particle of mass µ mov
ing in the effective potential
V
eff
(r) = −
e
2
4π
0
r
+
l (l +1) ¯h
2
2 µr
2
. (5.110)
The effective potential has a simple physical interpretation. The ﬁrst part is the
attractive Coulomb potential, and the second part corresponds to the repulsive
centrifugal force.
Let
a =
¸
¸
¸
¸
_
−¯h
2
2 µE
, (5.111)
and y = r/a, with
P(r) = f(y) exp(−y). (5.112)
Here, it is assumed that the energy eigenvalue E is negative. Equation (5.109)
transforms to
_
_
d
2
dy
2
−2
d
dy
−
l (l +1)
y
2
+
2 µe
2
a
4π
0
¯h
2
y
_
_
f = 0. (5.113)
86
5.6 Energy levels of the hydrogen atom 5 ANGULAR MOMENTUM
Let us look for a powerlaw solution of the form
f(y) =
n
c
n
y
n
. (5.114)
Substituting this solution into Eq. (5.113), we obtain
n
c
n
_
n(n −1) y
n−2
−2 ny
n−1
−l (l +1) y
n−2
+
2 µe
2
a
4π
0
¯h
2
y
n−1
_
= 0. (5.115)
Equating the coefﬁcients of y
n−2
gives
c
n
[n(n −1) −l (l +1)] = c
n−1
_
_
2 (n −1) −
2 µe
2
a
4π
0
¯h
2
_
_
. (5.116)
Now, the power law series (5.114) must terminate at small n, at some positive
value of n, otherwise f(y) behaves unphysically as y →0. This is only possible if
[n
min
(n
min
−1)−l (l+1)] = 0, where the ﬁrst term in the series is c
n
min
y
n
min
. There
are two possibilities: n
min
= −l or n
min
= l + 1. The former predicts unphysical
behaviour of the wavefunction at y = 0. Thus, we conclude that n
min
= l + 1.
Note that for an l = 0 state there is a ﬁnite probability of ﬁnding the electron at
the nucleus, whereas for an l > 0 state there is zero probability of ﬁnding the
electron at the nucleus (i.e., ψ
2
= 0 at r = 0, except when l = 0). Note, also,
that it is only possible to obtain sensible behaviour of the wavefunction as r →0
if l is an integer.
For large values of y, the ratio of successive terms in the series (5.114) is
c
n
y
c
n−1
=
2 y
n
, (5.117)
according to Eq. (5.116). This is the same as the ratio of successive terms in the
series
n
(2 y)
n
n!
, (5.118)
which converges to exp(2 y). We conclude that f(y) → exp(2 y) as y → ∞. It
follows from Eq. (5.112) that R(r) → exp(r/a)/r as r → ∞. This does not cor
respond to physically acceptable behaviour of the wavefunction, since
_
ψ
2
dV
must be ﬁnite. The only way in which we can avoid this unphysical behaviour is
87
5.6 Energy levels of the hydrogen atom 5 ANGULAR MOMENTUM
if the series (5.114) terminates at some maximum value of n. According to the
recursion relation (5.116), this is only possible if
µe
2
a
4π
0
¯h
2
= n, (5.119)
where the last term in the series is c
n
y
n
. It follows from Eq. (5.111) that the
energy eigenvalues are quantized, and can only take the values
E = −
µe
4
32π
2
2
0
¯h
2
n
2
. (5.120)
Here, n is a positive integer which must exceed the quantum number l, otherwise
there would be no terms in the series (5.114).
It is clear that the wavefunction for a hydrogen atom can be written
ψ(r, θ, ϕ) = R(r/a) Y
m
l
(θ, ϕ), (5.121)
where
a =
n4π
0
¯h
2
µe
2
= 5.3 10
−11
n meters, (5.122)
and R(x) is a wellbehaved solution of the differential equation
_
_
1
x
2
d
dx
x
2
d
dx
−
l (l +1)
x
2
+
2 n
x
−1
_
_
R = 0. (5.123)
Finally, the Y
m
l
are spherical harmonics. The restrictions on the quantum numbers
are m ≤ l < n. Here, n is a positive integer, l is a nonnegative integer, and m
is an integer.
The ground state of hydrogen corresponds to n = 1. The only permissible
values of the other quantum numbers are l = 0 and m = 0. Thus, the ground
state is a spherically symmetric, zero angular momentum state. The energy of
the ground state is
E
0
= −
µe
4
32π
2
2
0
¯h
2
= −13.6 electron volts. (5.124)
88
5.7 Spin angular momentum 5 ANGULAR MOMENTUM
The next energy level corresponds to n = 2. The other quantum numbers are
allowed to take the values l = 0, m = 0 or l = 1, m = −1, 0, 1. Thus, there are
n = 2 states with nonzero angular momentum. Note that the energy levels given
in Eq. (5.120) are independent of the quantum number l, despite the fact that l
appears in the radial eigenfunction equation (5.123). This is a special property
of a 1/r Coulomb potential.
In addition to the quantized negative energy state of the hydrogen atom, which
we have just found, there is also a continuum of unbound positive energy states.
5.7 Spin angular momentum
Up to now, we have tacitly assumed that the state of a particle in quantum me
chanics can be completely speciﬁed by giving the wavefunction ψ as a function
of the spatial coordinates x, y, and z. Unfortunately, there is a wealth of experi
mental evidence which suggests that this simplistic approach is incomplete.
Consider an isolated system at rest, and let the eigenvalue of its total angular
momentum be j (j +1) ¯h
2
. According to the theory of orbital angular momentum
outlined in Sects. 5.4 and 5.5, there are two possibilities. For a system consisting
of a single particle, j = 0. For a system consisting of two (or more) particles, j is
a nonnegative integer. However, this does not agree with observations, because
we often ﬁnd systems which appear to be structureless, and yet have j ,= 0. Even
worse, systems where j has halfinteger values abound in nature. In order to
explain this apparent discrepancy between theory and experiments, Gouldsmit
and Uhlenbeck (in 1925) introduced the concept of an internal, purely quantum
mechanical, angular momentum called spin. For a particle with spin, the total
angular momentum in the rest frame is nonvanishing.
Let us denote the three components of the spin angular momentum of a par
ticle by the Hermitian operators (S
x
, S
y
, S
z
) ≡ S. We assume that these operators
obey the fundamental commutation relations (5.8)–(5.10) for the components of
an angular momentum. Thus, we can write
S S = i ¯hS. (5.125)
89
5.7 Spin angular momentum 5 ANGULAR MOMENTUM
We can also deﬁne the operator
S
2
= S
2
x
+S
2
y
+S
2
z
. (5.126)
According to the quite general analysis of Sect. 5.1,
[S, S
2
] = 0. (5.127)
Thus, it is possible to ﬁnd simultaneous eigenstates of S
2
and S
z
. These are
denoted s, s
z
), where
S
z
s, s
z
) = s
z
¯hs, s
z
), (5.128)
S
2
s, s
z
) = s (s +1) ¯h
2
s, s
z
). (5.129)
According to the equally general analysis of Sect. 5.2, the quantum number s can,
in principle, take integer or halfinteger values, and the quantum number s
z
can
only take the values s, s −1 −s +1, −s.
Spin angular momentum clearly has many properties in common with orbital
angular momentum. However, there is one vitally important difference. Spin
angular momentum operators cannot be expressed in terms of position and mo
mentum operators, like in Eqs. (5.1)–(5.3), since this identiﬁcation depends on
an analogy with classical mechanics, and the concept of spin is purely quantum
mechanical: i.e., it has no analogy in classical physics. Consequently, the re
striction that the quantum number of the overall angular momentum must take
integer values is lifted for spin angular momentum, since this restriction (found
in Sects. 5.3 and 5.4) depends on Eqs. (5.1)–(5.3). In other words, the quantum
number s is allowed to take halfinteger values.
Consider a spin onehalf particle, for which
S
z
±) = ±
¯h
2
±), (5.130)
S
2
±) =
3 ¯h
2
4
±). (5.131)
Here, the ±) denote eigenkets of the S
z
operator corresponding to the eigen
values ±¯h/2. These kets are orthonormal (since S
z
is an Hermitian operator),
so
¸+−) = 0. (5.132)
90
5.8 Wavefunction of a spin onehalf particle 5 ANGULAR MOMENTUM
They are also properly normalized and complete, so that
¸++) = ¸−−) = 1, (5.133)
and
+)¸+ + −)¸− = 1. (5.134)
It is easily veriﬁed that the Hermitian operators deﬁned by
S
x
=
¯h
2
( +)¸− + −)¸+ ) , (5.135)
S
y
=
i ¯h
2
( − +)¸− + −)¸+ ) , (5.136)
S
z
=
¯h
2
( +)¸+ − −)¸− ) , (5.137)
satisfy the commutation relations (5.8)–(5.10) (with the L
j
replaced by the S
j
).
The operator S
2
takes the form
S
2
=
3 ¯h
2
4
. (5.138)
It is also easily demonstrated that S
2
and S
z
, deﬁned in this manner, satisfy the
eigenvalue relations (5.130)–(5.131). Equations (5.135)–(5.138) constitute a re
alization of the spin operators S and S
2
(for a spin onehalf particle) in spin space
(i.e., that Hilbert subspace consisting of kets which correspond to the different
spin states of the particle).
5.8 Wavefunction of a spin onehalf particle
The state of a spin onehalf particle is represented as a vector in ket space.
Let us suppose that this space is spanned by the basis kets x
, y
, z
, ±). Here,
x
, y
, z
, ±) denotes a simultaneous eigenstate of the position operators x, y, z,
and the spin operator S
z
, corresponding to the eigenvalues x
, y
, z
, and ±¯h/2,
respectively. The basis kets are assumed to satisfy the completeness relation
___
( x
, y
, z
, +)¸x
, y
, z
, + + x
, y
, z
, −)¸x
, y
, z
, − ) dx
dy
dz
= 1.
(5.139)
91
5.8 Wavefunction of a spin onehalf particle 5 ANGULAR MOMENTUM
It is helpful to think of the ket x
, y
, z
, +) as the product of two kets—a
position space ket x
, y
, z
), and a spin space ket +). We assume that such a
product obeys the commutative and distributive axioms of multiplication:
x
, y
, z
)+) = +)x
, y
, z
), (5.140)
(c
x
, y
, z
) +c
x
, y
, z
)) +) = c
x
, y
, z
)+)
+c
x
, y
, z
)+) (5.141)
x
, y
, z
) (c
+
+) +c
−
−)) = c
+
x
, y
, z
)+)
+c
−
x
, y
, z
)−), (5.142)
where the c’s are numbers. We can give meaning to any position space operator
(such as L
z
) acting on the product x
, y
, z
)+) by assuming that it operates only
on the x
, y
, z
) factor, and commutes with the +) factor. Similarly, we can give
a meaning to any spin operator (such as S
z
) acting on x
, y
, z
)+) by assuming
that it operates only on +), and commutes with x
, y
, z
). This implies that every
position space operator commutes with every spin operator. In this manner, we
can give meaning to the equation
x
, y
, z
, ±) = x
, y
, z
)±) = ±)x
, y
, z
). (5.143)
The multiplication in the above equation is of quite a different type to any
which we have encountered previously. The ket vectors x
, y
, z
) and ±) are
in two quite separate vector spaces, and their product x
, y
, z
)±) is in a third
vector space. In mathematics, the latter space is termed the product space of the
former spaces, which are termed factor spaces. The number of dimensions of a
product space is equal to the product of the number of dimensions of each of the
factor spaces. A general ket of the product space is not of the form (5.143), but
is instead a sum or integral of kets of this form.
A general state A of a spin onehalf particle is represented as a ket A)) in the
product of the spin and position spaces. This state can be completely speciﬁed by
two wavefunctions:
ψ
+
(x
, y
, z
) = ¸x
, y
, z
¸+A)), (5.144)
ψ
−
(x
, y
, z
) = ¸x
, y
, z
¸−A)). (5.145)
92
5.9 Rotation operators in spin space 5 ANGULAR MOMENTUM
The probability of observing the particle in the region x
to x
+dx
, y
to y
+dy
,
and z
to z
+ dz
, with s
z
= +1/2 is ψ
+
(x
, y
, z
)
2
dx
dy
dz
. Likewise, the
probability of observing the particle in the region x
to x
+ dx
, y
to y
+ dy
,
and z
to z
+dz
, with s
z
= −1/2 is ψ
−
(x
, y
, z
)
2
dx
dy
dz
. The normalization
condition for the wavefunctions is
___
_
ψ
+

2
+ ψ
−

2
_
dx
dy
dz
= 1. (5.146)
5.9 Rotation operators in spin space
Let us, for the moment, forget about the spatial position of the particle, and
concentrate on its spin state. A general spin state A is represented by the ket
A) = ¸+A)+) +¸−A)−) (5.147)
in spin space. In Sect. 5.3, we were able to construct an operator R
z
(∆ϕ) which
rotates the system by an angle ∆ϕ about the zaxis in position space. Can we also
construct an operator T
z
(∆ϕ) which rotates the system by an angle ∆ϕ about
the zaxis in spin space? By analogy with Eq. (5.62), we would expect such an
operator to take the form
T
z
(∆ϕ) = exp(−i S
z
∆ϕ/¯h). (5.148)
Thus, after rotation, the ket A) becomes
A
R
) = T
z
(∆ϕ)A). (5.149)
To demonstrate that the operator (5.148) really does rotate the spin of the
system, let us consider its effect on ¸S
x
). Under rotation, this expectation value
changes as follows:
¸S
x
) →¸A
R
S
x
A
R
) = ¸AT
†
z
S
x
T
z
A). (5.150)
Thus, we need to compute
exp( i S
z
∆ϕ/¯h) S
x
exp(−i S
z
∆ϕ/¯h). (5.151)
93
5.9 Rotation operators in spin space 5 ANGULAR MOMENTUM
This can be achieved in two different ways.
First, we can use the explicit formula for S
x
given in Eq. (5.135). We ﬁnd that
Eq. (5.151) becomes
¯h
2
exp( i S
z
∆ϕ/¯h) ( +)¸− + −)¸+ ) exp(−i S
z
∆ϕ/¯h), (5.152)
or
¯h
2
_
e
i ∆ϕ/2
+)¸− e
i ∆ϕ/2
+ e
−i ∆ϕ/2
−)¸+ e
−i ∆ϕ/2
_
, (5.153)
which reduces to
S
x
cos ∆ϕ−S
y
sin∆ϕ, (5.154)
where use has been made of Eqs. (5.135)–(5.137).
A second approach is to use the so called BakerHausdorff lemma. This takes
the form
exp( i Gλ) A exp(−i Gλ) = A+ i λ[G, A] +
_
_
i
2
λ
2
2!
_
_
[G, [G, A]] + (5.155)
+
_
i
n
λ
n
n!
_
[G, [G, [G, [G, A]]] ],
where G is a Hermitian operator, and λ is a real parameter. The proof of this
lemma is left as an exercise. Applying the BakerHausdorff lemma to Eq. (5.151),
we obtain
S
x
+
_
i ∆ϕ
¯h
_
[S
z
, S
x
] +
_
1
2!
_ _
i ∆ϕ
¯h
_
2
[S
z
, [S
z
, S
x
]] + , (5.156)
which reduces to
S
x
_
_
1 −
∆ϕ
2
2!
+
_
_
−S
y
_
_
ϕ−
∆ϕ
3
3!
+
_
_
, (5.157)
or
S
x
cos ∆ϕ−S
y
sin∆ϕ, (5.158)
where use has been made of Eq. (5.125). The second proof is more general than
the ﬁrst, since it only uses the fundamental commutation relation (5.125), and is,
therefore, valid for systems with spin angular momentum higher than onehalf.
94
5.9 Rotation operators in spin space 5 ANGULAR MOMENTUM
For a spin onehalf system, both methods imply that
¸S
x
) →¸S
x
) cos ∆ϕ−¸S
y
) sin∆ϕ (5.159)
under the action of the rotation operator (5.148). It is straightforward to show
that
¸S
y
) →¸S
y
) cos ∆ϕ+¸S
x
) sin∆ϕ. (5.160)
Furthermore,
¸S
z
) →¸S
z
), (5.161)
since S
z
commutes with the rotation operator. Equations (5.159)–(5.161) demon
strate that the operator (5.148) rotates the expectation value of S by an angle ∆ϕ
about the zaxis. In fact, the expectation value of the spin operator behaves like
a classical vector under rotation:
¸S
k
) →
l
R
kl
¸S
l
), (5.162)
where the R
kl
are the elements of the conventional rotation matrix for the rotation
in question. It is clear, from our second derivation of the result (5.159), that this
property is not restricted to the spin operators of a spin onehalf system. In fact,
we have effectively demonstrated that
¸J
k
) →
l
R
kl
¸J
l
), (5.163)
where the J
k
are the generators of rotation, satisfying the fundamental commuta
tion relation J J = i ¯hJ, and the rotation operator about the kth axis is written
R
k
(∆ϕ) = exp(−i J
k
∆ϕ/¯h).
Consider the effect of the rotation operator (5.148) on the state ket (5.147).
It is easily seen that
T
z
(∆ϕ)A) = e
−i ∆ϕ/2
¸+A)+) + e
i ∆ϕ/2
¸−A)−). (5.164)
Consider a rotation by 2π radians. We ﬁnd that
A) →T
z
(2π)A) = −A). (5.165)
95
5.10 Magnetic moments 5 ANGULAR MOMENTUM
Note that a ket rotated by 2π radians differs from the original ket by a minus
sign. In fact, a rotation by 4π radians is needed to transform a ket into itself.
The minus sign does not affect the expectation value of S, since S is sandwiched
between ¸A and A), both of which change sign. Nevertheless, the minus sign
does give rise to observable consequences, as we shall see presently.
5.10 Magnetic moments
Consider a particle of charge q and velocity v performing a circular orbit of radius
r in the xy plane. The charge is equivalent to a current loop of radius r in the
xy plane carrying current I = qv/2πr. The magnetic moment µ of the loop is of
magnitude πr
2
I and is directed along the zaxis. Thus, we can write
µ =
q
2
r v, (5.166)
where r and v are the vector position and velocity of the particle, respectively.
However, we know that p = v/m, where p is the vector momentum of the parti
cle, and m is its mass. We also know that L = rp, where L is the orbital angular
momentum. It follows that
µ =
q
2 m
L. (5.167)
Using the usual analogy between classical and quantum mechanics, we expect the
above relation to also hold between the quantum mechanical operators, µ and L,
which represent magnetic moment and orbital angular momentum, respectively.
This is indeed found to the the case.
Does spin angular momentum also give rise to a contribution to the magnetic
moment of a charged particle? The answer is “yes”. In fact, relativistic quantum
mechanics actually predicts that a charged particle possessing spin should also
possess a magnetic moment (this was ﬁrst demonstrated by Dirac). We can write
µ =
q
2 m
(L +gS) , (5.168)
where g is called the gyromagnetic ratio. For an electron this ratio is found to be
g
e
= 2
_
_
1 +
1
2π
e
2
4π
0
¯hc
_
_
. (5.169)
96
5.11 Spin precession 5 ANGULAR MOMENTUM
The factor 2 is correctly predicted by Dirac’s relativistic theory of the electron.
The small correction 1/(2π137), derived originally by Schwinger, is due to quan
tum ﬁeld effects. We shall ignore this correction in the following, so
µ · −
e
2 m
e
(L +2 S) (5.170)
for an electron (here, e > 0).
5.11 Spin precession
The Hamiltonian for an electron at rest in a zdirected magnetic ﬁeld, B = B ^z, is
H = −µB =
_
e
m
e
_
SB = ωS
z
, (5.171)
where
ω =
e B
m
e
. (5.172)
According to Eq. (4.28), the time evolution operator for this system is
T(t, 0) = exp(−i Ht/¯h) = exp(−i S
z
ωt/¯h). (5.173)
It can be seen, by comparison with Eq. (5.148), that the time evolution operator
is precisely the same as the rotation operator for spin, with ∆ϕ set equal to ωt.
It is immediately clear that the Hamiltonian (5.171) causes the electron spin to
precess about the zaxis with angular frequency ω. In fact, Eqs. (5.159)–(5.161)
imply that
¸S
x
)
t
= ¸S
x
)
t=0
cos ωt −¸S
y
)
t=0
sinωt, (5.174)
¸S
y
)
t
= ¸S
y
)
t=0
cos ωt +¸S
x
)
t=0
sinωt, (5.175)
¸S
z
)
t
= ¸S
z
)
t=0
. (5.176)
The time evolution of the state ket is given by analogy with Eq. (5.164):
A, t) = e
−i ωt/2
¸+A, 0)+) + e
i ωt/2
¸−A, 0)−). (5.177)
97
5.11 Spin precession 5 ANGULAR MOMENTUM
Note that it takes time t = 4π/ω for the state ket to return to its original state. By
contrast, it only takes times t = 2π/ω for the spin vector to point in its original
direction.
We now describe an experiment to detect the minus sign in Eq. (5.165). An
almost monoenergetic beam of neutrons is split in two, sent along two different
paths, A and B, and then recombined. Path A goes through a magnetic ﬁeld free
region. However, path B enters a small region where a static magnetic ﬁeld is
present. As a result, a neutron state ket going along path B acquires a phaseshift
exp(∓i ωT/2) (the ∓ signs correspond to s
z
= ±1/2 states). Here, T is the time
spent in the magnetic ﬁeld, and ω is the spin precession frequency
ω =
g
n
e B
m
p
. (5.178)
This frequency is deﬁned in an analogous manner to Eq. (5.172). The gyro
magnetic ratio for a neutron is found experimentally to be g
n
= −1.91. (The
magnetic moment of a neutron is entirely a quantum ﬁeld effect). When neu
trons from path A and path B meet they undergo interference. We expect the
observed neutron intensity in the interference region to exhibit a cos(±ωT/2+δ)
variation, where δ is the phase difference between paths A and B in the absence
of a magnetic ﬁeld. In experiments, the time of ﬂight T through the magnetic
ﬁeld region is kept constant, while the ﬁeldstrength B is varied. It follows that
the change in magnetic ﬁeld required to produce successive maxima is
∆B =
4π¯h
e g
n
¯ λ l
, (5.179)
where l is the pathlength through the magnetic ﬁeld region, and ¯ λ is the de
Broglie wavelength over 2π of the neutrons. The above prediction has been ver
iﬁed experimentally to within a fraction of a percent. This prediction depends
crucially on the fact that it takes a 4π rotation to return a state ket to its original
state. If it only took a 2π rotation then ∆B would be half of the value given above,
which does not agree with the experimental data.
98
5.12 Pauli twocomponent formalism 5 ANGULAR MOMENTUM
5.12 Pauli twocomponent formalism
We have seen, in Sect. 5.4, that the eigenstates of orbital angular momentum
can be conveniently represented as spherical harmonics. In this representation,
the orbital angular momentum operators take the form of differential operators
involving only angular coordinates. It is conventional to represent the eigenstates
of spin angular momentum as column (or row) matrices. In this representation,
the spin angular momentum operators take the form of matrices.
The matrix representation of a spin onehalf system was introduced by Pauli in
1926. Recall, from Sect. 5.9, that a general spin ket can be expressed as a linear
combination of the two eigenkets of S
z
belonging to the eigenvalues ±¯h/2. These
are denoted ±). Let us represent these basis eigenkets as column matrices:
+) →
_
_
1
0
_
_
≡ χ
+
, (5.180)
−) →
_
_
0
1
_
_
≡ χ
−
. (5.181)
The corresponding eigenbras are represented as row matrices:
¸+ → (1, 0) ≡ χ
†
+
, (5.182)
¸− → (0, 1) ≡ χ
†
−
. (5.183)
In this scheme, a general bra takes the form
A) = ¸+A)+) +¸−A)−) →
_
_
¸+A)
¸−A)
_
_
, (5.184)
and a general ket becomes
¸A = ¸A+)¸+ +¸A−)¸− →(¸A+), ¸A−)). (5.185)
The column matrix (5.184) is called a twocomponent spinor, and can be written
χ ≡
_
_
¸+A)
¸−A)
_
_
=
_
_
c
+
c
−
_
_
= c
+
χ
+
+c
−
χ
−
, (5.186)
99
5.12 Pauli twocomponent formalism 5 ANGULAR MOMENTUM
where the c
±
are complex numbers. The row matrix (5.185) becomes
χ
†
≡ (¸A+), ¸A−)) = (c
∗
+
, c
∗
−
) = c
∗
+
χ
†
+
+c
∗
−
χ
†
−
. (5.187)
Consider the ket obtained by the action of a spin operator on ket A:
A
) = S
k
A). (5.188)
This ket is represented as
A
) →
_
_
¸+A
)
¸−A
)
_
_
≡ χ
. (5.189)
However,
¸+A
) = ¸+S
k
+)¸+A) +¸+S
k
−)¸−A), (5.190)
¸−A
) = ¸−S
k
+)¸+A) +¸−S
k
−)¸−A), (5.191)
or
_
_
¸+A
)
¸−A
)
_
_
=
_
_
¸+S
k
+) ¸+S
k
−)
¸−S
k
+) ¸−S
k
−)
_
_
_
_
¸+A)
¸−A)
_
_
. (5.192)
It follows that we can represent the operator/ket relation (5.188) as the matrix
relation
χ
=
_
¯h
2
_
σ
k
χ, (5.193)
where the σ
k
are the matrices of the ¸±S
k
±) values divided by ¯h/2. These
matrices, which are called the Pauli matrices, can easily be evaluated using the
explicit forms for the spin operators given in Eqs. (5.135)–(5.137). We ﬁnd that
σ
1
=
_
_
0 1
1 0
_
_
, (5.194)
σ
2
=
_
_
0 −i
i 0
_
_
, (5.195)
σ
3
=
_
_
1 0
0 −1
_
_
. (5.196)
100
5.12 Pauli twocomponent formalism 5 ANGULAR MOMENTUM
Here, 1, 2, and 3 refer to x, y, and z, respectively. Note that, in this scheme, we
are effectively representing the spin operators in terms of the Pauli matrices:
S
k
→
_
¯h
2
_
σ
k
. (5.197)
The expectation value of S
k
can be written in terms of spinors and the Pauli
matrices:
¸S
k
) = ¸AS
k
A) =
±
¸A±)¸±S
k
±)¸±A) =
_
¯h
2
_
χ
†
σ
k
χ. (5.198)
The fundamental commutation relation for angular momentum, Eq. (5.125),
can be combined with (5.197) to give the following commutation relation for the
Pauli matrices:
σ σ = 2 i σ. (5.199)
It is easily seen that the matrices (5.194)–(5.196) actually satisfy these relations
(i.e., σ
1
σ
2
−σ
2
σ
1
= 2 i σ
3
, plus all cyclic permutations). It is also easily seen that
the Pauli matrices satisfy the anticommutation relations
{σ
i
, σ
j
} = 2 δ
ij
. (5.200)
Let us examine how the Pauli scheme can be extended to take into account the
position of a spin onehalf particle. Recall, from Sect. 5.8, that we can represent
a general basis ket as the product of basis kets in position space and spin space:
x
, y
, z
, ±) = x
, y
, z
)±) = ±)x
, y
, z
). (5.201)
The ket corresponding to state A is denoted A)), and resides in the product
space of the position and spin ket spaces. State A is completely speciﬁed by the
two wavefunctions
ψ
+
(x
, y
, z
) = ¸x
, y
, z
¸+A)), (5.202)
ψ
−
(x
, y
, z
) = ¸x
, y
, z
¸−A)). (5.203)
Consider the operator relation
A
)) = S
k
A)). (5.204)
101
5.12 Pauli twocomponent formalism 5 ANGULAR MOMENTUM
It is easily seen that
¸x
, y
, z
¸+A
)) = ¸+S
k
+)¸x
, y
, z
¸+A))
+¸+S
k
−)¸x
, y
, z
¸−A)), (5.205)
¸x
, y
, z
¸−A
)) = ¸−S
k
+)¸x
, y
, z
¸+A))
+¸−S
k
−)¸x
, y
, z
¸−A)), (5.206)
where use has been made of the fact that the spin operator S
k
commutes with
the eigenbras ¸x
, y
, z
. It is fairly obvious that we can represent the operator
relation (5.204) as a matrix relation if we generalize our deﬁnition of a spinor by
writing
A)) →
_
_
ψ
+
(r
)
ψ
−
(r
)
_
_
≡ χ, (5.207)
and so on. The components of a spinor are now wavefunctions, instead of com
plex numbers. In this scheme, the operator equation (5.204) becomes simply
χ
=
_
¯h
2
_
σ
k
χ. (5.208)
Consider the operator relation
A
)) = p
k
A)). (5.209)
In the Schr¨ odinger representation, we have
¸x
, y
, z
¸+A
)) = ¸x
, y
, z
p
k
¸+A))
= −i ¯h
∂
∂x
k
¸x
, y
, z
¸+A)), (5.210)
¸x
, y
, z
¸−A
)) = ¸x
, y
, z
p
k
¸−A))
= −i ¯h
∂
∂x
k
¸x
, y
, z
¸−A)), (5.211)
where use has been made of Eq. (3.78). The above equation reduces to
_
_
ψ
+
(r
)
ψ
−
(r
)
_
_
=
_
_
−i ¯h∂ψ
+
(r
)/∂x
k
−i ¯h∂ψ
−
(r
)/∂x
k
_
_
. (5.212)
102
5.12 Pauli twocomponent formalism 5 ANGULAR MOMENTUM
Thus, the operator equation (5.209) can be written
χ
= p
k
χ, (5.213)
where
p
k
→−i ¯h
∂
∂x
k
I. (5.214)
Here, I is the 2 2 unit matrix. In fact, any position operator (e.g., p
k
or L
k
)
is represented in the Pauli scheme as some differential operator of the position
eigenvalues multiplied by the 2 2 unit matrix.
What about combinations of position and spin operators? The most commonly
occurring combination is a dot product: e.g., S L = (¯h/2) σ L. Consider the
hybrid operator σa, where a ≡ (a
x
, a
y
, a
z
) is some vector position operator. This
quantity is represented as a 2 2 matrix:
σa ≡
k
a
k
σ
k
=
_
_
+a
3
a
1
− i a
2
a
1
+ i a
2
−a
3
_
_
. (5.215)
Since, in the Schr¨ odinger representation, a general position operator takes the
form of a differential operator in x
, y
, or z
, it is clear that the above quantity
must be regarded as a matrix differential operator which acts on spinors of the
general form (5.207). The important identity
(σa) (σb) = ab + i σ(a b) (5.216)
follows fromthe commutation and anticommutation relations (5.199) and (5.200).
Thus,
j
σ
j
a
j
k
σ
k
b
k
=
j
k
_
1
2
{σ
j
, σ
k
} +
1
2
[σ
j
, σ
k
]
_
a
j
b
k
=
j
k
(σ
jk
+ i
jkl
σ
l
) a
j
b
k
= ab + i σ(a b). (5.217)
A general rotation operator in spin space is written
T(∆φ) = exp(−i Sn∆ϕ/¯h) , (5.218)
103
5.12 Pauli twocomponent formalism 5 ANGULAR MOMENTUM
by analogy with Eq. (5.148), where n is a unit vector pointing along the axis of
rotation, and ∆ϕ is the angle of rotation. Here, n can be regarded as a trivial
position operator. The rotation operator is represented
exp(−i Sn∆ϕ/¯h) →exp(−i σn∆ϕ/2) (5.219)
in the Pauli scheme. The term on the righthand side of the above expression is
the exponential of a matrix. This can easily be evaluated using the Taylor series
for an exponential, plus the rules
(σn)
n
= 1 for n even, (5.220)
(σn)
n
= (σn) for n odd. (5.221)
These rules follow trivially from the identity (5.216). Thus, we can write
exp (−i σn∆ϕ/2) =
_
_
1 −
(σn)
2
2!
_
∆ϕ
2
_
2
+
(σn)
4
4!
_
∆ϕ
2
_
4
+
_
_
−i
_
_
(σ n)
_
∆ϕ
2
_
−
(σn)
3
3!
_
∆ϕ
2
_
3
+
_
_
= cos(∆ϕ/2) I − i sin(∆ϕ/2) σn. (5.222)
The explicit 2 2 form of this matrix is
_
_
cos(∆ϕ/2) − i n
z
sin(∆ϕ/2) (−i n
x
−n
y
) sin(∆ϕ/2)
(−i n
x
+n
y
) sin(∆ϕ/2) cos(∆ϕ/2) + i n
z
sin(∆ϕ/2)
_
_
. (5.223)
Rotation matrices act on spinors in much the same manner as the corresponding
rotation operators act on state kets. Thus,
χ
= exp(−i σn∆ϕ/2) χ, (5.224)
where χ
denotes the spinor obtained after rotating the spinor χ an angle ∆ϕ
about the naxis. The Pauli matrices remain unchanged under rotations. How
ever, the quantity χ
†
σ
k
χ is proportional to the expectation value of S
k
[see
Eq. (5.198)], so we would expect it to transform like a vector under rotation
(see Sect. 5.9). In fact, we require
(χ
†
σ
k
χ)
≡ (χ
†
)
σ
k
χ
=
l
R
kl
(χ
†
σ
l
χ), (5.225)
104
5.13 Spin greater than onehalf systems 5 ANGULAR MOMENTUM
where the R
kl
are the elements of a conventional rotation matrix. This is easily
demonstrated, since
exp
_
i σ
3
∆ϕ
2
_
σ
1
exp
_
−i σ
3
∆ϕ
2
_
= σ
1
cos ∆ϕ−σ
2
sin∆ϕ (5.226)
plus all cyclic permutations. The above expression is the 2 2 matrix analogue
of (see Sect. 5.9)
exp
_
i S
z
∆ϕ
¯h
_
S
x
exp
_
−i S
z
∆ϕ
¯h
_
= S
x
cos ∆ϕ−S
y
sin∆ϕ. (5.227)
The previous two formulae can both be validated using the BakerHausdorff
lemma, (5.156), which holds for Hermitian matrices, in addition to Hermitian
operators.
5.13 Spin greater than onehalf systems
In the absence of spin, the Hamiltonian can be written as some function of the
position and momentum operators. Using the Schr¨ odinger representation, in
which p →−i ¯h∇, the energy eigenvalue problem,
HE) = EE), (5.228)
can be transformed into a partial differential equation for the wavefunction
ψ(r
) ≡ ¸r
E). This function speciﬁes the probability density for observing the
particle at a given position, r
. In general, we ﬁnd
Hψ = Eψ, (5.229)
where H is now a partial differential operator. The boundary conditions (for a
bound state) are obtained from the normalization constraint
_
ψ
2
dV = 1. (5.230)
This is all very familiar. However, we now know how to generalize this scheme
to deal with a spin onehalf particle. Instead of representing the state of the par
ticle by a single wavefunction, we use two wavefunctions. The ﬁrst, ψ
+
(r
),
105
5.13 Spin greater than onehalf systems 5 ANGULAR MOMENTUM
speciﬁes the probability density of observing the particle at position r
with spin
angular momentum +¯h/2 in the zdirection. The second, ψ
−
(r
), speciﬁes the
probability density of observing the particle at position r
with spin angular mo
mentum −¯h/2 in the zdirection. In the Pauli scheme, these wavefunctions are
combined into a spinor, χ, which is simply the row vector of ψ
+
and ψ
−
. In gen
eral, the Hamiltonian is a function of the position, momentum, and spin opera
tors. Adopting the Schr¨ odinger representation, and the Pauli scheme, the energy
eigenvalue problem reduces to
Hχ = Eχ, (5.231)
where χ is a spinor (i.e., a 1 2 matrix of wavefunctions) and H is a 2 2 matrix
partial differential operator [see Eq. (5.215)]. The above spinor equation can
always be written out explicitly as two coupled partial differential equations for
ψ
+
and ψ
−
.
Suppose that the Hamiltonian has no dependence on the spin operators. In this
case, the Hamiltonian is represented as diagonal 2 2 matrix partial differential
operator in the Schr¨ odinger/Pauli scheme [see Eq. (5.214)]. In other words, the
partial differential equation for ψ
+
decouples from that for ψ
−
. In fact, both
equations have the same form, so there is only really one differential equation.
In this situation, the most general solution to Eq. (5.231) can be written
χ = ψ(r
)
_
_
c
+
c
−
_
_
. (5.232)
Here, ψ(r
) is determined by the solution of the differential equation, and the
c
±
are arbitrary complex numbers. The physical signiﬁcance of the above ex
pression is clear. The Hamiltonian determines the relative probabilities of ﬁnding
the particle at various different positions, but the direction of its spin angular
momentum remains undetermined.
Suppose that the Hamiltonian depends only on the spin operators. In this
case, the Hamiltonian is represented as a 2 2 matrix of complex numbers in
the Schr¨ odinger/Pauli scheme [see Eq. (5.197)], and the spinor eigenvalue equa
tion (5.231) reduces to a straightforward matrix eigenvalue problem. The most
106
5.13 Spin greater than onehalf systems 5 ANGULAR MOMENTUM
general solution can again be written
χ = ψ(r
)
_
_
c
+
c
−
_
_
. (5.233)
Here, the ratio c
+
/c
−
is determined by the matrix eigenvalue problem, and the
wavefunction ψ(r
) is arbitrary. Clearly, the Hamiltonian determines the direc
tion of the particle’s spin angular momentum, but leaves its position undeter
mined.
In general, of course, the Hamiltonian is a function of both position and
spin operators. In this case, it is not possible to decompose the spinor as in
Eqs. (5.232) and (5.233). In other words, a general Hamiltonian causes the di
rection of the particle’s spin angular momentum to vary with position in some
speciﬁed manner. This can only be represented as a spinor involving different
wavefunctions, ψ
+
and ψ
−
.
But, what happens if we have a spin one or a spin threehalves particle? It
turns out that we can generalize the Pauli twocomponent scheme in a fairly
straightforward manner. Consider a spins particle: i.e., a particle for which the
eigenvalue of S
2
is s (s + 1) ¯h
2
. Here, s is either an integer, or a halfinteger.
The eigenvalues of S
z
are written s
z
¯h, where s
z
is allowed to take the values
s, s − 1, , −s + 1, −s. In fact, there are 2 s + 1 distinct allowed values of s
z
.
Not surprisingly, we can represent the state of the particle by 2 s + 1 different
wavefunctions, denoted ψ
s
z
(r
). Here, ψ
s
z
(r
) speciﬁes the probability density
for observing the particle at position r
with spin angular momentum s
z
¯h in the
zdirection. More exactly,
ψ
s
z
(r
) = ¸r
¸s, s
z
A)), (5.234)
where A)) denotes a state ket in the product space of the position and spin op
erators. The state of the particle can be represented more succinctly by a spinor,
χ, which is simply the 2 s + 1 component row vector of the ψ
s
z
(r
). Thus, a spin
onehalf particle is represented by a twocomponent spinor, a spin one particle
by a threecomponent spinor, a spin threehalves particle by a fourcomponent
spinor, and so on.
107
5.13 Spin greater than onehalf systems 5 ANGULAR MOMENTUM
In this extended Schr¨ odinger/Pauli scheme, position space operators take the
form of diagonal (2 s + 1) (2 s + 1) matrix differential operators. Thus, we can
represent the momentum operators as [see Eq. (5.214)]
p
k
→−i ¯h
∂
∂x
k
I, (5.235)
where I is the (2 s +1) (2 s +1) unit matrix. We represent the spin operators as
S
k
→s ¯hσ
k
, (5.236)
where the (2 s +1) (2 s +1) extended Pauli matrix σ
k
has elements
(σ
k
)
jl
=
¸s, jS
k
s, l)
s ¯h
. (5.237)
Here, j, l are integers, or halfintegers, lying in the range −s to +s. But, how can
we evaluate the brackets ¸s, jS
k
s, l) and, thereby, construct the extended Pauli
matrices? In fact, it is trivial to construct the σ
z
matrix. By deﬁnition,
S
z
s, j) = j ¯hs, j). (5.238)
Hence,
(σ
z
)
jl
=
¸s, jS
z
s, l)
s ¯h
=
j
s
δ
jl
, (5.239)
where use has been made of the orthonormality property of the s, j). Thus, σ
z
is the suitably normalized diagonal matrix of the eigenvalues of S
z
. The matrix
elements of σ
x
and σ
y
are most easily obtained by considering the shift operators,
S
±
= S
x
±i S
y
. (5.240)
We know, from Eqs. (5.55)–(5.56), that
S
+
s, j) =
_
s (s +1) −j (j +1) ¯hs, j +1), (5.241)
S
−
s, j) =
_
s (s +1) −j (j −1) ¯hs, j −1). (5.242)
It follows from Eqs. (5.237), and (5.240)–(5.242), that
(σ
x
)
jl
=
_
s (s +1) −j (j −1) δ
j,l+1
2 s
108
5.13 Spin greater than onehalf systems 5 ANGULAR MOMENTUM
+
_
s (s +1) −j (j +1) δ
j,l−1
2 s
, (5.243)
(σ
y
)
jl
=
_
s (s +1) −j (j −1) δ
j,l+1
2 i s
−
_
s (s +1) −j (j +1) δ
j,l−1
2 i s
. (5.244)
According to Eqs. (5.239) and (5.243)–(5.244), the Pauli matrices for a spin one
half (s = 1/2) particle are
σ
x
=
_
_
0 1
1 0
_
_
, (5.245)
σ
y
=
_
_
0 −i
i 0
_
_
, (5.246)
σ
z
=
_
_
1 0
0 −1
_
_
, (5.247)
as we have seen previously. For a spin one (s = 1) particle, we ﬁnd that
σ
x
=
1
√
2
_
_
_
_
_
0 1 0
1 0 1
0 1 0
_
_
_
_
_
, (5.248)
σ
y
=
1
√
2
_
_
_
_
_
0 −i 0
i 0 −i
0 i 0
_
_
_
_
_
, (5.249)
σ
z
=
_
_
_
_
_
1 0 0
0 0 0
0 0 −1
_
_
_
_
_
. (5.250)
In fact, we can now construct the Pauli matrices for a spin anything particle.
This means that we can convert the general energy eigenvalue problem for a
spins particle, where the Hamiltonian is some function of position and spin op
erators, into 2 s + 1 coupled partial differential equations involving the 2 s + 1
wavefunctions ψ
s
z
(r
). Unfortunately, such a system of equations is generally
too complicated to solve exactly.
109
5.14 Addition of angular momentum 5 ANGULAR MOMENTUM
5.14 Addition of angular momentum
Consider a hydrogen atom in an l = 1 state. The electron possesses orbital angu
lar momentum of magnitude ¯h, and spin angular momentum of magnitude ¯h/2.
So, what is the total angular momentum of the system?
In order to answer this question, we are going to have to learn how to add
angular momentum operators. Let us consider the most general case. Suppose
that we have two sets of angular momentum operators, J
1
and J
2
. By deﬁnition,
these operators are Hermitian, and obey the fundamental commutation relations
J
1
J
1
= i ¯hJ
1
, (5.251)
J
2
J
2
= i ¯hJ
2
. (5.252)
We assume that the two groups of operators correspond to different degrees of
freedom of the system, so that
[J
1i
, J
2j
] = 0, (5.253)
where i, j stand for either x, y, or z. For instance, J
1
could be an orbital angular
momentum operator, and J
2
a spin angular momentum operator. Alternatively, J
1
and J
2
could be the orbital angular momentum operators of two different parti
cles in a multiparticle system. We know, from the general properties of angular
momentum, that the eigenvalues of J
2
1
and J
2
2
can be written j
1
(j
1
+ 1) ¯h
2
and
j
2
(j
2
+1) ¯h
2
, respectively, where j
1
and j
2
are either integers, or halfintegers. We
also know that the eigenvalues of J
1z
and J
2z
take the form m
1
¯h and m
2
¯h, respec
tively, where m
1
and m
2
are numbers lying in the ranges j
1
, j
1
−1, , −j
1
+1, −j
1
and j
2
, j
2
−1, , −j
2
+1, −j
2
, respectively.
Let us deﬁne the total angular momentum operator
J = J
1
+J
2
. (5.254)
Now J is an Hermitian operator, since it is the sum of Hermitian operators. Ac
cording to Eqs. (5.11) and (5.14), J satisﬁes the fundamental commutation rela
tion
J J = i ¯hJ. (5.255)
110
5.14 Addition of angular momentum 5 ANGULAR MOMENTUM
Thus, J possesses all of the expected properties of an angular momentum opera
tor. It follows that the eigenvalue of J
2
can be written j (j + 1) ¯h
2
, where j is an
integer, or a halfinteger. The eigenvalue of J
z
takes the form m¯h, where m lies
in the range j, j −1, , −j +1, −j. At this stage, we do not know the relationship
between the quantum numbers of the total angular momentum, j and m, and
those of the individual angular momenta, j
1
, j
2
, m
1
, and m
2
.
Now
J
2
= J
2
1
+J
2
2
+2 J
1
J
2
. (5.256)
We know that
[J
2
1
, J
1i
] = 0, (5.257)
[J
2
2
, J
2i
] = 0, (5.258)
and also that all of the J
1i
operators commute with the J
2i
operators. It follows
from Eq. (5.256) that
[J
2
, J
2
1
] = [J
2
, J
2
2
] = 0. (5.259)
This implies that the quantum numbers j
1
, j
2
, and j can all be measured simulta
neously. In other words, we can know the magnitude of the total angular momen
tum together with the magnitudes of the component angular momenta. However,
it is clear from Eq. (5.256) that
[J
2
, J
1z
] ,= 0, (5.260)
[J
2
, J
2z
] ,= 0. (5.261)
This suggests that it is not possible to measure the quantum numbers m
1
and
m
2
simultaneously with the quantum number j. Thus, we cannot determine the
projections of the individual angular momenta along the zaxis at the same time
as the magnitude of the total angular momentum.
It is clear, from the preceding discussion, that we can form two alternate
groups of mutually commuting operators. The ﬁrst group is J
2
1
, J
2
2
, J
1z
, and J
2z
.
The second group is J
2
1
, J
2
2
, J
2
, and J
z
. These two groups of operators are in
compatible with one another. We can deﬁne simultaneous eigenkets of each
operator group. The simultaneous eigenkets of J
2
1
, J
2
2
, J
1z
, and J
2z
are denoted
111
5.14 Addition of angular momentum 5 ANGULAR MOMENTUM
j
1
, j
2
; m
1
, m
2
), where
J
2
1
j
1
, j
2
; m
1
, m
2
) = j
1
(j
1
+1) ¯h
2
j
1
, j
2
; m
1
, m
2
), (5.262)
J
2
2
j
1
, j
2
; m
1
, m
2
) = j
2
(j
2
+1) ¯h
2
j
1
, j
2
; m
1
, m
2
), (5.263)
J
1z
j
1
, j
2
; m
1
, m
2
) = m
1
¯hj
1
, j
2
; m
1
, m
2
), (5.264)
J
2z
j
1
, j
2
; m
1
, m
2
) = m
2
¯hj
1
, j
2
; m
1
, m
2
). (5.265)
The simultaneous eigenkets of J
2
1
, J
2
2
, J
2
and J
z
are denoted j
1
, j
2
; j, m), where
J
2
1
j
1
, j
2
; j, m) = j
1
(j
1
+1) ¯h
2
j
1
, j
2
; j, m), (5.266)
J
2
2
j
1
, j
2
; j, m) = j
2
(j
2
+1) ¯h
2
j
1
, j
2
; j, m), (5.267)
J
2
j
1
, j
2
; j, m) = j (j +1) ¯h
2
j
1
, j
2
; j, m), (5.268)
J
z
j
1
, j
2
; j, m) = m¯hj
1
, j
2
; j, m). (5.269)
Each set of eigenkets are complete, mutually orthogonal (for eigenkets corre
sponding to different sets of eigenvalues), and have unit norms. Since the op
erators J
2
1
and J
2
2
are common to both operator groups, we can assume that the
quantum numbers j
1
and j
2
are known. In other words, we can always determine
the magnitudes of the individual angular momenta. In addition, we can either
know the quantum numbers m
1
and m
2
, or the quantum numbers j and m, but
we cannot know both pairs of quantum numbers at the same time. We can write
a conventional completeness relation for both sets of eigenkets:
m
1
m
2
j
1
, j
2
; m
1
, m
2
)¸j
1
, j
2
; m
1
, m
2
 = 1, (5.270)
j
m
j
1
, j
2
; j, m)¸j
1
, j
2
; j, m = 1, (5.271)
where the righthand sides denote the identity operator in the ket space corre
sponding to states of given j
1
and j
2
. The summation is over all allowed values of
m
1
, m
2
, j, and m.
The operator group J
2
1
, J
2
2
, J
2
, and J
z
is incompatible with the group J
2
1
, J
2
2
,
J
1z
, and J
2z
. This means that if the system is in a simultaneous eigenstate of the
former group then, in general, it is not in an eigenstate of the latter. In other
112
5.14 Addition of angular momentum 5 ANGULAR MOMENTUM
words, if the quantum numbers j
1
, j
2
, j, and m are known with certainty, then a
measurement of the quantum numbers m
1
and m
2
will give a range of possible
values. We can use the completeness relation (5.270) to write
j
1
, j
2
; j, m) =
m
1
m
2
¸j
1
, j
2
; m
1
, m
2
j
1
, j
2
; j, m)j
1
, j
2
; m
1
, m
2
). (5.272)
Thus, we can write the eigenkets of the ﬁrst group of operators as a weighted
sum of the eigenkets of the second set. The weights, ¸j
1
, j
2
; m
1
, m
2
j
1
, j
2
; j, m),
are called the ClebschGordon coefﬁcients. If the system is in a state where a
measurement of J
2
1
, J
2
2
, J
2
, and J
z
is bound to give the results j
1
(j
1
+1) ¯h
2
, j
2
(j
2
+
1) ¯h
2
, j (j + 1) ¯h
2
, and j
z
¯h, respectively, then a measurement of J
1z
and J
2z
will
give the results m
1
¯h and m
2
¯h with probability ¸j
1
, j
2
; m
1
, m
2
j
1
, j
2
; j, m)
2
.
The ClebschGordon coefﬁcients possess a number of very important proper
ties. First, the coefﬁcients are zero unless
m = m
1
+m
2
. (5.273)
To prove this, we note that
(J
z
−J
1z
−J
2z
)j
1
, j
2
; j, m) = 0. (5.274)
Forming the inner product with ¸j
1
, j
2
; m
1
, m
2
, we obtain
(m−m
1
−m
2
)¸j
1
, j
2
; m
1
, m
2
j
1
, j
2
; j, m) = 0, (5.275)
which proves the assertion. Thus, the zcomponents of different angular mo
menta add algebraically. So, an electron in an l = 1 state, with orbital angular
momentum ¯h, and spin angular momentum ¯h/2, projected along the zaxis, con
stitutes a state whose total angular momentum projected along the zaxis is 3¯h/2.
What is uncertain is the magnitude of the total angular momentum.
Second, the coefﬁcients vanish unless
j
1
−j
2
 ≤ j ≤ j
1
+j
2
. (5.276)
We can assume, without loss of generality, that j
1
≥ j
2
. We know, fromEq. (5.273),
that for given j
1
and j
2
the largest possible value of m is j
1
+ j
2
(since j
1
is the
113
5.14 Addition of angular momentum 5 ANGULAR MOMENTUM
largest possible value of m
1
, etc.). This implies that the largest possible value of j
is j
1
+j
2
(since, by deﬁnition, the largest value of m is equal to j). Now, there are
(2 j
1
+1) allowable values of m
1
and (2 j
2
+1) allowable values of m
2
. Thus, there
are (2 j
1
+ 1) (2 j
2
+ 1) independent eigenkets, j
1
, j
2
; m
1
, m
2
), needed to span the
ket space corresponding to ﬁxed j
1
and j
2
. Since the eigenkets j
1
, j
2
; j, m) span
the same space, they must also form a set of (2 j
1
+1) (2 j
2
+1) independent kets.
In other words, there can only be (2 j
1
+ 1) (2 j
2
+ 1) distinct allowable values of
the quantum numbers j and m. For each allowed value of j, there are 2 j + 1
allowed values of m. We have already seen that the maximum allowed value of
j is j
1
+ j
2
. It is easily seen that if the minimum allowed value of j is j
1
− j
2
then
the total number of allowed values of j and m is (2 j
1
+1) (2 j
2
+1): i.e.,
j
1
+j
2
j=j
1
−j
2
(2 j +1) ≡ (2 j
1
+1) (2 j
2
+1). (5.277)
This proves our assertion.
Third, the sum of the modulus squared of all of the ClebschGordon coefﬁ
cients is unity: i.e.,
m
1
m
2
¸j
1
, j
2
; m
1
, m
2
j
1
, j
2
; j, m)
2
= 1. (5.278)
This assertion is proved as follows:
¸j
1
, j
2
; j, mj
1
, j
2
; j, m) =
m
1
m
2
¸j
1
, j
2
; j, mj
1
, j
2
; m
1
, m
2
)¸j
1
, j
2
; m
1
, m
2
j
1
, j
2
; j, m)
=
m
1
m
2
¸j
1
, j
2
; m
1
, m
2
j
1
, j
2
; j, m)
2
= 1, (5.279)
where use has been made of the completeness relation (5.270).
Finally, the ClebschGordon coefﬁcients obey two recursion relations. To ob
tain these relations we start from
J
±
j
1
, j
2
; j, m) = (J
±
1
+J
±
2
) (5.280)
m
1
m
2
¸j
1
, j
2
; m
1
, m
2
j
1
, j
2
; j, m)j
1
, j
2
; m
1
, m
2
).
114
5.14 Addition of angular momentum 5 ANGULAR MOMENTUM
Making use of the wellknown properties of the shift operators, which are speci
ﬁed by Eqs. (5.55)–(5.56), we obtain
_
j (j +1) −m(m±1) j
1
, j
2
; j, m±1) =
m
1
m
2
_
_
j
1
(j
1
+1) −m
1
(m
1
±1) j
1
, j
2
; m
1
±1, m
2
)
+
_
j
2
(j
2
+1) −m
2
(m
2
±1) j
1
, j
2
; m
1
, m
2
±1)
_
¸j
1
, j
2
; m
1
, m
2
j
1
, j
2
; j, m). (5.281)
Taking the inner product with ¸j
1
, j
2
; m
1
, m
2
, and making use of the orthonormal
ity property of the basis eigenkets, we obtain the desired recursion relations:
_
j (j +1) −m(m±1) ¸j
1
, j
2
; m
1
, m
2
j
1
, j
2
; j, m±1) =
_
j
1
(j
1
+1) −m
1
(m
1
∓1) ¸j
1
, j
2
; m
1
∓1, m
2
j
1
, j
2
; j, m)
+
_
j
2
(j
2
+1) −m
2
(m
2
∓1) ¸j
1
, j
2
; m
1
, m
2
∓1j
1
, j
2
; j, m). (5.282)
It is clear, from the absence of complex coupling coefﬁcients in the above re
lations, that we can always choose the ClebschGordon coefﬁcients to be real
numbers. This is a convenient choice, since it ensures that the inverse Clebsch
Gordon coefﬁcients, ¸j
1
, j
2
; j, mj
1
, j
2
; m
1
, m
2
), are identical to the ClebschGordon
coefﬁcients. In other words,
¸j
1
, j
2
; j, mj
1
, j
2
; m
1
, m
2
) = ¸j
1
, j
2
; m
1
, m
2
j
1
, j
2
; j, m). (5.283)
The inverse ClebschGordon coefﬁcients are the weights in the expansion of the
j
1
, j
2
; m
1
, m
2
) in terms of the j
1
, j
2
; j, m):
j
1
, j
2
; m
1
, m
2
) =
j
m
¸j
1
, j
2
; j, mj
1
, j
2
; m
1
, m
2
)j
1
, j
2
; j, m). (5.284)
It turns out that the recursion relations (5.282), together with the normal
ization condition (5.278), are sufﬁcient to completely determine the Clebsch
Gordon coefﬁcients to within an arbitrary sign (multiplied into all of the coefﬁ
cients). This sign is ﬁxed by convention. The easiest way of demonstrating this
assertion is by considering some speciﬁc examples.
115
5.14 Addition of angular momentum 5 ANGULAR MOMENTUM
Let us add the angular momentum of two spin onehalf systems: e.g., two
electrons at rest. So, j
1
= j
2
= 1/2. We know, from general principles, that
m
1
 ≤ 1/2 and m
2
 ≤ 1/2. We also know, from Eq. (5.276), that 0 ≤ j ≤ 1,
where the allowed values of j differ by integer amounts. It follows that either
j = 0 or j = 1. Thus, two spin onehalf systems can be combined to form either a
spin zero system or a spin one system. It is helpful to arrange all of the possibly
nonzero ClebschGordon coefﬁcients in a table:
m
1
m
2
1/2 1/2 ? ? ? ?
1/2 1/2 ? ? ? ?
1/2 1/2 ? ? ? ?
1/2 1/2 ? ? ? ?
j
1
=1/2 j 1 1 1 0
j
2
=1/2 m 1 0 1 0
The box in this table corresponding to m
1
= 1/2, m
2
= 1/2, j = 1, m = 1
gives the ClebschGordon coefﬁcient ¸1/2, 1/2; 1/2, 1/21/2, 1/2; 1, 1), or the in
verse ClebschGordon coefﬁcient ¸1/2, 1/2; 1, 11/2, 1/2; 1/2, 1/2). All the boxes
contain question marks because we do not know any ClebschGordon coefﬁcients
at the moment.
A ClebschGordon coefﬁcient is automatically zero unless m
1
+ m
2
= m. In
other words, the zcomponents of angular momentum have to add algebraically.
Many of the boxes in the above table correspond to m
1
+m
2
,= m. We immediately
conclude that these boxes must contain zeroes: i.e.,
m
1
m
2
1/2 1/2 ? 0 0 0
1/2 1/2 0 ? 0 ?
1/2 1/2 0 ? 0 ?
1/2 1/2 0 0 ? 0
j
1
=1/2 j 1 1 1 0
j
2
=1/2 m 1 0 1 0
The normalization condition (5.278) implies that the sum of the squares of all
116
5.14 Addition of angular momentum 5 ANGULAR MOMENTUM
the rows and columns of the above table must be unity. There are two rows and
two columns which only contain a single nonzero entry. We conclude that these
entries must be ±1, but we have no way of determining the signs at present.
Thus,
m
1
m
2
1/2 1/2 ±1 0 0 0
1/2 1/2 0 ? 0 ?
1/2 1/2 0 ? 0 ?
1/2 1/2 0 0 ±1 0
j
1
=1/2 j 1 1 1 0
j
2
=1/2 m 1 0 1 0
Let us evaluate the recursion relation (5.282) for j
1
= j
2
= 1/2, with j = 1,
m = 0, m
1
= m
2
= ±1/2, taking the upper/lower sign. We ﬁnd that
¸1/2, −1/21, 0) +¸−1/2, 1/21, 0) =
√
2 ¸1/2, 1/21, 1) = ±
√
2, (5.285)
and
¸1/2, −1/21, 0) +¸−1/2, 1/21, 0) =
√
2 ¸−1/2, −1/21, −1) = ±
√
2. (5.286)
Here, the j
1
and j
2
labels have been suppressed for ease of notation. We also
know that
¸1/2, −1/21, 0)
2
+¸−1/2, 1/21, 0)
2
= 1, (5.287)
from the normalization condition. The only real solutions to the above set of
equations are
√
2 ¸1/2, −1/21, 0) =
√
2 ¸−1/2, 1/21, 0)
= ¸1/2, 1/21, 1) = ¸1/2, 1/21, −1) = ±1. (5.288)
The choice of sign is arbitrary—the conventional choice is a positive sign. Thus,
117
5.14 Addition of angular momentum 5 ANGULAR MOMENTUM
our table now reads
m
1
m
2
1/2 1/2 1 0 0 0
1/2 1/2 0 1/
√
2 0 ?
1/2 1/2 0 1/
√
2 0 ?
1/2 1/2 0 0 1 0
j
1
=1/2 j 1 1 1 0
j
2
=1/2 m 1 0 1 0
We could ﬁll in the remaining unknown entries of our table by using the re
cursion relation again. However, an easier method is to observe that the rows
and columns of the table must all be mutually orthogonal. That is, the dot prod
uct of a row with any other row must be zero. Likewise, for the dot product of
a column with any other column. This follows because the entries in the table
give the expansion coefﬁcients of one of our alternative sets of eigenkets in terms
of the other set, and each set of eigenkets contains mutually orthogonal vectors
with unit norms. The normalization condition tells us that the dot product of a
row or column with itself must be unity. The only way that the dot product of the
fourth column with the second column can be zero is if the unknown entries are
equal and opposite. The requirement that the dot product of the fourth column
with itself is unity tells us that the magnitudes of the unknown entries have to
be 1/
√
2. The unknown entries are undetermined to an arbitrary sign multiplied
into them both. Thus, the ﬁnal form of our table (with the conventional choice
of arbitrary signs) is
m
1
m
2
1/2 1/2 1 0 0 0
1/2 1/2 0 1/
√
2 0 1/
√
2
1/2 1/2 0 1/
√
2 0 1/
√
2
1/2 1/2 0 0 1 0
j
1
=1/2 j 1 1 1 0
j
2
=1/2 m 1 0 1 0
The table can be read in one of two ways. The columns give the expansions of
118
5.14 Addition of angular momentum 5 ANGULAR MOMENTUM
the eigenstates of overall angular momentum in terms of the eigenstates of the
individual angular momenta of the two component systems. Thus, the second
column tells us that
1, 0) =
1
√
2
( 1/2, −1/2) +  −1/2, 1/2) ) . (5.289)
The ket on the lefthand side is a j, m) ket, whereas those on the righthand side
are m
1
, m
2
) kets. The rows give the expansions of the eigenstates of individual
angular momentum in terms of those of overall angular momentum. Thus, the
second row tells us that
1/2, −1/2) =
1
√
2
( 1, 0) + 0, 0) ) . (5.290)
Here, the ket on the lefthand side is a m
1
, m
2
) ket, whereas those on the right
hand side are j, m) kets.
Note that our table is really a combination of two subtables, one involving
j = 0 states, and one involving j = 1 states. The ClebschGordon coefﬁcients cor
responding to two different choices of j are completely independent: i.e., there is
no recursion relation linking ClebschGordon coefﬁcients corresponding to differ
ent values of j. Thus, for every choice of j
1
, j
2
, and j we can construct a table of
ClebschGordon coefﬁcients corresponding to the different allowed values of m
1
,
m
2
, and m (subject to the constraint that m
1
+m
2
= m). A complete knowledge
of angular momentum addition is equivalent to a knowing all possible tables of
ClebschGordon coefﬁcients. These tables are listed (for moderate values of j
1
, j
2
and j) in many standard reference books.
119
6 APPROXIMATION METHODS
6 Approximation methods
6.1 Introduction
We have developed techniques by which the general energy eigenvalue problem
can be reduced to a set of coupled partial differential equations involving various
wavefunctions. Unfortunately, the number of such problems which yield exactly
soluble equations is comparatively small. Clearly, we need to develop some tech
niques for ﬁnding approximate solutions to otherwise intractable problems.
Consider the following problem, which is very common. The Hamiltonian of a
system is written
H = H
0
+H
1
. (6.1)
Here, H
0
is a simple Hamiltonian for which we know the exact eigenvalues and
eigenstates. H
1
introduces some interesting additional physics into the problem,
but it is sufﬁciently complicated that when we add it to H
0
we can no longer
ﬁnd the exact energy eigenvalues and eigenstates. However, H
1
can, in some
sense (which we shall specify more exactly later on), be regarded as being small
compared to H
0
. Can we ﬁnd the approximate eigenvalues and eigenstates of the
modiﬁed Hamiltonian, H
0
+H
1
, by performing some sort of perturbation analysis
about the eigenvalues and eigenstates of the original Hamiltonian, H
0
? Let us
investigate.
6.2 The twostate system
Let us begin by considering timeindependent perturbation theory, in which the
modiﬁcation to the Hamiltonian, H
1
, has no explicit dependence on time. It is
usually assumed that the unperturbed Hamiltonian, H
0
, is also timeindependent.
Consider the simplest nontrivial system, in which there are only two indepen
dent eigenkets of the unperturbed Hamiltonian. These are denoted
H
0
1) = E
1
1), (6.2)
120
6.2 The twostate system 6 APPROXIMATION METHODS
H
0
2) = E
2
2). (6.3)
It is assumed that these states, and their associated eigenvalues, are known. Since
H
0
is, by deﬁnition, an Hermitian operator, its two eigenkets are orthonormal and
form a complete set. The lengths of these eigenkets are both normalized to unity.
Let us now try to solve the modiﬁed energy eigenvalue problem
(H
0
+H
1
)E) = EE). (6.4)
In fact, we can solve this problem exactly. Since, the eigenkets of H
0
form a
complete set, we can write
E) = ¸1E)1) +¸2E)2). (6.5)
Rightmultiplication of Eq. (6.4) by ¸1 and ¸2 yields two coupled equations,
which can be written in matrix form:
_
_
E
1
−E +e
11
e
12
e
∗
12
E
2
−E +e
22
_
_
_
_
¸1E)
¸2E)
_
_
=
_
_
0
0
_
_
. (6.6)
Here,
e
11
= ¸1H
1
1), (6.7)
e
22
= ¸2H
1
2), (6.8)
e
12
= ¸1H
1
2). (6.9)
In the special (but common) case of a perturbing Hamiltonian whose diagonal
matrix elements (in the unperturbed eigenstates) are zero, so that
e
11
= e
22
= 0, (6.10)
the solution of Eq. (6.6) (obtained by setting the determinant of the matrix equal
to zero) is
E =
(E
1
+E
2
) ±
_
(E
1
−E
2
)
2
+4 e
12

2
2
. (6.11)
Let us expand in the supposedly small parameter
=
e
12

E
1
−E
2

. (6.12)
121
6.3 Nondegenerate perturbation theory 6 APPROXIMATION METHODS
We obtain
E ·
1
2
(E
1
+E
2
) ±
1
2
(E
1
−E
2
)(1 +2
2
+ ). (6.13)
The above expression yields the modiﬁcations to the energy eigenvalues due to
the perturbing Hamiltonian:
E
1
= E
1
+
e
12

2
E
1
−E
2
+ , (6.14)
E
2
= E
2
−
e
12

2
E
1
−E
2
+ . (6.15)
Note that H
1
causes the upper eigenvalue to rise, and the lower eigenvalue to
fall. It is easily demonstrated that the modiﬁed eigenkets take the form
1)
= 1) +
e
∗
12
E
1
−E
2
2) + , (6.16)
2)
= 2) −
e
12
E
1
−E
2
1) + . (6.17)
Thus, the modiﬁed energy eigenstates consist of one of the unperturbed eigen
states with a slight admixture of the other. Note that the series expansion in
Eq. (6.13) only converges if 2  < 1. This suggests that the condition for the
validity of the perturbation expansion is
e
12
 <
E
1
−E
2

2
. (6.18)
In other words, when we say that H
1
needs to be small compared to H
0
, what we
really mean is that the above inequality needs to be satisﬁed.
6.3 Nondegenerate perturbation theory
Let us now generalize our perturbation analysis to deal with systems possessing
more than two energy eigenstates. The energy eigenstates of the unperturbed
Hamiltonian, H
0
, are denoted
H
0
n) = E
n
n), (6.19)
122
6.3 Nondegenerate perturbation theory 6 APPROXIMATION METHODS
where n runs from 1 to N. The eigenkets n) are orthonormal, form a complete
set, and have their lengths normalized to unity. Let us now try to solve the energy
eigenvalue problem for the perturbed Hamiltonian:
(H
0
+H
1
)E) = EE). (6.20)
We can express E) as a linear superposition of the unperturbed energy eigenkets,
E) =
k
¸kE)k), (6.21)
where the summation is from k = 1 to N. Substituting the above equation into
Eq. (6.20), and rightmultiplying by ¸m, we obtain
(E
m
+e
mm
−E)¸mE) +
k=m
e
mk
¸kE) = 0, (6.22)
where
e
mk
= ¸mH
1
k). (6.23)
Let us now develop our perturbation expansion. We assume that
e
mk

E
m
−E
k
∼ O(), (6.24)
for all m ,= k, where ¸1 is our expansion parameter. We also assume that
e
mm

E
m
∼ O(), (6.25)
for all m. Let us search for a modiﬁed version of the nth unperturbed energy
eigenstate, for which
E = E
n
+O(), (6.26)
and
¸nE) = 1, (6.27)
¸mE) ∼ O(), (6.28)
for m ,= n. Suppose that we write out Eq. (6.22) for m ,= n, neglecting terms
which are O(
2
) according to our expansion scheme. We ﬁnd that
(E
m
−E
n
)¸mE) +e
mn
· 0, (6.29)
123
6.4 The quadratic Stark effect 6 APPROXIMATION METHODS
giving
¸mE) · −
e
mn
E
m
−E
n
. (6.30)
Substituting the above expression into Eq. (6.22), evaluated for m = n, and
neglecting O(
3
) terms, we obtain
(E
n
+e
nn
−E) −
k=n
e
nk

2
E
k
−E
n
= 0. (6.31)
Thus, the modiﬁed nth energy eigenstate possesses an eigenvalue
E
n
= E
n
+e
nn
+
k=n
e
nk

2
E
n
−E
k
+O(
3
), (6.32)
and a eigenket
n)
= n) +
k=n
e
kn
E
n
−E
k
k) +O(
2
). (6.33)
Note that
¸mn)
= δ
mn
+
e
∗
nm
E
m
−E
n
+
e
mn
E
n
−E
m
+O(
2
) = δ
mn
+O(
2
). (6.34)
Thus, the modiﬁed eigenkets remain orthonormal and properly normalized to
O(
2
).
6.4 The quadratic Stark effect
Suppose that a oneelectron atom [i.e., either a hydrogen atom, or an alkali metal
atom (which possesses one valance electron orbiting outside a closed, spher
ically symmetric shell)] is subjected to a uniform electric ﬁeld in the positive
zdirection. The Hamiltonian of the system can be split into two parts. The un
perturbed Hamiltonian,
H
0
=
p
2
2 m
e
+V(r), (6.35)
and the perturbing Hamiltonian,
H
1
= e E z. (6.36)
124
6.4 The quadratic Stark effect 6 APPROXIMATION METHODS
It is assumed that the unperturbed energy eigenvalues and eigenstates are
completely known. The electron spin is irrelevant in this problem (since the spin
operators all commute with H
1
), so we can ignore the spin degrees of freedom of
the system. This implies that the system possesses no degenerate energy eigen
values. This is not true for the n ,= 1 energy levels of the hydrogen atom, due to
the special properties of a pure Coulomb potential. It is necessary to deal with
this case separately, because the perturbation theory presented in Sect. 6.3 breaks
down for degenerate unperturbed energy levels.
An energy eigenket of the unperturbed Hamiltonian is characterized by three
quantum numbers—the radial quantum number n, and the two angular quantum
numbers l and m (see Sect. 5.6). Let us denote such a ket n, l, m), and let its
energy level be E
nlm
. According to Eq. (6.32), the change in this energy level
induced by a small electric ﬁeld is given by
∆E
nlm
= e E ¸n, l, mzn, l, m)
+e
2
E
2
n
,l
,m
=n,l,m
¸n, l, mzn,
l
, m
)
2
E
nlm
−E
n
l
m
. (6.37)
Now, since
L
z
= x p
y
−yp
x
, (6.38)
it follows that
[L
z
, z] = 0. (6.39)
Thus,
¸n, l, m[L
z
, z]n
, l
, m
) = 0, (6.40)
giving
(m−m
)¸n, l, mzn
, l
, m
) = 0, (6.41)
since n, l, m) is, by deﬁnition, an eigenstate of L
z
with eigenvalue m¯h. It is
clear, from the above relation, that the matrix element ¸n, l, mzn
, l
, m
) is zero
unless m
= m. This is termed the selection rule for the quantum number m.
Let us now determine the selection rule for l. We have
[L
2
, z] = [L
2
x
, z] + [L
2
y
, z]
125
6.4 The quadratic Stark effect 6 APPROXIMATION METHODS
= L
x
[L
x
, z] + [L
x
, z]L
x
+L
y
[L
y
, z] + [L
y
, z]L
y
= i ¯h(−L
x
y −yL
x
+L
y
x +x L
y
)
= 2 i ¯h(L
y
x −L
x
y + i ¯hz)
= 2 i ¯h(L
y
x −yL
x
) = 2 i ¯h(x L
y
−L
x
y), (6.42)
where use has been made of Eqs. (5.1)–(5.6). Similarly,
[L
2
, y] = 2 i ¯h(L
x
z −x L
z
), (6.43)
[L
2
, x] = 2 i ¯h(yL
z
−L
y
z). (6.44)
Thus,
[L
2
, [L
2
, z]] = 2 i ¯h
_
L
2
, L
y
x −L
x
y + i ¯hz
_
= 2 i ¯h
_
L
y
[L
2
, x] −L
x
[L
2
, y] + i ¯h[L
2
, z]
_
,
= −4 ¯h
2
L
y
(yL
z
−L
y
z) +4 ¯h
2
L
x
(L
x
z −x L
z
)
−2 ¯h
2
(L
2
z −z L
2
). (6.45)
This reduces to
[L
2
, [L
2
, z]] = −¯h
2
_
4 (L
x
x +L
y
y +L
z
z)L
z
−4 (L
2
x
+L
2
y
+L
2
z
) z
+2 (L
2
z −z L
2
)
_
. (6.46)
However, it is clear from Eqs. (5.1)–(5.3) that
L
x
x +L
y
y +L
z
z = 0. (6.47)
Hence, we obtain
[L
2
, [L
2
, z]] = 2 ¯h
2
(L
2
z +z L
2
). (6.48)
Finally, the above expression expands to give
L
4
z −2 L
2
z L
2
+z L
4
−2 ¯h
2
(L
2
z +z L
2
) = 0. (6.49)
Equation (6.49) implies that
¸n, l, mL
4
z −2 L
2
z L
2
+z L
4
−2 ¯h
2
(L
2
z +z L
2
)n
, l
, m
) = 0. (6.50)
126
6.4 The quadratic Stark effect 6 APPROXIMATION METHODS
This expression yields
_
l
2
(l +1)
2
−2 l (l +1) l
(l
+1) +l
2
(l
+1)
2
−2 l (l +1) −2 l
(l
+1)] ¸n, l, mzn
, l
, m
) = 0, (6.51)
which reduces to
(l +l
+2) (l +l
) (l −l
+1) (l −l
−1)¸n, l, mzn
, l
, m
) = 0. (6.52)
According to the above formula, the matrix element ¸n, l, mzn
, l
, m
) vanishes
unless l = l
= 0 or l
= l ±1. This matrix element can be written
¸n, l, mzn
, l
, m
) =
___
ψ
∗
nlm
(r
, θ
, ϕ
) r
cos θ
ψ
n
m
l
(r
, θ
, ϕ
) dV
, (6.53)
where ψ
nlm
(r
) = ¸r
n, l, m). Recall, however, that the wavefunction of an l = 0
state is spherically symmetric (see Sect. 5.3): i.e., ψ
n00
(r
) = ψ
n00
(r
). It follows
from Eq. (6.53) that the matrix element vanishes by symmetry when l = l
= 0.
In conclusion, the matrix element ¸n, l, mzn
, l
, m
) is zero unless l
= l ± 1.
This is the selection rule for the quantum number l.
Application of the selection rules to Eq. (6.37) yields
∆E
nlm
= e
2
E
2
n
l
=l±1
¸n, l, mzn
, l
, m)
2
E
nlm
−E
n
l
m
. (6.54)
Note that all of the terms in Eq. (6.37) which vary linearly with the electric ﬁeld
strength vanish by symmetry, according to the selection rules. Only those terms
which vary quadratically with the ﬁeldstrength survive. The polarizability of an
atom is deﬁned in terms of the energyshift of the atomic state as follows:
∆E = −
1
2
αE
2
. (6.55)
Consider the ground state of a hydrogen atom. (Recall, that we cannot address
the n > 1 excited states because they are degenerate, and our theory cannot
handle this at present). The polarizability of this state is given by
α = 2 e
2
n>1
¸1, 0, 0zn, 1, 0)
2
E
n00
−E
100
. (6.56)
127
6.4 The quadratic Stark effect 6 APPROXIMATION METHODS
Here, we have made use of the fact that E
n10
= E
n00
for a hydrogen atom.
The sum in the above expression can be evaluated approximately by noting
that [see Eq. (5.120)]
E
n00
= −
e
2
8π
0
a
0
n
2
(6.57)
for a hydrogen atom, where
a
0
=
4π
0
¯h
2
µe
2
= 5.3 10
−11
meters (6.58)
is the Bohr radius. We can write
E
n00
−E
100
≥ E
200
−E
100
=
3
4
e
2
8π
0
a
0
. (6.59)
Thus,
α <
16
3
4π
0
a
0
n>1
¸1, 0, 0zn, 1, 0)
2
. (6.60)
However,
n>1
¸1, 0, 0zn, 1, 0)
2
=
n
,l
,m
¸1, 0, 0zn
, l
, m
)¸n
, m
, l
z1, 0, 0)
= ¸1, 0, 0z
2
1, 0, 0), (6.61)
where we have made use of the fact that the wavefunctions of a hydrogen atom
form a complete set. It is easily demonstrated from the actual form of the ground
state wavefunction that
¸1, 0, 0z
2
1, 0, 0) = a
2
0
. (6.62)
Thus, we conclude that
α <
16
3
4π
0
a
3
0
· 5.3 4π
0
a
3
0
. (6.63)
The true result is
α =
9
2
4π
0
a
3
0
= 4.5 4π
0
a
3
0
. (6.64)
It is actually possible to obtain this answer, without recourse to perturbation the
ory, by solving Schr¨ odinger’s equation exactly in parabolic coordinates.
128
6.5 Degenerate perturbation theory 6 APPROXIMATION METHODS
6.5 Degenerate perturbation theory
Let us now consider systems in which the eigenstates of the unperturbed Hamil
tonian, H
0
, possess degenerate energy levels. It is always possible to represent
degenerate energy eigenstates as the simultaneous eigenstates of the Hamilto
nian and some other Hermitian operator (or group of operators). Let us denote
this operator (or group of operators) L. We can write
H
0
n, l) = E
n
n, l), (6.65)
and
Ln, l) = L
nl
n, l), (6.66)
where [H
0
, L] = 0. Here, the E
n
and the L
nl
are real numbers which depend on
the quantum numbers n, and n and l, respectively. It is always possible to ﬁnd a
sufﬁcient number of operators which commute with the Hamiltonian in order to
ensure that the L
nl
are all different. In other words, we can choose L such that
the quantum numbers n and l uniquely specify each eigenstate. Suppose that for
each value of n there are N
n
different values of l: i.e., the nth energy eigenstate
is N
n
fold degenerate.
In general, L does not commute with the perturbing Hamiltonian, H
1
. This
implies that the modiﬁed energy eigenstates are not eigenstates of L. In this situ
ation, we expect the perturbation to split the degeneracy of the energy levels, so
that each modiﬁed eigenstate n, l)
acquires a unique energy eigenvalue E
nl
. Let
us naively attempt to use the standard perturbation theory of Sect. 6.3 to evalu
ate the modiﬁed energy eigenstates and energy levels. A direct generalization of
Eqs. (6.32) and (6.33) yields
E
nl
= E
n
+e
nlnl
+
n
,l
=n,l
e
n
l
nl

2
E
n
−E
n
+O(
3
), (6.67)
and
n, l)
= n, l) +
n
,l
=n,l
e
n
l
nl
E
n
−E
n
n
, l
) +O(
2
), (6.68)
where
e
n
l
nl
= ¸n
, l
H
1
n, l). (6.69)
129
6.5 Degenerate perturbation theory 6 APPROXIMATION METHODS
It is fairly obvious that the summations in Eqs. (6.67) and (6.68) are not well
behaved if the nth energy level is degenerate. The problem terms are those
involving unperturbed eigenstates labeled by the same value of n, but different
values of l: i.e., those states whose unperturbed energies are E
n
. These terms
give rise to singular factors 1/(E
n
− E
n
) in the summations. Note, however, that
this problem would not exist if the matrix elements, e
nl
nl
, of the perturbing
Hamiltonian between distinct, degenerate, unperturbed energy eigenstates cor
responding to the eigenvalue E
n
were zero. In other words, if
¸n, l
H
1
n, l) = λ
nl
δ
ll
, (6.70)
then all of the singular terms in Eqs. (6.67) and (6.68) would vanish.
In general, Eq. (6.70) is not satisﬁed. Fortunately, we can always redeﬁne the
unperturbed energy eigenstates belonging to the eigenvalue E
n
in such a manner
that Eq. (6.70) is satisﬁed. Let us deﬁne N
n
new states which are linear combina
tions of the N
n
original degenerate eigenstates corresponding to the eigenvalue
E
n
:
n, l
(1)
) =
N
n
k=1
¸n, kn, l
(1)
)n, k). (6.71)
Note that these new states are also degenerate energy eigenstates of the unper
turbed Hamiltonian corresponding to the eigenvalue E
n
. The n, l
(1)
) are chosen
in such a manner that they are eigenstates of the perturbing Hamiltonian, H
1
.
Thus,
H
1
n, l
(1)
) = λ
nl
n, l
(1)
). (6.72)
The n, l
(1)
) are also chosen so that they are orthonormal, and have unit lengths.
It follows that
¸n, l
(1)
H
1
n, l
(1)
) = λ
nl
δ
ll
. (6.73)
Thus, if we use the new eigenstates, instead of the old ones, then we can employ
Eqs. (6.67) and (6.68) directly, since all of the singular terms vanish. The only
remaining difﬁculty is to determine the new eigenstates in terms of the original
ones.
130
6.5 Degenerate perturbation theory 6 APPROXIMATION METHODS
Now
N
n
l=1
n, l)¸n, l = 1, (6.74)
where 1 denotes the identity operator in the subspace of all unperturbed energy
eigenkets corresponding to the eigenvalue E
n
. Using this completeness relation,
the operator eigenvalue equation (6.72) can be transformed into a straightfor
ward matrix eigenvalue equation:
N
n
l
=1
¸n, l
H
1
n, l
)¸n, l
n, l
(1)
) = λ
nl
¸n, l
n, l
(1)
). (6.75)
This can be written more transparently as
Ux = λ x, (6.76)
where the elements of the N
n
N
n
Hermitian matrix U are
U
jk
= ¸n, jH
1
n, k). (6.77)
Provided that the determinant of U is nonzero, Eq. (6.76) can always be solved
to give N
n
eigenvalues λ
nl
(for l = 1 to N
n
), with N
n
corresponding eigenvectors
x
nl
. The eigenvectors specify the weights of the new eigenstates in terms of the
original eigenstates: i.e.,
(x
nl
)
k
= ¸n, kn, l
(1)
), (6.78)
for k = 1 to N
n
. In our new scheme, Eqs. (6.67) and (6.68) yield
E
nl
= E
n
+λ
nl
+
n
=n,l
e
n
l
nl

2
E
n
−E
n
+O(
3
), (6.79)
and
n, l
(1)
)
= n, l
(1)
) +
n
=n,l
e
n
l
nl
E
n
−E
n
n
, l
) +O(
2
). (6.80)
There are no singular terms in these expressions, since the summations are over
n
,= n: i.e., they speciﬁcally exclude the problematic, degenerate, unperturbed
energy eigenstates corresponding to the eigenvalue E
n
. Note that the ﬁrstorder
energy shifts are equivalent to the eigenvalues of the matrix equation (6.76).
131
6.6 The linear Stark effect 6 APPROXIMATION METHODS
6.6 The linear Stark effect
Let us examine the effect of an electric ﬁeld on the excited energy levels of a
hydrogen atom. For instance, consider the n = 2 states. There is a single l =
0 state, usually referred to as 2s, and three l = 1 states (with m = −1, 0, 1),
usually referred to as 2p. All of these states possess the same energy, E
200
=
−e
2
/(32π
0
a
0
). As in Sect. 6.4, the perturbing Hamiltonian is
H
1
= e E z. (6.81)
In order to apply perturbation theory, we have to solve the matrix eigenvalue
equation
Ux = λ x, (6.82)
where U is the array of the matrix elements of H
1
between the degenerate 2s and
2p states. Thus,
U = e E
_
_
_
_
_
_
_
_
0 ¸2, 0, 0z2, 1, 0) 0 0
¸2, 1, 0z2, 0, 0) 0 0 0
0 0 0 0
0 0 0 0
_
_
_
_
_
_
_
_
, (6.83)
where the rows and columns correspond to the 2, 0, 0), 2, 1, 0), 2, 1, 1), and
2, 1, −1) states, respectively. Here, we have made use of the selection rules,
which tell us that the matrix element of z between two hydrogen atom states is
zero unless the states possess the same m quantum number, and l quantum num
bers which differ by unity. It is easily demonstrated, from the exact forms of the
2s and 2p wavefunctions, that
¸2, 0, 0z2, 1, 0) = ¸2, 1, 0z2, 0, 0) = 3 a
0
. (6.84)
It can be seen, by inspection, that the eigenvalues of U are λ
1
= 3 e a
0
E,
λ
2
= −3 e a
0
E, λ
3
= 0, and λ
4
= 0. The corresponding eigenvectors are
x
1
=
_
_
_
_
_
_
_
_
1/
√
2
1/
√
2
0
0
_
_
_
_
_
_
_
_
, (6.85)
132
6.6 The linear Stark effect 6 APPROXIMATION METHODS
x
2
=
_
_
_
_
_
_
_
_
1/
√
2
−1/
√
2
0
0
_
_
_
_
_
_
_
_
, (6.86)
x
3
=
_
_
_
_
_
_
_
_
0
0
1
0
_
_
_
_
_
_
_
_
, (6.87)
x
4
=
_
_
_
_
_
_
_
_
0
0
0
1
_
_
_
_
_
_
_
_
. (6.88)
It follows from Sect. 6.5 that the simultaneous eigenstates of the unperturbed
Hamiltonian and the perturbing Hamiltonian take the form
1) =
2, 0, 0) + 2, 1, 0)
√
2
, (6.89)
2) =
2, 0, 0) − 2, 1, 0)
√
2
, (6.90)
3) = 2, 1, 1), (6.91)
4) = 2, 1, −1). (6.92)
In the absence of an electric ﬁeld, all of these states possess the same energy, E
200
.
The ﬁrstorder energy shifts induced by an electric ﬁeld are given by
∆E
1
= +3 e a
0
E, (6.93)
∆E
2
= −3 e a
0
E, (6.94)
∆E
3
= 0, (6.95)
∆E
4
= 0. (6.96)
Thus, the energies of states 1 and 2 are shifted upwards and downwards, respec
tively, by an amount 3 e a
0
E in the presence of an electric ﬁeld. States 1 and
2 are orthogonal linear combinations of the original 2s and 2p(m = 0) states.
133
6.7 Fine structure 6 APPROXIMATION METHODS
Note that the energy shifts are linear in the electric ﬁeldstrength, so this is a
much larger effect that the quadratic effect described in Sect. 6.4. The energies
of states 3 and 4 (which are equivalent to the original 2p(m = 1) and 2p(m = −1)
states, respectively) are not affected to ﬁrstorder. Of course, to secondorder the
energies of these states are shifted by an amount which depends on the square of
the electric ﬁeldstrength.
Note that the linear Stark effect depends crucially on the degeneracy of the 2s
and 2p states. This degeneracy is a special property of a pure Coulomb potential,
and, therefore, only applies to a hydrogen atom. Thus, alkali metal atoms do not
exhibit the linear Stark effect.
6.7 Fine structure
Let us now consider the energy levels of hydrogenlike atoms (i.e., alkali metal
atoms) in more detail. The outermost electron moves in a spherically symmetric
potential V(r) due to the nuclear charge and the charges of the other electrons
(which occupy spherically symmetric closed shells). The shielding effect of the
inner electrons causes V(r) to depart from the pure Coulomb form. This splits the
degeneracy of states characterized by the same value of n, but different values of
l. In fact, higher l states have higher energies.
Let us examine a phenomenon known as ﬁne structure, which is due to inter
action between the spin and orbital angular momenta of the outermost electron.
This electron experiences an electric ﬁeld
E =
∇V
e
. (6.97)
However, a charge moving in an electric ﬁeld also experiences an effective mag
netic ﬁeld
B = −v E. (6.98)
Now, an electron possesses a spin magnetic moment [see Eq. (5.170)]
µ = −
e S
m
e
. (6.99)
134
6.7 Fine structure 6 APPROXIMATION METHODS
We, therefore, expect a spinorbit contribution to the Hamiltonian of the form
H
LS
= −µB
= −
e S
m
e
v
_
1
e
r
r
dV
dr
_
=
1
m
2
e
r
dV
dr
LS, (6.100)
where L = m
e
rv is the orbital angular momentum. When the above expression
is compared to the observed spinorbit interaction, it is found to be too large by
a factor of two. There is a classical explanation for this, due to spin precession,
which we need not go into. The correct quantummechanical explanation requires
a relativistically covariant treatment of electron dynamics (this is achieved using
the socalled Dirac equation).
Let us now apply perturbation theory to a hydrogenlike atom, using H
LS
as
the perturbation (with H
LS
taking one half of the value given above), and
H
0
=
p
2
2 m
e
+V(r) (6.101)
as the unperturbed Hamiltonian. We have two choices for the energy eigenstates
of H
0
. We can adopt the simultaneous eigenstates of H
0
, L
2
, S
2
, L
z
and S
z
, or the
simultaneous eigenstates of H
0
, L
2
, S
2
, J
2
, and J
z
, where J = L + S is the total
angular momentum. Although the departure of V(r) from a pure 1/r form splits
the degeneracy of same n, different l, states, those states characterized by the
same values of n and l, but different values of m
l
, are still degenerate. (Here,
m
l
, m
s
, and m
j
are the quantum numbers corresponding to L
z
, S
z
, and J
z
, respec
tively.) Moreover, with the addition of spin degrees of freedom, each state is
doubly degenerate due to the two possible orientations of the electron spin (i.e.,
m
s
= ±1/2). Thus, we are still dealing with a highly degenerate system. We
know, from Sect. 6.6, that the application of perturbation theory to a degenerate
system is greatly simpliﬁed if the basis eigenstates of the unperturbed Hamilto
nian are also eigenstates of the perturbing Hamiltonian. Now, the perturbing
Hamiltonian, H
LS
, is proportional to LS, where
LS =
J
2
−L
2
−S
2
2
. (6.102)
135
6.7 Fine structure 6 APPROXIMATION METHODS
It is fairly obvious that the ﬁrst group of operators (H
0
, L
2
, S
2
, L
z
and S
z
) does not
commute with H
LS
, whereas the second group (H
0
, L
2
, S
2
, J
2
, and J
z
) does. In
fact, LS is just a combination of operators appearing in the second group. Thus,
it is advantageous to work in terms of the eigenstates of the second group of
operators, rather than those of the ﬁrst group.
We now need to ﬁnd the simultaneous eigenstates of H
0
, L
2
, S
2
, J
2
, and J
z
. This
is equivalent to ﬁnding the eigenstates of the total angular momentum resulting
from the addition of two angular momenta: j
1
= l, and j
2
= s = 1/2. According
to Eq. (5.276), the allowed values of the total angular momentum are j = l +1/2
and j = l −1/2. We can write
l +1/2, m) = cos αm−1/2, 1/2) + sinαm+1/2, −1/2), (6.103)
l −1/2, m) = −sinαm−1/2, 1/2) + cos αm+1/2, −1/2). (6.104)
Here, the kets on the lefthand side are j, m
j
) kets, whereas those on the right
hand side are m
l
, m
s
) kets (the j
1
, j
2
labels have been dropped, for the sake of
clarity). We have made use of the fact that the ClebschGordon coefﬁcients are
automatically zero unless m
j
= m
l
+ m
s
. We have also made use of the fact that
both the j, m
j
) and m
l
, m
s
) kets are orthonormal, and have unit lengths. We
now need to determine
cos α = ¸m−1/2, 1/2l +1/2, m), (6.105)
where the ClebschGordon coefﬁcient is written in ¸m
l
, m
s
j, m
j
) form.
Let us now employ the recursion relation for ClebschGordon coefﬁcients,
Eq. (5.282), with j
1
= l, j
2
= 1/2, j = l + 1/2, m
1
= m − 1/2, m
2
= 1/2 (lower
sign). We obtain
_
(l +1/2) (l +3/2) −m(m+1) ¸m−1/2, 1/2l +1/2, m)
=
_
l (l +1) − (m−1/2) (m+1/2) ¸m+1/2, 1/2l +1/2, m+1), (6.106)
which reduces to
¸m−1/2, 1/2l +1/2, m) =
¸
¸
¸
¸
_
l +m+1/2
l +m+3/2
¸m+1/2, 1/2l +1/2, m+1). (6.107)
136
6.7 Fine structure 6 APPROXIMATION METHODS
We can use this formula to successively increase the value of m
l
. For instance,
¸m−1/2, 1/2l +1/2, m) =
¸
¸
¸
¸
_
l +m+1/2
l +m+3/2
¸
¸
¸
¸
_
l +m+3/2
l +m+5/2
¸m+3/2, 1/2l +1/2, m+2). (6.108)
This procedure can be continued until m
l
attains its maximum possible value, l.
Thus,
¸m−1/2, 1/2l +1/2, m) =
¸
¸
¸
_
l +m+1/2
2 l +1
¸l, 1/2l +1/2, l +1/2). (6.109)
Consider the situation in which m
l
and m both take their maximum values,
l and 1/2, respectively. The corresponding value of m
j
is l + 1/2. This value is
possible when j = l +1/2, but not when j = l −1/2. Thus, the m
l
, m
s
) ket l, 1/2)
must be equal to the j, m
j
) ket l + 1/2, l + 1/2), up to an arbitrary phasefactor.
By convention, this factor is taken to be unity, giving
¸l, 1/2l +1/2, l +1/2) = 1. (6.110)
It follows from Eq. (6.109) that
cos α = ¸m−1/2, 1/2l +1/2, m) =
¸
¸
¸
_
l +m+1/2
2 l +1
. (6.111)
Now,
sin
2
α = 1 −
l +m+1/2
2 l +1
=
l −m+1/2
2 l +1
. (6.112)
We now need to determine the sign of sinα. A careful examination of the
recursion relation, Eq. (5.282), shows that the plus sign is appropriate. Thus,
l +1/2, m) =
¸
¸
¸
_
l +m+1/2
2 l +1
m−1/2, 1/2)
+
¸
¸
¸
_
l −m+1/2
2 l +1
m+1/2, −1/2), (6.113)
137
6.7 Fine structure 6 APPROXIMATION METHODS
l −1/2, m) = −
¸
¸
¸
_
l −m+1/2
2 l +1
m−1/2, 1/2)
+
¸
¸
¸
_
l +m+1/2
2 l +1
m+1/2, −1/2). (6.114)
It is convenient to deﬁne so called spinangular functions using the Pauli two
component formalism:
¸
j=l±1/2,m
l
= ±
¸
¸
¸
_
l ±m+1/2
2 l +1
Y
m−1/2
l
(θ, ϕ) χ
+
+
¸
¸
¸
_
l ∓m+1/2
2 l +1
Y
m+1/2
l
(θ, ϕ) χ
−
=
1
√
2 l +1
_
_
_
±
_
l ±m+1/2 Y
m−1/2
l
(θ, ϕ)
_
l ∓m+1/2 Y
m+1/2
l
(θ, ϕ)
_
_
_. (6.115)
These functions are eigenfunctions of the total angular momentum for spin one
half particles, just as the spherical harmonics are eigenfunctions of the orbital
angular momentum. A general wavefunction for an energy eigenstate in a
hydrogenlike atom is written
ψ
nlm±
= R
nl
(r) ¸
j=l±1/2,m
. (6.116)
The radial part of the wavefunction, R
nl
(r), depends on the radial quantum num
ber n and the angular quantum number l. The wavefunction is also labeled by
m, which is the quantum number associated with J
z
. For a given choice of l, the
quantum number j (i.e., the quantum number associated with J
2
) can take the
values l ±1/2.
The l ±1/2, m) kets are eigenstates of LS, according to Eq. (6.102). Thus,
LSj = l ±1/2, m
j
= m) =
¯h
2
2
[j (j +1) −l (l +1) −3/4] j, m), (6.117)
giving
LSl +1/2, m) =
l ¯h
2
2
l +1/2, m), (6.118)
LSl −1/2, m) = −
(l +1) ¯h
2
2
l −1/2, m). (6.119)
138
6.7 Fine structure 6 APPROXIMATION METHODS
It follows that
_
(¸
l+1/2,m
)
†
LS¸
l+1/2,m
dΩ =
l ¯h
2
2
, (6.120)
_
(¸
l−1/2,m
)
†
LS¸
l−1/2,m
dΩ = −
(l +1) ¯h
2
2
, (6.121)
where the integrals are over all solid angle.
Let us nowapply degenerate perturbation theory to evaluate the shift in energy
of a state whose wavefunction is ψ
nlm±
due to the spinorbit Hamiltonian H
LS
.
To ﬁrstorder, the energyshift is given by
∆E
nlm±
=
_
(ψ
nlm±
)
†
H
LS
ψ
nlm±
dV, (6.122)
where the integral is over all space. Equations (6.100) (remember the factor of
two), (6.116), and (6.120)–(6.121) yield
∆E
nlm+
= +
1
2 m
2
e
_
1
r
dV
dr
_
l ¯h
2
2
, (6.123)
∆E
nlm−
= −
1
2 m
2
e
_
1
r
dV
dr
_
(l +1) ¯h
2
2
, (6.124)
where
_
1
r
dV
dr
_
=
_
(R
nl
)
∗
1
r
dV
dr
R
nl
r
2
dr. (6.125)
Equations (6.123)–(6.124) are known as Lande’s interval rule.
Let us now apply the above result to the case of a sodium atom. In chemist’s
notation, the ground state is written
(1s)
2
(2s)
2
(2p)
6
(3s). (6.126)
The inner ten electrons effectively form a spherically symmetric electron cloud.
We are interested in the excitation of the eleventh electron from 3s to some
higher energy state. The closest (in energy) unoccupied state is 3p. This state
has a higher energy than 3s due to the deviations of the potential from the pure
139
6.8 The Zeeman effect 6 APPROXIMATION METHODS
Coulomb form. In the absence of spinorbit interaction, there are six degenerate
3p states. The spinorbit interaction breaks the degeneracy of these states. The
modiﬁed states are labeled (3p)
1/2
and (3p)
3/2
, where the subscript refers to the
value of j. The four (3p)
3/2
states lie at a slightly higher energy level than the two
(3p)
1/2
states, because the radial integral (6.125) is positive. The splitting of the
(3p) energy levels of the sodium atom can be observed using a spectroscope. The
wellknown sodium D line is associated with transitions between the 3p and 3s
states. The fact that there are two slightly different 3p energy levels (note that
spinorbit coupling does not split the 3s energy levels) means that the sodium D
line actually consists of two very closely spaced spectroscopic lines. It is easily
demonstrated that the ratio of the typical spacing of Balmer lines to the splitting
brought about by spinorbit interaction is about 1 : α
2
, where
α =
e
2
2
0
hc
=
1
137
(6.127)
is the ﬁne structure constant. Note that Eqs. (6.123)–(6.124) are not entirely
correct, since we have neglected an effect (namely, the relativistic mass correction
of the electron) which is the same order of magnitude as spinorbit coupling.
6.8 The Zeeman effect
Consider a hydrogenlike atom placed in a uniform zdirected magnetic ﬁeld. The
change in energy of the outermost electron is
H
B
= −µB, (6.128)
where
µ = −
e
2 m
e
(L +2 S) (6.129)
is its magnetic moment, including both the spin and orbital contributions. Thus,
H
B
=
e B
2 m
e
(L
z
+2 S
z
). (6.130)
Suppose that the energyshifts induced by the magnetic ﬁeld are much smaller
than those induced by spinorbit interaction. In this situation, we can treat H
B
as
140
6.8 The Zeeman effect 6 APPROXIMATION METHODS
a small perturbation acting on the eigenstates of H
0
+H
LS
. Of course, these states
are the simultaneous eigenstates of J
2
and J
z
. Let us consider one of these states,
labeled by the quantum numbers j and m, where j = l ± 1/2. From standard
perturbation theory, the ﬁrstorder energyshift in the presence of a magnetic
ﬁeld is
∆E
nlm±
= ¸l ±1/2, mH
B
l ±1/2, m). (6.131)
Since
L
z
+2 S
z
= J
z
+S
z
, (6.132)
we ﬁnd that
∆E
nlm±
=
e B
2 m
e
(m¯h +¸l ±1/2, mS
z
l ±1/2, m) ) . (6.133)
Now, from Eqs. (6.113)–(6.114),
l ±1/2, m) = ±
¸
¸
¸
_
l ±m+1/2
2 l +1
m−1/2, 1/2)
+
¸
¸
¸
_
l ∓m+1/2
2 l +1
m+1/2, −1/2). (6.134)
It follows that
¸l ±1/2, mS
z
l ±1/2, m) =
¯h
2 (2 l +1)
[(l ±m+1/2) − (l ∓m+1/2)]
= ±
m¯h
2 l +1
. (6.135)
Thus, we obtain Lande’s formula for the energyshift induced by a weak magnetic
ﬁeld:
∆E
nlm±
=
e ¯hB
2 m
e
m
_
1 ±
1
2 l +1
_
. (6.136)
Let us apply this theory to the sodium atom. We have already seen that the
nonCoulomb potential splits the degeneracy of the 3s and 3p states, the latter
states acquiring a higher energy. The spinorbit interaction splits the six 3p states
into two groups, with four j = 3/2 states lying at a slightly higher energy than
two j = 1/2 states. According to Eq. (6.136), a magnetic ﬁeld splits the (3p)
3/2
141
6.8 The Zeeman effect 6 APPROXIMATION METHODS
quadruplet of states, each state acquiring a different energy. In fact, the energy of
each state becomes dependent on the quantum number m, which measures the
projection of the total angular momentum along the zaxis. States with higher
m values have higher energies. A magnetic ﬁeld also splits the (3p)
1/2
doublet
of states. However, it is evident from Eq. (6.136) that these states are split by a
lesser amount than the j = 3/2 states.
Suppose that we increase the strength of the magnetic ﬁeld, so that the energy
shift due to the magnetic ﬁeld becomes comparable to the energyshift induced
by spinorbit interaction. Clearly, in this situation, it does not make much sense to
think of H
B
as a small interaction term operating on the eigenstates of H
0
+H
LS
.
In fact, this intermediate case is very difﬁcult to analyze. Let us consider the
extreme limit in which the energyshift due to the magnetic ﬁeld greatly exceeds
that induced by spinorbit effects. This is called the PaschenBack limit.
In the PaschenBack limit we can think of the spinorbit Hamiltonian, H
LS
, as
a small interaction term operating on the eigenstates of H
0
+ H
B
. Note that the
magnetic Hamiltonian, H
B
, commutes with L
2
, S
2
, L
z
, S
z
, but does not commute
with L
2
, S
2
, J
2
, J
z
. Thus, in an intense magnetic ﬁeld, the energy eigenstates of a
hydrogenlike atom are approximate eigenstates of the spin and orbital angular
momenta, but are not eigenstates of the total angular momentum. We can label
each state by the quantum numbers n (the energy quantum number), l, m
l
,
and m
s
. Thus, our energy eigenkets are written n, l, m
l
, m
s
). The unperturbed
Hamiltonian, H
0
, causes states with different values of the quantum numbers n
and l to have different energies. However, states with the same value of n and l,
but different values of m
l
and m
s
, are degenerate. The shift in energy due to the
magnetic ﬁeld is simply
∆E
nlm
l
m
s
= ¸n, l, m
l
, m
s
H
B
n, l, m
l
, m
s
)
=
e ¯hB
2 m
e
(m
l
+2 m
s
). (6.137)
Thus, states with different values of m
l
+2 m
s
acquire different energies.
Let us apply this result to a sodium atom. In the absence of a magnetic ﬁeld,
the six 3p states form two groups of four and two states, depending on the values
142
6.8 The Zeeman effect 6 APPROXIMATION METHODS
of their total angular momentum. In the presence of an intense magnetic ﬁeld
the 3p states are split into ﬁve groups. There is a state with m
l
+2 m
s
= 2, a state
with m
l
+2 m
s
= 1, two states with m
l
+2 m
s
= 0, a state with m
l
+2 m
s
= −1,
and a state with m
l
+2 m
s
= −2. These groups are equally spaced in energy, the
energy difference between adjacent groups being e ¯hB/2 m
e
.
The energyshift induced by the spinorbit Hamiltonian is given by
∆E
nl m
l
m
s
= ¸n, l, m
l
, m
s
H
LS
n, l, m
l
, m
s
), (6.138)
where
H
LS
=
1
2 m
2
e
1
r
dV
dr
LS. (6.139)
Now,
¸LS) = ¸ L
z
S
z
+ (L
+
S
−
+L
−
S
+
)/2 )
= ¯h
2
m
l
m
s
, (6.140)
since
¸L
±
) = ¸S
±
) = 0 (6.141)
for expectation values taken between the simultaneous eigenkets of L
z
and S
z
.
Thus,
∆E
nlm
l
m
s
=
¯h
2
m
l
m
s
2 m
2
e
_
1
r
dV
dr
_
. (6.142)
Let us apply the above result to a sodium atom. In the presence of an intense
magnetic ﬁeld, the 3p states are split into ﬁve groups with (m
l
, m
s
) quantum
numbers (1, 1/2), (0, 1/2), (1, −1/2), or (−1, 1/2), (0, −1/2), and (−1, −1/2), re
spectively, in order of decreasing energy. The spinorbit term increases the en
ergy of the highest energy state, does not affect the next highest energy state,
decreases, but does not split, the energy of the doublet, does not affect the next
lowest energy state, and increases the energy of the lowest energy state. The net
result is that the ﬁve groups of states are no longer equally spaced in energy.
The sort of magnetic ﬁeldstrength needed to get into the PaschenBach limit
is given by
B
PB
∼ α
2
e m
e
0
ha
0
· 25 tesla. (6.143)
143
6.9 Timedependent perturbation theory 6 APPROXIMATION METHODS
Obviuously, this is an extremely large ﬁeldstrength.
6.9 Timedependent perturbation theory
Suppose that the Hamiltonian of the system under consideration can be written
H = H
0
+H
1
(t), (6.144)
where H
0
does not contain time explicitly, and H
1
is a small timedependent
perturbation. It is assumed that we are able to calculate the eigenkets of the
unperturbed Hamiltonian:
H
0
n) = E
n
n). (6.145)
We know that if the system is in one of the eigenstates of H
0
then, in the ab
sence of the external perturbation, it remains in this state for ever. However,
the presence of a small timedependent perturbation can, in principle, give rise
to a ﬁnite probability that a system initially in some eigenstate i) of the unper
turbed Hamiltonian is found in some other eigenstate at a subsequent time (since
i) is no longer an exact eigenstate of the total Hamiltonian), In other words, a
timedependent perturbation causes the system to make transitions between its
unperturbed energy eigenstates. Let us investigate this effect.
Suppose that at t = t
0
the state of the system is represented by
A) =
n
c
n
n), (6.146)
where the c
n
are complex numbers. Thus, the initial state is some linear su
perposition of the unperturbed energy eigenstates. In the absence of the time
dependent perturbation, the time evolution of the system is given by
A, t
0
, t) =
n
c
n
exp([−i E
n
(t −t
0
)/¯h] n). (6.147)
Now, the probability of ﬁnding the system in state n) at time t is
P
n
(t) = c
n
exp[−i E
n
(t −t
0
)/¯h]
2
= c
n

2
= P
n
(t
0
). (6.148)
144
6.9 Timedependent perturbation theory 6 APPROXIMATION METHODS
Clearly, with H
1
= 0, the probability of ﬁnding the system in state n) at time t is
exactly the same as the probability of ﬁnding the system in this state at the initial
time t
0
. However, with H
1
,= 0, we expect P
n
(t) to vary with time. Thus, we can
write
A, t
0
, t) =
n
c
n
(t) exp[−i E
n
(t −t
0
)/¯h] n), (6.149)
where P
n
(t) = c
n
(t)
2
. Here, we have carefully separated the fast phase oscilla
tion of the eigenkets, which depends on the unperturbed Hamiltonian, from the
slow variation of the amplitudes c
n
(t), which depends entirely on the perturba
tion (i.e., c
n
is constant if H
1
= 0). Note that in Eq. (6.149) the eigenkets n) are
timeindependent (they are actually the eigenkets of H
0
evaluated at the time t
0
).
Schr¨ odinger’s time evolution equation yields
i ¯h
∂
∂t
A, t
0
, t) = HA, t
0
, t) = (H
0
+H
1
) A, t
0
, t). (6.150)
It follows from Eq. (6.149) that
(H
0
+H
1
)A, t
0
, t) =
m
c
m
(t) exp[−i E
m
(t −t
0
)/¯h] (E
m
+H
1
) m). (6.151)
We also have
i ¯h
∂
∂t
A, t
0
, t) =
m
_
i ¯h
dc
m
dt
+c
m
(t) E
m
_
exp[−i E
m
(t −t
0
)/¯h] m), (6.152)
where use has been made of the timeindependence of the kets m). According
to Eq. (6.150), we can equate the righthand sides of the previous two equations
to obtain
m
i ¯h
dc
m
dt
exp[−i E
m
(t −t
0
)/¯h]m) =
m
c
m
(t) exp[−i E
m
(t −t
0
)/¯h] H
1
m).
(6.153)
Leftmultiplication by ¸n yields
i ¯h
dc
n
dt
=
m
H
nm
(t) exp[i ω
nm
(t −t
0
)] c
m
(t), (6.154)
where
H
nm
(t) = ¸nH
1
(t)m), (6.155)
145
6.10 The twostate system 6 APPROXIMATION METHODS
and
ω
nm
=
E
n
−E
m
¯h
. (6.156)
Here, we have made use of the standard orthonormality result, ¸nm) = δ
nm
.
Suppose that there are N linearly independent eigenkets of the unperturbed
Hamiltonian. According to Eq. (6.154), the time variation of the coefﬁcients
c
n
, which specify the probability of ﬁnding the system in state n) at time t, is de
termined by N coupled ﬁrstorder differential equations. Note that Eq. (6.154) is
exact—we have made no approximations at this stage. Unfortunately, we cannot
generally ﬁnd exact solutions to this equation, so we have to obtain approximate
solutions via suitable expansions in small quantities. However, for the particu
larly simple case of a twostate system (i.e., N = 2), it is actually possible to
solve Eq. (6.154) without approximation. This solution is of enormous practical
importance.
6.10 The twostate system
Consider a system in which the timeindependent Hamiltonian possesses two
eigenstates, denoted
H
0
1) = E
1
1), (6.157)
H
0
2) = E
2
2). (6.158)
Suppose, for the sake of simplicity, that the diagonal matrix elements of the in
teraction Hamiltonian, H
1
, are zero:
¸1H
1
1) = ¸2H
1
2) = 0. (6.159)
The offdiagonal matrix elements are assumed to oscillate sinusoidally at some
frequency ω:
¸1H
1
2) = ¸2H
1
1)
∗
= γexp(i ωt), (6.160)
where γ and ω are real. Note that it is only the offdiagonal matrix elements
which give rise to the effect which we are interested in—namely, transitions be
tween states 1 and 2.
146
6.10 The twostate system 6 APPROXIMATION METHODS
For a twostate system, Eq. (6.154) reduces to
i ¯h
dc
1
dt
= γexp[+i (ω−ω
21
) t ] c
2
, (6.161)
i ¯h
dc
2
dt
= γexp[−i (ω−ω
21
) t ] c
1
, (6.162)
where ω
21
= (E
2
− E
1
)/¯h, and assuming that t
0
= 0. Equations (6.161) and
(6.162) can be combined to give a secondorder differential equation for the
time variation of the amplitude c
2
:
d
2
c
2
dt
2
+ i (ω−ω
21
)
dc
2
dt
+
γ
2
¯h
2
c
2
= 0. (6.163)
Once we have solved for c
2
, we can use Eq. (6.162) to obtain the amplitude c
1
.
Let us look for a solution in which the system is certain to be in state 1 at time
t = 0. Thus, our boundary conditions are c
1
(0) = 1 and c
2
(0) = 0. It is easily
demonstrated that the appropriate solutions are
c
2
(t) =
−i γ/¯h
_
γ
2
/¯h
2
+ (ω−ω
21
)
2
/4
exp[−i (ω−ω
21
) t/2]
sin
__
γ
2
/¯h
2
+ (ω−ω
21
)
2
/4 t
_
, (6.164)
c
1
(t) = exp[ i (ω−ω
21
) t/2] cos
__
γ
2
/¯h
2
+ (ω−ω
21
)
2
/4 t
_
−
i (ω−ω
21
)/2
_
γ
2
/¯h
2
+ (ω−ω
21
)
2
/4
exp[ i (ω−ω
21
) t/2]
sin
__
γ
2
/¯h
2
+ (ω−ω
21
)
2
/4 t
_
. (6.165)
Now, the probability of ﬁnding the system in state 1 at time t is simply P
1
(t) =
c
1

2
. Likewise, the probability of ﬁnding the system in state 2 at time t is P
2
(t) =
c
2

2
. It follows that
P
2
(t) =
γ
2
/¯h
2
γ
2
/¯h
2
+ (ω−ω
21
)
2
/4
147
6.10 The twostate system 6 APPROXIMATION METHODS
sin
2
__
γ
2
/¯h
2
+ (ω−ω
21
)
2
/4 t
_
, (6.166)
P
1
(t) = 1 −P
2
(t). (6.167)
This result is known as Rabi’s formula.
Equation (6.166) exhibits all the features of a classic resonance. At resonance,
when the oscillation frequency of the perturbation, ω, matches the frequency
ω
21
, we ﬁnd that
P
1
(t) = cos
2
(γt/¯h), (6.168)
P
2
(t) = sin
2
(γt/¯h). (6.169)
According to the above result, the system starts off at t = 0 in state 1. After a
time interval π¯h/2 γ it is certain to be in state 2. After a further time interval
π¯h/2 γ it is certain to be in state 1, and so on. Thus, the system periodically
ﬂipﬂops between states 1 and 2 under the inﬂuence of the timedependent per
turbation. This implies that the system alternatively absorbs and emits energy
from the source of the perturbation.
The absorptionemission cycle also take place away from the resonance, when
ω ,= ω
21
. However, the amplitude of oscillation of the coefﬁcient c
2
is reduced.
This means that the maximum value of P
2
(t) is no longer unity, nor is the mini
mum value of P
1
(t) zero. In fact, if we plot the maximum value of P
2
(t) as a func
tion of the applied frequency, ω, we obtain a resonance curve whose maximum
(unity) lies at the resonance, and whose fullwidth halfmaximum (in frequency)
is 4 γ/¯h. Thus, if the applied frequency differs from the resonant frequency by
substantially more than 2 γ/¯h then the probability of the system jumping from
state 1 to state 2 is very small. In other words, the timedependent perturbation
is only effective at causing transitions between states 1 and 2 if its frequency of
oscillation lies in the approximate range ω
21
± 2 γ/¯h. Clearly, the weaker the
perturbation (i.e., the smaller γ becomes), the narrower the resonance.
148
6.11 Spin magnetic resonance 6 APPROXIMATION METHODS
6.11 Spin magnetic resonance
Consider a spin onehalf system (e.g., a bound electron) placed in a uniform z
directed magnetic ﬁeld, and then subjected to a small timedependent magnetic
ﬁeld rotating in the xy plane. Thus,
B = B
0
^z +B
1
(cos ωt ^ x + sinωt ^ y), (6.170)
where B
0
and B
1
are constants, with B
1
¸ B
0
. The rotating magnetic ﬁeld usu
ally represents the magnetic component of an electromagnetic wave propagating
along the zaxis. In this system, the electric component of the wave has no effect.
The Hamiltonian is written
H = −µB = H
0
+H
1
, (6.171)
where
H
0
=
e B
0
m
e
S
z
, (6.172)
and
H
1
=
e B
1
m
e
(cos ωt S
x
+ sinωt S
y
) . (6.173)
The eigenstates of the unperturbed Hamiltonian are the ‘spin up’ and ‘spin
down’ states, denoted +) and −), respectively. Thus,
H
0
±) = ±
e ¯hB
0
2 m
e
±). (6.174)
The timedependent Hamiltonian can be written
H
1
=
e B
1
2 m
e
_
exp( i ωt) S
−
+ exp(−i ωt) S
+
_
, (6.175)
where S
+
and S
−
are the conventional raising and lowering operators for the spin
angular momentum. It follows that
¸+H
1
+) = ¸−H
1
−) = 0, (6.176)
and
¸−H
1
+) = ¸+H
1
−)
∗
=
e ¯hB
1
2 m
e
exp( i ωt). (6.177)
149
6.12 The Dyson series 6 APPROXIMATION METHODS
It can be seen that this system is exactly the same as the twostate system
discussed in the previous section, provided that we make the identiﬁcations
1) → −), (6.178)
2) → +), (6.179)
ω
21
→
e B
0
m
e
, (6.180)
γ →
e ¯hB
1
2 m
e
. (6.181)
The resonant frequency, ω
21
, is simply the spin precession frequency for an elec
tron in a uniform magnetic ﬁeld of strength B
0
. In the absence of the perturba
tion, the expectation values of S
x
and S
y
oscillate because of the spin precession,
but the expectation value of S
z
remains invariant. If we now apply a magnetic
perturbation rotating at the resonant frequency then, according to the analysis of
the previous section, the system undergoes a succession of spinﬂops, +)
÷
÷−),
in addition to the spin precession. We also know that if the oscillation frequency
of the applied ﬁeld is very different from the resonant frequency then there is
virtually zero probability of the ﬁeld triggering a spinﬂop. The width of the
resonance (in frequency) is determined by the strength of the oscillating mag
netic perturbation. Experimentalist are able to measure the magnetic moments
of electrons, and other spin onehalf particles, to a high degree of accuracy by
placing the particles in a magnetic ﬁeld, and subjecting them to an oscillating
magnetic ﬁeld whose frequency is gradually scanned. By determining the reso
nant frequency (i.e., the frequency at which the particles absorb energy from the
oscillating ﬁeld), it is possible to calculate the magnetic moment.
6.12 The Dyson series
Let us now try to ﬁnd approximate solutions of Eq. (6.154) for a general system.
It is convenient to work in terms of the time evolution operator, U(t
0
, t), which is
deﬁned
A, t
0
, t) = U(t
0
, t) A). (6.182)
150
6.12 The Dyson series 6 APPROXIMATION METHODS
Here, A, t
0
, t) is the state ket of the system at time t, given that the state ket at
the initial time t
0
is A). It is easily seen that the time evolution operator satisﬁes
the differential equation
i ¯h
∂U(t
0
, t)
∂t
= (H
0
+H
1
) U(t
0
, t), (6.183)
subject to the boundary condition
U(t
0
, t
0
) = 1. (6.184)
In the absence of the external perturbation, the time evolution operator re
duces to
U(t
0
, t) = exp[−i H
0
(t −t
0
)/¯h]. (6.185)
Let us switch on the perturbation and look for a solution of the form
U(t
0
, t) = exp[−i H
0
(t −t
0
)/¯h] U
I
(t
0
, t). (6.186)
It is readily demonstrated that U
I
satisﬁes the differential equation
i ¯h
∂U
I
(t
0
, t)
∂t
= H
I
(t
0
, t) U
I
(t
0
, t), (6.187)
where
H
I
(t
0
, t) = exp[+i H
0
(t −t
0
)/¯h] H
1
exp[−i H
0
(t −t
0
)/¯h], (6.188)
subject to the boundary condition
U
I
(t
0
, t
0
) = 1. (6.189)
Note that U
I
speciﬁes that component of the time evolution operator which is due
to the timedependent perturbation. Thus, we would expect U
I
to contain all of
the information regarding transitions between different eigenstates of H
0
caused
by the perturbation.
Suppose that the system starts off at time t
0
in the eigenstate i) of the un
perturbed Hamiltonian. The subsequent evolution of the state ket is given by
Eq. (6.149),
i, t
0
, t) =
m
c
m
(t) exp[−i E
m
(t −t
0
)/¯h] m). (6.190)
151
6.12 The Dyson series 6 APPROXIMATION METHODS
However, we also have
i, t
0
, t) = exp[−i H
0
(t −t
0
)/¯h] U
I
(t
0
, t) i). (6.191)
It follows that
c
n
(t) = ¸nU
I
(t
0
, t)i), (6.192)
where use has been made of ¸nm) = δ
nm
. Thus, the probability that the system
is found in state n) at time t, given that it is deﬁnitely in state i) at time t
0
, is
simply
P
i→n
(t
0
, t) = ¸nU
I
(t
0
, t)i)
2
. (6.193)
This quantity is usually termed the transition probability between states i) and
n).
Note that the differential equation (6.187), plus the boundary condition (6.189),
are equivalent to the following integral equation,
U
I
(t
0
, t) = 1 −
i
¯h
_
t
t
0
H
I
(t
0
, t
) U
I
(t
0
, t
) dt
. (6.194)
We can obtain an approximate solution to this equation by iteration:
U
I
(t
0
, t) · 1 −
i
¯h
_
t
t
0
H
I
(t
0
, t
)
_
_
1 −
i
¯h
_
t
t
0
H
I
(t
0
, t
) U
I
(t
0
, t
)
_
_
dt
· 1 −
i
¯h
_
t
t
0
H
I
(t
0
, t
) dt
+
_
−i
¯h
_
2
_
t
t
0
dt
_
t
t
0
H
I
(t
0
, t
) H
I
(t
0
, t
) dt
+ . (6.195)
This expansion is known as the Dyson series. Let
c
n
= c
(0)
n
+c
(1)
n
+c
(2)
n
+ , (6.196)
where the superscript
(1)
refers to a ﬁrstorder term in the expansion, etc. It
follows from Eqs. (6.192) and (6.195) that
c
(0)
n
(t) = δ
in
, (6.197)
152
6.12 The Dyson series 6 APPROXIMATION METHODS
c
(1)
n
(t) = −
i
¯h
_
t
t
0
¸nH
I
(t
0
, t
)i) dt
, (6.198)
c
(2)
n
(t) =
_
−i
¯h
_
2
_
t
t
0
dt
_
t
t
0
¸nH
I
(t
0
, t
) H
I
(t
0
, t
)i) dt
. (6.199)
These expressions simplify to
c
(0)
n
(t) = δ
in
, (6.200)
c
(1)
n
(t) = −
i
¯h
_
t
t
0
exp[ i ω
ni
(t
−t
0
)] H
ni
(t
) dt
, (6.201)
c
(2)
n
(t) =
_
−i
¯h
_
2
m
_
t
t
0
dt
_
t
t
0
dt
exp[ i ω
nm
(t
−t
0
)]
H
nm
(t
) exp[ i ω
mi
(t
−t
0
)] H
mi
(t
), (6.202)
where
ω
nm
=
E
n
−E
m
¯h
, (6.203)
and
H
nm
(t) = ¸nH
1
(t)m). (6.204)
The transition probability between states i and n is simply
P
i→n
(t
0
, t) = c
(0)
n
+c
(1)
n
+c
(2)
n
+ 
2
. (6.205)
According to the above analysis, there is no chance of a transition between
states i) and n) (i ,= n) to zerothorder (i.e., in the absence of the perturbation).
To ﬁrstorder, the transition probability is proportional to the time integral of the
matrix element ¸nH
1
i), weighted by some oscillatory phasefactor. Thus, if the
matrix element is zero, then there is no chance of a ﬁrstorder transition between
states i) and n). However, to secondorder, a transition between states i) and
n) is possible even when the matrix element ¸nH
1
i) is zero.
153
6.13 Constant perturbations 6 APPROXIMATION METHODS
6.13 Constant perturbations
Consider a constant perturbation which is suddenly switched on at time t = 0:
H
1
(t) = 0 for t < 0
H
1
(t) = H
1
for t ≥ 0, (6.206)
where H
1
is timeindependent, but is generally a function of the position, mo
mentum, and spin operators. Suppose that the system is deﬁnitely in state i) at
time t = 0. According to Eqs. (6.200)–(6.202) (with t
0
= 0),
c
(0)
n
(t) = δ
in
, (6.207)
c
(1)
n
(t) = −
i
¯h
H
ni
_
t
0
exp[ i ω
ni
(t
−t)] dt
=
H
ni
E
n
−E
i
[1 − exp( i ω
ni
t)], (6.208)
giving
P
i→n
(t) · c
(1)
n

2
=
4 H
ni

2
E
n
−E
i

2
sin
2
_
_
(E
n
−E
i
) t
2 ¯h
_
_
, (6.209)
for i ,= n. The transition probability between states i) and n) can be written
P
i→n
(t) =
H
ni

2
t
2
¯h
2
sinc
2
_
_
(E
n
−E
i
) t
2 ¯h
_
_
, (6.210)
where
sinc(x) =
sinx
x
. (6.211)
The sinc function is highly oscillatory, and decays like 1/x at large x. It is a
good approximation to say that sinc(x) is small except when x
<
∼
π. It follows
that the transition probability, P
i→n
, is small except when
E
n
−E
i

<
∼
2π¯h
t
. (6.212)
Note that in the limit t → ∞ only those transitions which conserve energy (i.e.,
E
n
= E
i
) have an appreciable probability of occurrence. At ﬁnite t, is is possible
to have transitions which do not exactly conserve energy, provided that
∆E∆t
<
∼
¯h, (6.213)
154
6.13 Constant perturbations 6 APPROXIMATION METHODS
where ∆E = E
n
− E
i
 is change in energy of the system associated with the
transition, and ∆t = t is the time elapsed since the perturbation was switched
on. Clearly, this result is just a manifestation of the wellknown uncertainty re
lation for energy and time. This uncertainty relation is fundamentally different
to the positionmomentum uncertainty relation, since in nonrelativistic quan
tum mechanics position and momentum are operators, whereas time is merely a
parameter.
The probability of a transition which conserves energy (i.e., E
n
= E
i
) is
P
i→n
(t) =
H
in

2
t
2
¯h
2
, (6.214)
where use has been made of sinc(0) = 1. Note that this probability grows quadrat
ically with time. This result is somewhat surprising, since it implies that the prob
ability of a transition occurring in a ﬁxed time interval, t to t +dt, grows linearly
with t, despite the fact that H
1
is constant for t > 0. In practice, there is usually
a group of ﬁnal states, all possessing nearly the same energy as the energy of the
initial state i). It is helpful to deﬁne the density of states, ρ(E), where the num
ber of ﬁnal states lying in the energy range E to E+dE is given by ρ(E) dE. Thus,
the probability of a transition from the initial state i to any of the continuum of
possible ﬁnal states is
P
i→
(t) =
_
P
i→n
(t) ρ(E
n
) dE
n
, (6.215)
giving
P
i→
(t) =
2 t
¯h
_
H
ni

2
ρ(E
n
) sinc
2
(x) dx, (6.216)
where
x = (E
n
−E
i
) t/2 ¯h, (6.217)
and use has been made of Eq. (6.210). We know that in the limit t → ∞ the
function sinc(x) is only nonzero in an inﬁnitesimally narrow range of ﬁnal ener
gies centred on E
n
= E
i
. It follows that, in this limit, we can take ρ(E
n
) and H
ni

2
out of the integral in the above formula to obtain
P
i→[n]
(t) =
2π
¯h
H
ni

2
ρ(E
n
) t
¸
¸
¸
¸
¸
E
n
E
i
, (6.218)
155
6.13 Constant perturbations 6 APPROXIMATION METHODS
where P
i→[n]
denotes the transition probability between the initial state i) and
all ﬁnal states n) which have approximately the same energy as the initial state.
Here, H
ni

2
is the average of H
ni

2
over all ﬁnal states with approximately the
same energy as the initial state. In deriving the above formula, we have made
use of the result
_
∞
−∞
sinc
2
(x) dx = π. (6.219)
Note that the transition probability, P
i→[n]
, is now proportional to t, instead of t
2
.
It is convenient to deﬁne the transition rate, which is simply the transition
probability per unit time. Thus,
w
i→[n]
=
dP
i→[n]
dt
, (6.220)
giving
w
i→[n]
=
2π
¯h
H
ni

2
ρ(E
n
)
¸
¸
¸
¸
¸
E
n
E
i
. (6.221)
This appealingly simple result is known as Fermi’s golden rule. Note that the
transition rate is constant in time (for t > 0): i.e., the probability of a transition
occurring in the time interval t to t + dt is independent of t for ﬁxed dt. Fermi’s
golden rule is sometimes written
w
i→n
=
2π
¯h
H
ni

2
δ(E
n
−E), (6.222)
where it is understood that this formula must be integrated with
_
ρ(E
n
) dE
n
to
obtain the actual transition rate.
Let us now calculate the secondorder term in the Dyson series, using the
constant perturbation (6.206). From Eq. (6.202) we ﬁnd that
c
(2)
n
(t) =
_
−i
¯h
_
2
m
H
nm
H
mi
_
t
0
dt
exp( i ω
nm
t
)
_
t
0
dt
exp( i ω
mi
t )
=
i
¯h
m
H
nm
H
mi
E
m
−E
i
_
t
0
[exp( i ω
ni
t
) − exp( i ω
nm
t
] ) dt
=
i t
¯h
m
H
nm
H
mi
E
m
−E
i
[exp( i ω
ni
t/2) sinc(ω
ni
t/2)
156
6.13 Constant perturbations 6 APPROXIMATION METHODS
−exp( i ω
nm
t/2) sinc(ω
nm
t/2)] . (6.223)
Thus,
c
n
(t) = c
(1)
n
+c
(2)
n
=
i t
¯h
exp( i ω
ni
t/2)
_
_
_
_
H
ni
+
m
H
nm
H
mi
E
m
−E
i
_
_
sinc(ω
ni
t/2)
−
m
H
nm
H
mi
E
m
−E
i
exp( i ω
im
t/2) sinc(ω
nm
t/2)
_
_
, (6.224)
where use has been made of Eq. (6.208). It follows, by analogy with the previous
analysis, that
w
i→[n]
=
2π
¯h
¸
¸
¸
¸
¸
¸
H
ni
+
m
H
nm
H
mi
E
m
−E
i
¸
¸
¸
¸
¸
¸
2
ρ(E
n
)
¸
¸
¸
¸
¸
¸
¸
E
n
E
i
, (6.225)
where the transition rate is calculated for all ﬁnal states, n), with approximately
the same energy as the initial state, i), and for intermediate states, m) whose
energies differ from that of the initial state. The fact that E
m
,= E
i
causes the
last term on the righthand side of Eq. (6.224) to average to zero (due to the
oscillatory phasefactor) during the evaluation of the transition probability.
According to Eq. (6.225), a secondorder transition takes place in two steps.
First, the system makes a nonenergyconserving transition to some intermedi
ate state m). Subsequently, the system makes another nonenergyconserving
transition to the ﬁnal state n). The net transition, from i) to n), conserves en
ergy. The nonenergyconserving transitions are generally termed virtual transi
tions, whereas the energy conserving ﬁrstorder transition is termed a real transi
tion. The above formula clearly breaks down if H
nm
H
mi
,= 0 when E
m
= E
i
.
This problem can be avoided by gradually turning on the perturbation: i.e.,
H
1
→ exp(ηt) H
1
(where η is very small). The net result is to change the en
ergy denominator in Eq. (6.225) from E
i
−E
m
to E
i
−E
m
+ i ¯hη.
157
6.14 Harmonic perturbations 6 APPROXIMATION METHODS
6.14 Harmonic perturbations
Consider a perturbation which oscillates sinusoidally in time. This is usually
called a harmonic perturbation. Thus,
H
1
(t) = V exp( i ωt) +V
†
exp(−i ωt), (6.226)
where V is, in general, a function of position, momentum, and spin operators.
Let us initiate the system in the eigenstate i) of the unperturbed Hamiltonian,
H
0
, and switch on the harmonic perturbation at t = 0. It follows from Eq. (6.201)
that
c
(1)
n
=
−i
¯h
_
t
0
_
V
ni
exp(i ωt
) +V
†
ni
exp(−i ωt
)
_
exp( i ω
ni
t
) dt
, (6.227)
=
1
¯h
_
_
1 − exp[ i (ω
ni
+ω) t]
ω
ni
+ω
V
ni
+
1 − exp[ i (ω
ni
−ω) t]
ω
ni
−ω
V
†
ni
_
_
,
where
V
ni
= ¸nVi), (6.228)
V
†
ni
= ¸nV
†
i) = ¸iVn)
∗
. (6.229)
This formula is analogous to Eq. (6.208), provided that
ω
ni
=
E
n
−E
i
¯h
→ω
ni
±ω. (6.230)
Thus, it follows fromthe previous analysis that the transition probability P
i→n
(t) =
c
(1)
n

2
is only appreciable in the limit t →∞if
ω
ni
+ω · 0 or E
n
· E
i
− ¯hω, (6.231)
ω
ni
−ω · 0 or E
n
· E
i
+ ¯hω. (6.232)
Clearly, (6.231) corresponds to the ﬁrst termon the righthand side of Eq. (6.227),
and (6.232) corresponds to the second term. The former term describes a process
by which the system gives up energy ¯hω to the perturbing ﬁeld, whilst making a
transition to a ﬁnal state whose energy level is less than that of the initial state
158
6.15 Absorption and stimulated emission of radiation 6 APPROXIMATION METHODS
by ¯hω. This process is known as stimulated emission. The latter term describes
a process by which the system gains energy ¯hω from the perturbing ﬁeld, whilst
making a transition to a ﬁnal state whose energy level exceeds that of the initial
state by ¯hω. This process is known as absorption. In both cases, the total energy
(i.e., that of the system plus the perturbing ﬁeld) is conserved.
By analogy with Eq. (6.221),
w
i→[n]
=
2π
¯h
V
ni

2
ρ(E
n
)
¸
¸
¸
¸
¸
E
n
=E
i
−¯hω
, (6.233)
w
i→[n]
=
2π
¯h
V
†
ni

2
ρ(E
n
)
¸
¸
¸
¸
¸
E
n
=E
i
+¯hω
. (6.234)
Equation (6.233) speciﬁes the transition rate for stimulated emission, whereas
Eq. (6.234) gives the transition rate for absorption. These equations are more
usually written
w
i→n
=
2π
¯h
V
ni

2
δ(E
n
−E
i
+ ¯hω), (6.235)
w
i→n
=
2π
¯h
V
†
ni

2
δ(E
n
−E
i
− ¯hω). (6.236)
It is clear from Eqs. (6.228)(6.229) that V
†
ni

2
= V
ni

2
. It follows from
Eqs. (6.233)–(6.234) that
w
i→[n]
ρ(E
n
)
=
w
n→[i]
ρ(E
i
)
. (6.237)
In other words, the rate of stimulated emission, divided by the density of ﬁnal
states for stimulated emission, equals the rate of absorption, divided by the den
sity of ﬁnal states for absorption. This result, which expresses a fundamental
symmetry between absorption and stimulated emission, is known as detailed bal
ancing, and is very important in statistical mechanics.
6.15 Absorption and stimulated emission of radiation
Let us use some of the results of timedependent perturbation theory to inves
tigate the interaction of an atomic electron with classical (i.e., nonquantized)
159
6.15 Absorption and stimulated emission of radiation 6 APPROXIMATION METHODS
electromagnetic radiation.
The unperturbed Hamiltonian is
H
0
=
p
2
2 m
e
+V
0
(r). (6.238)
The standard classical prescription for obtaining the Hamiltonian of a particle of
charge q in the presence of an electromagnetic ﬁeld is
p → p +qA, (6.239)
H → H−qφ, (6.240)
where A(r) is the vector potential and φ(r) is the scalar potential. Note that
E = −∇φ −
∂A
∂t
, (6.241)
B = ∇A. (6.242)
This prescription also works in quantum mechanics. Thus, the Hamiltonian of an
atomic electron placed in an electromagnetic ﬁeld is
H =
(p −e A)
2
2 m
e
+e φ +V
0
(r), (6.243)
where A and φ are functions of the position operators. The above equation can
be written
H =
_
p
2
−e Ap −e pA +e
2
A
2
_
2 m
e
+e φ +V
0
(r). (6.244)
Now,
pA = Ap, (6.245)
provided that we adopt the gauge ∇A = 0. Hence,
H =
p
2
2 m
e
−
e Ap
m
e
+
e
2
A
2
2 m
e
+e φ +V
0
(r). (6.246)
Suppose that the perturbation corresponds to a monochromatic planewave,
for which
φ = 0, (6.247)
A = 2 A
0
cos
_
ω
c
nr −ωt
_
, (6.248)
160
6.15 Absorption and stimulated emission of radiation 6 APPROXIMATION METHODS
where and n are unit vectors which specify the direction of polarization and
the direction of propagation, respectively. Note that n = 0. The Hamiltonian
becomes
H = H
0
+H
1
(t), (6.249)
with
H
0
=
p
2
2 m
e
+V(r), (6.250)
and
H
1
· −
e Ap
m
e
, (6.251)
where the A
2
term, which is second order in A
0
, has been neglected.
The perturbing Hamiltonian can be written
H
1
= −
e A
0
p
m
e
(exp[ i (ω/c) nr − i ωt] + exp[−i (ω/c) nr + i ωt]) . (6.252)
This has the same form as Eq. (6.226), provided that
V = −
e A
0
p
m
e
exp[−i (ω/c) nr ] (6.253)
It is clear, by analogy with the previous analysis, that the ﬁrst term on the right
hand side of Eq. (6.252) describes the absorption of a photon of energy ¯hω,
whereas the second term describes the stimulated emission of a photon of energy
¯hω. It follows from Eq. (6.236) that the rate of absorption is
w
i→n
=
2π
¯h
e
2
m
2
e
A
0

2
¸n exp[ i (ω/c) nr] pi)
2
δ(E
n
−E
i
− ¯hω). (6.254)
The absorption crosssection is deﬁned as the ratio of the power absorbed by
the atom to the incident power per unit area in the electromagnetic ﬁeld. Now
the energy density of an electromagnetic ﬁeld is
U =
1
2
_
_
0
E
2
0
2
+
B
2
0
2 µ
0
_
_
, (6.255)
161
6.16 The electric dipole approximation 6 APPROXIMATION METHODS
where E
0
and B
0
= E
0
/c = 2 A
0
ω/c are the peak electric and magnetic ﬁeld
strengths, respectively. The incident power per unit area of the electromagnetic
ﬁeld is
c U = 2
0
c ω
2
A
0

2
. (6.256)
Now,
σ
abs
=
¯hωw
i→n
c U
, (6.257)
so
σ
abs
=
πe
2
0
m
2
e
ωc
¸n exp[ i (ω/c) nr] pi)
2
δ(E
n
−E
i
− ¯hω). (6.258)
6.16 The electric dipole approximation
In general, the wavelength of the type of electromagnetic radiation which in
duces, or is emitted during, transitions between different atomic energy levels is
much larger than the typical size of a light atom. Thus,
exp[ i (ω/c) nr] = 1 + i
ω
c
nr + , (6.259)
can be approximated by its ﬁrst term, unity (remember that ω/c = 2π/λ). This
approximation is known as the electric dipole approximation. It follows that
¸n exp[ i (ω/c) nr] pi) · ¸npi). (6.260)
It is readily demonstrated that
[r, H
0
] =
i ¯hp
m
e
, (6.261)
so
¸npi) = −i
m
e
¯h
¸n[r, H
0
]i) = i m
e
ω
ni
¸nri). (6.262)
Using Eq. (6.258), we obtain
σ
abs
= 4π
2
αω
ni
¸nri)
2
δ(ω−ω
ni
), (6.263)
162
6.16 The electric dipole approximation 6 APPROXIMATION METHODS
where α = e
2
/(2
0
hc) = 1/137 is the ﬁne structure constant. It is clear that if
the absorption crosssection is regarded as a function of the applied frequency,
ω, then it exhibits a sharp maximum at ω = ω
ni
= (E
n
−E
i
)/¯h.
Suppose that the radiation is polarized in the zdirection, so that = ^z. We
have already seen, from Sect. 6.4, that ¸nzi) = 0 unless the initial and ﬁnal
states satisfy
∆l = ±1, (6.264)
∆m = 0. (6.265)
Here, l is the quantum number describing the total orbital angular momentum
of the electron, and m is the quantum number describing the projection of the
orbital angular momentum along the zaxis. It is easily demonstrated that ¸nxi)
and ¸nyi) are only nonzero if
∆l = ±1, (6.266)
∆m = ±1. (6.267)
Thus, for generally directed radiation ¸nri) is only nonzero if
∆l = ±1, (6.268)
∆m = 0, ±1. (6.269)
These are termed the selection rules for electric dipole transitions. It is clear, for
instance, that the electric dipole approximation allows a transition from a 2p
state to a 1s state, but disallows a transition from a 2s to a 1s state. The latter
transition is called a forbidden transition.
Forbidden transitions are not strictly forbidden. Instead, they take place at a
far lower rate than transitions which are allowed according to the electric dipole
approximation. After electric dipole transitions, the next most likely type of tran
sition is a magnetic dipole transition, which is due to the interaction between the
electron spin and the oscillating magnetic ﬁeld of the incident electromagnetic
radiation. Magnetic dipole transitions are typically about 10
5
times more unlikely
than similar electric dipole transitions. The ﬁrstorder term in Eq. (6.259) yields
163
6.16 The electric dipole approximation 6 APPROXIMATION METHODS
socalled electric quadrupole transitions. These are typically about 10
8
times more
unlikely than electric dipole transitions. Magnetic dipole and electric quadrupole
transitions satisfy different selection rules than electric dipole transitions: for
instance, the selection rules for electric quadrupole transitions are ∆l = 0, ±2.
Thus, transitions which are forbidden as electric dipole transitions may well be
allowed as magnetic dipole or electric quadrupole transitions.
Integrating Eq. (6.263) over all possible frequencies of the incident radiation
yields
_
σ
abs
(ω) dω =
n
4π
2
αω
ni
¸nri)
2
. (6.270)
Suppose, for the sake of deﬁniteness, that the incident radiation is polarized in
the xdirection. It is easily demonstrated that
[x, [x, H
0
] ] = −
¯h
2
m
e
. (6.271)
Thus,
¸i[x, [x, H
0
] ]i) = ¸ix
2
H
0
+H
0
x
2
−2 x H
0
xi) = −
¯h
2
m
e
, (6.272)
giving
2
n
(¸ixn)E
i
¸nxi) −¸ixn)E
n
¸nxi)) = −
¯h
2
m
e
. (6.273)
It follows that
2 m
e
¯h
n
ω
ni
¸nxi)
2
= 1. (6.274)
This is known as the ThomasReicheKuhn sum rule. According to this rule,
Eq. (6.270) reduces to
_
σ
abs
(ω) dω =
2π
2
α¯h
m
e
=
πe
2
2
0
m
e
c
. (6.275)
Note that ¯h has dropped out of the ﬁnal result. In fact, the above formula is ex
actly the same as that obtained classically by treating the electron as an oscillator.
164
6.17 Energyshifts and decaywidths 6 APPROXIMATION METHODS
6.17 Energyshifts and decaywidths
We have examined how a state n), other than the initial state i), becomes popu
lated as a result of some timedependent perturbation applied to the system. Let
us now consider how the initial state becomes depopulated.
It is convenient to gradually turn on the perturbation from zero at t = −∞.
Thus,
H
1
(t) = exp(ηt) H
1
, (6.276)
where η is small and positive, and H
1
is a constant.
In the remote past, t →−∞, the system is assumed to be in the initial state i).
Thus, c
i
(t → −∞) = 1, and c
n=i
(t → −∞) = 0. Basically, we want to calculate
the time evolution of the coefﬁcient c
i
(t). First, however, let us check that our
previous Fermi golden rule result still applies when the perturbing potential is
turned on slowly, instead of very suddenly. For c
n=i
(t) we have fromEqs. (6.200)–
(6.201) that
c
(0)
n
(t) = 0, (6.277)
c
(1)
n
(t) = −
i
¯h
H
ni
_
t
−∞
exp[ (η + i ω
ni
)t
] dt
= −
i
¯h
H
ni
exp[ (η + i ω
ni
)t ]
η + i ω
ni
, (6.278)
where H
ni
= ¸nH
1
i). It follows that, to ﬁrstorder, the transition probability
from state i) to state n) is
P
i→n
(t) = c
(1)
n

2
=
H
ni

2
¯h
2
exp(2 ηt)
η
2
+ω
2
ni
. (6.279)
The transition rate is given by
w
i→n
(t) =
dP
i→n
dt
=
2 H
ni

2
¯h
2
ηexp(2 ηt)
η
2
+ω
2
ni
. (6.280)
Consider the limit η →0. In this limit, exp(ηt) →1, but
lim
η→0
η
η
2
+ω
2
ni
= πδ(ω
ni
) = π¯hδ(E
n
−E
i
). (6.281)
165
6.17 Energyshifts and decaywidths 6 APPROXIMATION METHODS
Thus, Eq. (6.280) yields the standard Fermi golden rule result
w
i→n
=
2π
¯h
H
ni

2
δ(E
n
−E
i
). (6.282)
It is clear that the deltafunction in the above formula actually represents a func
tion which is highly peaked at some particular energy. The width of the peak is
determined by how fast the perturbation is switched on.
Let us now calculate c
i
(t) using Eqs. (6.200)–(6.202). We have
c
(0)
i
(t) = 1, (6.283)
c
(1)
i
(t) = −
i
¯h
H
ii
_
t
−∞
exp(ηt
) dt
= −
i
¯h
H
ii
exp(ηt)
η
, (6.284)
c
(2)
i
(t) =
_
−i
¯h
_
2
m
H
mi

2
_
t
−∞
dt
_
t
−∞
dt
exp[ (η + i ω
im
)t
] exp[ (η + i ω
mi
)t
],
=
_
−i
¯h
_
2
m
H
mi

2
exp(2 ηt)
2 η(η + i ω
mi
)
. (6.285)
Thus, to secondorder we have
c
i
(t) · 1 +
_
−i
¯h
_
H
ii
exp(ηt)
η
+
_
−i
¯h
_
2
H
ii

2
exp(2 ηt)
2 η
2
+
_
−i
¯h
_
m=i
H
mi

2
exp(2 ηt)
2 η(E
i
−E
m
+ i ¯hη)
. (6.286)
Let us now consider the ratio ˙ c
i
/c
i
, where ˙ c
i
≡ dc
i
/dt. Using Eq. (6.286), we
can evaluate this ratio in the limit η →0. We obtain
˙ c
i
c
i
·
_
¸
_
_
−i
¯h
_
H
ii
+
_
−i
¯h
_
2
H
ii

2
η
+
_
−i
¯h
_
m=i
H
mi

2
E
i
−E
m
+ i ¯hη
_
¸
_
__
1 −
i
¯h
H
ii
η
_
·
_
−i
¯h
_
H
ii
+ lim
η→0
_
−i
¯h
_
m=i
H
mi

2
E
i
−E
m
+ i ¯hη
. (6.287)
166
6.17 Energyshifts and decaywidths 6 APPROXIMATION METHODS
This result is formally correct to secondorder in perturbed quantities. Note that
the righthand side of Eq. (6.287) is independent of time. We can write
˙ c
i
c
i
=
_
−i
¯h
_
∆
i
, (6.288)
where
∆
i
= H
ii
+ lim
η→0
m=i
H
mi

2
E
i
−E
m
+ i ¯hη
(6.289)
is a constant. According to a wellknown result in pure mathematics,
lim
→0
1
x + i
= P
1
x
− i πδ(x), (6.290)
where > 0, and P denotes the principle part. It follows that
∆
i
= H
ii
+P
m=i
H
mi

2
E
i
−E
m
− i π
m=i
H
mi

2
δ(E
i
−E
m
). (6.291)
It is convenient to normalize the solution of Eq. (6.288) so that c
i
(0) = 1.
Thus, we obtain
c
i
(t) = exp
_
−i ∆
i
t
¯h
_
. (6.292)
According to Eq. (6.149), the time evolution of the initial state ket i) is given by
i, t) = exp[−i (∆
i
+E
i
) t/¯h] i). (6.293)
We can rewrite this result as
i, t) = exp(−i [E
i
+ Re(∆
i
) ] t/¯h) exp[ Im(∆
i
) t/¯h] i). (6.294)
It is clear that the real part of ∆
i
gives rise to a simple shift in energy of state i),
whereas the imaginary part of ∆
i
governs the growth or decay of this state. Thus,
i, t) = exp[−i (E
i
+∆E
i
) t/¯h] exp(−Γ
i
t/2¯h) i), (6.295)
where
∆E
i
= Re(∆
i
) = H
ii
+P
m=i
H
mi

2
E
i
−E
m
, (6.296)
167
6.17 Energyshifts and decaywidths 6 APPROXIMATION METHODS
and
Γ
i
¯h
= −
2 Im(∆
i
)
¯h
=
2π
¯h
m=i
H
mi

2
δ(E
i
−E
m
). (6.297)
Note that the energyshift ∆E
i
is the same as that predicted by standard time
independent perturbation theory.
The probability of observing the system in state i) at time t > 0, given that it
is deﬁnately in state i) at time t = 0, is given by
P
i→i
(t) = c
i

2
= exp(−Γ
i
t/¯h), (6.298)
where
Γ
i
¯h
=
m=i
w
i→m
. (6.299)
Here, use has been made of Eq. (6.222). Clearly, the rate of decay of the initial
state is a simple function of the transition rates to the other states. Note that the
system conserves probability up to secondorder in perturbed quantities, since
c
i

2
+
m=i
c
m

2
· (1 −Γ
i
t/¯h) +
m=i
w
i→m
t = 1. (6.300)
The quantity ∆
i
is called the decaywidth of state i). It is closely related to the
mean lifetime of this state,
τ
i
=
¯h
Γ
i
, (6.301)
where
P
i→i
= exp(−t/τ
i
). (6.302)
According to Eq. (6.294), the amplitude of state i) both oscillates and decays as
time progresses. Clearly, state i) is not a stationary state in the presence of the
timedependent perturbation. However, we can still represent it as a superposi
tion of stationary states (whose amplitudes simply oscillate in time). Thus,
exp[−i (E
i
+∆E
i
) t/¯h] exp(−Γ
i
t/2¯h) =
_
f(E) exp(−i Et/¯h) dE, (6.303)
168
6.17 Energyshifts and decaywidths 6 APPROXIMATION METHODS
where f(E) is the weight of the stationary state with energy E in the superposition.
The Fourier inversion theorem yields
f(E)
2
∝
1
(E − [E
i
+ Re(∆
i
)])
2
+Γ
2
i
/4
. (6.304)
In the absence of the perturbation, f(E)
2
is basically a deltafunction centred on
the unperturbed energy E
i
of state i). In other words, state i) is a stationary state
whose energy is completely determined. In the presence of the perturbation, the
energy of state i) is shifted by Re(∆
i
). The fact that the state is no longer station
ary (i.e., it decays in time) implies that its energy cannot be exactly determined.
Indeed, the energy of the state is smeared over some region of width (in energy)
Γ
i
centred around the shifted energy E
i
+Re(∆
i
). The faster the decay of the state
(i.e., the larger Γ
i
), the more its energy is spread out. This effect is clearly a man
ifestation of the energytime uncertainty relation ∆E∆t ∼ ¯h. One consequence of
this effect is the existence of a natural width of spectral lines associated with the
decay of some excited state to the ground state (or any other lower energy state).
The uncertainty in energy of the excited state, due to its propensity to decay, gives
rise to a slight smearing (in wavelength) of the spectral line associated with the
transition. Strong lines, which correspond to fast transitions, are smeared out
more that weak lines. For this reason, spectroscopists generally favour forbid
den lines for Doppler shift measurements. Such lines are not as bright as those
corresponding to allowed transitions, but they are a lot sharper.
169
7 SCATTERING THEORY
7 Scattering theory
7.1 Introduction
Historically, data regarding quantum phenomena has been obtained from two
main sources—the study of spectroscopic lines, and scattering experiments. We
have already developed theories which account for some aspects of the spectra of
hydrogenlike atoms. Let us now examine the quantum theory of scattering.
7.2 The LipmannSchwinger equation
Consider timeindependent scattering theory, for which the Hamiltonian of the
system is written
H = H
0
+H
1
, (7.1)
where H
0
is the Hamiltonian of a free particle of mass m,
H
0
=
p
2
2 m
, (7.2)
and H
1
represents the nontimevarying source of the scattering. Let φ) be an
energy eigenket of H
0
,
H
0
φ) = Eφ), (7.3)
whose wavefunction ¸r
φ) is φ(r
). This state is a planewave state or, possibly,
a sphericalwave state. Schr¨ odinger’s equation for the scattering problem is
(H
0
+H
1
)ψ) = Eψ), (7.4)
where ψ) is an energy eigenstate of the total Hamiltonian whose wavefunction
¸r
ψ) is ψ(r
). In general, both H
0
and H
0
+ H
1
have continuous energy spectra:
i.e., their energy eigenstates are unbound. We require a solution of Eq. (7.4)
which satisﬁes the boundary condition ψ) → φ) as H
1
→ 0. Here, φ) is a
solution of the free particle Schr¨ odinger equation, (7.3), corresponding to the
same energy eigenvalue.
170
7.2 The LipmannSchwinger equation 7 SCATTERING THEORY
Formally, the desired solution can be written
ψ) = φ) +
1
E −H
0
H
1
ψ). (7.5)
Note that we can recover Eq. (7.4) by operating on the above equation with
E − H
0
, and making use of Eq. (7.3). Furthermore, the solution satisﬁes the
boundary condition ψ) →φ) as H
1
→0. Unfortunately, the operator (E −H
0
)
−1
is singular: i.e., it produces inﬁnities when it operates on an eigenstate of H
0
corresponding to the eigenvalue E. We need a prescription for dealing with these
inﬁnities, otherwise the above solution is useless. The standard prescription is to
make the energy eigenvalue E slightly complex. Thus,
ψ
±
) = φ) +
1
E −H
0
±i
H
1
ψ
±
), (7.6)
where is real, positive, and small. Equation (7.6) is called the LipmannSchwinger
equation, and is nonsingular as long as > 0. The physical signiﬁcance of the ±
signs will become apparent later on.
The LipmannSchwinger equation can be converted into an integral equation
via left multiplication by ¸r. Thus,
ψ
±
(r) = φ(r) +
_
_
r
¸
¸
¸
¸
¸
1
E −H
0
±i
¸
¸
¸
¸
¸
r
_
¸r
H
1
ψ
±
) d
3
r
. (7.7)
Adopting the Schr¨ odinger representation, we can write the scattering problem
(7.4) in the form
(∇
2
+k
2
) ψ(r) =
2 m
¯h
2
¸rH
1
ψ), (7.8)
where
E =
¯h
2
k
2
2 m
. (7.9)
This equation is called Helmholtz’s equation, and can be inverted using standard
Green’s function techniques. Thus,
ψ(r) = φ(r) +
2 m
¯h
2
_
G(r, r
)¸r
H
1
ψ) d
3
r
, (7.10)
171
7.2 The LipmannSchwinger equation 7 SCATTERING THEORY
where
(∇
2
+k
2
) G(r, r
) = δ(r −r
). (7.11)
Note that the solution (7.10) satisﬁes the boundary condition ψ) →φ) as H
1
→
0. As is wellknown, the Green’s function for the Helmholtz problem is given by
G(r, r
) = −
exp(±i k r −r
 )
4πr −r

. (7.12)
Thus, Eq. (7.10) becomes
ψ
±
(r) = φ(r) −
2 m
¯h
2
_
exp(±i k r −r
 )
4πr −r

¸r
H
1
ψ) d
3
r
. (7.13)
A comparison of Eqs. (7.7) and (7.13) suggests that the kernel to Eq. (7.7) takes
the form
_
r
¸
¸
¸
¸
¸
1
E −H
0
±i
¸
¸
¸
¸
¸
r
_
= −
2 m
¯h
2
exp(±i k r −r
 )
4πr −r

. (7.14)
It is not entirely clear that the ± signs correspond on both sides of this equation.
In fact, they do, as is easily proved by a more rigorous derivation of this result.
Let us suppose that the scattering Hamiltonian, H
1
, is only a function of the
position operators. This implies that
¸r
H
1
r) = V(r) δ(r −r
). (7.15)
We can write
¸r
H
1
ψ
±
) =
_
¸r
H
1
r
)¸r
ψ
±
) d
3
r
= V(r
) ψ
±
(r
). (7.16)
Thus, the integral equation (7.13) simpliﬁes to
ψ
±
(r) = φ(r) −
2 m
¯h
2
_
exp(±i k r −r
)
4πr −r

V(r
) ψ
±
(r
) d
3
r
. (7.17)
Suppose that the initial state φ) is a planewave with wavevector k (i.e., a
stream of particles of deﬁnite momentum p = ¯hk). The ket corresponding to
this state is denoted k). The associated wavefunction takes the form
¸rk) =
exp( i kr)
(2π)
3/2
. (7.18)
172
7.2 The LipmannSchwinger equation 7 SCATTERING THEORY
The wavefunction is normalized such that
¸kk
) =
_
¸kr)¸rk
) d
3
r
=
_
exp[−i r(k −k
)]
(2π)
3
d
3
r = δ(k −k
). (7.19)
Suppose that the scattering potential V(r) is only nonzero in some relatively
localized region centred on the origin (r = 0). Let us calculate the wavefunction
ψ(r) a long way from the scattering region. In other words, let us adopt the
ordering r ¸r
. It is easily demonstrated that
r −r
 · r − ^rr
(7.20)
to ﬁrstorder in r
/r, where
^r =
r
r
(7.21)
is a unit vector which points from the scattering region to the observation point.
Let us deﬁne
k
= k^r. (7.22)
Clearly, k
is the wavevector for particles which possess the same energy as the
incoming particles (i.e., k
= k), but propagate from the scattering region to the
observation point. Note that
exp(±i k r −r
 ) · exp(±i k r) exp(∓i k
r
). (7.23)
In the larger limit, Eq. (7.17) reduces to
ψ(r)
±
·
exp( i kr)
(2π)
3/2
−
m
2π¯h
2
exp(±i k r)
r
_
exp(∓i k
r
) V(r
) ψ
±
(r
) d
3
r
. (7.24)
The ﬁrst term on the righthand side is the incident wave. The second term
represents a spherical wave centred on the scattering region. The plus sign (on
ψ
±
) corresponds to a wave propagating away from the scattering region, whereas
the minus sign corresponds to a wave propagating towards the scattering region.
173
7.2 The LipmannSchwinger equation 7 SCATTERING THEORY
It is obvious that the former represents the physical solution. Thus, the wave
function a long way from the scattering region can be written
ψ(r) =
1
(2π)
3/2
_
_
exp( i kr) +
exp( i kr)
r
f(k
, k)
_
_
, (7.25)
where
f(k
, k) = −
(2π)
2
m
¯h
2
_
exp(−i k
r
)
(2π)
3/2
V(r
) ψ(r
) d
3
r
= −
(2π)
2
m
¯h
2
¸k
H
1
ψ). (7.26)
Let us deﬁne the differential crosssection dσ/dΩ as the number of particles
per unit time scattered into an element of solid angle dΩ, divided by the incident
ﬂux of particles. Recall, from Sect. 4, that the probability ﬂux (i.e., the particle
ﬂux) associated with a wavefunction ψ is
j =
¯h
m
Im(ψ
∗
∇ψ). (7.27)
Thus, the probability ﬂux associated with the incident wavefunction,
exp( i kr)
(2π)
3/2
, (7.28)
is
j
inci
=
¯h
(2π)
3
m
k. (7.29)
Likewise, the probability ﬂux associated with the scattered wavefunction,
exp( i k r)
(2π)
3/2
f(k
, k)
r
, (7.30)
is
j
scat
=
¯h
(2π)
3
m
f(k
, k)
2
r
2
k^r. (7.31)
Now,
dσ
dΩ
dΩ =
r
2
dΩj
scat

j
inci

, (7.32)
174
7.3 The Born approximation 7 SCATTERING THEORY
giving
dσ
dΩ
= f(k
, k)
2
. (7.33)
Thus, f(k
, k)
2
gives the differential crosssection for particles with incident mo
mentum ¯hk to be scattered into states whose momentum vectors are directed in
a range of solid angles dΩ about ¯hk
. Note that the scattered particles possess
the same energy as the incoming particles (i.e., k
= k). This is always the case
for scattering Hamiltonians of the form shown in Eq. (7.15).
7.3 The Born approximation
Equation (7.33) is not particularly useful, as it stands, because the quantity
f(k
, k) depends on the unknown ket ψ). Recall that ψ(r) = ¸rψ) is the solu
tion of the integral equation
ψ(r) = φ(r) −
m
2π¯h
2
exp( i k r)
r
_
exp(−i k
r
) V(r
) ψ(r
) d
3
r
, (7.34)
where φ(r) is the wavefunction of the incident state. According to the above
equation the total wavefunction is a superposition of the incident wavefunction
and lots of sphericalwaves emitted from the scattering region. The strength of
the sphericalwave emitted at a given point is proportional to the local value of
the scattering potential, V, as well as the local value of the wavefunction, ψ.
Suppose that the scattering is not particularly strong. In this case, it is reason
able to suppose that the total wavefunction, ψ(r), does not differ substantially
from the incident wavefunction, φ(r). Thus, we can obtain an expression for
f(k
, k) by making the substitution
ψ(r) →φ(r) =
exp( i kr)
(2π)
3/2
. (7.35)
This is called the Born approximation.
The Born approximation yields
f(k
, k) · −
m
2π¯h
2
_
exp[ i (k −k
)r
] V(r
) d
3
r
. (7.36)
175
7.3 The Born approximation 7 SCATTERING THEORY
Thus, f(k
, k) is proportional to the Fourier transform of the scattering potential
V(r) with respect to the wavevector q ≡ k −k
.
For a spherically symmetric potential,
f(k
, k) · −
m
2π¯h
2
___
exp( i qr
cos θ
) V(r
) r
2
dr
sinθ
dθ
dφ
, (7.37)
giving
f(k
, k) · −
2 m
¯h
2
q
_
∞
0
r
V(r
) sin(qr
) dr
. (7.38)
Note that f(k
, k) is just a function of q for a spherically symmetric potential. It is
easily demonstrated that
q ≡ k −k
 = 2 k sin(θ/2), (7.39)
where θ is the angle subtended between the vectors k and k
. In other words, θ
is the angle of scattering. Recall that the vectors k and k
have the same length
by energy conservation.
Consider scattering by a Yukawa potential
V(r) =
V
0
exp(−µr)
µr
, (7.40)
where V
0
is a constant and 1/µ measures the “range” of the potential. It follows
from Eq. (7.38) that
f(θ) = −
2 mV
0
¯h
2
µ
1
q
2
+µ
2
, (7.41)
since
_
∞
0
exp(−µr
) sin(qr
) dr
=
q
µ
2
+q
2
. (7.42)
Thus, in the Born approximation, the differential crosssection for scattering by a
Yukawa potential is
dσ
dΩ
·
_
_
2 mV
0
¯h
2
µ
_
_
2
1
[2 k
2
(1 − cos θ) +µ
2
]
2
, (7.43)
given that
q
2
= 4 k
2
sin
2
(θ/2) = 2 k
2
(1 − cos θ). (7.44)
176
7.3 The Born approximation 7 SCATTERING THEORY
The Yukawa potential reduces to the familiar Coulomb potential as µ → 0,
provided that V
0
/µ → ZZ
e
2
/4π
0
. In this limit the Born differential cross
section becomes
dσ
dΩ
·
_
_
2 mZZ
e
2
4π
0
¯h
2
_
_
2
1
16 k
4
sin
4
(θ/2)
. (7.45)
Recall that ¯hk is equivalent to p, so the above equation can be rewritten
dσ
dΩ
·
_
_
ZZ
e
2
16π
0
E
_
_
2
1
sin
4
(θ/2)
, (7.46)
where E = p
2
/2 m is the kinetic energy of the incident particles. Equation (7.46)
is the classical Rutherford scattering crosssection formula.
The Born approximation is valid provided that ψ(r) is not too different from
φ(r) in the scattering region. It follows, from Eq. (7.17), that the condition for
ψ(r) · φ(r) in the vicinity of r = 0 is
¸
¸
¸
¸
¸
¸
m
2π¯h
2
_
exp( i k r
)
r
V(r
) d
3
r
¸
¸
¸
¸
¸
¸
¸1. (7.47)
Consider the special case of the Yukawa potential. At low energies, (i.e., k ¸ µ)
we can replace exp( i k r
) by unity, giving
2 m
¯h
2
V
0

µ
2
¸1 (7.48)
as the condition for the validity of the Born approximation. The condition for the
Yukawa potential to develop a bound state is
2 m
¯h
2
V
0

µ
2
≥ 2.7, (7.49)
where V
0
is negative. Thus, if the potential is strong enough to form a bound
state then the Born approximation is likely to break down. In the highk limit,
Eq. (7.47) yields
2 m
¯h
2
V
0

µk
¸1. (7.50)
This inequality becomes progressively easier to satisfy as k increases, implying
that the Born approximation is more accurate at high incident particle energies.
177
7.4 Partial waves 7 SCATTERING THEORY
7.4 Partial waves
We can assume, without loss of generality, that the incident wavefunction is
characterized by a wavevector k which is aligned parallel to the zaxis. The
scattered wavefunction is characterized by a wavevector k
which has the same
magnitude as k, but, in general, points in a different direction. The direction of k
is speciﬁed by the polar angle θ (i.e., the angle subtended between the two wave
vectors), and an azimuthal angle ϕ about the zaxis. Equation (7.38) strongly
suggests that for a spherically symmetric scattering potential [i.e., V(r) = V(r)]
the scattering amplitude is a function of θ only:
f(θ, ϕ) = f(θ). (7.51)
It follows that neither the incident wavefunction,
φ(r) =
exp( i k z)
(2π)
3/2
=
exp( i k r cos θ)
(2π)
3/2
, (7.52)
nor the total wavefunction,
ψ(r) =
1
(2π)
3/2
_
_
exp( i k r cos θ) +
exp( i k r) f(θ)
r
_
_
, (7.53)
depend on the azimuthal angle ϕ.
Outside the range of the scattering potential, both φ(r) and ψ(r) satisfy the
free space Schr¨ odinger equation
(∇
2
+k
2
) ψ = 0. (7.54)
What is the most general solution to this equation in spherical polar coordinates
which does not depend on the azimuthal angle ϕ? Separation of variables yields
ψ(r, θ) =
l
R
l
(r) P
l
(cos θ), (7.55)
since the Legendre functions P
l
(cos θ) form a complete set in θspace. The Leg
endre functions are related to the spherical harmonics introduced in Sect. 5 via
P
l
(cos θ) =
¸
¸
¸
_
4π
2 l +1
Y
0
l
(θ, ϕ). (7.56)
178
7.4 Partial waves 7 SCATTERING THEORY
Equations (7.54) and (7.55) can be combined to give
r
2
d
2
R
l
dr
2
+2 r
dR
l
dr
+ [k
2
r
2
−l (l +1)]R
l
= 0. (7.57)
The two independent solutions to this equation are called a spherical Bessel func
tion, j
l
(k r), and a Neumann function, η
l
(k r). It is easily demonstrated that
j
l
(y) = y
l
_
−
1
y
d
dy
_
l
siny
y
, (7.58)
η
l
(y) = −y
l
_
−
1
y
d
dy
_
l
cos y
y
. (7.59)
Note that spherical Bessel functions are wellbehaved in the limit y →0 , whereas
Neumann functions become singular. The asymptotic behaviour of these func
tions in the limit y →∞is
j
l
(y) →
sin(y −l π/2)
y
, (7.60)
η
l
(y) → −
cos(y −l π/2)
y
. (7.61)
We can write
exp( i k r cos θ) =
l
a
l
j
l
(k r) P
l
(cos θ), (7.62)
where the a
l
are constants. Note there are no Neumann functions in this expan
sion, because they are not wellbehaved as r → 0. The Legendre functions are
orthonormal,
_
1
−1
P
n
(µ) P
m
(µ) dµ =
δ
nm
n +1/2
, (7.63)
so we can invert the above expansion to give
a
l
j
l
(k r) = (l +1/2)
_
1
−1
exp( i k r µ) P
l
(µ) dµ. (7.64)
It is wellknown that
j
l
(y) =
(−i)
l
2
_
1
−1
exp( i yµ) P
l
(µ) dµ, (7.65)
179
7.4 Partial waves 7 SCATTERING THEORY
where l = 0, 1, 2, [see Abramowitz and Stegun (Dover, New York NY, 1965),
Eq. 10.1.14]. Thus,
a
l
= i
l
(2 l +1), (7.66)
giving
exp( i k r cos θ) =
l
i
l
(2 l +1) j
l
(k r) P
l
(cos θ). (7.67)
The above expression tells us how to decompose a planewave into a series of
sphericalwaves (or “partial waves”).
The most general solution for the total wavefunction outside the scattering
region is
ψ(r) =
1
(2π)
3/2
l
[A
l
j
l
(k r) +B
l
η
l
(k r)] P
l
(cos θ), (7.68)
where the A
l
and B
l
are constants. Note that the Neumann functions are allowed
to appear in this expansion, because its region of validity does not include the
origin. In the larger limit, the total wavefunction reduces to
ψ(r) ·
1
(2π)
3/2
l
_
_
A
l
sin(k r −l π/2)
k r
−B
l
cos(k r −l π/2)
k r
_
_
P
l
(cos θ), (7.69)
where use has been made of Eqs. (7.60)–(7.61). The above expression can also
be written
ψ(r) ·
1
(2π)
3/2
l
C
l
sin(k r −l π/2 +δ
l
)
k r
P
l
(cos θ), (7.70)
where the sine and cosine functions have been combined to give a sine function
which is phaseshifted by δ
l
.
Equation (7.70) yields
ψ(r) ·
1
(2π)
3/2
l
C
l
exp[ i (k r −l π/2 +δ
l
)] − exp[−i (k r −l π/2 +δ
l
)]
2 i k r
P
l
(cos θ), (7.71)
which contains both incoming and outgoing sphericalwaves. What is the source
of the incoming waves? Obviously, they must be part of the larger asymptotic
180
7.5 The optical theorem 7 SCATTERING THEORY
expansion of the incident wavefunction. In fact, it is easily seen that
φ(r) ·
1
(2π)
3/2
l
i
l
(2l +1)
exp[ i (k r −l π/2)] − exp[−i (k r −l π/2)]
2 i k r
P
l
(cos θ) (7.72)
in the larger limit. Now, Eqs. (7.52) and (7.53) give
(2π)
3/2
[ψ(r) −φ(r)] =
exp( i k r)
r
f(θ). (7.73)
Note that the righthand side consists only of an outgoing spherical wave. This
implies that the coefﬁcients of the incoming spherical waves in the larger ex
pansions of ψ(r) and φ(r) must be equal. It follows from Eqs. (7.71) and (7.72)
that
C
l
= (2 l +1) exp[ i (δ
l
+l π/2)]. (7.74)
Thus, Eqs. (7.71)–(7.73) yield
f(θ) =
∞
l=0
(2 l +1)
exp( i δ
l
)
k
sinδ
l
P
l
(cos θ). (7.75)
Clearly, determining the scattering amplitude f(θ) via a decomposition into par
tial waves (i.e., sphericalwaves) is equivalent to determining the phaseshifts δ
l
.
7.5 The optical theorem
The differential scattering crosssection dσ/dΩ is simply the modulus squared of
the scattering amplitude f(θ). The total crosssection is given by
σ
total
=
_
f(θ)
2
dΩ
=
1
k
2
_
dϕ
_
1
−1
dµ
l
l
(2 l +1) (2 l
+1) exp[ i (δ
l
−δ
l
]
sinδ
l
sinδ
l
P
l
(µ) P
l
(µ), (7.76)
181
7.6 Determination of phaseshifts 7 SCATTERING THEORY
where µ = cos θ. It follows that
σ
total
=
4π
k
2
l
(2 l +1) sin
2
δ
l
, (7.77)
where use has been made of Eq. (7.63). A comparison of this result with Eq. (7.75)
yields
σ
total
=
4π
k
Im[f(0)] , (7.78)
since P
l
(1) = 1. This result is known as the optical theorem. It is a reﬂection
of the fact that the very existence of scattering requires scattering in the forward
(θ = 0) direction in order to interfere with the incident wave, and thereby reduce
the probability current in this direction.
It is usual to write
σ
total
=
∞
l=0
σ
l
, (7.79)
where
σ
l
=
4π
k
2
(2 l +1) sin
2
δ
l
(7.80)
is the lth partial crosssection: i.e., the contribution to the total crosssection from
the lth partial wave. Note that the maximumvalue for the lth partial crosssection
occurs when the phaseshift δ
l
takes the value π/2.
7.6 Determination of phaseshifts
Let us now consider how the phaseshifts δ
l
can be evaluated. Consider a spher
ically symmetric potential V(r) which vanishes for r > a, where a is termed
the range of the potential. In the region r > a, the wavefunction ψ(r) satisﬁes
the freespace Schr¨ odinger equation (7.54). The most general solution which is
consistent with no incoming sphericalwaves is
ψ(r) =
1
(2π)
3/2
∞
l=0
i
l
(2 l +1) A
l
(r) P
l
(cos θ), (7.81)
182
7.6 Determination of phaseshifts 7 SCATTERING THEORY
where
A
l
(r) = exp( i δ
l
) [ cos δ
l
j
l
(k r) − sinδ
l
η
l
(k r) ] . (7.82)
Note that Neumann functions are allowed to appear in the above expression,
because its region of validity does not include the origin (where V ,= 0). The
logarithmic derivative of the lth radial wavefunction A
l
(r) just outside the range
of the potential is given by
β
l+
= k a
_
_
cos δ
l
j
l
(k a) − sinδ
l
η
l
(k a)
cos δ
l
j
l
(k a) − sinδ
l
η
l
(k a)
_
_
, (7.83)
where j
l
(x) denotes dj
l
(x)/dx, etc. The above equation can be inverted to give
tanδ
l
=
k aj
l
(k a) −β
l+
j
l
(k a)
k aη
l
(k a) −β
l+
η
l
(k a)
. (7.84)
Thus, the problem of determining the phaseshift δ
l
is equivalent to that of ob
taining β
l+
.
The most general solution to Schr¨ odinger’s equation inside the range of the
potential (r < a) which does not depend on the azimuthal angle ϕ is
ψ(r) =
1
(2π)
3/2
∞
l=0
i
l
(2 l +1) R
l
(r) P
l
(cos θ), (7.85)
where
R
l
(r) =
u
l
(r)
r
, (7.86)
and
d
2
u
l
dr
2
+
_
_
k
2
−
2m
¯h
2
V −
l (l +1)
r
2
_
_
u
l
= 0. (7.87)
The boundary condition
u
l
(0) = 0 (7.88)
ensures that the radial wavefunction is wellbehaved at the origin. We can
launch a wellbehaved solution of the above equation from r = 0, integrate out
to r = a, and form the logarithmic derivative
β
l−
=
1
(u
l
/r)
d(u
l
/r)
dr
¸
¸
¸
¸
¸
¸
r=a
. (7.89)
183
7.7 Hard sphere scattering 7 SCATTERING THEORY
Since ψ(r) and its ﬁrst derivatives are necessarily continuous for physically ac
ceptible wavefunctions, it follows that
β
l+
= β
l−
. (7.90)
The phaseshift δ
l
is obtainable from Eq. (7.84).
7.7 Hard sphere scattering
Let us test out this scheme using a particularly simple example. Consider scatter
ing by a hard sphere, for which the potential is inﬁnite for r < a, and zero for
r > a. It follows that ψ(r) is zero in the region r < a, which implies that u
l
= 0
for all l. Thus,
β
l−
= β
l+
= ∞, (7.91)
for all l. It follows from Eq. (7.84) that
tanδ
l
=
j
l
(k a)
η
l
(k a)
. (7.92)
Consider the l = 0 partial wave, which is usually referred to as the swave.
Equation (7.92) yields
tanδ
0
=
sin(k a)/k a
−cos(k a)/ka
= −tank a, (7.93)
where use has been made of Eqs. (7.58)–(7.59). It follows that
δ
0
= −k a. (7.94)
The swave radial wave function is
A
0
(r) = exp(−i k a)
[cos k a sink r − sink a cos k r]
k r
= exp(−i k a)
sin[k (r −a)]
k r
. (7.95)
The corresponding radial wavefunction for the incident wave takes the form
˜
A
0
(r) =
sink r
k r
. (7.96)
184
7.7 Hard sphere scattering 7 SCATTERING THEORY
It is clear that the actual l = 0 radial wavefunction is similar to the incident l = 0
wavefunction, except that it is phaseshifted by k a.
Let us consider the low and high energy asymptotic limits of tanδ
l
. Low energy
means k a ¸ 1. In this regime, the spherical Bessel functions and Neumann
functions reduce to:
j
l
(k r) ·
(k r)
l
(2 l +1)!!
, (7.97)
η
l
(k r) · −
(2 l −1)!!
(k r)
l+1
, (7.98)
where n!! = n(n −2) (n −4) 1. It follows that
tanδ
l
=
−(k a)
2 l+1
(2 l +1) [(2 l −1)!!]
2
. (7.99)
It is clear that we can neglect δ
l
, with l > 0, with respect to δ
0
. In other words,
at low energy only swave scattering (i.e., spherically symmetric scattering) is
important. It follows from Eqs. (7.33), (7.75), and (7.94) that
dσ
dΩ
=
sin
2
k a
k
2
· a
2
(7.100)
for k a ¸1. Note that the total crosssection
σ
total
=
_
dσ
dΩ
dΩ = 4πa
2
(7.101)
is four times the geometric crosssection πa
2
(i.e., the crosssection for classical
particles bouncing off a hard sphere of radius a). However, low energy scattering
implies relatively long wavelengths, so we do not expect to obtain the classical
result in this limit.
Consider the high energy limit k a ¸ 1. At high energies, all partial waves
up to l
max
= k a contribute signiﬁcantly to the scattering crosssection. It follows
from Eq. (7.77) that
σ
total
=
4π
k
2
l
max
l=0
(2 l +1) sin
2
δ
l
. (7.102)
185
7.8 Low energy scattering 7 SCATTERING THEORY
With so many l values contributing, it is legitimate to replace sin
2
δ
l
by its average
value 1/2. Thus,
σ
total
=
ka
l=0
2π
k
2
(2 l +1) · 2πa
2
. (7.103)
This is twice the classical result, which is somewhat surprizing, since we might ex
pect to obtain the classical result in the short wavelength limit. For hard sphere
scattering, incident waves with impact parameters less than a must be deﬂected.
However, in order to produce a “shadow” behind the sphere, there must be scat
tering in the forward direction (recall the optical theorem) to produce destruc
tive interference with the incident planewave. In fact, the interference is not
completely destructive, and the shadow has a bright spot in the forward direc
tion. The effective crosssection associated with this bright spot is πa
2
which,
when combined with the crosssection for classical reﬂection, πa
2
, gives the ac
tual crosssection of 2πa
2
.
7.8 Low energy scattering
At low energies (i.e., when 1/k is much larger than the range of the potential)
partial waves with l > 0, in general, make a negligible contribution to the scatter
ing crosssection. It follows that, at these energies, with a ﬁnite range potential,
only swave scattering is important.
As a speciﬁc example, let us consider scattering by a ﬁnite potential well, char
acterized by V = V
0
for r < a, and V = 0 for r ≥ a. Here, V
0
is a constant.
The potential is repulsive for V
0
> 0, and attractive for V
0
< 0. The outside
wavefunction is given by [see Eq. (7.82)]
A
0
(r) = exp( i δ
0
) [j
0
(k r) cos δ
0
−η
0
(k r) sinδ
0
] (7.104)
=
exp( i δ
0
) sin(k r +δ
0
)
k r
, (7.105)
where use has been made of Eqs. (7.58)–(7.59). The inside wavefunction follows
186
7.8 Low energy scattering 7 SCATTERING THEORY
from Eq. (7.87). We obtain
A
0
(r) = B
sink
r
r
, (7.106)
where use has been made of the boundary condition (7.88). Here, B is a constant,
and
E −V
0
=
¯h
2
k
2
2 m
. (7.107)
Note that Eq. (7.106) only applies when E > V
0
. For E < V
0
, we have
A
0
(r) = B
sinhκ r
r
, (7.108)
where
V
0
−E =
¯h
2
κ
2
2 m
. (7.109)
Matching A
0
(r), and its radial derivative at r = a, yields
tan(k a +δ
0
) =
k
k
tank
a (7.110)
for E > V
0
, and
tan(k a +δ
0
) =
k
κ
tanhκ a (7.111)
for E < V
0
.
Consider an attractive potential, for which E > V
0
. Suppose that V
0
 ¸ E
(i.e., the depth of the potential well is much larger than the energy of the inci
dent particles), so that k
¸ k. It follows from Eq. (7.110) that, unless tank
a
becomes extremely large, the righthand side is much less that unity, so replacing
the tangent of a small quantity with the quantity itself, we obtain
k a +δ
0
·
k
k
tank
a. (7.112)
This yields
δ
0
· k a
_
_
tank
a
k
a
−1
_
_
. (7.113)
187
7.9 Resonances 7 SCATTERING THEORY
According to Eq. (7.102), the scattering crosssection is given by
σ
total
·
4π
k
2
sin
2
δ
0
= 4πa
2
_
_
tank
a
k
a
−1
_
_
2
. (7.114)
Now
k
a =
¸
¸
¸
_
k
2
a
2
+
2 mV
0
 a
2
¯h
2
, (7.115)
so for sufﬁciently small values of k a,
k
a ·
¸
¸
¸
_
2 mV
0
 a
2
¯h
2
. (7.116)
It follows that the total (swave) scattering crosssection is independent of the
energy of the incident particles (provided that this energy is sufﬁciently small).
Note that there are values of k
a (e.g., k
a · 4.49) at which δ
0
→ π, and the
scattering crosssection (7.114) vanishes, despite the very strong attraction of the
potential. In reality, the crosssection is not exactly zero, because of contributions
from l > 0 partial waves. But, at low incident energies, these contributions are
small. It follows that there are certain values of V
0
and k which give rise to almost
perfect transmission of the incident wave. This is called the RamsauerTownsend
effect, and has been observed experimentally.
7.9 Resonances
There is a signiﬁcant exception to the independence of the crosssection on en
ergy. Suppose that the quantity
_
2 mV
0
 a
2
/¯h
2
is slightly less than π/2. As the
incident energy increases, k
a, which is given by Eq. (7.115), can reach the value
π/2. In this case, tank
a becomes inﬁnite, so we can no longer assume that the
righthand side of Eq. (7.110) is small. In fact, at the value of the incident energy
when k
a = π/2, it follows from Eq. (7.110) that k a + δ
0
= π/2, or δ
0
· π/2
(since we are assuming that k a ¸1). This implies that
σ
total
=
4π
k
2
sin
2
δ
0
= 4πa
2
_
1
k
2
a
2
_
. (7.117)
188
7.9 Resonances 7 SCATTERING THEORY
Note that the crosssection now depends on the energy. Furthermore, the mag
nitude of the crosssection is much larger than that given in Eq. (7.114) for
k
a ,= π/2 (since k a ¸1).
The origin of this rather strange behaviour is quite simple. The condition
¸
¸
¸
_
2 mV
0
 a
2
¯h
2
=
π
2
(7.118)
is equivalent to the condition that a spherical well of depth V
0
possesses a bound
state at zero energy. Thus, for a potential well which satisﬁes the above equation,
the energy of the scattering system is essentially the same as the energy of the
bound state. In this situation, an incident particle would like to form a bound
state in the potential well. However, the bound state is not stable, since the
system has a small positive energy. Nevertheless, this sort of resonance scattering
is best understood as the capture of an incident particle to form a metastable
bound state, and the subsequent decay of the bound state and release of the
particle. The crosssection for resonance scattering is generally far higher than
that for nonresonance scattering.
We have seen that there is a resonant effect when the phaseshift of the swave
takes the value π/2. There is nothing special about the l = 0 partial wave, so it
is reasonable to assume that there is a similar resonance when the phaseshift of
the lth partial wave is π/2. Suppose that δ
l
attains the value π/2 at the incident
energy E
0
, so that
δ
l
(E
0
) =
π
2
. (7.119)
Let us expand cot δ
l
in the vicinity of the resonant energy:
cot δ
l
(E) = cot δ
l
(E
0
) +
_
dcot δ
l
dE
_
E=E
0
(E −E
0
) + (7.120)
= −
_
_
1
sin
2
δ
l
dδ
l
dE
_
_
E=E
0
(E −E
0
) + . (7.121)
Deﬁning
_
_
dδ
l
(E)
dE
_
_
E=E
0
=
2
Γ
, (7.122)
189
7.9 Resonances 7 SCATTERING THEORY
we obtain
cot δ
l
(E) = −
2
Γ
(E −E
0
) + . (7.123)
Recall, from Eq. (7.80), that the contribution of the lth partial wave to the scat
tering crosssection is
σ
l
=
4π
k
2
(2 l +1) sin
2
δ
l
=
4π
k
2
(2 l +1)
1
1 + cot
2
δ
l
. (7.124)
Thus,
σ
l
·
4π
k
2
(2 l +1)
Γ
2
/4
(E −E
0
)
2
+Γ
2
/4
. (7.125)
This is the famous BreitWigner formula. The variation of the partial crosssection
σ
l
with the incident energy has the form of a classical resonance curve. The
quantity Γ is the width of the resonance (in energy). We can interpret the Breit
Wigner formula as describing the absorption of an incident particle to form a
metastable state, of energy E
0
, and lifetime τ = ¯h/Γ (see Sect. 6.17).
190
Contents
1 Introduction 1.1 Major sources . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2 Fundamental concepts 2.1 The breakdown of classical physics . . . . . . . . . 2.2 The polarization of photons . . . . . . . . . . . . . 2.3 The fundamental principles of quantum mechanics 2.4 Ket space . . . . . . . . . . . . . . . . . . . . . . . 2.5 Bra space . . . . . . . . . . . . . . . . . . . . . . . 2.6 Operators . . . . . . . . . . . . . . . . . . . . . . . 2.7 The outer product . . . . . . . . . . . . . . . . . . 2.8 Eigenvalues and eigenvectors . . . . . . . . . . . . 2.9 Observables . . . . . . . . . . . . . . . . . . . . . . 2.10 Measurements . . . . . . . . . . . . . . . . . . . . 2.11 Expectation values . . . . . . . . . . . . . . . . . . 2.12 Degeneracy . . . . . . . . . . . . . . . . . . . . . . 2.13 Compatible observables . . . . . . . . . . . . . . . 2.14 The uncertainty relation . . . . . . . . . . . . . . . 2.15 Continuous spectra . . . . . . . . . . . . . . . . . . 3 Position and momentum 3.1 Introduction . . . . . . . . . . . 3.2 Poisson brackets . . . . . . . . . 3.3 Wavefunctions . . . . . . . . . . 3.4 Schr¨dinger’s representation  I . o 3.5 Schr¨dinger’s representation  II o 3.6 The momentum representation . 3.7 The uncertainty relation . . . . . 3.8 Displacement operators . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5 5 6 6 7 9 10 14 17 19 20 21 24 25 26 27 28 31 33 33 33 37 39 43 46 48 50
4 Quantum dynamics 55 4.1 Schr¨dinger’s equations of motion . . . . . . . . . . . . . . . . . . 55 o 4.2 Heisenberg’s equations of motion . . . . . . . . . . . . . . . . . . . 59
2
4.3 4.4
Ehrenfest’s theorem . . . . . . . . . . . . . . . . . . . . . . . . . . 61 Schr¨dinger’s waveequation . . . . . . . . . . . . . . . . . . . . . 65 o . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 71 71 74 78 81 84 86 89 91 93 96 97 99 105 110 120 120 120 122 124 129 132 134 140 144 146 149 150 154 158 159
5 Angular momentum 5.1 Orbital angular momentum . . . . . . . . . . 5.2 Eigenvalues of angular momentum . . . . . . 5.3 Rotation operators . . . . . . . . . . . . . . . 5.4 Eigenfunctions of orbital angular momentum 5.5 Motion in a central ﬁeld . . . . . . . . . . . . 5.6 Energy levels of the hydrogen atom . . . . . . 5.7 Spin angular momentum . . . . . . . . . . . 5.8 Wavefunction of a spin onehalf particle . . . 5.9 Rotation operators in spin space . . . . . . . 5.10 Magnetic moments . . . . . . . . . . . . . . . 5.11 Spin precession . . . . . . . . . . . . . . . . . 5.12 Pauli twocomponent formalism . . . . . . . . 5.13 Spin greater than onehalf systems . . . . . . 5.14 Addition of angular momentum . . . . . . . .
6 Approximation methods 6.1 Introduction . . . . . . . . . . . . . . . . . . . . 6.2 The twostate system . . . . . . . . . . . . . . . . 6.3 Nondegenerate perturbation theory . . . . . . . 6.4 The quadratic Stark effect . . . . . . . . . . . . . 6.5 Degenerate perturbation theory . . . . . . . . . . 6.6 The linear Stark effect . . . . . . . . . . . . . . . 6.7 Fine structure . . . . . . . . . . . . . . . . . . . . 6.8 The Zeeman effect . . . . . . . . . . . . . . . . . 6.9 Timedependent perturbation theory . . . . . . . 6.10 The twostate system . . . . . . . . . . . . . . . . 6.11 Spin magnetic resonance . . . . . . . . . . . . . 6.12 The Dyson series . . . . . . . . . . . . . . . . . . 6.13 Constant perturbations . . . . . . . . . . . . . . . 6.14 Harmonic perturbations . . . . . . . . . . . . . . 6.15 Absorption and stimulated emission of radiation
3
. 7. . . . . . . . . . . . . . . . . . . 162 6. . . . . . . 7. . . . . . . . . . . . . . 165 7 Scattering theory 7. . . . . . . . . . . . . . 7. . . . . . . . . . . . . . . . . . . . . . . . . . 7. . . . . . . . 7. .17 Energyshifts and decaywidths . . . . . . . . . . . . . . . . . . . .6. .4 Partial waves . . . . . . . . . . . . . . . .3 The Born approximation . . . . . . . . . . . . . . . . . . . . . . . . . . . .2 The LipmannSchwinger equation 7. . . . . . . . . . . . . . . . . . . . . . . . .16 The electric dipole approximation . . . 170 170 170 175 178 181 182 184 186 188
4
. . . . . . 7. . . . . . . .7 Hard sphere scattering . . . . . . .5 The optical theorem . . . . . . . . . . . . . . .6 Determination of phaseshifts . . . . . . . .9 Resonances . .1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7. . . . . . . . .8 Low energy scattering .
P Feynman. Dirac. Modern quantum mechanics. Volume III (AddisonWesley. E. Sands. 1985).
5
. 1970). Oxford. (OxThe principles of quantum mechanics.J. J. R. P ford University Press. 2nd Edition (John Wiley & Sons. and M.1 Major sources The textbooks which I have consulted most frequently while developing course material are: .B. 4th Edition (revised). Quantum mechanics. Merzbacher. Menlo Park CA.M.A. New York NY. 1958). .1 INTRODUCTION
1
Introduction
1. The Feynman lectures on physics. Sakurai. Reading MA. UK. 1965). Leighton. R. (Benjamin/Cummings.
Waveparticle duality: Classical physics can deal with waves or particles. Stories that physicists at the start of the twentieth century thought that classical physics explained everything. Experimentally. there is no such divergence. Cha. this fundamental problem with classical physics was known and appreciated in the middle of the nineteenth century.2 FUNDAMENTAL CONCEPTS
2
Fundamental concepts
2. electron diffraction) show quite clearly that waves sometimes act as if they were streams of particles.
6
. Experimentally.1 The breakdown of classical physics The necessity for a departure from classical mechanics is clearly demonstrated by: 1. and that there was nothing left to discover. The anomalous stability of atoms and molecules: According to classical physics. Vol.. Incidentally. this is not observed to happen. In fact. where R is the ideal gas constant. 3. each degree of freedom of an atomic or molecular system should contribute R/2 to its molar speciﬁc heat. However. I. The ultraviolet catastrophe: According to classical physics. and the total energy density is ﬁnite. and streams of particles sometimes act as if they were waves. The vibrational degrees of freedom appear to make no contribution at all (except at high temperatures). the energy density of an electromagnetic ﬁeld in vacuum is inﬁnite due to a divergence of energy carried by short wavelength modes. an electron orbiting a nucleus should lose energy by emission of synchrotron radiation. 4. The anomalously low speciﬁc heats of atoms and molecules: According to the equipartition theorem of classical physics. are largely apocryphal (see Feynman.g. various experiments (e. 40). the photoelectric effect. This is completely inexplicable within the framework of classical physics. and gradually spiral in towards the nucleus. 2. only the translational and some rotational degrees of freedom seem to contribute. light interference.
in the latter case. This picture leads to no difﬁculty if the plane of polarization lies parallel or perpendicular to the optic axis of the polaroid. Classical electromagnetic wave theory tells us that if the beam is polarized perpendicular to the optic axis then all of the light is transmitted. which are more usually associated with its wavelike behaviour.2. all of the photons are transmitted. and then look to see whether or not it emerges from the other side. But. or no photon is observed. If we repeat the experiment a great number of times then. In the former case. Let us try to account for these observations at the individual photon level. the polarization properties of light. A beam of light which is plane polarized in a certain direction is made up of a stream of photons which are each plane polarized in that direction. A beam of plane polarized light is passed through a polaroid ﬁlm. Clearly. and a fraction
7
. and. what happens in the case of an obliquely polarized incident beam? The above question is not very precise. and if the light is polarized at an angle α to the axis then a fraction sin 2 α of the beam is transmitted. a fraction sin2 α of the photons are transmitted through the ﬁlm. In particular.2 The polarization of photons It is known experimentally that when plane polarized light is used to eject photoelectrons there is a preferred direction of emission of the electrons. which has the property that it is only transparent to light whose plane of polarization lies perpendicular to its optic axis. is observed. none of the photons are transmitted.2 The polarization of photons
2 FUNDAMENTAL CONCEPTS
2. a polarization can be ascribed to each individual photon in a beam of light. Any photon which is transmitted though the ﬁlm must be polarized perpendicular to the optic axis. Let us reformulate it as a question relating to the result of some experiment which we could perform. Furthermore. The possible results of the experiment are that either a whole photon. also extend to its particlelike behaviour. whose energy is equal to the energy of the incident photon. it is impossible to imagine (in physics) ﬁnding part of a photon on the other side of the ﬁlm. Suppose that we were to ﬁre a single photon at a polaroid ﬁlm. on average. if the beam is polarized parallel to the optic axis then none of the light is transmitted. Consider the following wellknown experiment.
some further description is needed in order to allow the results of this experiment to be correlated with the results of other experiments which can be performed using photons. The further description provided by quantum mechanics is as follows. The above discussion about the results of an experiment with a single obliquely polarized photon incident on a polaroid ﬁlm answers all that can be legitimately asked about what happens to the photon when it reaches the ﬁlm. Since there is nothing special about the orientation of the optic axis in our experiment. we conclude that a photon has a probability sin 2 α of being transmitted as a photon polarized in the plane perpendicular to the optic axis. the oblique polarization state is some sort of superposition of two states of parallel and perpendicular polarization. direction of propagation. or how it changes its direction of polarization. are illegitimate. normally incident on a polaroid ﬁlm. since they do not relate to the outcome of a possible experiment. by abandoning the determinacy of classical theory. but recall that the state of a photon is fully speciﬁed once its energy.2 The polarization of photons
2 FUNDAMENTAL CONCEPTS
cos2 α are absorbed. in all cases. This is a fairly sweeping statement. Questions as to what decides whether the photon is transmitted or not. We have no way of knowing whether an individual obliquely polarized photon is going to be absorbed by or transmitted through a polaroid ﬁlm. In other words. we must conclude that any state of polarization can be regarded as a superposition of two mutually perpendicular states of polarization. with a particular oblique polarization.
8
. It is supposed that a photon polarized obliquely to the optic axis can be regarded as being partly in a state of polarization parallel to the axis. and partly in a state of polarization perpendicular to the axis. and a probability cos2 α of being absorbed. If we imagine performing experiments using monochromatic light. Nevertheless. and adopting a fundamentally probabilistic approach. Note that we have only been able to preserve the individuality of photons. Thus. These values for the probabilities lead to the correct classical limit for a beam containing a large number of photons. then the state of each individual photon in the beam is completely speciﬁed.2. and there is nothing left over to uniquely determine whether the photon is transmitted or absorbed by the ﬁlm. We only know the probability of each event occurring. and polarization are known.
Feynman. In other words.3 The fundamental principles of quantum mechanics There is nothing special about the transmission and absorption of photons through a polaroid ﬁlm.e. In other words. the introduction of indeterminacy into the problem is clearly connected with the act of observation. the indeterminacy is related to the inevitable disturbance of the system associated with the act of observation. such as the interference of photons (see Dirac. Exactly the same conclusions as those outlined above are obtained by studying other simple experiments. an atom. 3. In other words. molecule. Cha. Cha. The effect of making this observation is to force the photon entirely into a state of parallel or perpendicular polarization. Any other questions lie beyond the realms of physics. or particle) in a given state can be regarded as being partly in each of two or more other states. Sect. 2. any state can be regarded as a superposition of two or more other states.3). we are subjecting it to an observation. Otherwise. 5). Note that. we are observing whether it is polarized parallel or perpendicular to the optic axis. Dirac’s razor: Quantum mechanics can only answer questions regarding the outcome of possible experiments. and the SternGerlach experiment (see Sakurai. Such superpositions can be performed in an inﬁnite number of different ways. but is governed by probability laws. The principle of indeterminacy: An observation made on a microscopic system causes it to jump into one or more particular states (which are related to
9
. Which of the two states it will jump into cannot be predicted. In fact. it is transmitted. the photon has to jump suddenly from being partly in each of these two states to being entirely in one or the other of them.3 The fundamental principles of quantum mechanics
2 FUNDAMENTAL CONCEPTS
When we make the photon encounter a polaroid ﬁlm. The principle of superposition of states: Any microscopic system (i.. If the photon jumps into a state of parallel polarization then it is absorbed.2. The study of these simple experiments leads us to formulate the following fundamental principles of quantum mechanics: 1.
2. in this example. I. 1.
1)
Suppose that state A is. the superposition of two different states. Let us term each such motion a state of the system. The ﬁrst of these principles was formulated by quantum physicists (such as Dirac) in the 1920s to fend off awkward questions such as “How can a system suddenly jump from one state into another?”.4 Ket space
2 FUNDAMENTAL CONCEPTS
the type of observation). or “How does a system decide which state to jump into?”. residing in some vector space. According to the principle of superposition of states. There will be various possible motions of the particles or bodies consistent with the laws of force.
2.2. as well as the probability of the system making a particular jump. Thus. The state vector A is conventionally written A . etc. We can represent this state as a particular vector.) interacting according to speciﬁc laws of force. in fact. where the other elements of the space represent all of the other possible states of the system. The ﬁnal principle is still rather vague. any given state can be regarded as a superposition of two or more other states.4 Ket space Consider a microscopic system composed of particles or bodies with speciﬁc properties (mass. which we also label A. a photon with a particular energy. Let us consider a particular microscopic system in a particular state. B and
10
. which we label A: e.. It is impossible to predict into which ﬁnal state a particular system will jump. As we shall see. Such a space is called a ket space (after Dirac). the second principle is the basis for the mathematical formulation of quantum mechanics. The most obvious examples of such quantities are vectors. however the probability of a given system jumping into a given ﬁnal state can be predicted. and polarization. moment of inertia. momentum.g. states must be related to mathematical quantities of a kind which can be added together to give other quantities of the same kind. We need to extend it so that we can predict which possible states a system can jump into after a particular type of observation. (2.
and sin α of state C.2)
where B is the vector relating to the state B. This interrelation is represented in ket space by writing A = B + C . are physically irrelevant. etc.2. By analogy with classical physics. no distinction being made between the directions of the ket vectors A and −A . This new state is represented by cos α B + sin α C (2.4 Ket space
2 FUNDAMENTAL CONCEPTS
C.3)
in ket space. The absence of a state is represented by the null vector 0 in ket space. and plane polarized in the xdirection. In this case. a photon polarized in the ydirection superposed with another photon polarized in the ydirection (with the same energy and momentum) gives the same photon.e. For instance.. however. This implies that the ket vector c1 A + c2 A = (c1 + c2 )A (2. If c1 + c2 = 0 then the superposition process yields nothing at all: i. ket vectors differ from conventional vectors in that their magnitudes. We can do this via a suitably weighted superposition of states B and C. we require cos α of state B. and state C might represent a similar photon plane polarized in the ydirection.5)
for any vector A . For instance.e. the fact
11
. The fact that ket vectors pointing in the same direction represent the same state relates ultimately to the quantization of matter: i. Suppose that we want to construct a state whose plane of polarization makes an arbitrary angle α with the xdirection. All the states of the system are in one to one correspondence with all the possible directions of vectors in the ket space. (2.. Note that we cannot form a new state by superposing a state with itself.and ydirections (by analogy with classical physics). There is. one caveat to the above statements. state B might represent a photon propagating in the zdirection. Thus. the sum of these two states represents a photon whose plane of polarization makes an angle of 45◦ with both the x.4)
corresponds to the same state that A does. The null vector has the fairly obvious property that A + 0 = A . This latter state is represented by B + C in ket space. (2. no state. or lengths.
6)
where c1 and c2 are complex numbers. in Eq. plane polarized in orthogonal directions. if we observe a wave then the amplitude of the wave can take any value between zero and inﬁnity. if we were to represent a classical wave by a vector. a photon. we know from classical physics that a circularly polarized wave is a superposition of two waves of equal amplitude.. or an atom. A general elliptically polarized photon is represented by c1 B + c2 C .7) (2. If we observe a microscopic system then we either see a state (i.3). Well. etc. Thus. Thus.4 Ket space
2 FUNDAMENTAL CONCEPTS
that it comes in irreducible packets called photons. so that R = c1 A + c2 B . then the magnitude. or length. or a molecule. which are in phase quadrature. This suggests that a circularly polarized photon is the superposition of a photon polarized in the xdirection (state B) and a photon polarized in the ydirection (state C). and the direction would correspond to the frequency and wavelength. a circularly polarized photon is represented by B + i C in ket space. (2. Suppose that we want to construct a circularly polarized photon state.2. atoms. In classical physics. with equal weights given to the two states. Suppose that the ket R is expressible linearly in terms of the kets A and B . of the vector would correspond to the amplitude of the wave. (2. We have seen. By analogy with classical physics. We conclude that a ket space must be a complex vector space if it is to properly represent the mutual interrelations between the possible states of a microscopic system. so that two vectors of different lengths pointing in the same direction would represent different wave states.e.) or we see nothing—we can never see a fraction or a multiple of a state. but with the proviso that state C is 90 ◦ out of phase with state B. that any plane polarized state of a photon can be represented as a linear superposition of two orthogonal polarization states in which the weights are real numbers.8)
12
. etc. electrons. we can use complex numbers to simultaneously represent the weighting and relative phase in a linear superposition. (2.
any ket vector (or state) which is expressible linearly in terms of certain others is said to be dependent on them. The possible states of such a system are represented as a ket space whose dimensions are nondenumerably inﬁnite. Some microscopic systems have a ﬁnite number of independent states (e. Such a space can be treated in more or less the same manner as a ﬁnitedimensional space.. a set of ket vectors (or states) are termed independent if none of them are expressible linearly in terms of the others.and ydirections. Such a space is termed a Hilbert space by mathematicians. the ket space which represents the possible polarization states of a photon propagating in the zdirection is twodimensional (the two independent vectors correspond to photons plane polarized in the x. It follows that the state R can be regarded as a linear superposition of the states A and B. Unfortunately. some microscopic systems have a nondenumerably inﬁnite number of independent states (e. or denumerably inﬁnite.4 Ket space
2 FUNDAMENTAL CONCEPTS
We say that R is dependent on A and B .g. the spin states of an electron in a magnetic ﬁeld). dimensions. respectively). Likewise. The possible states of such a system are represented as a ket space whose dimensions are denumerably inﬁnite. the dimensionality of a ket space is equivalent to the number of independent ket vectors it contains.2. The dimensionality of a conventional vector space is deﬁned as the number of independent vectors contained in the space.g. a particle in an inﬁnitely deep. In conclusion. In fact. we can also say that state R is dependent on states A and B. the states of a general microscopic system can be represented as a complex vector space of (possibly) inﬁnite dimensions. If there are N independent states. then the possible states of the system are represented as an Ndimensional ket space.
13
. a free particle). onedimensional potential well). This type of space requires a slightly different treatment to spaces of ﬁnite.. Thus. So.g. Some microscopic systems have a denumerably inﬁnite number of independent states (e.. Likewise.
space]. (2.5 Bra space
2 FUNDAMENTAL CONCEPTS
2.e. where A and B are any two kets in a given ket space. and spitting out a general complex number φA .. labeled A.11)
where the αi are an arbitrary set of complex numbers.10)
A =
i=1
αi i . but since a ket space must be complete if it is to represent the states of a microscopic system. N → ∞). labeled F. It also does so in a deterministic manner: i. a ﬁnitedimensional. The only way the functional F can satisfy Eq.5 Bra space A snack machine inputs coins plus some code entered on a key pad. Only a special subset of denumerably inﬁnite dimensional spaces have this property (i. and (hopefully) outputs a snack.. Note that the input and output of the machine have completely different natures. the same money plus the same code produces the same snack (or the same error message) time after time. satisﬁes F(A + B ) = F(A ) + F(B ). This process is represented mathematically by writing F(A ) = φA . A general linear functional. (2. acting on a general ket vector. this is only strictly true for ﬁnitedimensional spaces.e.
14
. or denumerably inﬁnite dimensional (i.2. Not surprisingly. such functionals are termed linear functionals.e. Imagine a general functional.12)
Actually.9)
Let us narrow our focus to those functionals which preserve the linear dependencies of the ket vectors upon which they operate. we need only consider this special subset.e.
(2.
(2.. We can imagine building a rather abstract snack machine which inputs ket vectors and outputs complex numbers in a deterministic fashion. A general ket vector can be written1
N
(2. labeled F. Consider an Ndimensional ket space [i. Let the i (where i runs from 1 to N) represent N independent ket vectors in this space..10) for all vectors in the ket space is if
N
F(A ) =
i=1
1
fi αi . Mathematicians call such a machine a functional. they are complete).
5 Bra space
2 FUNDAMENTAL CONCEPTS
where the fi are a set of complex numbers relating to the functional. these vectors are written in mirror image notation. speciﬁed by Eq. This type of vector space is called a bra space (after Dirac).17)
. the corresponding bra vector is written
N
A =
i=1
α∗ i. So. There is a one to one correspondence between the elements of the ket space and those of the related bra space. A ←→ A. It follows. · · ·  and  · · · . there is a corresponding element.
There are an inﬁnite number of ways of setting up the correspondence between vectors in a ket space and those in the related bra space.11). where c is a complex number..14)
But.
(2. For a general ket vector A. i
(2. only one of these has any physical signiﬁcance. Let us deﬁne N basis functionals i which satisfy i(j ) = δij .13)
F =
i=1
fi i. More generally. this implies that the set of all possible linear functionals acting on an Ndimensional ket space is itself an Ndimensional vector space.
DC
(2.e.2. Note that bra vectors are quite different in nature to ket vectors (hence. it is dual to the original ket space). that the dual to c A is c∗ A . so that they can never be confused). A is termed the dual vector to A . from the above. Bra space is an example of what mathematicians call a dual vector space (i. (2. However.16)
where the are the complex conjugates of the αi . That is. It follows from the previous three equations that
N
(2. 1 2
15
DC
α∗ i
(2. for every element A of the ket space. c1 A + c2 B ←→ c∗ A + c∗ B. in the bra space.15)
where DC stands for dual correspondence. which it is also convenient to label A. and its constituent vectors (which are actually functionals of the ket space) are called bra vectors.
It follows from Eqs. (2. This expression can be further simpliﬁed to give BA . so the operation can also be written BA .18)
acting on the ket vector A . i
(2.. as will become apparent later. and spits out a complex number.23) A = AA
We can now appreciate the elegance of Dirac’s notation.21)
(2.5 Bra space
2 FUNDAMENTAL CONCEPTS
Recall that a bra vector is a functional which acts on a general ket vector.20) Consider the special case where B → A . where 1 ˜ A .12) and (2. however.e.22)
16
. ˜ Given a ket A which is not the null ket.11)]. It is easily demonstrated that BA = AB ∗ . and (2. According to Eqs. that we can omit the round brackets without causing any ambiguity. if all of the α i are zero in Eq.2. we can deﬁne a normalized ket  A .
(2. 2 An inner product is (almost) analogous to a scalar product between a covariant and contravariant vector in some curvilinear space. (2. This operation is denoted B(A ).
2
AA ≥ 0. (2. (2. (2.19)
Mathematicians term BA the inner product of a bra and a ket. Two kets A and B are said to be orthogonal if AB = 0. (2. The combination of a bra and a ket yields a “bra(c)ket” (which is just a number). which also implies that BA = 0. (2.16). This property of bra and ket vectors is essential for the probabilistic interpretation of quantum mechanics.18). and that The equality sign only holds if A is the null ket [i.
N
BA =
i=1
β∗ α i .20) that AA is a real number. Consider the functional which is dual to the ket vector
N
B =
i=1
βi i
(2.11). Note.12).
25)
17
. Such operators are termed linear operators. of a conventional vector. and X(cA ) = cXA .27) (2. Since A and cA represent the same physical state. Consider a somewhat different machine which inputs a ket vector and spits out another ket vector in a deterministic fashion. and is analogous to the length.6 Operators
2 FUNDAMENTAL CONCEPTS
with the property ˜ ˜ AA = 1. More of this later. it makes sense to require that all kets corresponding to physical states have unit norms. AA is known as the norm or “length” of A .24) Here. for all ket vectors A and B . Kronecker deltafunctions become Dirac deltafunctions. Operator X is linear provided that X(A + B ) = XA + XB . Operators X and Y are said to be equal if XA = YA (2. for all complex numbers c.26) (2. The main differences are that summations over discrete labels become integrations over continuous labels. It is possible to deﬁne a dual bra space for a ket space of nondenumerably inﬁnite dimensions in much the same manner as that described above. or magnitude. Consider an operator labeled X. completeness must be assumed (it cannot be proved). and the normalization convention is somewhat different. We are only interested in operators which preserve the linear dependencies of the ket vectors upon which they act.2.
2.6 Operators We have seen that a functional is a machine which inputs a ket vector and spits out a complex number. Suppose that when this operator acts on a general ket vector A it spits out a new ket vector which is denoted XA . Mathematicians call such a machine an operator. (2.
so we may look on it as the result of some linear operator applied to B. This operator is uniquely determined by the original operator X. Operator X is termed the null operator if XA = 0 (2. Consider the inner product of a general bra B with the ket XA . and A can be written BXA without ambiguity.29) (2. it may be considered to be the inner product of A with some bra. The equation which deﬁnes this vector is ( BX)A = B(XA ) (2.31) (2. Thus. The multiplication is associative: X(YA ) = (X Y)A = X YA . the operator in the middle. However. The triple product of B. Such addition is deﬁned to obey a commutative and associate algebra: X + Y = Y + X. X + (Y + Z) = (X + Y) + Z.28)
for all ket vectors in the space. we have only considered linear operators acting on ket vectors. We can also give a meaning to their operating on bra vectors. Consider the dual bra to XA . provided we adopt the convention that the bra vector always goes on the left. so we might as well call it the same operator acting on B . in general. it may be regarded as the result of some
18
.2. A suitable notation to use for the resulting bra when X operates on B is BX.6 Operators
2 FUNDAMENTAL CONCEPTS
for all kets in the ket space in question. This bra depends antilinearly on A and must therefore depend linearly on A.33) (2. This bra depends linearly on B. it is noncommutative: X Y = Y X.32) (2. X(Y Z) = (X Y)Z = X Y Z. This product is a number which depends linearly on A .30)
So far. Operators can also be multiplied. (2. X. and the ket vector on the right.34)
for any A and B. Operators can be added together. Thus.
(2. Are there any other products we are allowed to form? How about B A ? (2. A Hermitian operator ξ has the special property that it is its own adjoint: i. BXA .38) (2. Thus. It is also easily demonstrated that (B A)† = A B.7 The outer product So far we have formed the following products: BA .37) It is also easily seen that the adjoint of the adjoint of a linear operator is equivalent to the original operator. the product B A is a linear operator. XA .35) It is readily demonstrated that BX† A = AXB ∗ . DC XA ←→ AX† . and is denoted X† . B A acting on a general ket C yields another ket. (2.39)
This clearly depends linearly on the ket A and the bra B . The outer product should not be confused with the inner product.2.. X Y.41)
Mathematicians term the operator B A the outer product of B and A. Suppose that we rightmultiply the above product by the general ket C .40)
since AC is just a number.39) by a general bra C. We obtain B AC = AC B .36)
2. Clearly. (2. AB . (2. as is easily demonstrated by leftmultiplying the expression (2. which is just a number. Thus.e. ξ = ξ† .
19
. (2. This operator also acts on bras. plus (X Y)† = Y † X† . AX.7 The outer product
2 FUNDAMENTAL CONCEPTS
linear operator applied to A. This operator is termed the adjoint of X.
and take the difference. the ket XA is not a constant multiple of A . Suppose that the eigenvalues ξ and ξ are different. are numbers called eigenvalues.
(2..44) where ξ is the eigenket associated with the eigenvalue ξ . x and have the property Xx = x x . (2.8 Eigenvalues and eigenvectors In general. These are denoted ξξ = ξ ξ .. Consider the eigenkets and eigenvalues of a Hermitian operator ξ. there are some special kets known as the eigenkets of operator X. Three important results are readily deduced: (i) The eigenvalues are all real numbers. and the eigenkets corresponding to different eigenvalues are orthogonal. (2.44) (for the eigenvalue ξ ) reads ξ ξ = ξ
∗
ξ . (2.47) where we have used the fact that ξ is not the null ket. (2. we obtain (ξ − ξ ∗ ) ξ ξ = 0. These are denoted x .. (2. (2. rightmultiply the above equation by ξ .. Clearly. x ..45)
If we leftmultiply Eq. x . . This proves that the eigenvalues are real numbers. Since ξ is Hermitian. (2.2.8 Eigenvalues and eigenvectors
2 FUNDAMENTAL CONCEPTS
2. It follows that ξ ξ = 0.46)
Suppose that the eigenvalues ξ and ξ are the same.43) . the dual equation to Eq. . It follows from the above that ξ = ξ ∗.42)
where x . Xx = x x . applying X to one of its eigenkets yields the same eigenket multiplied by the associated eigenvalue.48)
20
. .44) by ξ .. (2. However.
49) (iii) The dual of any eigenket is an eigenbra belonging to the same eigenvalue. We have already seen that kets can be used to represent the possible states of a microscopic system. assume that the dynamical variables of a microscopic system are represented as linear operators acting on the bras and kets which correspond to the various possible states of the system. by placing a polaroid ﬁlm in its path.2.
2. and conversely. and linear operators.g. so we must conclude that bras could just as well be used to represent the states of a microscopic system. there is a one to one correspondence between the elements of a ket space and its dual bra space. (2. Since the lengths of bras and kets have no physical signiﬁcance.9 Observables We have developed a mathematical formalism which comprises three types of objects—bras. We.9 Observables
2 FUNDAMENTAL CONCEPTS
which demonstrates that eigenkets corresponding to different eigenvalues are orthogonal. its position. spin. (ii) The eigenvalues associated with eigenkets are the same as the eigenvalues associated with eigenbras.. However. kets. the only objects we have left over are operators. In general. we cannot predict into which state a given photon will jump
21
. it is reasonable to suppose that nonlinear operators are also without physical signiﬁcance. and the latter state is transmitted (which is how we tell them apart). We have seen that if we observe the polarization state of a photon. An eigenbra of ξ corresponding to an eigenvalue ξ is deﬁned ξ ξ = ξ ξ .)? How can these be represented in our formalism? Well. etc. momentum. in general. spit out bras/kets pointing in different directions when fed bras/kets pointing in the same direction but differing in length. the result is to cause the photon to jump into a state of polarization parallel or perpendicular to the optic axis of the ﬁlm. The former state is absorbed. energy. Note that the operators have to be linear. therefore. otherwise they would. What about the dynamical variables of the system (e.
e. Finally. more generally. Furthermore.9 Observables
2 FUNDAMENTAL CONCEPTS
(except in a statistical sense). Not surprisingly. In general. different results of the measurement) are orthogonal is in accordance with our earlier requirement that the states into which the system jumps should be mutually
22
. Note that the result of the measurement must be a real number (there are no measurement machines which output complex numbers). and yield the same value for the dynamical variable. We can make a second observation of the polarization state of such a photon by placing an identical polaroid ﬁlm (with the same orientation of the optic axis) immediately behind the ﬁrst ﬁlm. There is nothing special about the polarization states of a photon. with one particular value for the dynamical variable. So. and the system is found to be a one particular ﬁnal state.e. by a fairly nonobvious leap of intuition.. The fact that the eigenkets of a Hermitian operator corresponding to different eigenvalues (i. The fact that the result of the measurement must be a real number implies that dynamical variables can only be represented by Hermitian operators (since only Hermitian operators are guaranteed to have real eigenvalues). However. It is clear that the photon will deﬁnitely be transmitted through the second ﬁlm. made immediately after the ﬁrst one. we are going to assert that a measurement of a dynamical variable corresponding to an operator X in ket space causes the system to jump into a state corresponding to one of the eigenkets of X. and if it is initially polarized perpendicular to the axis then it will deﬁnitely be transmitted. each of these ﬁnal states is associated with a different result of the measurement: i. will deﬁnitely ﬁnd the system in the same state. we can say that when a dynamical variable of a microscopic system is measured the system is caused to jump into one of a number of independent states (note that the perpendicular and parallel polarization states of our photon are linearly independent). How can we represent all of these facts in our mathematical formalism? Well. we do know that if the photon is initially polarized parallel to the optic axis then it will deﬁnitely be absorbed. then a second observation.. We also known that after passing though the ﬁlm a photon must be in a state of polarization perpendicular to the optic axis (otherwise it would not have been transmitted).2. such a state is termed an eigenstate. the result of the measurement is the eigenvalue associated with the eigenket into which the system jumps. a different value of the dynamical variable. if an observation is made.
e. Conversely. We can conclude that the result of a measurement of a dynamical variable represented by a Hermitian operator ξ must be one of the eigenvalues of ξ. A Hermitian operator which satisﬁes this condition is termed an observable.9 Observables
2 FUNDAMENTAL CONCEPTS
independent. This follows because the system cannot jump into an eigenstate corresponding to a different eigenvalue of ξ.
23
. a general ket must always be dependent on the eigenkets of ξ. This can only be the case if the eigenkets form a complete set (i. First. since such a state is not dependent on the original state. In other words. However. if the system is in an eigenstate of ξ. independent of) any other eigenstate corresponding to a different eigenvalue. and a dynamical variable and its representative operator. immediately after an observation whose result is a particular eigenvalue ξ .2. any observable quantity must be a Hermitian operator with a complete set of eigenstates. In other words. every eigenvalue of ξ is a possible result of a measurement made on the corresponding dynamical variable. Furthermore. corresponding to an eigenvalue ξ . This gives us the physical signiﬁcance of the eigenvalues. This fairly innocuous statement has two very important corollaries.. they span ket space). the distinction between a state and its representative ket vector.) It is reasonable to suppose that if a certain dynamical variable ξ is measured with the system in a particular state. it stands to reason that a measurement of ξ must always yield some result. then a measurement of ξ is bound to give the result ξ . It follows that no matter what the initial state of the system. for the sake of simplicity. Second. this eigenstate is orthogonal to (i. Thus. Conversely.. the second measurement is bound to give the same result as the ﬁrst.e. It follows that a second measurement made immediately after the ﬁrst one must leave the system in an eigenstate corresponding to the eigenvalue ξ . (From now on. the system is left in the associated eigenstate. it must always be able to jump into one of the eigenstates of ξ. then the states into which the system may jump on account of the measurement are such that the original state is dependent on them. will be dropped. in order for a Hermitian operator ξ to be observable its eigenkets must form a complete set.
It is impossible to determine into which eigenstate a given system will jump. It is convenient to normalize our eigenkets such that they all have unit norms.10 Measurements
2 FUNDAMENTAL CONCEPTS
2. and the transition probability to the same eigenstate ξ is unity. In fact.  Aξ 2 ? This quantity is deﬁnitely a positive number (so it could be a probability). For the moment. So. since Aξ is. Let us try again.10 Measurements We have seen that a measurement of some observable ξ of a microscopic system causes the system to jump into one of the eigenstates of ξ. (2. as a result of a measurement made on the system? Let us start with the simplest case. Can we identify the transition probability to a ﬁnal eigenstate ξ with the inner product Aξ ? The straight answer is “no”. it is the correct guess. in general.51)
. This guess also gives the right answer for the transition probabilities between eigenstates.
24
(2. a complex number. The result of the measurement is the associated eigenvalue (or some function of this quantity).2. and complex probabilities do not make much sense. Since the eigenstates of an observable ξ form a complete set. but it is possible to predict the probability of such a transition. we can express any given state A as a linear combination of them. and zero otherwise. How about if we identify the transition probability with the modulus squared of the inner product. It is easily demonstrated that A =
ξ
ξ
ξ A . It follows from the orthogonality property of the eigenkets that ξ ξ = δξ ξ . Can we use this correspondence to obtain a general rule for calculating transition probabilities? Well. If the system is initially in an eigenstate ξ then the transition probability to a eigenstate ξ corresponding to a different eigenvalue is zero.50)
where δξ ξ is unity if ξ = ξ . we are assuming that the eigenvalues of ξ are all different. Note that the probability of a transition from an initial eigenstate ξ to a ﬁnal eigenstate ξ is the same as the value of the inner product ξ ξ . what is the probability that a system in some initial state A makes a transition to an eigenstate ξ of an observable ξ. suppose that the system is initially in a state A which is not an eigenstate of ξ.
and use has been made of Eq. then this probability simply reduces to P(ξ ) =  Aξ 2 .52)  Aξ 2 . is given by ξ =
ξ
ξ P(ξ ) =
ξ
ξ  Aξ 2
25
. Suppose a measurement of the observable ξ is made on each system. (2. ξ A =
ξ
(2. (2.53)
AA
=
where the summation is over all the different eigenvalues of ξ.20). The relative probability of a transition to an eigenstate ξ . = 2 AA ξ  Aξ  (2.55)
If the ket A is normalized such that its norm is unity.11 Expectation values Consider an ensemble of microscopic systems prepared in the same initial state A .54)
ξ
where 1 denotes the identity operator. which is generally referred to as the expectation value of ξ.11 Expectation values
2 FUNDAMENTAL CONCEPTS
A =
ξ
Aξ Aξ
ξ
ξ .2. Note that all of the above results follow from the extremely useful (and easily proved) result ξ ξ  = 1. What is the mean value of the measurement? This quantity. is P(ξ ) ∝  Aξ 2 . which is equivalent to the relative probability of a measurement of ξ yielding the result ξ . (2.57)
2. and the fact that the eigenstates are mutually orthogonal.56) (2. We know that each measurement yields the value ξ with probability P(ξ ). (2. The absolute probability is clearly P(ξ ) =  Aξ 2  Aξ 2 .
All states are eigenstates of this operator with the eigenvalue unity.16).61) (2.. that we adopted earlier. they are not orthogonal to each other (i.12 Degeneracy
2 FUNDAMENTAL CONCEPTS
=
ξ
ξ Aξ
ξ A =
ξ
Aξξ
ξ A . Note that it is only possible to normalize a given ket A such that Eq. (2. between the elements of a ket space and those of its dual bra space. (2. Degenerate eigenstates are necessarily orthogonal to any eigenstates corresponding to different eigenvalues. since much of the previous formalism depends crucially on the mutual orthogonality of the different eigenstates of an observable.2. ξ1 ξ2 = ξa .21) of the norm. This property depends on the particular correspondence (2. that any linear combination of ξa and ξb is also an eigenstate corresponding to the eigenvalue ξ . (2.12 Degeneracy Suppose that two different eigenstates ξa and ξb of ξ correspond to the same eigenvalue ξ .e. Consider the identity operator. in general. It follows that we can always construct two mutually orthogonal degenerate eigenstates. Note. = ξb − ξa ξb ξa .60) is satisﬁed because of the more general property (2.8 does not work in this case). however. but.54). A1A = AA = 1.
(2.62)
26
.60) for all A . the expectation value of this operator is always unity: i.e.59)
2. the proof of orthogonality given in Sect.. These are termed degenerate eigenstates. (2. 1. For instance. This is unfortunate. Thus.58)
which reduces to ξ = AξA with the aid of Eq. 1 −  ξa ξb 2 (2. 2.
1. ξ and η. What happens if we now make a measurement of η? Well. and another which can measure η. with eigenvalue ξ . For instance. however. Clearly. Is it still possible to measure both observables simultaneously? Let us again make an observation of ξ which throws the system into an eigenstate ξ . Sect. Such observables are termed compatible. if all eigenstates of ξ are also eigenstates of η then it is always possible to make a simultaneous measurement of ξ and η. In principle. We conclude that it is always possible to construct a complete set of mutually orthogonal eigenstates for any given observable. with the eigenvalue η . with eigenvalue η . Suppose.1). suppose that the eigenstate ξ is also an eigenstate of η. Suppose that the system is thrown into an eigenstate η .
2. Suppose that we make a measurement of ξ. We can now make a second observation to determine η. A second measurement of ξ will deﬁnitely give the result ξ . Each eigenstate is again associated with a different possible result of the measurement. In this case. ξ .and zdirections of the spin angular momentum of a spin onehalf particle. the two observables in question might be the projection in the x. This will throw the system into one of the (many) eigenstates of η which depend on ξ . In this sense.13 Compatible observables
2 FUNDAMENTAL CONCEPTS
This result is easily generalized to the case of more than two degenerate eigenstates.2. with eigenvalue ξ . respectively. we can say that the observables ξ and η simultaneously have the values ξ and η .13 Compatible observables Suppose that we wish to simultaneously measure two observables. each of these eigenstates is associated with a different result of the measurement. a measurement of η will deﬁnitely give the result η . Another measurement of ξ will throw the system into one of the (many) eigenstates of ξ which depend on η . that the eigenstates of ξ are not eigenstates of η. of a microscopic system? Let us assume that we possess an apparatus which is capable of measuring ξ. These could be measured using appropriate SternGerlach apparatuses (see Sakurai. It is clear that if the observables ξ and η do not possess simultaneous eigenstates then if the value
27
. and the system is consequently thrown into one of the eigenstates of ξ. and so on.
We say that the two observables are incompatible. Thus. It is convenient to denote this simultaneous eigenstate ξ η . the system is in an eigenstate of ξ) then the value of η is uncertain (i.65)
for each simultaneous eigenstate. the system is not in an eigenstate of η). with eigenvalue η . with eigenvalue ξ .
2. and the second equation by ξ.. The only way that this can be true is if ξ η = η ξ.67) (2.. where A is a general ket. and vice versa.e. (2. Suppose that this is the case. = η ξ η . It is
28
.2. Let a general eigenstate of ξ. and then take the difference. and vice versa. It follows that the simultaneous eigenstates of two observables must also form a complete set.66)
Thus. every eigenstate of ξ should also be an eigenstate of η). the above equation implies that (ξ η − η ξ)A = 0 . Recall that the eigenstates of an observable must form a complete set.e. also be an eigenstate of η.14 The uncertainty relation
2 FUNDAMENTAL CONCEPTS
of ξ is known (i. We have seen that the condition for two observables ξ and η to be simultaneously measurable is that they should possess simultaneous eigenstates (i.e.14 The uncertainty relation We have seen that if ξ and η are two noncommuting observables. We have ξξ η ηξ η = ξ ξ η .63) (2.64)
We can leftmultiply the ﬁrst equation by η. (2. The result is (ξ η − η ξ)ξ η = 0 (2.. the condition for two observables ξ and η to be simultaneously measurable is that they should commute. then a determination of the value of ξ leaves the value of η uncertain.
(2. which is analogous to in Euclidian space. The expectation value of (∆ξ)2 ≡ ∆ξ ∆ξ is termed the variance of ξ. it is easily demonstrated that (∆ξ)2 = ξ2 − ξ 2 .74) (2. we can deﬁne a Hermitian operator ∆ξ = ξ − ξ . If c takes the special value − BA / BB then the above inequality reduces to AA BB −  AB 2 ≥ 0.. which is the same as the Schwarz inequality. ξ . In fact. and is.73)
29
. it is a measure of the width of the distribution of likely values of ξ about the expectation value).69)
The variance of ξ is a measure of the uncertainty in the value of ξ for the particular state in question (i. and a measurement of ξ is bound to give the expectation value.14 The uncertainty relation
2 FUNDAMENTAL CONCEPTS
possible to quantify this uncertainty. = ∆η . (2.70)
where c is any complex number.68) where the expectation value is taken over the particular physical state under consideration. (2. This inequality can be proved by noting that ( A + c∗ B)(A + cB ) ≥ 0. in general.e.71) (2.72) a2  b2 ≥ a · b2 (2. For a general observable ξ.2. nonzero. It is obvious that the expectation value of ∆ξ is zero. Let us substitute A B = ∆ξ . (2. If the variance is zero then there is no uncertainty. Consider the Schwarz inequality AA BB ≥  AB 2 .75) (2.
78) (2. ∆η] ≡ ∆ξ ∆η − ∆η ∆ξ.79)
into the Schwarz inequality. ∆η}. According to this relation. The one exception to this rule is when ξ and η commute. We ﬁnd (∆ξ)2 (∆η)2 ≥  ∆ξ ∆η 2 . The ﬁnal term in the above expression is positive deﬁnite.83) 4 where use has been made of Eq. η] 2 +  {∆ξ.14 The uncertainty relation
2 FUNDAMENTAL CONCEPTS
where use has been made of the fact that ∆ξ and ∆η are Hermitian operators. ∆η} . It is clear that the right hand side of 1 1 [∆ξ. ∆η] . ∆η} .76)
{∆ξ. ∆η} 2 . whereas the expectation value of an antiHermitian operator is a pure imaginary number. and vice versa. η] 2 .82) 4 4 where use has been made of ∆ξ = 0. ∆η} ≡ ∆ξ ∆η + ∆η ∆ξ.2.81) 2 2 consists of the sum of a purely real and a purely imaginary number. Taking the modulus squared of both sides gives ∆ξ ∆η = 1 1  ∆ξ ∆η 2 =  [ξ. an exact knowledge of the value of ξ implies no knowledge whatsoever of the value of η.
30
. ∆η]. The commutator is clearly antiHermitian. in which case exact knowledge of ξ does not necessarily imply no knowledge of η. (2. and the anticommutator.80)
whereas the anticommutator is obviously Hermitian. (2. The above expression is termed the uncertainty relation. (2. [∆ξ. Note that 1 1 (2. etc.
([∆ξ. ∆η] + {∆ξ. Now. (2.76). it is easily demonstrated that the expectation value of a Hermitian operator is a real number. ∆η])† = (∆ξ ∆η − ∆η ∆ξ)† = ∆η ∆ξ − ∆ξ ∆η = − [∆ξ.77) ∆ξ ∆η = [∆ξ. ∆η] + {∆ξ.
(2. {∆ξ. so we can write 1 (∆ξ)2 (∆η)2 ≥  [ξ. are deﬁned [∆ξ. (2. (2. 2 2 where the commutator. where the blank ket  stands for any general ket.
In fact. we have studiously avoided dealing with observables possessing eigenvalues which lie in a continuous range.. Let us assume. ξ can now take a continuous range of values. that eigenstates corresponding to a continuous range of eigenvalues cannot be normalized so that they have unit norms. Note that there are clearly a nondenumerably inﬁnite number of mutually orthogonal eigenstates of ξ. for the sake of simplicity. (2.e. (2.85) where δ(x) denotes the famous Dirac deltafunction. rather than having discrete values. Fortunately. continuous eigenvalues are unavoidable in quantum mechanics. these eigenstates have inﬁnite norms: i. (2.84)
But. Suppose that ξ is an observable with continuous eigenvalues. namely position and momentum.86)
Note that a summation over discrete eigenvalues goes over into an integral over a continuous range of eigenvalues. the dimensionality of ket space is nondenumerably inﬁnite.2. Hence. Note. We can still write the eigenvalue equation as ξξ = ξ ξ . Unfortunately. The extremely useful relation (2. This is the major difference between eigenstates in a ﬁnitedimensional and an inﬁnitedimensional ket space.50) generalizes to ξ ξ = δ(ξ − ξ ).54) generalizes to dξ ξ ξ  = 1. they are inﬁnitely long. generally have continuous eigenvalues. The eigenstates ξ must form a complete set if ξ is to be an observable. The orthogonality condition (2. In fact. It follows that any general ket can be expanded in
31
.15 Continuous spectra
2 FUNDAMENTAL CONCEPTS
2. The reason for this is because continuous eigenvalues imply a ket space of nondenumerably inﬁnite dimension. many of the results we obtained previously for a ﬁnitedimensional ket space with discrete eigenvalues can be generalized to ket spaces of nondenumerably inﬁnite dimensions. also.15 Continuous spectra Up to now. that ξ can take any value. the most important observables of all.
86).89)
A = AA =
dξ Aξ dξ Aξ
These results also follow simply from Eq. Fortunately. or observables whose eigenvalues take on a ﬁnite range of values plus a set of discrete values. ξ A = dξ  Aξ 2 . For instance.51)–(2.88) (2. II and III). In fact.15 Continuous spectra
2 FUNDAMENTAL CONCEPTS
terms of the ξ . In fact. the normalization condition can be written AA = dξ  Aξ 2 = 1. (2. We have seen that it is not possible to normalize the eigenstates ξ such that they have unit norms.2. Both of these cases can be dealt with using a fairly straightforward generalization of the previous analysis (see Dirac. this convenient normalization is still possible for a general state vector. according to Eq. There are number of other cases we could look at.90)
We have now studied observables whose eigenvalues can take a discrete number of values as well as those whose eigenvalues can take any value. (2. Cha.87) (2. (2.89). observables whose eigenvalues can only take a ﬁnite range of values. the expansions (2.53) generalize to A = dξ ξ ξ A . ξ . (2.
32
.
Let us investigate the role of such variables in quantum mechanics. the position q and momentum p of some component of a dynamical system are represented as real numbers which. As is wellknown. Do the position and momentum operators commute? If not. In quantum mechanics.1 Introduction So far. t) is the energy of the system at time t expressed in terms of the classical coordinates and canonical momenta.2)
where the function H(qi . dqi ∂H = . in classical mechanics the most important dynamical variables are those involving position and momentum.3 POSITION AND MOMENTUM
3
Position and momentum
3. etc. dt ∂qi (3. by deﬁnition. For instance. commute.1) (3. This function is
33
. However. these quantities are represented as noncommuting linear Hermitian operators acting in a ket space which represents all of the possible states of the system. the behaviour of a classical system can be speciﬁed in terms of Lagrangian or Hamiltonian dynamics. in Hamiltonian dynamics. a Cartesian coordinate has an associated linear momentum. we have considered general dynamical variables represented by general linear operators acting in ket space. Associated with each generalized coordinate qi is a classical canonical momentum pi .2 Poisson brackets Consider a dynamic system whose state at a particular time t is fully speciﬁed by N independent classical coordinates qi (where i runs from 1 to N). an angular coordinate has an associated angular momentum. For instance. pi . what is the value of q p − p q?
3. Our ﬁrst task is to discover a quantum mechanical replacement for the classical result q p − p q = 0. In classical mechanics. dt ∂pi dpi ∂H = − .
v] =
i=1
∂u ∂v ∂u ∂v − .4) (3. (3. [pi . It is easily demonstrated that [qi .
Can we construct a quantum mechanical Poisson bracket in which u and v are noncommuting operators. [qi .5) (3. u]. v] = −[v. pj ] = 0. ∂qi ∂pi ∂pi ∂qi
(3. There is.6)
The time evolution of a dynamical variable can also be written in terms of a Poisson bracket by noting that du = dt =
i=1 N i=1 N
∂u dpi ∂u dqi + ∂qi dt ∂pi dt ∂u ∂H ∂u ∂H − ∂qi ∂pi ∂pi ∂qi (3. qj ] = 0. instead of functions? Well. where use has been made of Hamilton’s equations. The Poisson bracket of two dynamical variables u and v is deﬁned
N
[u.8)
.7)
= [u. indeed. H].2 Poisson brackets
3 POSITION AND MOMENTUM
usually referred to as the Hamiltonian of the system. one wellknown construct in classical dynamics which involves products of dynamical variables. If such a construct exists we hope to generalize it somehow to obtain a rule describing how dynamical variables commute with one another in quantum mechanics.3.
34
(3.3)
where u and v are regarded as functions of the coordinates and momenta qi and pi . the main properties of the classical Poisson bracket are as follows: [u. We are interested in ﬁnding some construct of classical dynamics which consists of products of dynamical variables. pj ] = δij .
v] = [u1 . v]] = 0. and [u.13) ﬁrst. h u2 v2 − v2 u2 = i ¯ [u2 . v2 ] [u1 u2 . v2 ]. etc.. v1 v2 ] = [u1 . it follows that u1 v1 − v1 u1 = i ¯ [u1 . represent dynamical variables.10) (3. v] = [u1 . v] + [u2 . v].2 Poisson brackets
3 POSITION AND MOMENTUM
[u.11) (3. v1 ](u2 v2 − v2 u2 ) = (u1 v1 − v1 u1 )[u2 . v2 ]} = [u1 . [u1 u2 . [u. v2 ]u2 + u1 [u2 . h
35
(3. v2 ]. Thus.14)
The last relation is known as the Jacobi identity. v2 ]u2 + v1 u1 [u2 . w. v1 v2 ] = [u1 u2 . u.
(3. c] = 0. etc. v1 ]v2 + v1 [u1 . v1 ] + [u.13) (3.15)
= [u1 . v1 ]v2 + u1 v1 [u2 . we can evaluate the Poisson bracket [u1 u2 .18) (3. v1 + v2 ] = [u. v1 ]v2 + v1 [u1 u2 . which satisﬁes all of the above relations? Well. and c represents a number. v]. v1 ]u2 v2 + u1 [u2 . v1 v2 ] = {[u1 . w]] + [v. [u. [v. In the above. u]] + [w. v1 v2 ] = [u. v1 ]v2 u2 + v1 [u1 . v1 ]v2 + v1 [u2 . v2 ] (3.. v2 ]. Equating the above two results yields [u1 . [w. v1 v2 ] in two different ways. v.3. since they now represent noncommuting operators. v1 v2 ]u2 + u1 [u2 .12) (3. v1 ]v2 + v1 [u1 . Note that the order of the various factors has been preserved. [u. Can we ﬁnd some combination of noncommuting operators u and v.12) or (3.16) (3.19)
. [u1 + u2 . and [u1 u2 .17)
Since this relation must hold for u1 and v1 quite independent of u2 and v2 . (3. v2 ]. v]u2 + u1 [u2 . v2 ].9) (3. v1 ]. v1 ]v2 + v1 [u. v2 ]} u2 + u1 {[u2 . since we can use either of the formulae (3.
the notation [u. v]quantum to denote the quantum Poisson bracket. at least for the simplest cases. We h want the quantum mechanical Poisson bracket of two Hermitian operators to be an Hermitian operator itself.24) (3. v1 . (3.2 Poisson brackets
3 POSITION AND MOMENTUM
where ¯ does not depend on u1 .4)–(3. Quantum mechanics agrees with h experiments provided that ¯ takes the value h/2π. leads us to make the assumption that the quantum mechanical bracket has the same value as the corresponding classical bracket. it follows that ¯ is just a number.22) i¯ h It is easily demonstrated that the quantum mechanical Poisson bracket. [pi . v]quantum = The strong analogy we have found between the classical Poisson bracket. where h h = 6. since the classical Poisson bracket of two real dynamical variables is real.14). satisﬁes all of the relations (3. as deﬁned above. v] . qj ] = 0. [u.6261 × 10−34 J s (3. For instance. (3. pj ] = i ¯ δij .21)
is Planck’s constant. [u.3). (3. This argument yields the fundamental commutation relations [qi . if two dynamical variables. deﬁned in Eq. ξ
36
. and the quantum mechanical Poisson bracket. v2 .25)
These results provide us with the basis for calculating commutation relations between general dynamical variables.20) i¯ h where ¯ is a new universal constant of nature. (3. pj ] = 0. (3. etc. the quantum mechanical Poisson bracket of two dynamical variables u and v is given by uv − vu [u. we are assuming that Eqs. v] = . are quite general operators. Somewhat confusingly.6) hold for quantum mechanical as well as classical Poisson brackets.8)–(3. [qi .23) (3. h Thus. Thus..3. u2 . h (3. deﬁned in Eq. v] is conventionally reserved for the commutator u v − v u in quantum mechanics. This requirement is satisﬁed if ¯ is a real number. In other words. and also commutes with (u1 v1 −v1 u1 ).22). h Since u1 . We will use [u.
It is clear from Eqs. Moreover. Note that the classical result (that everything commutes) is obtained in the limit ¯ → 0.86)]
+∞
dx x
−∞
x  = 1. It is only those variables corresponding to the same degree of freedom which may fail to commute.3 Wavefunctions Consider a simple system with one classical degree of freedom. Suppose that x is free to take any value (e. Equations (3.25) that in quantum mechanics the dynamical variables corresponding to different degrees of freedom all commute. (3. Moreover.
3.26)
The eigenkets satisfy the extremely useful relation [see Eq. qi and pi .25) provide the foundation for the analogy between quantum mechanics and classical mechanics.25). (2.3. η] to be expressed in terms of the fundamental commutation relations (3.23)–(3. An eigenket corresponding to the eigenvalue x is denoted x .23)–(3.27)
This formula expresses the fact that the eigenkets are complete.3 Wavefunctions
3 POSITION AND MOMENTUM
and η.. x could be the position of a free particle). (2. can both be written as a power series in the qi and pi . which corresponds to the Cartesian coordinate x. classical mechanics can be h regarded as the limiting case of quantum mechanics when ¯ goes to zero. [see Eq.
(3. In classih cal mechanics.85)] x x = δ(x − x ).
37
. the operator x possesses eigenvalues x lying in the continuous range −∞ < x < +∞ (since the eigenvalues correspond to all the possible results of a measurement of x). Thus.8)–(3. The classical dynamical variable x is represented in quantum mechanics as a linear Hermitian operator which is also called x. correspond to a different classical degree of freedom of the system. mutually orthogonal.g.13) allows [ξ. then repeated application of Eqs. We can span ket space using the suitably normalized eigenkets of x. (3. each pair of generalized coordinate and its conjugate momentum. (3. and suitably normalized.23)–(3.
e. It is clear that the wavefunction of state A is simply the collection of the weights of the corresponding state ket A . Thus. (3.30)
This formula is only valid if the state ket A is properly normalized: i. from Sect.3 Wavefunctions
3 POSITION AND MOMENTUM
A state ket A (which represents a general state A of the system) can be expressed as a linear superposition of the eigenkets of the position operator using Eq.10. (3. We can write x A = ψA (x )..31)
Consider a second state B represented by a state ket B and a wavefunction ψB (x ). (3. the probability of a measurement of position yielding a result in the range x to x + dx when the wavefunction of the system is ψA (x ) is P(x .28)
The quantity x A is a complex function of the position eigenvalue x .
+∞
A =
−∞
dx x A x
(3.
−∞
(3.3. Note that state A is completely speciﬁed by its wavefunction ψA (x ) [since the wavefunction can be used to reconstruct the state ket A using Eq. Recall.29) Here. In fact. ψA (x ) is the famous wavefunction of quantum mechanics. B
(3. if AA = 1. assuming that the eigenvalues of ξ are discrete. the probability of a measurement of x yielding a result lying in the range x to x + dx when the system is in a state A is  x A 2 dx . when it is expanded in terms of the eigenkets of the position operator. 2.27). (3. In other words. The inner product BA can be written
+∞ +∞
BA =
−∞
dx Bx
x A =
−∞
ψ∗ (x ) ψA (x ) dx . that the probability of a measurement of a dynamical variable ξ yielding the result ξ when the system is in state A is given by  ξ A 2 . dx ) = ψA (x )2 dx . The corresponding normalization for the wavefunction is
+∞
ψA (x )2 dx = 1.28)].32)
38
. This result is easily generalized to dynamical variables possessing continuous eigenvalues.
and ψ(x ) is the associated wavefunction. f(x) = x 2 ].27) and (3.e.. leaving the dependence on the position operator x tacitly understood. Consider the ket whose wavefunction
39
.. Here. Thus. If B = f(x)A then it follows that
+∞
ψB (x ) = =
x f(x)
−∞ +∞
dx ψA (x )x (3. from the above result. It follows.3.
−∞
giving ψB (x ) = f(x ) ψA (x ).
3. if f(x) = x 2 then f(x ) = x 2 .29).4 Schr¨ dinger’s representation . and the ket has the wavefunction ψ(x ) = 1. that ψA (x) is often shortened to ψA .36)
Note. (3.I o Consider the simple system described in the previous section. and is denoted . (3.g. (3. Consider a general function f(x) of the observable x [e.I o
3 POSITION AND MOMENTUM
where use has been made of Eqs. the inner product of two states is related to the overlap integral of their wavefunctions.35) where ψA (x) is the same function of the operator x that the wavefunction ψ A (x ) is of the position eigenvalue x . (3. DC
(3. The dual of the standard ket is termed the standard bra. f(x ) is the same function of the position eigenvalue x that f(x) is of the position operator x: i.4 Schr¨dinger’s representation . The ket is termed the standard ket. where ψ(x) is a general function of the position operator x.34) where use has been made of Eq. A general state ket can be written ψ(x) . that a general state ket A can be written A = ψA (x) . ﬁnally.33)
dx f(x ) ψA (x ) x x .26). It is easily seen that
∗ ψA (x) ←→ ψA (x) .
37)
Any linear operator which acts on ket vectors can also act on bra vectors.38) φ dx dx Making use of Eqs. (3.36).39)
The righthand side can be transformed via integration by parts to give
+∞ −∞
d φ x dx ψ(x ) = − dx
+∞ −∞
dφ(x ) dx ψ(x ). (3.27) and (3.40) is reasonable because physical wavefunctions are squareintegrable [see Eq. so we can think of it as the result of some linear operator acting on ψ . The new ket is clearly a linear function of the original ket.44) dx dx
40
. by comparison with Eqs.4 Schr¨dinger’s representation . we can write
+∞ −∞
d φ x dx ψ(x ) = dx
+∞
φ(x ) dx
−∞
dψ(x ) . It follows. (3.31)].42). (2. It follows that d dψ ψ = . (3.40)
assuming that the contributions from the limits of integration vanish.34). dx dx (3. Let us denote this operator d/dx. Consider d/dx acting on a general bra φ(x). (3. It follows that dφ(x ) d . dx
(3. (2.35) and (3. Note that d dψ DC dψ∗ d ψ = ←→ = − ψ∗ . the bra φ d/dx satisﬁes d d ψ = φ ψ . (3.29).41) φ x = − dx dx which implies d dφ φ =− .3.43) dx dx dx dx where use has been made of Eq. that d † d =− . (3. (3.42) dx dx The neglect of contributions from the limits of integration in Eq. (3. This ket is denoted dψ/dx . dx
(3. According to Eq.I o
3 POSITION AND MOMENTUM
is dψ(x )/dx .
3.4 Schr¨dinger’s representation  I o
3 POSITION AND MOMENTUM
Thus, d/dx is an antiHermitian operator. Let us evaluate the commutation relation between the operators x and d/dx. We have d d(x ψ) d xψ = =x ψ +ψ . (3.45) dx dx dx Since this holds for any ket ψ , it follows that d d x−x = 1. (3.46) dx dx Let p be the momentum conjugate to x (for the simple system under consideration p is a straightforward linear momentum). According to Eq. (3.25), x and p satisfy the commutation relation xp − px = i¯. h (3.47)
It can be seen, by comparison with Eq. (3.46), that the Hermitian operator −i ¯ d/dx satisﬁes the same commutation relation with x that p does. The most h general conclusion which may be drawn from a comparison of Eqs. (3.46) and (3.47) is that d p = −i ¯ h + f(x), (3.48) dx since (as is easily demonstrated) a general function f(x) of the position operator automatically commutes with x. We have chosen to normalize the eigenkets and eigenbras of the position operator so that they satisfy the normalization condition (3.26). However, this choice of normalization does not uniquely determine the eigenkets and eigenbras. Suppose that we transform to a new set of eigenbras which are related to the old set via x new = e i γ x old , (3.49) where γ ≡ γ(x ) is a real function of x . This transformation amounts to a rearrangement of the relative phases of the eigenbras. The new normalization condition is x x
new
=
x e i γ e−i γ x
old
= e i (γ −γ
)
x x
old
= e i (γ −γ ) δ(x − x ) = δ(x − x ).
41
(3.50)
3.4 Schr¨dinger’s representation  I o
3 POSITION AND MOMENTUM
Thus, the new eigenbras satisfy the same normalization condition as the old eigenbras. By deﬁnition, the standard ket satisﬁes x  = 1. It follows from Eq. (3.49) that the new standard ket is related to the old standard ket via
new
= e−i γ
old ,
(3.51)
where γ ≡ γ(x) is a real function of the position operator x. The dual of the above equation yields the transformation rule for the standard bra,
new = old e iγ
.
(3.52)
The transformation rule for a general operator A follows from Eqs. (3.51) and (3.52), plus the requirement that the triple product A remain invariant (this must be the case, otherwise the probability of a measurement yielding a certain result would depend on the choice of eigenbras). Thus, Anew = e−i γ Aold e i γ . (3.53)
Of course, if A commutes with x then A is invariant under the transformation. In fact, d/dx is the only operator (we know of) which does not commute with x, so Eq. (3.53) yields d dγ d d +i , (3.54) = e−i γ e i γ = dx new dx dx dx where the subscript “old” is taken as read. It follows, from Eq. (3.48), that the momentum operator p can be written p = −i ¯ h Thus, the special choice ¯ γ(x) = h yields p = −i ¯ h d dx .
new
d dx
−¯ h
new x
dγ + f(x). dx
(3.55)
f(x) dx
(3.56)
(3.57)
42
3.5 Schr¨dinger’s representation  II o
3 POSITION AND MOMENTUM
Equation (3.56) ﬁxes γ to within an arbitrary additive constant: i.e., the special eigenkets and eigenbras for which Eq. (3.57) is true are determined to within an arbitrary common phasefactor. In conclusion, it is possible to ﬁnd a set of basis eigenkets and eigenbras of the position operator x which satisfy the normalization condition (3.26), and for which the momentum conjugate to x can be represented as the operator d . (3.58) dx A general state ket is written ψ(x) , where the standard ket satisﬁes x  = 1, and where ψ(x ) = x ψ(x) is the wavefunction. This scheme of things is known as Schr¨dinger’s representation, and is the basis of wave mechanics. o p = −i ¯ h
3.5 Schr¨ dinger’s representation  II o o In the preceding sections, we have developed Schr¨dinger’s representation for the case of a single operator x corresponding to a classical Cartesian coordinate. However, this scheme can easily be extended. Consider a system with N generalized coordinates, q1 · · · qN , which can all be simultaneously measured. These are represented as N commuting operators, q1 · · · qN , each with a continuous range of eigenvalues, q1 · · · qN . Ket space is conveniently spanned by the simultaneous eigenkets of q1 · · · qN , which are denoted q1 · · · qN . These eigenkets must form a complete set, otherwise the q1 · · · qN would not be simultaneously observable. The orthogonality condition for the eigenkets [i.e., the generalization of Eq. (3.26)] is The completeness condition [i.e., the generalization of Eq. (3.27)] is
+∞ −∞ +∞
q1 · · · qN q1 · · · qN = δ(q1 − q1 ) δ(q2 − q2 ) · · · δ(qN − qN ). ··· dq1 · · · dqN q1 · · · qN q1 · · · qN  = 1.
(3.59)
(3.60)
−∞
The standard ket is deﬁned such that q1 · · · qN  = 1.
43
(3.61)
42)] φ ∂ ∂φ =− .67)
The N linear operators ∂/∂qi (where i runs from 1 to N) are deﬁned ∂ψ ∂ ψ = . the second coordinate in the range q2 to q2 + dq2 .3. ∂qi ∂qi (3.46). is P(q1 · · · qN .68)
These linear operators can also act on bras (provided the associated wavefunctions are square integrable) in accordance with [see Eq.65)
φ(q1 · · · qN ) = φq1 · · · qN .66)
···
−∞
ψ(q1 · · · qN )2 dq1 · · · dqN = 1.69)
Corresponding to Eq.. etc.II o
3 POSITION AND MOMENTUM
The standard bra is the dual of the standard ket.63)
(3. where The probability of an observation of the system ﬁnding the ﬁrst coordinate in the range q1 to q1 + dq1 . (3. A general state ket is written ψ(q1 · · · qN ) . Finally. Likewise.70)
.62)
(3. dq1 · · · dqN ) = ψ(q1 · · · qN )2 dq1 · · · dqN .
(3. The associated wavefunction is ψ(q1 · · · qN ) = q1 · · · qN ψ . (3.64) (3. we can derive the commutation relations ∂ ∂ qj − q j = δij . ∂qi ∂qi
44
(3.
(3. ∂qi ∂qi (3.5 Schr¨dinger’s representation . a general state bra is written φ(q1 · · · qN ). the normalization condition for a physical wavefunction is
+∞ −∞ +∞
(3.
73)
However.23)–(3.25). and Dirac). o It follows from Eqs.61).74)
This is the generalized Schr¨dinger representation. the function F can be transformed away via a suitable readjustment of the phases of the basis eigenkets (see Sect. the standard ket in Schr¨dinger’s representation is a simultaneous eigenket o of all the momentum operators belonging to the eigenvalue zero. we can always construct a set of simultaneous eigenkets of q1 · · · qN for which pi = −i ¯ h ∂ .76) ∂qi ∂qi ∂qi ∂qi 1 ∂ ∂ = q · · · qN . Note that q1 · · · q N  Hence.71)
It can be seen.75)
Thus.4.II o
3 POSITION AND MOMENTUM
It is also clear that ∂2 ψ ∂ ∂ ∂ ∂ ψ = = ψ. ∂qi ∂qj ∂qj ∂qi (3. ∂qi 1 (3.77)
(3.78)
. by comparison with Eqs.3. (3. q1 · · · q N  so that h q1 · · · qN pi = −i ¯
45
∂ψ ∂ψ(q1 · · · qN ) ∂ ∂ ψ = q1 · · · qN  = = q · · · qN ψ . ∂qi ∂qj ∂qi ∂qj ∂qj ∂qi showing that ∂ ∂ ∂ ∂ = . 3. (3. (3.5 Schr¨dinger’s representation . (3.74) that pi = 0. The most general conclusion we can draw from this coincidence of commutation relations is (see Dirac) pi = −i ¯ h ∂F(q1 · · · qN ) ∂ + . and (3. (3. Thus. ∂qi ∂qi (3. ∂qi (3. ∂qi ∂qi 1 ∂ q · · · qN . that the linear operators −i ¯ ∂/∂qi satisfy the same commutation relations with the q’s and with h each other that the p’s do.72) (3.68).
80)
and the corresponding completeness relation is dp p
−∞
p  = 1. We have seen that it is possible to represent the system in terms of the eigenkets of x.
(3.
46
(3.
(3. This is termed Schr¨dinger’s representation. it is o also possible to represent the system in terms of the eigenkets of p.3.82)
Note that the standard ket in this representation is quite different to that in Schr¨dinger’s representation. (3.79)
= δ(p − p ).6 The momentum representation
3 POSITION AND MOMENTUM
The dual of the above equation gives pi q1 · · · qN = i ¯ h 3. describable in terms of a coordinate x and its conjugate momentum p. dp ) = φ(p )2 dp .6 The momentum representation Consider a system with one degree of freedom. Consider the eigenkets of p which belong to the eigenvalues p . The orthogonality relation for the momentum eigenkets is p p
+∞
∂ q · · · qN . However.84) The probability that a measurement of the momentum yields a result lying in the range p to p + dp is given by P(p .85)
. ∂qi 1
(3.81)
A general state ket can be written φ(p) where the standard ket satisﬁes p  = 1. (3. The momentum space wavefunction φ(p ) sato isﬁes φ(p ) = p φ .83) (3. both of which have a continuous range of eigenvalues. These are denoted p .
3.6 The momentum representation
3 POSITION AND MOMENTUM
Finally, the normalization condition for a physical momentum space wavefunction is +∞ φ(p )2 dp = 1.
−∞
(3.86)
The fundamental commutation relations (3.23)–(3.25) exhibit a particular symmetry between coordinates and their conjugate momenta. If all the coordinates are transformed into their conjugate momenta, and vice versa, and i is then replaced by −i, the commutation relations are unchanged. It follows from this symmetry that we can always choose the eigenkets of p in such a manner that the coordinate x can be represented as (see Sect. 3.4) x = i¯ h d . dp (3.87)
This is termed the momentum representation. The above result is easily generalized to a system with more than one degree of freedom. Suppose the system is speciﬁed by N coordinates, q1 · · · qN , and N conjugate momenta, p1 · · · pN . Then, in the momentum representation, the coordinates can be written as ∂ qi = i ¯ h . (3.88) ∂pi We also have qi = 0, (3.89) and h p1 · · · pN qi = i ¯ ∂ p · · · pN . ∂pi 1 (3.90)
The momentum representation is less useful than Schr¨dinger’s representao tion for a very simple reason. The energy operator (i.e., the Hamiltonian) of most simple systems takes the form of a sum of quadratic terms in the momenta (i.e., the kinetic energy) plus a complicated function of the coordinates (i.e., the potential energy). In Schr¨dinger’s representation, the eigenvalue problem for o the energy translates into a secondorder differential equation in the coordinates, with a complicated potential function. In the momentum representation, the
47
3.7 The uncertainty relation
3 POSITION AND MOMENTUM
problem transforms into a highorder differential equation in the momenta, with a quadratic potential. With the mathematical tools at our disposal, we are far better able to solve the former type of problem than the latter. Hence, Schr¨dinger’s o representation is generally more useful than the momentum representation.
3.7 The uncertainty relation How is a momentum space wavefunction related to the corresponding coordinate space wavefunction? To answer this question, let us consider the representative x p of the momentum eigenkets p in Schr¨dinger’s representation for o a system with a single degree of freedom. This representative satisﬁes p x p = x pp = −i ¯ h d x p , dx (3.91)
where use has been made of Eq. (3.78) (for the case of a system with one degree of freedom). The solution of the above differential equation is x p = c exp(i p x /¯ ), h where c = c (p ). It is easily demonstrated that
+∞
(3.92)
p p
=
−∞
p x dx x p
=c c
∗
−∞
∞
exp[−i (p − p ) x /¯ ] dx . h
(3.93)
The wellknown mathematical result
+∞
exp(i a x) dx = 2π δ(a),
−∞
(3.94)
yields p p = c 2 h δ(p − p ). (3.95) This is consistent with Eq. (3.80), provided that c = h−1/2 . Thus, x p = h−1/2 exp(i p x /¯ ). h (3.96)
48
3.7 The uncertainty relation
3 POSITION AND MOMENTUM
Consider a general state ket A whose coordinate wavefunction is ψ(x ), and whose momentum wavefunction is Ψ(p ). In other words, ψ(x ) = Ψ(p ) = It is easily demonstrated that
+∞
x A , p A .
(3.97) (3.98)
ψ(x ) =
−∞
dp x p
+∞
p A (3.99)
1 = 1/2 h and
+∞
Ψ(p ) exp(i p x /¯ ) dp h
−∞
Ψ(p ) =
−∞
dx p x
+∞
x A (3.100)
1 = 1/2 h
ψ(x ) exp(−i p x /¯ ) dx , h
−∞
where use has been made of Eqs. (3.27), (3.81), (3.94), and (3.96). Clearly, the momentum space wavefunction is the Fourier transform of the coordinate space wavefunction. Consider a state whose coordinate space wavefunction is a wavepacket. In other words, the wavefunction only has nonnegligible amplitude in some spatially localized region of extent ∆x. As is wellknow, the Fourier transform of a wavepacket ﬁlls up a wavenumber band of approximate extent δk ∼ 1/∆x. Note that in Eq. (3.99) the role of the wavenumber k is played by the quantity p /¯ . It h follows that the momentum space wavefunction corresponding to a wavepacket in coordinate space extends over a range of momenta ∆p ∼ ¯ /∆x. Clearly, a meah surement of x is almost certain to give a result lying in a range of width ∆x. Likewise, measurement of p is almost certain to yield a result lying in a range of width ∆p. The product of these two uncertainties is ∆x ∆p ∼ ¯ . h This result is called Heisenberg’s uncertainty principle.
49
(3.101)
The ﬁnal state of the system is completely determined by its initial state. Even if we adopt the convention that all state kets have unit norms. together with the direction and magnitude of the displacement. We obtain (∆x)
2
(∆p)
2
¯2 h ≥ 4
(3..103)
in the undisplaced system. if the system is subject to an external potential.3. It is easily demonstrated that the minimum uncertainty states. then the potential must be displaced.
50
(3. The ﬁnal state of the system only determines the direction of the displaced state ket.
3. (2. So.8 Displacement operators Consider a system with one degree of freedom corresponding to the Cartesian coordinate x. if R = A + B (3.83) and the commutation relation (3.47).8 Displacement operators
3 POSITION AND MOMENTUM
Actually. Thus. it is possible to write Heisenberg’s uncertainty principle more exactly by making use of Eq.102)
for any general state. for which the equality sign holds in the above relation. correspond to Gaussian wavepackets in both coordinate and momentum space. This follows because the superposition relations have a physical signiﬁcance which is unaffected by a displacement of the system. However. Note that the type of displacement we are considering is one in which everything to do with the system is displaced. Suppose that we displace this system some distance along the xaxis. the ﬁnal ket is still not completely determined. The situation is not so clear with state kets. and we just give it a little push. etc. and the displacement causes ket R to transform to ket Rd . since it can be multiplied by a constant phasefactor. We could imagine that the system is on wheels. then in the displaced system we have Rd = Ad + Bd .104)
. we know that the superposition relations between states remain invariant under the displacement.
In other words.3. (3. this determines the displaced kets to within a single arbitrary phasefactor to be multiplied into all of them. Since Eq.
51
(3.107)
Hence.111)
. Ad = DA and Ad = AD† . we must have AdAd = 1. Now. and dynamical variables which possesses a physical signiﬁcance is invariant under a displacement of the system. The normalization condition AA = 1 (3. so AD† DA = 1. (3. kets.109) (3. (3. We now adopt the ansatz that any combination of bras. it follows that the displaced ket Rd must be the result of some linear operator acting on the undisplaced ket R .8 Displacement operators
3 POSITION AND MOMENTUM
Incidentally. The displaced kets cannot be multiplied by individual phasefactors. (3.110) The equation v A = B .105)
where D an operator which depends only on the nature of the displacement. the displacement operator is unitary. it follows that D† D = 1. (3.106)
for a state ket A certainly has a physical signiﬁcance.108) (3.103) holds in the undisplaced system. Thus.104) holds in the displaced system whenever Eq. Note that the above relation implies that A = D† Ad . Since this must hold for any state ket A . Rd = DR . The arbitrary phasefactor by which all displaced kets may be multiplied results in D being undetermined to an arbitrary multiplicative constant of modulus unity. because this would wreck the superposition relations.
8 Displacement operators
3 POSITION AND MOMENTUM
where the operator v represents a dynamical variable.117)
where ax is the limit of γ/δx. Since this is true for any ket Ad . For small δx. We have assumed.109) and (3. we have D = 1 + δx dx . we have vd = D v D † . where γ is a real phaseangle. we expect the limit Ad − A D−1 = lim A δx→0 δx→0 δx δx lim dx = lim to exist.118)
. (3. Suppose. that the system is displaced an inﬁnitesimal distance δx along the xaxis.112)
Note that the arbitrary numerical factor in D does not affect either of the results (3.116) δx→0 δx where dx is denoted the displacement operator along the xaxis. has some physical signiﬁcance. now. Thus.
52
(3.115)
D−1 . we require that vd Ad = Bd . Thus. implies that d x can be replaced by D exp(i γ) − 1 D − 1 + iγ = lim = dx + i a x . δx→0 δx→0 δx δx lim (3.114) (3. It follows that vd Ad = DB = D v A = D v D† Ad . (3.113) (3. as seems reasonable.114).3. We expect that the displaced ket Ad should approach the undisplaced ket A in the limit as δx → 0. that γ tends to zero as δx → 0. The fact that D can be replaced by D exp(i γ). where vd is the displaced operator. It is clear that the displacement operator is undetermined to an arbitrary imaginary additive constant. Let
(3.
8 Displacement operators
3 POSITION AND MOMENTUM
It follows from Eq.109) that (1 + δx dx† )(1 + δx dx ) = 1.e. (3. xd = x − δx.3. h (3.122). (3. the fact that dx is undetermined to an arbitrary additive imaginary constant (which could be a function of x) enables us to transform the function f out of the above equation.125)
where f is Hermitian (since px is Hermitian). It can be seen that the new wavefunction is obtained from the old wavefunction according to the prescription x → x − δx. A comparison with Eq. leaving px = i ¯ d x . δx→0 δx lim (3. Actually. yields dx x − x dx = −1.114). Thus.25)].122)
Let us consider a speciﬁc example.121)
(3.
(3. If the system is displaced a distance δx along the xaxis then the new wavefunction is ψ(x − δx) (i. h
53
(3.124)
It follows that i ¯ dx obeys the same commutation relation with x that px .123)
(3. The most general conclusion we can draw from this observation is that px = i ¯ dx + f(x). (3. the same shape shifted in the xdirection by a distance δx). Substituting into Eq.120) (3. (3. we ﬁnd that vd = (1 + δx dx ) v (1 − δx dx ) = v + δx (dx v − v dx ). Neglecting order (δx)2 . we obtain dx† + dx = 0. (3.119)
Thus. which implies vd − v = dx v − v d x . using x = v.126)
. does [see Eq. and again neglecting order (δx)2 . the new wavefunction can be multiplied by an arbitrary number of modulus unity. However. the h momentum conjugate to x.. Suppose that a state has a wavefunction ψ(x ). the displacement operator is antiHermitian.
3. the displacement operator in the xdirection is proportional to the momentum conjugate to x.8 Displacement operators
3 POSITION AND MOMENTUM
Thus. (3.24) and (3. We say that px is the generator of translations along the xaxis. A ﬁnite translation along the xaxis can be constructed from a series of very many inﬁnitesimal translations. (3. (3.and zaxes.128)
54
. We can also construct displacement operators which translate the system along the y.126).
(3.128)]. and then ∆y along the yaxis then it ends up in the same state as if it were moved ∆y along the yaxis. It follows that D(∆x) = exp (−i px ∆x/¯ ) .118) and (3. the operator D(∆x) which translates the system a distance ∆x along the xaxis is written D(∆x) = lim ∆x px 1−i N ¯ h
N
N→∞
. The fact that translations in independent directions commute is clearly associated with the fact that the conjugate momentum operators associated with these directions also commute [see Eqs. if the system is moved ∆x along the xaxis. In other words. Note that a displacement a distance ∆x along the xaxis commutes with a displacement a distance ∆y along the yaxis. h The unitary nature of the operator is now clearly apparent. and then ∆x along the xaxis.127)
where use has been made of Eqs. Thus.
1) and (4.2). The label t is needed to distinguish the different states of the system at different times.3)
. then we should have Rt = At + Bt . we expect that if a superposition relation holds for certain states at time t0 then the same relation should hold between the corresponding timeevolved states at time t. (4.1)
This rule determines the timeevolved kets to within a single arbitrary phasefactor to be multiplied into all of them. i.1 Schr¨ dinger’s equations of motion o Up to now. the ﬁnal ket is still not completely determined..4 QUANTUM DYNAMICS
4
Quantum dynamics
4. However. the ﬁnal ket can be regarded as the result of some linear operator acting on the initial ket: .e. (4. if Rt0 = At0 + Bt0 for any three kets. since this would invalidate the superposition relation at later times. Thus. At time t the state of the system is represented by the ket At . According to Eqs. Let us now investigate how quantum mechanical systems evolve with time. we have only considered systems at one particular instant of time. The ﬁnal state of the system at time t is completely determined by its initial state at time t0 plus the time interval t − t0 (assuming that the system is left undisturbed during this time interval). However.2) (4. The evolved kets cannot be multiplied by individual phasefactors. assuming that the system is left undisturbed between times t0 and t.
55
(4. the ﬁnal ket Rt depends linearly on the initial ket Rt0 . Even if we adopt the convention that all state kets have unit norms. The label A is needed to distinguish the ket from any other ket (Bt . say) which is evolving in time. Consider a system in a state A which evolves in time. In other words. the ﬁnal state only determines the direction of the ﬁnal state ket. Rt = T Rt0 . since it can be multiplied by an arbitrary phasefactor.
4. it make sense to deﬁne the time evolution operator T in such a manner that it preserves the length of any ket upon which it acts (i.5)
. if a ket is properly normalized at time t then it will remain normalized at all subsequent times t > t0 ). The triple product AξA can evolve either because the ket A evolves and the operator ξ stays constant. However. the time evolution operator T is a unitary operator.3). Since we have adopted a convention in which the norm of any state ket is unity. Since we are already committed to evolving state kets. that as t → t0 then At → At0 for any ket A. the limit T −1 At − At0 = lim At0 (4. We expect. there are some important differences between time evolution and spatial displacement. This is always possible. In general. the time evolution operator T looks very much like the spatial displacement operator D introduced in the previous section. Hence. The arbitrary phasefactor by which all time evolved kets may be multiplied results in T (t. t0 ) being undetermined to an arbitrary multiplicative constant of modulus unity. according to Eq. let us assume that the time evolution operator T can be chosen in such a manner that the operators representing the dynamical variables of the system do not evolve in time (unless they contain some speciﬁc time dependence). Thus.4)
(4. Up to now.1 Schr¨dinger’s equations of motion o
4 QUANTUM DYNAMICS
where T is a linear operator which depends only on the times t and t0 . we do expect the expectation value of some observable ξ to evolve with time. from physical continuity. the ket A stays constant and the operator ξ evolves..e. or both the ket A and the operator ξ evolve. since the length of a ket possesses no physical signiﬁcance.6) lim t→t0 t − t0 t→t0 t − t0
56
(4. time evolution has no meaning unless something observable changes with time). (4. we require that At0 At0 = AtAt for any ket A. Thus. which immediately yields T † T = 1. even if the system is left in a state of undisturbed motion (after all.
Let us deﬁne the Hermitian operator H(t0 ) = i ¯ τ. t0 ) evolves the system in time from t0 to t then H(t0 ) = i ¯ lim h
57
T (t.9)
When written for general t this equation becomes i¯ h dAt = H(t)At . It involves a o Hermitian operator H(t) which is. h t→t0 dt0 t − t0 (4. x0 ) − 1 . It follows from Eqs.7) τ(t0 ) = lim t→t0 t − t0 It is easily demonstrated from Eq. x0 ) displaces the system along the xaxis from x0 to x then px = i ¯ x→x h lim D(x. Note that this limit is simply the derivative of At 0 with respect to t0 .8)
The fact that T can be replaced by T exp(i γ) (where γ is real) implies that τ is undetermined to an arbitrary imaginary additive constant (see previous section). τ† + τ = 0.11)
where px is the operator representing the momentum conjugate to x.6) and (4. 0 x − x0 (4. We saw.10)
Equation (4.5) that τ is antiHermitian: i. (4.e. t0 ) − 1 . This equation is denoted Schr¨dinger’s equation of motion. in the previous section. (4. Let T (t. a characteristic of the dynamical system under investigation. presumably.12)
. This operator is undetermined h to an arbitrary real additive constant. (4. dt (4. We now have that if the operator T (t.7) that i¯ h dAt0 At − At0 = i ¯ lim h = i ¯ τ(t0 )At0 = H(t0 )At0 . t→t0 t − t0
(4. that if the operator D(x. (4.. t0 ) − 1 .4.1 Schr¨dinger’s equations of motion o
4 QUANTUM DYNAMICS
should exist.10) gives the general law for the time evolution of a state ket in a scheme in which the operators representing the dynamical variables remain ﬁxed.
) The operator H(t) is usually called the Hamiltonian of the system.5) and (4.e. dt
t t0
(4.4. dt (4. but time is only a label.10) yields i¯ h dT At0 = H(t) T At0 .13)
Since this must hold for any initial state At0 we conclude that i¯ h dT = H(t) T. time and space coordinates are treated on the same footing by relegating position from being an operator to being just a label.6). time is not an operator (we cannot observe time.
58
. in classical physics. (4. if the equations of motion of a system are invariant under an xdisplacement of the system then this implies that the system conserves momentum in the xdirection. the dynamical variable corresponding to the operator H stands to time t as the momentum px stands to the coordinate x. (4.15)
where use has been made of Eqs. In relativistic quantum mechanics.1 Schr¨dinger’s equations of motion o
4 QUANTUM DYNAMICS
Thus. The fact that the Hamiltonian is undetermined to an arbitrary real additive constant is related to the wellknown phenomenon that energy is undetermined to an arbitrary additive constant in physics (i. Note that. Substituting At = T At0 into Eq. it is just a parameter (or. we assume that Hamiltonian operators evaluated at different times commute with one another).. h
(4. By analogy with classical physics. in the above analysis. Since we are only dealing with nonrelativistic quantum mechanics. more accurately.14)
This equation can be integrated to give T (t. Likewise. if the equations of motion are invariant under a temporal displacement then this implies that the system conserves energy. It is now clear how the fact that H is undetermined to an arbitrary real additive constant leaves T undetermined to a phasefactor. (Here. as such). the zero of potential energy is not welldeﬁned). need not worry us unduly. the fact that position is an operator. a continuous label). this suggests that H(t) is the operator representing the total energy of the system. (Recall that. t0 ) = exp −i
H(t ) dt /¯ .
which is outlined in Sect. this is not the only way in which to represent the time evolution of the system. as opposed to the Schr¨dinger picture. t0 )At0 = At t0 . The time evolution of the transformed state ket is given by At t = T † (t. t0 ) T (t.5). whereas the state kets evolve according to Eq. the transformation (4. Suppose that a general state ket A is subject to the transformation At = T † (t.18)
The transformation rule for a general observable v is obtained from the requirement that the expectation value AvA should remain invariant. (4. This is termed the Heisenberg picture.17)
where use has been made of Eqs.2 Heisenberg’s equations of motion We have seen that in Schr¨dinger’s scheme the dynamical variables of the system o remain ﬁxed during a period of undisturbed motion.2 Heisenberg’s equations of motion
4 QUANTUM DYNAMICS
4. However.16) yields At  = AT. The dual of Eq. t 0 ) obviously depends on time. 4. which corresponds to a ﬁxed linear operator in Schr¨o dinger’s scheme. a dynamical variable. and the fact that T (t0 . The subscript t is used to remind us that the transformation is timedependent. It is easily seen that vt = T † v T. (4. Clearly.10).16) has the effect of bringing all kets representing states of undisturbed motion of the system to rest. (4. t0 ) = 1. t0 )A . t0 )At = T † (t. the transformed state ket does not evolve in time.4.1. since the operator T (t. (4. corresponds to a moving linear operator in this new scheme. and the dynamical variables are represented by moving linear operators. Thus. (4. (4. (4.3). o
59
.19) Thus. It is clear that the transformation (4.16) leads us to a scenario in which the state of the system is represented by a ﬁxed vector. (4. The transformation must also be applied to bras.16)
This is a timedependent transformation.
dt dt dt With the help of Eq. the Heisenberg equation of motion can be written dvt = [vt .4. Ht ]quantum .. According to Eq.25) (4.24) dvt = v H T. dt (4. (4.19).14). It is denoted Heisenberg’s equation of motion.26) dt where [· · ·]quantum denotes the quantum Poisson bracket. dt
Equation (4. Note that the timevarying dynamical variables in the Heisenberg picture are usually called Heisenberg dynamical variables to distinguish them from Schr¨dinger dynamical o variables (i.7)] dv = [v. Ht ].22) (4. (3. dt
60
(4. the corresponding variables in the Schr¨dinger picture). this reduces to H T vt + i ¯ T h or i¯ h where Ht = T † H T. Differentiation with respect to time yields dvt dT dT vt + T =v . Equation (4.e.21) (4.23)
dvt = T † v H T − T † H T v t = v t H t − H t vt .2 Heisenberg’s equations of motion
4 QUANTUM DYNAMICS
Consider a dynamical variable v corresponding to a ﬁxed linear operator in the Schr¨dinger picture. H]classical . we can write o T vt = v T. (4.22). which can be written in the form [see Eq.25) shows how the dynamical variables of the system evolve in the Heisenberg picture. (4. which do o not evolve in time.23) can be written i¯ h dvt = [vt . (3.27)
. According to Eq.20)
(4. Let us compare this equation with the classical time evolution equation for a general dynamical variable v. dt (4.
it is represented by the same ﬁxed operator in both the Schr¨dinger o and Heisenberg pictures). i¯ h Suppose that v is an observable which commutes with the Hamiltonian (and. Note that if the Hamiltonian does not explicitly depend on time (i. H] = 0.28)
4. H] = 0.4. the system is not subject to some timedependent external force) then Eq.31) dt Thus.30) dt Thus. we
61
.26) and (4. and H denotes the classical Hamiltonian. [· · ·]classical is the classical Poisson bracket.25) gives dH = [H. For a physical system which possess a classical analogue.19) that vt = v. It follows from Eq. (4.29) (4. if the energy of the system has no explicit timedependence then it is represented by the same nontimevarying operator H in both the Schr¨dinger and o Heisenberg pictures.e. (4. The only thing which is lacking is some rule to determine the form of the quantum mechanical Hamiltonian. hence. h This operator manifestly commutes with H. Only those observables which do not commute with the Hamiltonian evolve in time in the Heisenberg picture. Eq.3 Ehrenfest’s theorem
4 QUANTUM DYNAMICS
Here. The strong resemblance between Eqs.27) provides us with further justiﬁcation for our identiﬁcation of the linear operator H with the energy of the system in quantum mechanics. (4. (4.3 Ehrenfest’s theorem We have now derived all of the basic elements of quantum mechanics. t0 ) = exp [−i H (t − t0 )/¯ ] . (4. with the time evolution operator T ).15) yields T (t.. so Ht = T † H T = H. Heisenberg’s equation of motion yields dv i¯ h = [v. (4. any observable which commutes with the Hamiltonian is a constant of the motion (hence. Furthermore. (4.
p2 . ∂xi [xi .35) H= pi 2 .. For instance.33) (4. this can usually be resolved by requiring the Hamiltonian H to be an Hermitian operator. we would write the quantum mechanical analogue of the classical product x p. with three corresponding conjugate momenta pi . we replace the classical coordinates and conjugate momenta by the corresponding quantum mechanical operators). G(x)] = −i ¯ h . These are represented by three commuting position operators xi . The following useful formulae. p3 ) as p.e.3 Ehrenfest’s theorem
4 QUANTUM DYNAMICS
generally assume that the Hamiltonian has the same form as in classical physics (i. appearing in the Hamiltonian. ∂pi ∂G [pi . 2 m 2 m i=1
62
.32). The Hamiltonian is assumed to have the same form as in classical physics: 3 1 p2 = (4. respectively.32) It is helpful to denote (x1 . Whenever an ambiguity arises because of noncommuting observables. as the Hermitian product (1/2)(x p + p x).4. x2 . Let us now consider the threedimensional motion of a free particle of mass m in the Heisenberg picture. This scheme guarantees that quantum mechanics yields the correct classical equations of motion in the classical limit. Consider a threedimensional system characterized by three independent Cartesian position coordinates xi (where i runs from 1 to 3). pj ] = i ¯ δij . and three commuting momentum operators pi . x3 ) as x and (p1 .34)
where F and G are functions which can be expanded as power series. F(p)] = i ¯ h (4. are easily proved using the fundamental commutation relations Eq. When the system in question has no classical analogue then we are reduced to guessing a form for H which reproduces the observed behaviour of the system. (4. The commutation relations satisﬁed by the position and momentum operators are [see Eq. ∂F . h (4. (3.25)] [xi .
3 Ehrenfest’s theorem
4 QUANTUM DYNAMICS
In the following. Note that even though [xi (0). pj 2 = dt i¯ h i¯ 2m h ∂pi j=1 m m where use has been made of Eq. The time evolution of the position operator xi is given by 1 1 1 ∂ pi (0) pi dxi = [xi . xj (0)] = 0. H] = i¯ h = . The time evolution of the momentum operator pi follows from Heisenberg’s equation of motion (4.36)
since pi automatically commutes with any function of the momentum operators. It follows that pi (0) t. (4.40)
Combining the above commutation relation with the uncertainty relation (2. xi (0)] = . We ﬁnd that dpi 1 = [pi . (4. all dynamical variables are assumed to be Heisenberg dynamical variables.41) 4 m2 This result implies that even if a particle is welllocalized at t = 0. xi (t) = xi (0) + m
3
(4. although we will omit the subscript t for the sake of clarity.4.38)
which is analogous to the equation of motion of a classical free particle.
63
. H] = 0.39) where the position operators are evaluated at equal times. For instance. xi (0) = . its position becomes progressively more uncertain with time. (4. which means that pi (t) = pi (0) at all times t (for i is 1 to 3). pi (0) t −i ¯ t h [xi (t). This conclusion can also be obtained by studying the propagation of wavepackets in wave mechanics. dt i¯ h (4. m m
(4. for a free particle the momentum operators are constants of the motion. Thus.33).37)
(4.25). the xi do not commute when evaluated at different times.83) yields ¯ 2 t2 h 2 2 (∆xi ) t (∆xi ) t=0 ≥ .
the operator equation (4. (4. Heisenberg’s equation of motion gives dpi 1 ∂V(x) = [pi . (4.45) dt2 i ¯ dt h i¯ m h m dt m ∂xi In vectorial form. the result dxi pi = dt m (4. (4. In fact. H= 2m (4. because the xi all commute with the new term V(x) in the Hamiltonian. it guarantees to us that the centre of a wavepacket h always moves like a classical particle.42)
Here.H = . m 2 = dt dt (4.43) dt i¯ h ∂xi where use has been made of Eq.44)
still holds.46)
This is the quantum mechanical equivalent of Newton’s second law of motion. Taking the expectation values of both sides with respect to a Heisenberg state ket that does not move with time. On the other hand.46) only holds if x and o p are understood to be Heisenberg dynamical variables. (4. this result is independent of whether we are using the Heisenberg or Schr¨dinger picture.4. Note that Eq. (4. this equation becomes d2 x dp = − V(x).H = =− .3 Ehrenfest’s theorem
4 QUANTUM DYNAMICS
Let us now add a potential V(x) to our free particle Hamiltonian: p2 + V(x). We can use the Heisenberg equation of motion a second time to deduce that 1 dxi 1 pi 1 dpi 1 ∂V(x) d2 x i = .34).47) has no dependence on ¯ . V is some function of the xi operators. When written in terms of expectation values.
64
.47)
This is known as Ehrenfest’s theorem. V(x)] = − . In contrast. we obtain d2 x dp m = =− dt2 dt V(x) .
We adopt Schr¨do inger’s representation in which the momentum conjugate to the position operator xi is written [see Eq.4. x3 ). t) = x At . x2 . x2 .74)] ∂ pi = −i ¯ h . p2 . ≡ (∂/∂x . since the position operators are ﬁxed in the Schr¨dinger picture.53)
65
. (4. (4. The ﬁxed dynamical variables of the system are the position operators x ≡ (x1 . (3.52) x 2m 2m where use has been made of Eq. ∂/∂z ) denotes the gradient operator written in terms of the position eigenvalues. p3 ).10) yields o i¯ h ∂ x At = x HAt .49) (4. Here.4 Schr¨ dinger’s waveequation o Let us now consider the motion of a particle in three dimensions in the Schr¨dinger o picture. The state of the system is represented as some time evolving ket At . (4. We can also write x V(x) = V(x ) x .48)
Schr¨dinger’s equation of motion (4. ∂/∂y .50)
where use has been made of the time independence of the x .4 Schr¨dinger’s waveequation o
4 QUANTUM DYNAMICS
4. Let x represent a simultaneous eigenket of the position operators belonging to the eigenvalues x ≡ (x1 .78).51) ∂xi Thus. (3. 2m (4. ∂t (4. and the momentum operators p ≡ (p1 . ¯2 2 h p2 At = − x At . x3 ). The Hamiltonian of the system is taken to be H= p2 + V(x). Note that. The o wavefunction of the system at time t is deﬁned ψ(x . we do not expect the x to evolve in time.
It just happens to give a type of equation which we know how to solve. t). (4. dt (4.50). instead of the Heisenberg picture.58) dAt = H At .4.53). Suppose that the ket At is an eigenket of the Hamiltonian belonging to the eigenvalue H : HAt = H At . we have chosen to work in the Schr¨dinger picture.49). In deriving the waveequation. We have also ﬁxed the relative phases of the x o according to Schr¨dinger’s representation. and is the basis of wave mechanics.
(4. (4.10) yields o i¯ h This can be integrated to give At = exp[−i H (t − t0 )/¯ ]At0 . and (4. (4. This suggests that if the system is initially in an eigenstate of the Hamiltonian then it remains in this state for ever.
(4. (4.57)
Note that At only differs from At0 by a phasefactor. instead of those of the momentum operators. h (4. o Note. however.56) Schr¨dinger’s equation of motion (4.51) is valid.52).4 Schr¨dinger’s waveequation o
4 QUANTUM DYNAMICS
where V(x ) is a scalar function of the position eigenvalues. in which the opposite is true. t) + V(x ) ψ(x . Finally. we obtain ¯2 h ∂ x At i¯ h =− ∂t 2m
2
x At + V(x ) x At . Combining Eqs. that the waveequation is just one of many possible representations of quantum mechanics. (4. we have chosen to represent the system in terms of the eigenkets of the position operators. so that Eq. in which state kets evolve and o dynamical variables are ﬁxed. as long as the
66
.54)
which can also be written ∂ψ(x . The direction of the vector remains ﬁxed in ket space. t) ¯2 h i¯ h =− ∂t 2m
2
ψ(x .55)
This is Schr¨dinger’s famous waveequation.
30). t) d3 x if the wavefunction is properly normalized.63) is termed the probability density. (4. h (4. The probability is equal to ρ(x .65) ∂t
67
.62)
E < lim V(x ). A bound state solution of the o above equation.
Since it is conventional to set the potential at inﬁnity equal to zero.64)
Schr¨dinger’s timedependent waveequation. t0 ) exp[−i H (t − t0 )/¯ ]. and E = H is the energy of the system. satisﬁes the boundary condition Such a solution is only possible if ψ0 (x ) → 0
x →∞
as x  → ∞.4 Schr¨dinger’s waveequation o
4 QUANTUM DYNAMICS
system is undisturbed.60) − 2m where ψ0 (x ) ≡ ψ(x . t) = ψ(x . the above relation implies that bound states are equivalent to negative energy states. from Eq.59)
Substituting the above relation into Schr¨dinger’s wave equation (4. (4. t) = ψ(x . (4. deﬁned by ρ(x .61) (4. so that ρ(x . The wavefunction of a stationary state satisﬁes ψ(x .4. t).
(4. Such a state is called a stationary state. t0 ).55). (4. t)2 . can easily be written in o the form of a conservation equation for the probability density: ∂ρ + · j = 0. (3. in which the particle is conﬁned within a ﬁnite region of space. t) d3 x = 1.61) is sufﬁcient to uniquely specify the solution of Eq. The boundary condition (4. This is Schr¨dinger’s timeindependent waveequation. that the probability of observing the particle in some volume element d3 x around position x is proportional to ρ(x . t) d3 x . The quantity ρ(x . we o obtain ¯2 2 h ψ0 (x ) + (V(x ) − E) ψ0 (x ) = 0. (4. Recall.55). (4.60).
∂t ¯ h Thus. Clearly.66)
We can integrate Eq. t) = − i¯ h [ψ∗ 2m ψ−( ψ∗ ) ψ] = ¯ h Im(ψ∗ m ψ). Schr¨dinger’s waveequation conserves probability. (4. (4.64). t) d3 x .69)
(4. In this case.71)
. if the o wavefunction starts off properly normalized.67)
Thus. Thus.4 Schr¨dinger’s waveequation o
4 QUANTUM DYNAMICS
The probability current j takes the form j(x .65) over all space. to obtain ∂ ρ(x . using the divergence theorem. the probability current is indirectly related to the particle momentum. ψ(x . that the potential has an imaginary component.65) generalizes to ∂ρ + ∂t giving ∂ 2 (4. t) exp ¯ h
68
·j=
2 Im(V) ρ. according to Eq. t) d3 x = 0. The wavefunction can always be written in the form i S(x . (4. assumed that the potential V(x ) is real. Eq. In particular. t) d3 x = Im V(x ) ρ(x . t) d3 x = . an imaginary potential can be used to account for the disappearance of a particle. Suppose. naturally. then it remains properly normalized at all subsequent times.4. if Im(V) < 0 then the total probability of observing the particle anywhere in space decreases monotonically with time. ¯ h
(4. however. Such a potential is often employed to model nuclear reactions in which incident particles can be absorbed by nuclei. and the boundary condition ρ → 0 as x  → ∞.65) we have.70) ρ(x . t) . It is easily demonstrated that pt j(x . ∂t
(4. (4. In deriving Eq. t) = ρ(x .68) m where p t denotes the expectation value of the momentum evaluated at time t. (4. (4.
t) = H(x . To lowest order. wave mechanics h reduces to classical mechanics. In classical mechanics. Let us substitute Eq. (4. (4.77)
.. In particular. in the limit ¯ → 0.4 Schr¨dinger’s waveequation o
4 QUANTUM DYNAMICS
where ρ and S are both real functions. t).4. o We obtain − 1 ¯2 h 2m
2√
j=
ρ + 2i ¯ h
√ ( ρ)·
S−
√
ρ
√ √ S2 + i ¯ ρ 2 S + ρ V h √ ∂ ρ √ ∂S = i¯ h .71).75)
where H(x. (4. and is one of the many forms in which we can write the equations of classical mechanics. the probability current is locally normal to the contours of the phasefunction S.66) that .e.76)
(4.74) yields h − 1 ∂S(x . the gradient of the phase of the wavefunction determines the direction of the probability current.72)
It follows from Eq.
(4. the pathintegral of the Lagrangian). What is the interpretation of S? Note that ψ∗ ψ= √ ρ √ ( ρ) + ρ S i ρ ¯ h S. (4. S is the action (i. S. p. (4.74) provided that h ¯ h Note that. λ= ¯ ¯ h .71) into Schr¨dinger’s timedependent waveequation. The above equation is known as the HamiltonJacobi equation. Thus.73) m Thus. t) =  ∂t 2m S(x . It is a good approximation to neglect the terms involving ¯ in Eq. according to Eq. t) is the Hamiltonian operator. The interpretation of ρ as a probability density has already been given.74) − ρ ∂t ∂t
Let us treat ¯ as a small quantity. (4.  S
69
2
S

S2 . (4. Eq. (4. t)2 + V(x . (4.
70
. quantum mechanics reduces to classical mechanics whenever the de Broglie wavelength is small compared to the characteristic distance over which things (other than the quantum phase) vary.4 Schr¨dinger’s waveequation o
4 QUANTUM DYNAMICS
where λ is the de Broglie wavelength divided by 2π.4. The inequality (4.76) is ¯ equivalent to  λ ¯ 1. (4. This distance is usually set by the variation scalelength of the potential.78) In other words.
[xi . also. we are going to assume that the above equations specify the angular momentum operators in terms of the position and linear momentum operators. in principle. The fundamental commutation relations satisﬁed by the position and linear momentum operators are [see Eqs.3). since all of the products consist of operators which commute. Consider the commutator of the operators Lx and Lz : [Lx . (z px − x pz )] = y [pz . [pi . that there is no ambiguity regarding the order in which operators appear in products on the righthand sides of Eqs. Ly . pz ) ≡ p. The classical deﬁnition of the orbital angular momentum of such a particle about the origin is L = r × p.5) (5. Ly ] = [(y pz − z py ).3)
Let us assume that the operators (Lx .1) (5.25)] [xi . Note that Lx . y. h h
71
(5.2) (5. Lz ) ≡ L which represent the components of orbital angular momentum in quantum mechanics can be deﬁned in an analogous manner to the corresponding components of classical angular momentum. (5. pj ] = i ¯ δij . or z. py . h (5. so they represent things which can. pz ] = i ¯ (−y px + x py ) = i ¯ Lz . and Lz are Hermitian. z) ≡ r and their conjugate momenta (px . y. Note.7)
. pj ] = 0. Lz = x p y − y p x .1)–(5. (5. (3.1 Orbital angular momentum Consider a particle described by the Cartesian coordinates (x.5 ANGULAR MOMENTUM
5
Angular momentum
5. In other words.4) (5.23)–(3. z] px + x py [z.6)
where i and j stand for either x. Ly = z p x − x p z . Ly . giving Lx = y p z − z p y . xj ] = 0. be measured.
5. Each of these vectors satisﬁes Eq. h [Ly .13)
N i=1 Li . Whenever we encounter three operators having these commutation relations. h
. Lx ] = i ¯ Ly .11) (5.14)
72
= i¯ h
i=1
Li = i ¯ L. (5.9) (5. we can write Li × Lj + Lj × Li = 0. we know that the dynamical variables which they represent have identical properties to those of the components of an angular momentum (which we are about to derive).8)–(5. So.10) are the foundation for the whole theory of angular momentum in quantum mechanics.13) that
N N N N
However. with angular momentum vectors Li (where i runs from 1 to N). Suppose that there are N particles in the system. Lz ] = i ¯ Lx .10)
The three commutation relations (5. h (5.j=1 (5. h These can be summed up more succinctly by writing L × L = i ¯ L.11). we shall assume that any three operators which satisfy the commutation relations (5. we expect the angular momentum operators belonging to different particles to commute.8) (5. L = clear from Eqs. Consider the total angular momentum of the system. h [Lz .10) represent the components of an angular momentum. In fact. Ly ] = i ¯ Lz .8)– (5. since they represent different degrees of freedom of the system.12) and (5. (5.1 Orbital angular momentum
5 ANGULAR MOMENTUM
The cyclic permutations of the above result yield the fundamental commutation relations satisﬁed by the components of an angular momentum: [Lx . h (5. (5.
It is
L×L =
i=1
Li ×
N
Lj =
j=1 i=1
1 Li × L i + (Li × Lj + Lj × Li ) 2 i.12)
for i = j. so that Li × L i = i ¯ L i .
Lz ] = 0. Lz ] = −¯ L+ .8)–(5.15)
(5. h
73
(5.1 Orbital angular momentum
5 ANGULAR MOMENTUM
Thus. Lz ]. Note that [L+ . Lz ] + [Ly2 . It is conventional to specify the zcomponent.21)
(5. so [L2 . we conclude that L 2 also commutes with Lx and Ly . It is convenient to deﬁne the shift operators L+ and L− : L+ = L x + i L y .16) (5. In fact. (5.11).5. Lz ] = [Lx2 .17) (5. the zcomponent). Lz ] + [Lz2 . Lz ] = +i ¯ (Lx Ly + Ly Lx ).22)
.19) Since there is nothing special about the zaxis.18)
(5. Lz ] = 0. The commutator of L2 and Lz is written [L2 . [Lz2 . Lz . L2 ≡ Lx2 + Ly2 + Lz2 . the values of other two components become uncertain. It is clear from Eqs. L− = L x − i L y . It is easily demonstrated that h [Lx2 . once we have speciﬁed one component.10) and (5. the total angular momentum of the system satisﬁes the commutation relation (5.20) (5. In particular. The immediate conclusion which can be drawn from the commutation relations (5. Lz ] = −i ¯ (Lx Ly + Ly Lx ). (5.10) is that the three components of an angular momentum vector cannot be speciﬁed (or measured) simultaneously. h [Ly2 .19) that the best we can do in quantum mechanics is to specify the magnitude of an angular momentum vector along with one of its components (by convention.8)–(5. Consider the magnitude squared of the angular momentum vector. the sum of two or more angular momentum vectors satisﬁes the same commutation relation as a primitive angular momentum vector.
25) Thus. m = [f(l. m . m have unit norms. m is the eigenvalue of Lz divided by ¯ . The quantum number m is deﬁned by Lz l. m . We can write L2 l. (5. mL2 − Lz2 l. m l. Lz ] = +¯ L− . m = f(l. m = = It is easily demonstrated that Aξ2 A ≥ 0. h [L+ . h (5. However. Now. assuming that the l.27)
where A is a general ket. m . l. m) − m2 ]¯ 2 .2 Eigenvalues of angular momentum Suppose that the simultaneous eigenkets of L2 and Lz are completely speciﬁed by two quantum numbers. These kets are denoted l.5. mf(l.
(5. m = m ¯ l.2 Eigenvalues of angular momentum
5 ANGULAR MOMENTUM
[L− . m) is some real dimensionless function of l and m. m) ¯ 2 − m2 ¯ 2 l. where f(l. m . mLx2 + Ly2 l. we will show that f(l. L− ] = 2 ¯ Lz . since Lz is an Hermitian operator. that both shift operators commute with L2 . Note that m is a h real number.29) l. The proof follows from the observation that Aξ2 A = Aξ† ξA = BB . m) = l (l + 1). Later on. (5. mLy2 l. It is possible to write such an h equation because ¯ has the dimensions of angular momentum. h (5. m) ¯ 2 l.28) (5. mLx2 l. also.
74
(5. l and m. mL2 − Lz2 l. and ξ is an Hermitian operator.23) (5. m = l.30)
. m + l. h Note. h h h l.26) without loss of generality.24)
5.
m + 1 . m = c+ ¯ l. (5.29) that m2 ≤ f(l.m h Hence. l. It is easily demonstrated that L2 (L+ l. that any value of m can be obtained by applying the shift operators a sufﬁcient number of times.34)
where c+ is a number. the eigenvalue of L2 remains unchanged.33)
where use has been made of Eq. m) (L+ l. m = c− ¯ l.27)–(5. h Using similar arguments to those given above. m = (L+ Lz + [Lz . m = (L+ Lz + ¯ L+ )l. m . It follows from Eqs. L+ is called a raising operator. according to Eq. L+ ])l.m neous eigenstate of L2 and Lz . h (5.21)]. For this reason.32)
where use has been made of Eq. m . (5. It is clear that when the operator L+ acts on a simultal. This bound is determined by the
75
. We can write L+ l. m). It follows that the ket L+ l.31). m h = (m + 1) ¯ L+ l. m . m is proportional to l.22). m has the same eigenvalue of L2 as the ket l. l. m ) = ¯ 2 f(l.2 Eigenvalues of angular momentum
5 ANGULAR MOMENTUM
where B = ξA . (5. However. (5. at ﬁrst sight.35) L− l. plus the fact that BB ≥ 0 for a general ket B [see Eq. plus the fact that L2 and Lz commute. Note that Lz L+ l.31)
Consider the effect of the shift operator L+ on the eigenket l. m + 1 . m − 1 . the shift operator L+ does not affect the magnitude of the angular momentum of any eigenket it acts upon. (2.26). L− is called a lowering operator. The shift operators step the value of m up and down by unity each time they operate on one of the simultaneous eigenkets of L2 and Lz . it is possible to demonstrate that (5. but the eigenvalue of Lz is increased by ¯ . m ). there is a deﬁnite upper bound to the values that m2 can take. Thus. h (5.m h (5. The above equation implies that L+ l.5. It would appear. (5.
namely mmax (mmax + 1).39) (5.37)
But. we can give the unknown quantum number l the value mmax . n. h so Eq. mmax ) = mmax (mmax + 1). L2 l.42)
At this stage.40) (5. mmax = 0 .26) yields the result f(l. (5. h
76
(5. Since there is no state with m > mmax . m = mmax (mmax + 1) ¯ 2 l. L− L+ = Lx2 + Ly2 + i [Lx . m .44) (5. h The above equation can be rearranged to give L2 l. etc. l.36) This implies that L− L+ l. However.5. m = ¯ 2 [l (l + 1) − m (m + 1)]. m = (L2 − Lz2 − ¯ Lz )l.45)
. mmax it generates n. mmax − 1 . mmax = 0 . we must have L+ l. (5. Suppose that we attempt to raise the value of m above its maximum value mmax . m . m . h (5. mmax = (Lz2 + ¯ Lz )l. (5. m = ¯ 2 [l (l + 1) − m (m + 1)]l. h It is easily seen that L− L+ l. (5. without loss of generality. Since the lowering operator does not change the eigenvalue of L2 . h h Thus. h h Comparison of this equation with Eq.37) yields (L2 − Lz2 − ¯ Lz )l.41) (5. mmax . when L− operates on n. m = l (l + 1) ¯ 2 l. Thus. (5. mmax = mmax (mmax + 1) ¯ 2 l.43)
(5. mL− L+ l. all of these states must correspond to the same value of f.26)]. We can also write the above equation in the form L2 l. mmax = 0 .38) (5. Ly ] = L2 − Lz2 − ¯ Lz . mmax − 2 .2 Eigenvalues of angular momentum
5 ANGULAR MOMENTUM
eigenvalue of L2 [see Eq. It follows that there is a maximum and a minimum possible value which m can take.
c− = (c+ )∗ = l (l + 1) − m (m − 1).47) and (5.47)
= ( l.m l.m l.m The solution of the above equation is c+ = l (l + 1) − m (m + 1).
∗
(5. where γ is real. mL− l.m (5. mLy l. m + 1Lx l. m + 1Lx l. (5. m . m + 1L+ l. mL− ¯ c+ l. m )∗ (5.49).m tive.2 Eigenvalues of angular momentum
5 ANGULAR MOMENTUM
However.e. l.50)
Note that c+ is undetermined to an arbitrary phasefactor [i. m + i l.31) implies that there is a maximum and a minimum possible value of m. m + 1Ly l. m + 1 = ¯ 2 c+ c− . l.35).m+1 = [l (l + 1) − m (m + 1)]. l. by c+ exp(i γ). m ∗ − i l. but convenient. and we still satisfy Eq.m (5.m . (5. (5. mLx l. m + 1 − i l.m+1 = (cl.35).51) (5.46)
(5.48)
where use has been made of the fact that Lx and Ly are Hermitian. l.m+1 where use has been made of Eqs..34) and (5. It follows that c+ c− l. given above.m + cl. m + 1 l.49) c− l. m + 1Ly l.m We have made the arbitrary.m−1 l.34) and (5. m = l. According to Eq. The above equation reduces to + ∗ (5.m h l. choice that c+ is real and posil.5. Equations (5. m + 1 = = = l.52)
We have already seen that the inequality (5.m ) with the aid of Eqs. we also know that l. This is equivalent to choosing the relative phases of the eigenkets l. Consider the following: l. h l. we can replace l. mL− L+ l. m ∗ . m l. (5. The maximum value of m is denoted
77
.50)].49) can be combined to give c+ 2 = [l (l + 1) − m (m + 1)].
Ly . ϕ). we have shown that the eigenvalues of L2 ≡ Lx2 + Ly2 + Lz2 can be written l (l + 1) ¯ 2 . l = 2. We have shown that L+ l.. in quantum mechanics we can
78
. plus the fact that Lx .5. (5.m (5. 1. We have also demonstrated that the eigenvalues of L z can only take the values m ¯ . Lz . or it is a halfinteger (e. using just the fundamental commutation relations (5. θ. According to Sect. 3.55) (5. where l is an integer.8)–(5. −1. m L− l. The classical momentum conjugate to the azimuthal angle ϕ is the zcomponent of angular momentum.3 Rotation operators Consider a particle described by the spherical polar coordinates (r. Since there is no state with m < mmin . m + 1 h l (l + 1) − m (m − 1) ¯ l. h (5. In summary.10). 2). −l + 1. mmin = 0. 5. l.g.g. (5. m = = l (l + 1) − m (m + 1) ¯ l.5. we must have L− l.35). Let l. −1/2. and the bottom rung is −l. each rung differing from its immediate neighbours by unity.54)
It can be seen from Eq. · · · l − 1. h or a halfinteger. We will prove in the next section that an orbital angular momentum can only take integer values of l. l. There are only two possible choices for l. which allows m to take the values −2. The top rung is l. where m lies in the range −l..56)
where L± = Lx ± i Ly are the socalled shift operators. m − 1 . this implies that c− min = 0. l = 3/2.3 Rotation operators
5 ANGULAR MOMENTUM
l. m h denote a properly normalized simultaneous eigenket of L2 and Lz . 1/2. What is the minimum value? Suppose that we try to lower the value of m below its minimum value mmin . respectively. and Lz are Hermitian operators.52) that mmin = −l. belonging to h h the eigenvalues l (l + 1) ¯ 2 and m ¯ . We conclude that m can take a “ladder” of discrete values. which allows m to take the values −3/2. (5.53) According to Eq. Either it is an integer (e. 3/2). 0.
58)
where px is the linear momentum conjugate to x.5.60)
in the limit δϕ → 0. introduced in Sect. the angular momentum operator Lz can be used to rotate the system about the zaxis by an inﬁnitesimal amount. (5. We conclude that Lz = i ¯ lim h R(δϕ) − 1 . δx→0 δx (5.8.61) N→∞ N ¯ h which reduces to R(∆ϕ) = exp(−i Lz ∆ϕ/¯ ).5 which speciﬁes that we have to use Cartesian coordinates—the representation (3. we can write R(δϕ) = 1 − i Lz δϕ/¯ h (5. which translates the system a distance ∆x along the x axis.74) works for any welldeﬁned set of coordinates. δϕ→0 δϕ (5. θ. (5. (5. for which ket space is spanned by the o simultaneous eigenkets of the position operators r. h (5.59). Consider an operator R(∆ϕ) which rotates the system an angle ∆ϕ about the zaxis. We were able to demonstrate in Sect. In other words. Thus. 3.59)
According to Eq.62)
79
. 3.8 that px = i ¯ lim h D(δx) − 1 . There is nothing in our derivation of this result which speciﬁes that x has to be a Cartesian coordinate. 3.57) ∂ϕ We can do this because there is nothing in Sect. The above equation implies that ∆ϕ Lz N R(∆ϕ) = lim 1 − i . and L z takes the form ∂ Lz = −i ¯ h . the result should apply just as well to an angular coordinate. and φ.3 Rotation operators
5 ANGULAR MOMENTUM
always adopt Schr¨dinger’s representation. This operator is very similar to the operator D(∆x). We say that Lz is the generator of rotations about the zaxis.
Rx (∆ϕx ) denotes an operator which rotates the system by an angle ∆ϕx about the xaxis. Consider the action of the rotation operator R(∆ϕ) on an eigenstate possessing zero angular momentum about the zaxis (i.. m = l.e.67)
Thus. Here.e. and the eigenvalue of Lz is m ¯ . We have R(∆ϕ)l. m . the eigenstate is invariant to rotations about the zaxis. This implies. h Rz (∆ϕy ) = exp(−i Lz ∆ϕz /¯ ). an m = 0 state). orbital angular momentum can only take on integer values of the quantum numbers l and m. Thus. etc. (5.70)
by analogy with Eq.69) (5. m = exp(−i 2 π m)l. R(∆ϕ1 ) R(∆ϕ2 ) = R(∆ϕ1 + ∆ϕ2 ).65)
Suppose that the system is in a simultaneous eigenstate of L2 and Lz . Thus. As before. 0 .66)
We conclude that m must be an integer. We expect the wavefunction to h h remain unaltered if we rotate the system 2π degrees about the zaxis. (5. 0 = exp(0)l. We know that the system is also in an eigenstate of zero orbital angular momentum about any particular axis. its wavefunction must be symmetric about the zaxis. (5. according
80
. an l = 0 state).. h Ry (∆ϕy ) = exp(−i Ly ∆ϕy /¯ ). from the previous section. R(2π)l. Suppose that the system is in an eigenstate of zero overall orbital angular momentum (i. 0 = l. that l must also be an integer.5. There is nothing special about the zaxis. where the eigenvalue of L 2 is l (l + 1) ¯ 2 .64) (5. m . h (5.68) (5. m = exp(−i Lz 2π/¯ )l. R(∆ϕ) R(−∆ϕ) = 1. h (5.3 Rotation operators
5 ANGULAR MOMENTUM
Note that R(∆ϕ) has all of the properties we would expect of a rotation operator R(0) = 1. so we can write Rx (∆ϕx ) = exp(−i Lx ∆ϕx /¯ ). This follows because l = 0 implies m = 0.62).63) (5. this state is represented by the eigenket l. Clearly.
0 Rz (∆ϕz )0.75) (5. Note that a rotation about the xaxis does not commute with a rotation about the yaxis. this implies that Ry (∆ϕy ) Rx (∆ϕx ) = Rx (∆ϕx ) Ry (∆ϕy ).
5. 0 = 0. a zero angular momentum state is invariant to rotations about any axis. Thus.76)
using the Schr¨dinger representation. In quantum mechanics.72) (5. = exp(0)0.77)
. Such a state must possess a spherically symmetric wavefunction.71) (5. and we can choose the zaxis to point in any direction.74) (5. and then ∆ϕx about the xaxis. In other words.5. and then ∆ϕy about the yaxis. −x ∂x ∂z ∂ ∂ = −i ¯ x h −y . 0 = exp(0)0. [see Eqs. it ends up in a different state to that obtained by rotating an angle ∆ϕy about the yaxis. Thus.68)–(5.70)]. if the system is rotated an angle ∆ϕx about the xaxis. 0 = 0. = exp(0)0. 0 . 0 . x = r sin θ cos ϕ. Rx (∆ϕx )0. or Ly Lx = Lx Ly .4 Eigenfunctions of orbital angular momentum
5 ANGULAR MOMENTUM
to the previous section. 0 .4 Eigenfunctions of orbital angular momentum In Cartesian coordinates.
81
(5. 0 Ry (∆ϕy )0. (5. 0 = 0.73)
Clearly. the noncommuting nature of the angular momentum operators is a direct consequence of the fact that rotations do not commute. ∂z ∂y ∂ ∂ = −i ¯ z h . (5. Transforming to standard spherical polar o coordinates. ∂y ∂x (5. the three components of orbital angular momentum can be written Lx = −i ¯ y h Ly Lz ∂ ∂ −z .
5.4 Eigenfunctions of orbital angular momentum
5 ANGULAR MOMENTUM
y = r sin θ sin ϕ, z = r cos θ, we obtain
(5.78) (5.79)
∂ ∂ + cot θ cos ϕ (5.80) ∂θ ∂ϕ ∂ ∂ − cot θ sin ϕ Ly = −i ¯ cos ϕ h (5.81) ∂θ ∂ϕ ∂ . (5.82) Lz = −i ¯ h ∂ϕ Note that Eq. (5.82) accords with Eq. (5.57). The shift operators L± = Lx ± i Ly become ∂ ∂ L± = ±¯ exp(±i ϕ) h ± i cot θ . (5.83) ∂θ ∂ϕ Now, L2 = Lx2 + Ly2 + Lz2 = Lz2 + (L+ L− + L− L+ )/2, (5.84) Lx = i ¯ sin ϕ h so ∂ ∂ 1 ∂2 2 1 2 L = −¯ h sin θ + . sin θ ∂θ ∂θ sin2 θ ∂ϕ2
(5.85)
The eigenvalue problem for L2 takes the form L2 ψ = λ ¯ 2 ψ, h where ψ(r, θ, ϕ) is the wavefunction, and λ is a number. Let us write ψ(r, θ, ϕ) = R(r) Y(θ, ϕ). Equation (5.86) reduces to ∂ 1 ∂2 1 ∂ sin θ + Y + λ Y = 0, sin θ ∂θ ∂θ sin2 θ ∂ϕ2
(5.86) (5.87)
(5.88)
where use has been made of Eq. (5.85). As is wellknown, square integrable solutions to this equation only exist when λ takes the values l (l + 1), where l is an integer. These solutions are known as spherical harmonics, and can be written Ylm (θ, ϕ) = 2 l + 1 (l − m)! m (−1)m e i m ϕ Pl (cos ϕ), 4π (l + m)!
82
(5.89)
5.4 Eigenfunctions of orbital angular momentum
5 ANGULAR MOMENTUM
m where m is a positive integer lying in the range 0 ≤ m ≤ l. Here, Pl (ξ) is an associated Legendre function satisfying the equation m m2 m d 2 dPl m (1 − ξ ) − P + l (l + 1) Pl = 0. 2 l dξ dξ 1−ξ
(5.90)
We deﬁne Yl−m = (−1)m (Ylm )∗ , (5.91) which allows m to take the negative values −l ≤ m < 0. The spherical harmonics are orthogonal functions, and are properly normalized with respect to integration over the entire solid angle:
π 0 2π
Ylm∗ (θ, ϕ) Ylm (θ, ϕ) sin θ dθ dϕ = δll δmm .
0
(5.92)
The spherical harmonics also form a complete set for representing general functions of θ and ϕ. By deﬁnition, L2 Ylm = l (l + 1) ¯ 2 Ylm , h where l is an integer. It follows from Eqs. (5.82) and (5.89) that h Lz Ylm = m ¯ Ylm , (5.94) (5.93)
where m is an integer lying in the range −l ≤ m ≤ l. Thus, the wavefunction ψ(r, θ, ϕ) = R(r) Ylm (θ, φ), where R is a general function, has all of the expected features of the wavefunction of a simultaneous eigenstate of L2 and Lz belonging to the quantum numbers l and m. The wellknown formula
m dPl 1 mξ m m+1 = √ P Pl − dξ 1 − ξ2 l 1 − ξ2 mξ m (l + m)(l − m + 1) m−1 √ + Pl = − P 1 − ξ2 l 1 − ξ2
(5.95)
can be combined with Eqs. (5.83) and (5.89) to give h L+ Ylm = l (l + 1) − m (m + 1) ¯ Ylm+1 , h L− Ylm = l (l + 1) − m (m − 1) ¯ Ylm−1 .
83
(5.96) (5.97)
5.5 Motion in a central ﬁeld
5 ANGULAR MOMENTUM
These equations are equivalent to Eqs. (5.55)–(5.56). Note that a spherical harmonic wavefunction is symmetric about the zaxis (i.e., independent of ϕ) when√ 0 ever m = 0, and is spherically symmetric whenever l = 0 (since Y0 = 1/ 4π). In summary, by solving directly for the eigenfunctions of L2 and Lz in Schr¨do inger’s representation, we have been able to reproduce all of the results of Sect. 5.2. Nevertheless, the results of Sect. 5.2 are more general than those obtained in this section, because they still apply when the quantum number l takes on halfinteger values.
5.5 Motion in a central ﬁeld Consider a particle of mass M moving in a spherically symmetric potential. The Hamiltonian takes the form p2 + V(r). H= 2M (5.98)
Adopting Schr¨dinger’s representation, we can write p = −(i/¯ ) . Hence, o h ¯2 h H=− 2M
2
+ V(r).
(5.99)
When written in spherical polar coordinates, the above equation becomes 1 ∂ ∂ 1 ∂2 ¯2 1 ∂ 2 ∂ h + V(r). r + sin θ + H=− 2 M r2 ∂r ∂r r2 sin θ ∂θ ∂θ r2 sin2 θ ∂ϕ2
(5.100)
Comparing this equation with Eq. (5.85), we ﬁnd that ¯2 1 ∂ 2 ∂ h L2 H= − 2 r + 2 2 + V(r). 2M r ∂r ∂r ¯ r h
(5.101)
Now, we know that the three components of angular momentum commute with L2 (see Sect. 5.1). We also know, from Eqs. (5.80)–(5.82), that Lx , Ly , and Lz take the form of partial derivative operators of the angular coordinates,
84
Lz . (5. In general. from the general properties of this type of equation. the previous two equations ensure that the angular momentum L and its magnitude squared L2 are both constants of the motion.105) into Eq. the energy eigenvalues depend on the quantum number l. H] = 0. 4. it is always possible to represent the state of the system in terms of the simultaneous eigenstates of L2 . (5. (5.5 Motion in a central ﬁeld
5 ANGULAR MOMENTUM
when written in terms of spherical polar coordinates using Schr¨dinger’s repreo sentation. These are the energy eigenvalues.101). (5. We know.102)
According to Sect. H] = 0.2.101) that all three components of the angular momentum commute with the Hamiltonian: [L. and making use of Eq. θ.5. This is as expected for a spherically symmetric potential. It is also easily seen that L2 commutes with the Hamiltonian: [L2 .
85
.93). But. and H. we obtain ¯2 1 d 2 d h l (l + 1) + (5.106) − 2 r + V(r) − E R = 0. It follows from Eq. (5.104)
where E is a number. ϕ). but are independent of the quantum number m.103) (5. Since L2 and Lz commute with each other and the Hamiltonian. 2M r dr dr r2
This is a SturmLiouville equation for the function R(r). we already know that the most general form for the wavefunction of a simultaneous eigenstate of L2 and Lz is (see previous section) ψ(r. (5. (5.105) Substituting Eq. ϕ) = R(r) Ylm (θ. that if R(r) is required to be wellbehaved at r = 0 and as r → ∞ then solutions only exist for a discrete set of values of E. Consider the energy eigenvalue problem H ψ = E ψ.
The ﬁrst part is the attractive Coulomb potential.6 Energy levels of the hydrogen atom
5 ANGULAR MOMENTUM
5. Let a= and y = r/a.110)
The effective potential has a simple physical interpretation. 2µE
(5.107)
The radial eigenfunction R(r) satisﬁes Eq. (5. which takes into account the fact that the electron (of mass me ) and the proton (of mass mp ) both rotate about a common centre.109) − 2 − − E P = 0.6 Energy levels of the hydrogen atom Consider a hydrogen atom. which is equivalent to a particle of mass µ rotating about a ﬁxed point. 4π 0 r 2 µ r2 (5. The above equation transforms to e2 d2 P 2 µ l (l + 1)¯ 2 h (5. it is assumed that the energy eigenvalue E is negative. (5. for which the potential takes the speciﬁc form V(r) = − e2 . 4π 0 r
(5.112) Here. Let us write the product r R(r) as the function P(r). with P(r) = f(y) exp(−y).5.111)
. which can be written e2 l (l + 1) ¯2 1 d 2 d h − + − E R = 0.108)
Here.109) transforms to d l (l + 1) 2 µ e2 a d2 −2 − + f = 0. µ = me mp /(me +mp ) is the reduced mass. and the second part corresponds to the repulsive centrifugal force.113) dy2 dy y2 4π 0 ¯ 2 y h
86
−¯ 2 h .106). − 2 r 2 2µ r dr dr r 4π 0 r
(5. dr2 2 µ r2 4π 0 r ¯ h which is the onedimensional Schr¨dinger equation for a particle of mass µ movo ing in the effective potential e2 l (l + 1) ¯ 2 h Veff (r) = − + . Equation (5. (5.
. except when l = 0).e.114) is cn y 2y = . we conclude that nmin = l + 1. (5. where the ﬁrst term in the series is cnmin ynmin .5. Note. This does not correspond to physically acceptable behaviour of the wavefunction. cn [n (n − 1) − l (l + 1)] = cn−1 4π 0 ¯ 2 h
(5. that it is only possible to obtain sensible behaviour of the wavefunction as r → 0 if l is an integer.114)
Substituting this solution into Eq. (5. We conclude that f(y) → exp(2 y) as y → ∞. This is the same as the ratio of successive terms in the series (2 y)n . (5.118) n! n which converges to exp(2 y).
(5. The former predicts unphysical behaviour of the wavefunction at y = 0.6 Energy levels of the hydrogen atom
5 ANGULAR MOMENTUM
Let us look for a powerlaw solution of the form f(y) =
n
c n yn . This is only possible if [nmin (nmin −1)−l (l+1)] = 0. whereas for an l > 0 state there is zero probability of ﬁnding the electron at the nucleus (i. Thus. cn−1 n (5. For large values of y. the power law series (5. The only way in which we can avoid this unphysical behaviour is
87
. also. at some positive value of n. since ψ2 dV must be ﬁnite. There are two possibilities: nmin = −l or nmin = l + 1. we obtain cn n (n − 1) y
n n−2
− 2ny
n−1
− l (l + 1) y
n−2
2 µ e2 a n−1 y + 4π 0 ¯ 2 h
= 0.114) must terminate at small n. It follows from Eq. (5. (5.112) that R(r) → exp(r/a)/r as r → ∞. otherwise f(y) behaves unphysically as y → 0.116)
Now.115)
Equating the coefﬁcients of yn−2 gives 2 µ e2 a 2 (n − 1) − .117)
according to Eq. Note that for an l = 0 state there is a ﬁnite probability of ﬁnding the electron at the nucleus.116). the ratio of successive terms in the series (5.113). ψ2 = 0 at r = 0.
Here. l is a nonnegative integer. The only permissible values of the other quantum numbers are l = 0 and m = 0. zero angular momentum state. Thus. E=− h 32π2 02 ¯ 2 n2 (5.122)
Finally. the Ylm are spherical harmonics.116). θ. and m is an integer. It is clear that the wavefunction for a hydrogen atom can be written ψ(r. n is a positive integer which must exceed the quantum number l. 2 µe and R(x) is a wellbehaved solution of the differential equation
(5. The energy of the ground state is µ e4 E0 = − = −13. It follows from Eq.119)
where the last term in the series is cn yn .111) that the energy eigenvalues are quantized. 32π2 02 ¯ 2 h
88
l (l + 1) 2 n 1 d 2d x − + − 1 R = 0.121)
(5.6 Energy levels of the hydrogen atom
5 ANGULAR MOMENTUM
if the series (5. ϕ).123)
(5.114).6 electron volts.120)
Here. and can only take the values µ e4 . ϕ) = R(r/a) Ylm (θ. the ground state is a spherically symmetric. According to the recursion relation (5.3 × 10−11 n meters. where n 4π 0 ¯ 2 h a= = 5. The restrictions on the quantum numbers are m ≤ l < n. x2 dx dx x2 x
(5. 4π 0 ¯ 2 h (5. The ground state of hydrogen corresponds to n = 1. n is a positive integer.114) terminates at some maximum value of n.5. otherwise there would be no terms in the series (5. (5. this is only possible if µ e2 a = n.124)
.
m = −1.120) are independent of the quantum number l.8)–(5. Even worse. h
89
(5. Note that the energy levels given in Eq. Consider an isolated system at rest.10) for the components of an angular momentum.5. Unfortunately. the total angular momentum in the rest frame is nonvanishing. This is a special property of a 1/r Coulomb potential. and z. angular momentum called spin.125)
. purely quantum mechanical. we can write S × S = i ¯ S.
5.5. For a particle with spin. For a system consisting of a single particle. Let us denote the three components of the spin angular momentum of a particle by the Hermitian operators (Sx . 1. j is a nonnegative integer. and yet have j = 0. Gouldsmit and Uhlenbeck (in 1925) introduced the concept of an internal. this does not agree with observations. (5. In order to explain this apparent discrepancy between theory and experiments. 5. there are n = 2 states with nonzero angular momentum. Sz ) ≡ S. According to the theory of orbital angular momentum h outlined in Sects.7 Spin angular momentum Up to now. m = 0 or l = 1. For a system consisting of two (or more) particles. Thus. Thus. In addition to the quantized negative energy state of the hydrogen atom. which we have just found. j = 0. because we often ﬁnd systems which appear to be structureless. y. However. systems where j has halfinteger values abound in nature.123).7 Spin angular momentum
5 ANGULAR MOMENTUM
The next energy level corresponds to n = 2. and let the eigenvalue of its total angular momentum be j (j + 1) ¯ 2 . despite the fact that l appears in the radial eigenfunction equation (5. 0. The other quantum numbers are allowed to take the values l = 0. there is a wealth of experimental evidence which suggests that this simplistic approach is incomplete. We assume that these operators obey the fundamental commutation relations (5. Sy . there is also a continuum of unbound positive energy states. we have tacitly assumed that the state of a particle in quantum mechanics can be completely speciﬁed by giving the wavefunction ψ as a function of the spatial coordinates x.4 and 5. there are two possibilities.
for which ¯ h Sz ± = ± ± .4) depends on Eqs. [S. it has no analogy in classical physics.3). S2 ] = 0.2.. −s. Consider a spin onehalf particle. Spin angular momentum clearly has many properties in common with orbital angular momentum. (5.126)
According to the equally general analysis of Sect. take integer or halfinteger values. (5. sz = sz ¯ s.3). 5.1.128) (5. h (5. the ± denote eigenkets of the Sz operator corresponding to the eigenvalues ±¯ /2. and the concept of spin is purely quantum mechanical: i. s − 1 · · · − s + 1. These are denoted s. h so +− = 0.e. sz . However. in principle. the quantum number s is allowed to take halfinteger values. where Sz s. since this restriction (found in Sects.3 and 5. sz . like in Eqs. sz S2 s. 5. (5. Consequently. (5. there is one vitally important difference.127) Thus. According to the quite general analysis of Sect.1)–(5. In other words. since this identiﬁcation depends on an analogy with classical mechanics.130) 2 3¯2 h S2 ± = ± .7 Spin angular momentum
5 ANGULAR MOMENTUM
We can also deﬁne the operator S2 = Sx2 + Sy2 + Sz2 . h = s (s + 1) ¯ 2 s. it is possible to ﬁnd simultaneous eigenstates of S2 and Sz .131) 4 Here.132)
90
.1)–(5. Spin angular momentum operators cannot be expressed in terms of position and momentum operators. 5. the quantum number s can.5. (5. These kets are orthonormal (since Sz is an Hermitian operator).129) (5. (5. the restriction that the quantum number of the overall angular momentum must take integer values is lifted for spin angular momentum. sz . and the quantum number s z can only take the values s.
y . Equations (5.135)–(5. The basis kets are assumed to satisfy the completeness relation ( x .e. y. z . z . (5. ± . y . h respectively. z . corresponding to the eigenvalues x . Here. (5. (5. (5. that Hilbert subspace consisting of kets which correspond to the different spin states of the particle). deﬁned in this manner.130)–(5. − ) dx dy dz = 1. z . − x . y .134) (5. + x .139)
91
. x . satisfy the eigenvalue relations (5. z . y . and ±¯ /2. y .136) Sy = 2 ¯ h ( + + − − − ) .133)
It is easily veriﬁed that the Hermitian operators deﬁned by ¯ h ( + − + − + ) . z. z . + + x . y . and + + + − − = 1.131).138) 4 It is also easily demonstrated that S2 and Sz . z .10) (with the Lj replaced by the Sj ).. y . ± denotes a simultaneous eigenstate of the position operators x.137) Sz = 2 satisfy the commutation relations (5. and the spin operator Sz .138) constitute a realization of the spin operators S and S2 (for a spin onehalf particle) in spin space (i.8)–(5.5.135) Sx = 2 i¯ h ( − + − + − + ) .8 Wavefunction of a spin onehalf particle
5 ANGULAR MOMENTUM
They are also properly normalized and complete. (5. so that ++ = −− = 1.8 Wavefunction of a spin onehalf particle The state of a spin onehalf particle is represented as a vector in ket space. Let us suppose that this space is spanned by the basis kets x .
2
5. (5. The operator S2 takes the form 3¯2 h S = .
z . z ± = ± x . = c x . y . z  −A . which are termed factor spaces.143)
The multiplication in the above equation is of quite a different type to any which we have encountered previously. y . A general ket of the product space is not of the form (5. y . we can give meaning to the equation x . y . y . z + by assuming that it operates only on + . y . and a spin space ket + . y . z + x . y .8 Wavefunction of a spin onehalf particle
5 ANGULAR MOMENTUM
It is helpful to think of the ket x . z + +c− x . ± = x . z − . The ket vectors x . y .145)
. + as the product of two kets—a position space ket x . but is instead a sum or integral of kets of this form.143). z ) = x . we can give a meaning to any spin operator (such as Sz ) acting on x . A general state A of a spin onehalf particle is represented as a ket A in the product of the spin and position spaces.5. y . x .140)
where the c’s are numbers. y . z . z ) = ψ− (x . y . y . Similarly. y . This state can be completely speciﬁed by two wavefunctions: ψ+ (x . z + (c x . y . (5. z . y . z + +c x . and commutes with the + factor. y .144) (5. z  +A . This implies that every position space operator commutes with every spin operator. z factor. and commutes with x . y . y . z and ± are in two quite separate vector spaces. y . z ± is in a third vector space. z + by assuming that it operates only on the x . z .
92
(5.141) (5. y . z ) + = + x . the latter space is termed the product space of the former spaces. y . y . y . In this manner. We assume that such a product obeys the commutative and distributive axioms of multiplication: x . In mathematics. (5. The number of dimensions of a product space is equal to the product of the number of dimensions of each of the factor spaces. z + c x . and their product x . z . We can give meaning to any position space operator (such as Lz ) acting on the product x . z (c+ + + c− − ) = c+ x .142) (5. z .
150)
Thus. (5. for the moment. let us consider its effect on Sx . (5. y to y + dy . z )2 dx dy dz . and z to z + dz .62).3. Likewise.148)
To demonstrate that the operator (5. y . after rotation. y to y + dy . we would expect such an operator to take the form Tz (∆ϕ) = exp(−i Sz ∆ϕ/¯ ). and concentrate on its spin state.146)
5. h Thus. forget about the spatial position of the particle.151)
. this expectation value changes as follows:
† Sx → AR Sx AR = ATz Sx Tz A . with sz = +1/2 is ψ+ (x .
(5.147)
in spin space. z )2 dx dy dz .5. we need to compute
exp( i Sz ∆ϕ/¯ ) Sx exp(−i Sz ∆ϕ/¯ ). we were able to construct an operator Rz (∆ϕ) which rotates the system by an angle ∆ϕ about the zaxis in position space. y .9 Rotation operators in spin space
5 ANGULAR MOMENTUM
The probability of observing the particle in the region x to x + dx . and z to z + dz . (5. The normalization condition for the wavefunctions is ψ+ 2 + ψ− 2 dx dy dz = 1. Under rotation. 5.148) really does rotate the spin of the system. with sz = −1/2 is ψ− (x .149) (5. the ket A becomes AR = Tz (∆ϕ)A . Can we also construct an operator Tz (∆ϕ) which rotates the system by an angle ∆ϕ about the zaxis in spin space? By analogy with Eq. the probability of observing the particle in the region x to x + dx . In Sect. A general spin state A is represented by the ket A = +A + + −A − (5. h h
93
(5.9 Rotation operators in spin space Let us.
154)
i n λn ··· + [G. therefore. This takes the form i 2 λ2 [G.5. where use has been made of Eqs. (5.
(5. A]]] · · ·].
(5. (5. we can use the explicit formula for Sx given in Eq. Sx ] + ¯ h 2! which reduces to ∆ϕ2 ∆ϕ3 1 − − S ϕ − Sx + ··· + · · · . h h 2 or ¯ i ∆ϕ/2 h e + − e i ∆ϕ/2 + e −i ∆ϕ/2 − + e −i ∆ϕ/2 .151) becomes ¯ h exp( i Sz ∆ϕ/¯ ) ( + − + − + ) exp(−i Sz ∆ϕ/¯ ). (5. and λ is a real parameter.9 Rotation operators in spin space
5 ANGULAR MOMENTUM
This can be achieved in two different ways.135)–(5. · · · [G.157)
or
Sx cos ∆ϕ − Sy sin ∆ϕ. [G. [Sz .151). Applying the BakerHausdorff lemma to Eq. (5. 2 which reduces to Sx cos ∆ϕ − Sy sin ∆ϕ.152) (5. Sx ]] + · · · . A second approach is to use the so called BakerHausdorff lemma. (5. since it only uses the fundamental commutation relation (5.155) exp( i G λ) A exp(−i G λ) = A + i λ[G. [G.153) (5.135). We ﬁnd that Eq.125).158)
where use has been made of Eq. The second proof is more general than the ﬁrst.
94
. n!
where G is a Hermitian operator. [G.137). and is. y 2! 3!
i ∆ϕ ¯ h
2
[Sz . valid for systems with spin angular momentum higher than onehalf. A]] + (5.156)
(5. we obtain i ∆ϕ 1 Sx + [Sz .125). A] + 2!
(5. The proof of this lemma is left as an exercise. First.
5.148) rotates the expectation value of S by an angle ∆ϕ about the zaxis.161) demonstrate that the operator (5. In fact. (5. Equations (5. It is straightforward to show that Sy → Sy cos ∆ϕ + Sx sin ∆ϕ.
l
(5.165)
. h Consider the effect of the rotation operator (5.
(5.148) on the state ket (5.148). the expectation value of the spin operator behaves like a classical vector under rotation: Sk → Rkl Sl .159). It is easily seen that Tz (∆ϕ)A = e−i ∆ϕ/2 +A + + e i ∆ϕ/2 −A − . and the rotation operator about the kth axis is written h Rk (∆ϕ) = exp(−i Jk ∆ϕ/¯ ).159)
under the action of the rotation operator (5.163)
where the Jk are the generators of rotation. We ﬁnd that A → Tz (2π)A = −A .161)
since Sz commutes with the rotation operator.164)
(5. satisfying the fundamental commutation relation J × J = i ¯ J. we have effectively demonstrated that Jk → Rkl Jl . It is clear. that this property is not restricted to the spin operators of a spin onehalf system.9 Rotation operators in spin space
5 ANGULAR MOMENTUM
For a spin onehalf system. both methods imply that Sx → Sx cos ∆ϕ − Sy sin ∆ϕ (5. In fact.
l
(5. Consider a rotation by 2π radians.160) Furthermore.
95
(5. from our second derivation of the result (5. Sz → S z .147).162)
where the Rkl are the elements of the conventional rotation matrix for the rotation in question.159)–(5.
168) 2m where g is called the gyromagnetic ratio. we expect the above relation to also hold between the quantum mechanical operators. In fact. Does spin angular momentum also give rise to a contribution to the magnetic moment of a charged particle? The answer is “yes”. The minus sign does not affect the expectation value of S.167) µ= 2m Using the usual analogy between classical and quantum mechanics. In fact. respectively. This is indeed found to the the case. relativistic quantum mechanics actually predicts that a charged particle possessing spin should also possess a magnetic moment (this was ﬁrst demonstrated by Dirac). (5. (5.
5. which represent magnetic moment and orbital angular momentum. we know that p = v/m. Thus. The charge is equivalent to a current loop of radius r in the xy plane carrying current I = q v/2π r. We can write q µ= (L + g S) . (5. µ and L. ge = 2 2π 4π 0 ¯ c h
(5.169)
96
.10 Magnetic moments
5 ANGULAR MOMENTUM
Note that a ket rotated by 2π radians differs from the original ket by a minus sign. since S is sandwiched between A and A .5. and m is its mass.10 Magnetic moments Consider a particle of charge q and velocity v performing a circular orbit of radius r in the xy plane. respectively. The magnetic moment µ of the loop is of magnitude π r2 I and is directed along the zaxis. where p is the vector momentum of the particle. a rotation by 4π radians is needed to transform a ket into itself. However. We also know that L = r × p. Nevertheless. For an electron this ratio is found to be e2 1 1 + . we can write q µ = r × v.166) 2 where r and v are the vector position and velocity of the particle. both of which change sign. the minus sign does give rise to observable consequences. as we shall see presently. where L is the orbital angular momentum. It follows that q L.
B = B ^. derived originally by Schwinger. In fact. with ∆ϕ set equal to ω t. e > 0). t=0 sin ωt.
(5. t = e−i ω t/2 +A.175) (5. is z H = −µ·B = where e S·B = ω Sz . We shall ignore this correction in the following. 0) = exp(−i Ht/¯ ) = exp(−i Sz ω t/¯ ). − e (L + 2 S) 2 me (5.11 Spin precession The Hamiltonian for an electron at rest in a zdirected magnetic ﬁeld. (5.161) imply that Sx Sy Sz
t t t
= = =
Sx Sy Sz
t=0 cos ωt t=0 cos ωt t=0 .159)–(5. (4. The small correction 1/(2π 137).28).
− Sy + Sx
t=0 sin ωt. so µ for an electron (here.148).171) causes the electron spin to precess about the zaxis with angular frequency ω. is due to quantum ﬁeld effects.171)
eB . that the time evolution operator is precisely the same as the rotation operator for spin. by comparison with Eq.172)
(5.174) (5.5.170)
5.176)
The time evolution of the state ket is given by analogy with Eq. It is immediately clear that the Hamiltonian (5. me (5.164): A.177)
. the time evolution operator for this system is ω= T (t. Eqs.173)
It can be seen. me According to Eq.
97
(5. 0 + + e i ω t/2 −A. (5.11 Spin precession
5 ANGULAR MOMENTUM
The factor 2 is correctly predicted by Dirac’s relativistic theory of the electron. h h
(5. (5. 0 − .
The gyromagnetic ratio for a neutron is found experimentally to be gn = −1. it only takes times t = 2π/ω for the spin vector to point in its original direction. (5.178)
This frequency is deﬁned in an analogous manner to Eq. while the ﬁeldstrength B is varied. In experiments. sent along two different paths. Here. The above prediction has been veriﬁed experimentally to within a fraction of a percent. and ω is the spin precession frequency ω= gn e B . As a result. mp (5. We expect the observed neutron intensity in the interference region to exhibit a cos(±ω T/2 + δ) variation. e gn λ l ¯ (5. a neutron state ket going along path B acquires a phaseshift exp( i ω T/2) (the signs correspond to sz = ±1/2 states). An almost monoenergetic beam of neutrons is split in two. A and B.
98
.179)
where l is the pathlength through the magnetic ﬁeld region.165). where δ is the phase difference between paths A and B in the absence of a magnetic ﬁeld. This prediction depends crucially on the fact that it takes a 4π rotation to return a state ket to its original state. We now describe an experiment to detect the minus sign in Eq. If it only took a 2π rotation then ∆B would be half of the value given above. and λ is the de ¯ Broglie wavelength over 2π of the neutrons. When neutrons from path A and path B meet they undergo interference. the time of ﬂight T through the magnetic ﬁeld region is kept constant.11 Spin precession
5 ANGULAR MOMENTUM
Note that it takes time t = 4π/ω for the state ket to return to its original state. By contrast. path B enters a small region where a static magnetic ﬁeld is present. and then recombined.91. (The magnetic moment of a neutron is entirely a quantum ﬁeld effect). It follows that the change in magnetic ﬁeld required to produce successive maxima is ∆B = 4π ¯ h .5. (5. Path A goes through a magnetic ﬁeld free region. However. which does not agree with the experimental data. T is the time spent in the magnetic ﬁeld.172).
1) ≡ χ† . that a general spin ket can be expressed as a linear combination of the two eigenkets of Sz belonging to the eigenvalues ±¯ /2. In this representation.182) (5. that the eigenstates of orbital angular momentum can be conveniently represented as spherical harmonics. The matrix representation of a spin onehalf system was introduced by Pauli in 1926. + + − − c−
99
(5.12 Pauli twocomponent formalism We have seen. + (5. the orbital angular momentum operators take the form of differential operators involving only angular coordinates. a general bra takes the form
− → (0. Let us represent these basis eigenkets as column matrices: + − 1 → ≡ χ+ .186)
.180) (5. 5.4. −
and a general ket becomes
A = +A + + −A − →
+A −A
. the spin angular momentum operators take the form of matrices. A− ). It is conventional to represent the eigenstates of spin angular momentum as column (or row) matrices.184) is called a twocomponent spinor.181)
The corresponding eigenbras are represented as row matrices: + → (1.183)
0 → ≡ χ− .184)
The column matrix (5. In this representation. 5. +A −A
(5.5. 1
In this scheme. 0) ≡ χ† . These h are denoted ± .9. in Sect. from Sect.185)
c = + = c χ +c χ . Recall.
(5.12 Pauli twocomponent formalism
5 ANGULAR MOMENTUM
5. 0
(5. and can be written χ≡
A = A+ + + A− − → ( A+ .
195) (5. = i 0
. We ﬁnd that σ1 σ2 σ3 0 1 = .12 Pauli twocomponent formalism
5 ANGULAR MOMENTUM
where the c± are complex numbers. These h matrices. A− ) = (c+ . can easily be evaluated using the explicit forms for the spin operators given in Eqs.194) (5. (5.
(5.135)–(5.191)
+A −A
It follows that we can represent the operator/ket relation (5.192)
(5.185) becomes
∗ ∗ ∗ ∗ χ† ≡ ( A+ .190) (5.193) 2 where the σk are the matrices of the ±Sk ± values divided by ¯ /2.
(5. 1 0
=
+Sk + −Sk +
+Sk − −Sk −
+A −A
. + −
(5. = =
+Sk + +A + +Sk − −A .188) as the matrix relation ¯ h χ = σk χ.
(5.189)
However.188)
+A −A
≡χ. −Sk + +A + −Sk − −A . This ket is represented as A → +A −A or
(5.187)
Consider the ket obtained by the action of a spin operator on ket A: A = Sk A . (5. The row matrix (5. which are called the Pauli matrices.5.196)
1 0 = .137). 0 −1
100
0 −i . c− ) = c+ χ† + c− χ† .
we are effectively representing the spin operators in terms of the Pauli matrices: Sk → ¯ h σk . z . and 3 refer to x. plus all cyclic permutations). 2 (5. that we can represent a general basis ket as the product of basis kets in position space and spin space: x .125).5. z  +A . State A is completely speciﬁed by the two wavefunctions ψ+ (x . in this scheme. (5. (5. y .
101
x . ± = x . respectively.. from Sect.
(5. σ1 σ2 − σ2 σ1 = 2 i σ3 .204)
. σj } = 2 δij . (5. y .197)
The expectation value of Sk can be written in terms of spinors and the Pauli matrices: ¯ h † Sk = ASk A = A± ±Sk ± ±A = χ σk χ. z ± = ± x . It is also easily seen that the Pauli matrices satisfy the anticommutation relations {σi .197) to give the following commutation relation for the Pauli matrices: σ × σ = 2 i σ. z  −A .196) actually satisfy these relations (i. Note that. 2.198) 2 ± The fundamental commutation relation for angular momentum. y .202) (5. Eq.8. (5.194)–(5. y . and resides in the product space of the position and spin ket spaces. and z. x . Recall. y . y.203)
(5.e. z ) = Consider the operator relation A = Sk A . y . z .12 Pauli twocomponent formalism
5 ANGULAR MOMENTUM
Here. z ) = ψ− (x .
(5. 5. can be combined with (5. y . 1.200)
Let us examine how the Pauli scheme can be extended to take into account the position of a spin onehalf particle.201)
The ket corresponding to state A is denoted A .199)
It is easily seen that the matrices (5.
y . y .207) ψ− (r ) and so on. = −i ¯ h ∂xk x .5.209) ¯ h σk χ.208)
In the Schr¨dinger representation. y . z pk +A ∂ x .78).206) (5. y . ∂xk
(5. (5. y .204) becomes simply χ = Consider the operator relation A = pk A . It is fairly obvious that we can represent the operator relation (5.12 Pauli twocomponent formalism
5 ANGULAR MOMENTUM
It is easily seen that x . y . (5. z . The components of a spinor are now wavefunctions. y . The above equation reduces to
h ψ (r ) −i ¯ ∂ψ+ (r )/∂xk + . y . (5. z  +A = x .212)
. z  +A + −Sk − x . z  −A .205)
where use has been made of the fact that the spin operator Sk commutes with the eigenbras x . y . y . 2
(5. In this scheme. y . z  −A = = +Sk + x . instead of complex numbers. z  −A . the operator equation (5. z  +A .211)
where use has been made of Eq. we have o x . z  −A
=
(5. z  +A + +Sk − x .210)
x . z  +A x . −Sk + x . y . (3.204) as a matrix relation if we generalize our deﬁnition of a spinor by writing ψ+ (r ) A → ≡ χ. = −i ¯ ∂ψ− (r )/∂xk h ψ− (r )
102
(5. y . z  −A . z pk −A ∂ = −i ¯ h x .
.
What about combinations of position and spin operators? The most commonly occurring combination is a dot product: e. where pk → −i ¯ h (5.209) can be written χ = pk χ. ak σ k = a1 + i a 2 −a3
(5.200). This quantity is represented as a 2 × 2 matrix: σ·a ≡
k
+a3 a1 − i a 2 .207).215)
o Since. it is clear that the above quantity must be regarded as a matrix differential operator which acts on spinors of the general form (5. where a ≡ (ax . h
103
(5. the operator equation (5.. Thus.5. The important identity (σ·a) (σ·b) = a·b + i σ·(a × b) (5.216)
follows from the commutation and anticommutation relations (5. A general rotation operator in spin space is written T (∆φ) = exp (−i S·n ∆ϕ/¯ ) .218)
.g.199) and (5. σk } + [σj . in the Schr¨dinger representation. In fact. σk ] aj bk 2 2 (σjk + i
jkl σl ) aj bk
=
j k
= a·b + i σ·(a × b). az ) is some vector position operator. I is the 2 × 2 unit matrix. (5.213)
∂ I.g. y . any position operator (e. or z . S · L = (¯ /2) σ · L. σ j aj
j k
σk bk =
j k
1 1 {σj .12 Pauli twocomponent formalism
5 ANGULAR MOMENTUM
Thus.214) ∂xk Here. Consider the h hybrid operator σ·a. p k or Lk ) is represented in the Pauli scheme as some differential operator of the position eigenvalues multiplied by the 2 × 2 unit matrix. a general position operator takes the form of a differential operator in x .217)
(5. ay .
(5. Thus. where n is a unit vector pointing along the axis of rotation. This can easily be evaluated using the Taylor series for an exponential. n can be regarded as a trivial position operator.198)].216).219)
in the Pauli scheme.225)
. and ∆ϕ is the angle of rotation. Here. we can write (σ·n)2 ∆ϕ 1 − exp (−i σ·n ∆ϕ/2) = 2! 2
2
(σ·n)4 ∆ϕ + 4! 2
3
4
= cos(∆ϕ/2) I − i sin(∆ϕ/2) σ·n. The term on the righthand side of the above expression is the exponential of a matrix. However. The explicit 2 × 2 form of this matrix is
(σ·n)3 ∆ϕ ∆ϕ (σ· n) − −i 2 3! 2
+ · · ·
+ · · ·
(5. (5.220) (5.224)
.222)
cos(∆ϕ/2) − i nz sin(∆ϕ/2) (−i nx − ny ) sin(∆ϕ/2) (−i nx + ny ) sin(∆ϕ/2) cos(∆ϕ/2) + i nz sin(∆ϕ/2)
Rotation matrices act on spinors in much the same manner as the corresponding rotation operators act on state kets. (5.12 Pauli twocomponent formalism
5 ANGULAR MOMENTUM
by analogy with Eq. we require (χ† σk χ) ≡ (χ† ) σk χ =
104
Rkl (χ† σl χ). (5. the quantity χ† σk χ is proportional to the expectation value of Sk [see Eq. The Pauli matrices remain unchanged under rotations. The rotation operator is represented exp (−i S·n ∆ϕ/¯ ) → exp (−i σ·n ∆ϕ/2) h (5. plus the rules (σ·n)n = 1 (σ·n)n = (σ·n) for n even. so we would expect it to transform like a vector under rotation (see Sect.221)
These rules follow trivially from the identity (5. In fact.
(5.148).5. 5. χ = exp (−i σ·n ∆ϕ/2) χ.
l
(5. Thus.223)
where χ denotes the spinor obtained after rotating the spinor χ an angle ∆ϕ about the naxis.9). for n odd.
r . The boundary conditions (for a bound state) are obtained from the normalization constraint ψ2 dV = 1.13 Spin greater than onehalf systems
5 ANGULAR MOMENTUM
where the Rkl are the elements of a conventional rotation matrix. (5. However.156). in addition to Hermitian operators.13 Spin greater than onehalf systems In the absence of spin.227)
The previous two formulae can both be validated using the BakerHausdorff lemma. which holds for Hermitian matrices. (5. This function speciﬁes the probability density for observing the particle at a given position.230)
This is all very familiar. can be transformed into a partial differential equation for the wavefunction ψ(r ) ≡ r E . we now know how to generalize this scheme to deal with a spin onehalf particle. Using the Schr¨dinger representation. Instead of representing the state of the particle by a single wavefunction.5. since exp i σ3 ∆ϕ −i σ3 ∆ϕ σ1 exp = σ1 cos ∆ϕ − σ2 sin ∆ϕ 2 2 (5. This is easily demonstrated. in which p → −i ¯ .228)
where H is now a partial differential operator.
105
.226)
plus all cyclic permutations. the Hamiltonian can be written as some function of the o position and momentum operators.9) exp −i Sz ∆ϕ i Sz ∆ϕ Sx exp = Sx cos ∆ϕ − Sy sin ∆ϕ. we ﬁnd H ψ = E ψ. we use two wavefunctions. ψ+ (r ).229)
(5. The ﬁrst. (5. 5. In general. The above expression is the 2 × 2 matrix analogue of (see Sect. ¯ h ¯ h (5. h H E = E E .
5. the energy eigenvalue problem.
The most
106
. Adopting the Schr¨dinger representation. and the c± are arbitrary complex numbers. (5.232)
Here. The physical signiﬁcance of the above expression is clear. both equations have the same form. the Hamiltonian is represented as diagonal 2 × 2 matrix partial differential operator in the Schr¨dinger/Pauli scheme [see Eq. c−
(5. the Hamiltonian is a function of the position.215)]. The above spinor equation can always be written out explicitly as two coupled partial differential equations for ψ+ and ψ− ..197)]. which is simply the row vector of ψ+ and ψ− .13 Spin greater than onehalf systems
5 ANGULAR MOMENTUM
speciﬁes the probability density of observing the particle at position r with spin angular momentum +¯ /2 in the zdirection. In other words. the most general solution to Eq.214)]. the Hamiltonian is represented as a 2 × 2 matrix of complex numbers in the Schr¨dinger/Pauli scheme [see Eq. In fact. In this situation.e. and the spinor eigenvalue equao tion (5. the o partial differential equation for ψ+ decouples from that for ψ− . (5. Suppose that the Hamiltonian depends only on the spin operators.5. χ.231) can be written c χ = ψ(r ) + . momentum. ψ− (r ).231)
where χ is a spinor (i. so there is only really one differential equation. Suppose that the Hamiltonian has no dependence on the spin operators. (5. but the direction of its spin angular momentum remains undetermined. ψ(r ) is determined by the solution of the differential equation. (5. The Hamiltonian determines the relative probabilities of ﬁnding the particle at various different positions. (5. speciﬁes the h probability density of observing the particle at position r with spin angular momentum −¯ /2 in the zdirection. and the Pauli scheme. In the Pauli scheme. In this case. these wavefunctions are h combined into a spinor. The second. a 1 × 2 matrix of wavefunctions) and H is a 2 × 2 matrix partial differential operator [see Eq.231) reduces to a straightforward matrix eigenvalue problem. In general. and spin operao tors. the energy eigenvalue problem reduces to H χ = E χ. In this case.
or a halfinteger. The state of the particle can be represented more succinctly by a spinor. sz A . In this case. we can represent the state of the particle by 2 s + 1 different wavefunctions.234)
where A denotes a state ket in the product space of the position and spin operators.
107
. Here. · · · . a particle for which the eigenvalue of S2 is s (s + 1) ¯ 2 . (5.233)
Here. −s. where sz is allowed to take the values h s. s − 1. In fact. In other words.233). of course. ψ+ and ψ− .. it is not possible to decompose the spinor as in Eqs.e. This can only be represented as a spinor involving different wavefunctions. Here. but leaves its position undetermined. c−
(5. Clearly. Thus. the ratio c+ /c− is determined by the matrix eigenvalue problem. a spin threehalves particle by a fourcomponent spinor. In general. the Hamiltonian is a function of both position and spin operators. ψsz (r ) = r  s. the Hamiltonian determines the direction of the particle’s spin angular momentum. there are 2 s + 1 distinct allowed values of s z . (5.5. a spin one particle by a threecomponent spinor. what happens if we have a spin one or a spin threehalves particle? It turns out that we can generalize the Pauli twocomponent scheme in a fairly straightforward manner. Consider a spins particle: i.13 Spin greater than onehalf systems
5 ANGULAR MOMENTUM
general solution can again be written c χ = ψ(r ) + . a general Hamiltonian causes the direction of the particle’s spin angular momentum to vary with position in some speciﬁed manner. −s + 1. and the wavefunction ψ(r ) is arbitrary. ψsz (r ) speciﬁes the probability density for observing the particle at position r with spin angular momentum sz ¯ in the h zdirection. But. More exactly. a spin onehalf particle is represented by a twocomponent spinor. denoted ψsz (r ).232) and (5. h The eigenvalues of Sz are written sz ¯ . Not surprisingly. and so on. s is either an integer. which is simply the 2 s + 1 component row vector of the ψsz (r ). χ.
it is trivial to construct the σz matrix.240) (5. position space operators take the o form of diagonal (2 s + 1) × (2 s + 1) matrix differential operators. σ z is the suitably normalized diagonal matrix of the eigenvalues of Sz . j. h Hence. j .237).240)–(5.242) (5. h (5. thereby. We know. Sz s. that (σx )jl = s (s + 1) − j (j − 1) δj.55)–(5. (5. s. we can represent the momentum operators as [see Eq. The matrix elements of σx and σy are most easily obtained by considering the shift operators. and (5. that S+ s.238)
It follows from Eqs. l and. or halfintegers.13 Spin greater than onehalf systems
5 ANGULAR MOMENTUM
In this extended Schr¨dinger/Pauli scheme. jSk s. jSk s.235)
where I is the (2 s + 1) × (2 s + 1) unit matrix. (σz )jl = S± = S x ± i S y .241) (5. But. j S− s. j − 1 .56). (5. lying in the range −s to +s. l j = δjl . Thus. We represent the spin operators as where the (2 s + 1) × (2 s + 1) extended Pauli matrix σk has elements (σk )jl = s. construct the extended Pauli matrices? In fact. j = j ¯ s. By deﬁnition.242). s¯ h (5. j .239) s¯ h s where use has been made of the orthonormality property of the s. jSz s. h s (s + 1) − j (j − 1) ¯ s.5. h (5.l+1 2s
108
.236)
Here. j + 1 . how can we evaluate the brackets s.237) Sk → s ¯ σ k .214)] pk → −i ¯ h ∂ I. l . l are integers. from Eqs. Thus. (5. ∂xk (5. (5. j = = s (s + 1) − j (j + 1) ¯ s.
243)–(5. (5. = 0 −1
0 −i . 2 0 1 0
1 0 .244). For a spin one (s = 1) particle.243) 2s s (s + 1) − j (j − 1) δj.l+1 (σy )jl = 2is s (s + 1) − j (j + 1) δj.
109
.l−1 .239) and (5. 0 0 −1
(5.244) − 2is According to Eqs. = i 0
(5.248)
σy
0 −i 0 1 = √ i 0 −i .l−1 . where the Hamiltonian is some function of position and spin operators. into 2 s + 1 coupled partial differential equations involving the 2 s + 1 wavefunctions ψsz (r ). such a system of equations is generally too complicated to solve exactly. Unfortunately. the Pauli matrices for a spin onehalf (s = 1/2) particle are + σx σy σz 0 1 = .13 Spin greater than onehalf systems
5 ANGULAR MOMENTUM
s (s + 1) − j (j + 1) δj. (5. (5.249)
σz
(5. This means that we can convert the general energy eigenvalue problem for a spins particle. 1 0
(5.246) (5.250)
In fact. 2 0 i 0 1 0 0 = 0 0 0 .5. we can now construct the Pauli matrices for a spin anything particle.245) (5. we ﬁnd that σx 0 1 0 1 = √ 1 0 1.247)
as we have seen previously.
respech h tively. (5. j stand for either x.14 Addition of angular momentum
5 ANGULAR MOMENTUM
5.252)
J2 × J 2 = i ¯ J 2 . where j1 and j2 are either integers. −j2 . respectively. J1 could be an orbital angular momentum operator.5. and J2 a spin angular momentum operator. According to Eqs. By deﬁnition. we are going to have to learn how to add angular momentum operators. J2j ] = 0.14 Addition of angular momentum Consider a hydrogen atom in an l = 1 state. what is the total angular momentum of the system? In order to answer this question. these operators are Hermitian.14). or z. that the eigenvalues of J12 and J22 can be written j1 (j1 + 1) ¯ 2 and h 2 j2 (j2 + 1) ¯ .251) (5. Alternatively. Let us consider the most general case.255)
110
. y. j1 − 1. h h So. · · · . where m1 and m2 are numbers lying in the ranges j1 .11) and (5. J1 and J2 . and spin angular momentum of magnitude ¯ /2. −j2 + 1. h
We assume that the two groups of operators correspond to different degrees of freedom of the system. and obey the fundamental commutation relations J1 × J 1 = i ¯ J 1 . Let us deﬁne the total angular momentum operator J = J 1 + J2 . · · · . The electron possesses orbital angular momentum of magnitude ¯ . For instance. j2 − 1. h (5. Suppose that we have two sets of angular momentum operators. −j1 and j2 . J satisﬁes the fundamental commutation relation J × J = i ¯ J. so that [J1i . respectively. −j1 + 1.254)
Now J is an Hermitian operator. We know. (5. (5. or halfintegers. since it is the sum of Hermitian operators. We h also know that the eigenvalues of J1z and J2z take the form m1 ¯ and m2 ¯ . from the general properties of angular momentum.253)
where i. J1 and J2 could be the orbital angular momentum operators of two different particles in a multiparticle system. h (5.
The ﬁrst group is J12 . J22 . j − 1.260) (5. J1z ] = 0. [J2 . J1z . It is clear. · · · . In other words. It follows that the eigenvalue of J 2 can be written j (j + 1) ¯ 2 . j and m. At this stage. that we can form two alternate groups of mutually commuting operators. J2i ] = 0. The eigenvalue of Jz takes the form m ¯ . J1i ] = 0. However. J possesses all of the expected properties of an angular momentum operator. and those of the individual angular momenta. We can deﬁne simultaneous eigenkets of each operator group. [J12 . (5. J22 ] = 0. J 2 . (5.258)
and also that all of the J1i operators commute with the J2i operators.261)
This suggests that it is not possible to measure the quantum numbers m1 and m2 simultaneously with the quantum number j. (5.14 Addition of angular momentum
5 ANGULAR MOMENTUM
Thus. and j can all be measured simultaneously. Now We know that J 2 = J12 + J22 + 2 J1 ·J2 .259) This implies that the quantum numbers j1 . j2 . where m lies h in the range j. and J2z . m1 . and J2z are denoted
111
.256) that [J2 .256)
(5.256) that [J 2 . J22 . It follows from Eq. J2z ] = 0. we do not know the relationship between the quantum numbers of the total angular momentum. [J22 . from the preceding discussion. Thus. J1z . it is clear from Eq. we cannot determine the projections of the individual angular momenta along the zaxis at the same time as the magnitude of the total angular momentum. and m2 . we can know the magnitude of the total angular momentum together with the magnitudes of the component angular momenta. (5. (5. J22 . The simultaneous eigenkets of J12 . −j + 1. and Jz . These two groups of operators are incompatible with one another.257) (5. The second group is J12 .5. where j is an h integer. j1 . J12 ] = [J 2 . −j. j2 . or a halfinteger.
J 2 and Jz are denoted j1 . but we cannot know both pairs of quantum numbers at the same time. m1 . we can assume that the quantum numbers j1 and j2 are known. j2 . j. j2 . j2 . or the quantum numbers j and m.267) (5. m2 . h = m2 ¯ j1 . j2 . j2 . m1 . m1 . J 2 . j. and m. h = m ¯ j1 . j2 .
m1 m2
(5. it is not in an eigenstate of the latter.270) (5. j. h = j (j + 1) ¯ 2 j1 . h = j2 (j2 + 1) ¯ 2 j1 . m1 . m2 J2z j1 . j2 . and J2z . In addition. m2 = j1 (j1 + 1) ¯ 2 j1 .268) (5. m J22 j1 . J22 .266) (5.271)
j1 . m . j.265)
The simultaneous eigenkets of J12 . j. m1 . Since the operators J12 and J22 are common to both operator groups. In other words. j2 .14 Addition of angular momentum
5 ANGULAR MOMENTUM
j1 . J22 . m2 . j2 . j. m . m Jz j1 . h = j2 (j2 + 1) ¯ 2 j1 . m1 . In other
112
. m1 . j2 . j2 . j2 . h (5. The summation is over all allowed values of m1 . m2 J22 j1 . m2 . j2 . m j1 . j. j.5. and Jz is incompatible with the group J12 . m1 . j2 . m2  = 1. j2 . j2 . m1 . h (5. This means that if the system is in a simultaneous eigenstate of the former group then. where J12 j1 . j2 . m . m1 . and have unit norms.269)
Each set of eigenkets are complete. j2 . m = 1. We can write a conventional completeness relation for both sets of eigenkets: j1 .262) (5. j.264) (5. j. m2 J1z j1 .263) (5. m J2 j1 . m2 . j2 . we can always determine the magnitudes of the individual angular momenta. h = m1 ¯ j1 . we can either know the quantum numbers m1 and m2 . mutually orthogonal (for eigenkets corresponding to different sets of eigenvalues). m = j1 (j1 + 1) ¯ 2 j1 . j. The operator group J12 . J22 . j2 . m . m1 . m . j2 . m2 . J1z . m2 . where J12 j1 . in general. j2 . j.
j m
where the righthand sides denote the identity operator in the ket space corresponding to states of given j1 and j2 . m2 j1 .
274) (5. respectively.275) (5. from Eq. m1 . and jz ¯ . we obtain (m − m1 − m2 ) j1 . So. without loss of generality.270) to write j1 . m2 j1 . the zcomponents of different angular momenta add algebraically. then a measurement of the quantum numbers m1 and m2 will give a range of possible values. j1 . and m are known with certainty. J22 . h What is uncertain is the magnitude of the total angular momentum. j2 . that j1 ≥ j2 . Forming the inner product with j1 . j.
(5. m = 0. (5. m = 0. Second. j2 . The weights. Thus. If the system is in a state where a measurement of J12 . m =
m1 m2
j1 .273)
which proves the assertion. j2 . are called the ClebschGordon coefﬁcients. that for given j1 and j2 the largest possible value of m is j1 + j2 (since j1 is the
113
. To prove this. j2 . j2 . (5. projected along the zaxis. and Jz is bound to give the results j1 (j1 + 1) ¯ 2 . m1 . j2 . j2 . j. m2 . m2 j1 . j2 . We know. m2 j1 . j (j + 1) ¯ . and spin angular momentum ¯ /2. the coefﬁcients are zero unless m = m 1 + m2 . with orbital angular momentum ¯ . (5. m1 . j. m1 . j. h h The ClebschGordon coefﬁcients possess a number of very important properties. m2 j1 . if the quantum numbers j1 . First. j. m1 . j2 . j2 . J 2 . j2 (j2 + h 2 2 1) ¯ . m . an electron in an l = 1 state. the coefﬁcients vanish unless j1 − j2  ≤ j ≤ j1 + j2 . we can write the eigenkets of the ﬁrst group of operators as a weighted sum of the eigenkets of the second set. we note that (Jz − J1z − J2z )j1 . j. j2 .5. then a measurement of J1z and J2z will h h h give the results m1 ¯ and m2 ¯ with probability  j1 . conh h stitutes a state whose total angular momentum projected along the zaxis is 3¯ /2. m j1 .273).14 Addition of angular momentum
5 ANGULAR MOMENTUM
words.276)
We can assume.272)
Thus. We can use the completeness relation (5. j2 . j2 . m2 . m1 . m 2 . j.
j2 . This implies that the largest possible value of j is j1 + j2 (since. j2 . j2 . j2 . j. by deﬁnition. j. m1 . m1 . m = j1 . mj1 .e. j2 . m2 j1 . there are (2 j1 + 1) allowable values of m1 and (2 j2 + 1) allowable values of m2 . mj1 .
(5. Finally.. j2 .5. j2 .277)
This proves our assertion. etc.  j1 . m j1 . the sum of the modulus squared of all of the ClebschGordon coefﬁcients is unity: i.e.278)
This assertion is proved as follows: j1 . m 2 = 1.
m1 m2
(5.14 Addition of angular momentum
5 ANGULAR MOMENTUM
largest possible value of m1 . j. j2 . m
m1 m2
=
m1 m2
 j1 . Third. j2 . j2 . m1 . m2 j1 . m1 . needed to span the ket space corresponding to ﬁxed j1 and j2 . there can only be (2 j1 + 1) (2 j2 + 1) distinct allowable values of the quantum numbers j and m. j2 . Since the eigenkets j1 . there are 2 j + 1 allowed values of m. they must also form a set of (2 j1 + 1) (2 j2 + 1) independent kets.
(5. j2 .
j1 +j2
j=j1 −j2
(2 j + 1) ≡ (2 j1 + 1) (2 j2 + 1). j2 . m1 . m span the same space.
m1 m2
(5. m1 . In other words. the largest value of m is equal to j). m2 j1 . j. j2 .). j. the ClebschGordon coefﬁcients obey two recursion relations. m2 j1 . there are (2 j1 + 1) (2 j2 + 1) independent eigenkets. j1 .270). m1 . j.. j. m2 . m 2 = 1. Thus.279)
where use has been made of the completeness relation (5. To obtain these relations we start from J± j1 . Now. j2 . m2 . m = (J± + J± ) 2 1 × j1 . j2 . j. m2 j1 . For each allowed value of j. It is easily seen that if the minimum allowed value of j is j1 − j2 then the total number of allowed values of j and m is (2 j1 + 1) (2 j2 + 1): i. We have already seen that the maximum allowed value of j is j1 + j2 . j.280)
114
.
(5.278). which are speciﬁed by Eqs. j2 . m2 . j2 . m . This is a convenient choice. and making use of the orthonormality property of the basis eigenkets.56). j2 .5. j. m1 ± 1.
115
. (5. j2 . are identical to the ClebschGordon coefﬁcients. (5. m2 j1 . j2 . m1 . j2 . j2 . j2 . j2 . m . since it ensures that the inverse ClebschGordon coefﬁcients. m : j1 . m1 . m1 1. m2 =
j m
j1 . j2 . m1 . m1 . mj1 . j. m2 j1 . m2 j1 . m . j2 . j. m1 . m2 ± 1 × j1 . m2 in terms of the j1 . j2 . j2 . m2 j1 . j1 .14 Addition of angular momentum
5 ANGULAR MOMENTUM
Making use of the wellknown properties of the shift operators. (5. m1 . m2
+ j2 (j2 + 1) − m2 (m2 ± 1) j1 . m ± 1 =
m1 m2
j1 (j1 + 1) − m1 (m1 ± 1) j1 . m2 . m1 . j. m2
It is clear. m 1j1 . j2 . we obtain the desired recursion relations: j (j + 1) − m (m ± 1) j1 . j. j.55)–(5. mj1 . m2 = j1 . m1 . we obtain j (j + 1) − m (m ± 1) j1 . together with the normalization condition (5. j. j2 . j. m1 . j2 . m ± 1 = j1 (j1 + 1) − m1 (m1 + j2 (j2 + 1) − m2 (m2 1) j1 . The easiest way of demonstrating this assertion is by considering some speciﬁc examples. j. j2 . j2 . j2 . mj1 . that we can always choose the ClebschGordon coefﬁcients to be real numbers.284)
It turns out that the recursion relations (5.
(5. j2 . m . In other words. j. m2 j1 .283)
The inverse ClebschGordon coefﬁcients are the weights in the expansion of the j1 . j2 . m1 . j2 . j2 .281) Taking the inner product with j1 .282) 1) j1 . j2 . from the absence of complex coupling coefﬁcients in the above relations. This sign is ﬁxed by convention. m1 . j1 . j. are sufﬁcient to completely determine the ClebschGordon coefﬁcients to within an arbitrary sign (multiplied into all of the coefﬁcients).282).
1/2 .14 Addition of angular momentum
5 ANGULAR MOMENTUM
Let us add the angular momentum of two spin onehalf systems: e.g. where the allowed values of j differ by integer amounts. All the boxes contain question marks because we do not know any ClebschGordon coefﬁcients at the moment.. m2 = 1/2. two electrons at rest.278) implies that the sum of the squares of all
116
. 1. It is helpful to arrange all of the possibly nonzero ClebschGordon coefﬁcients in a table: m1 m2 1/2 1/2 ? ? ? ? 1/2 1/2 ? ? ? ? 1/2 1/2 ? ? ? ? 1/2 1/2 ? ? ? ? j1 =1/2 j 1 1 1 0 j2 =1/2 m 1 0 1 0 The box in this table corresponding to m1 = 1/2. 11/2. 1/2.. A ClebschGordon coefﬁcient is automatically zero unless m1 + m2 = m. m1 m2 1/2 1/2 ? 0 1/2 1/2 0 ? 1/2 1/2 0 ? 1/2 1/2 0 0 j1 =1/2 j 1 1 j2 =1/2 m 1 0 0 0 0 ? 1 1 0 ? ? 0 0 0
The normalization condition (5. from Eq. 1/21/2.e. that m1  ≤ 1/2 and m2  ≤ 1/2. 1/2. j = 1. We know. or the inverse ClebschGordon coefﬁcient 1/2. the zcomponents of angular momentum have to add algebraically.276). j1 = j2 = 1/2. We immediately conclude that these boxes must contain zeroes: i. In other words. 1 . It follows that either j = 0 or j = 1. (5. 1/2. Many of the boxes in the above table correspond to m1 +m2 = m.5. 1/2. We also know. that 0 ≤ j ≤ 1. from general principles. Thus. 1/2. So. two spin onehalf systems can be combined to form either a spin zero system or a spin one system. 1. m = 1 gives the ClebschGordon coefﬁcient 1/2. 1/2.
(5. −1/21.5. 1/21. 1/21. m1 = m2 = ±1/2. the j1 and j2 labels have been suppressed for ease of notation. 0 + −1/2. m1 m2 1/2 1/2 ±1 0 0 0 1/2 1/2 0 ? 0 ? 1/2 1/2 0 ? 0 ? 1/2 1/2 0 0 ±1 0 j1 =1/2 j 1 1 1 0 j2 =1/2 m 1 0 1 0 Let us evaluate the recursion relation (5. 0 = 2 1/2. −1 = ± 2.286)
Here. m = 0. 0 = √ √ 2 −1/2. −1/21. 1 = 1/2. 1/21.14 Addition of angular momentum
5 ANGULAR MOMENTUM
the rows and columns of the above table must be unity. (5. but we have no way of determining the signs at present. The only real solutions to the above set of equations are √ √ 2 1/2. with j = 1. 0 = 1/2. 0 2 = 1.288)
The choice of sign is arbitrary—the conventional choice is a positive sign. and 1/2. 0 + −1/2. 1/21. We ﬁnd that √ √ (5.
117
.287) from the normalization condition. −1 = ±1.285) 1/2. There are two rows and two columns which only contain a single nonzero entry. 1/21. 0 = 2 −1/2. −1/21. Thus. (5. 1 = ± 2. Thus. 1/21. We also know that 1/2. We conclude that these entries must be ±1. −1/21.282) for j1 = j2 = 1/2. 0 2 + −1/2. taking the upper/lower sign. −1/21. 1/21.
The requirement that the dot product of the fourth column with itself is unity tells us that the magnitudes of the unknown entries have to √ be 1/ 2.5. However. This follows because the entries in the table give the expansion coefﬁcients of one of our alternative sets of eigenkets in terms of the other set.14 Addition of angular momentum
5 ANGULAR MOMENTUM
our table now reads m1 m2 1/2 1/2 1 0 √ 1/2 1/2 0 1/√2 1/2 1/2 0 1/ 2 1/2 1/2 0 0 j1 =1/2 j 1 1 j2 =1/2 m 1 0 0 0 0 1 1 1 0 ? ? 0 0 0
We could ﬁll in the remaining unknown entries of our table by using the recursion relation again. Thus. Likewise. The normalization condition tells us that the dot product of a row or column with itself must be unity. the dot product of a row with any other row must be zero. and each set of eigenkets contains mutually orthogonal vectors with unit norms. an easier method is to observe that the rows and columns of the table must all be mutually orthogonal. for the dot product of a column with any other column. The only way that the dot product of the fourth column with the second column can be zero is if the unknown entries are equal and opposite. That is. The columns give the expansions of
118
. The unknown entries are undetermined to an arbitrary sign multiplied into them both. the ﬁnal form of our table (with the conventional choice of arbitrary signs) is m1 m2 1/2 1/2 1 0 0 √ 0 √ 1/2 1/2 0 1/√2 0 1/ √2 1/2 1/2 0 1/ 2 0 1/ 2 1/2 1/2 0 0 1 0 j1 =1/2 j 1 1 1 0 j2 =1/2 m 1 0 1 0 The table can be read in one of two ways.
the ket on the lefthand side is a m1 . for every choice of j1 . m2 ket. The rows give the expansions of the eigenstates of individual angular momentum in terms of those of overall angular momentum. Note that our table is really a combination of two subtables. m2 . A complete knowledge of angular momentum addition is equivalent to a knowing all possible tables of ClebschGordon coefﬁcients. m kets. 2 (5. 0 = √ ( 1/2. 0 ) . m ket.e. j2 and j) in many standard reference books.5. 1/2 ) . one involving j = 0 states. the second row tells us that 1 1/2.
119
. 2 (5. Thus. whereas those on the righthand side are m1 . and j we can construct a table of ClebschGordon coefﬁcients corresponding to the different allowed values of m 1 .290)
Here. Thus. These tables are listed (for moderate values of j 1 .14 Addition of angular momentum
5 ANGULAR MOMENTUM
the eigenstates of overall angular momentum in terms of the eigenstates of the individual angular momenta of the two component systems. m2 kets. The ClebschGordon coefﬁcients corresponding to two different choices of j are completely independent: i. −1/2 = √ ( 1. and m (subject to the constraint that m1 + m2 = m). there is no recursion relation linking ClebschGordon coefﬁcients corresponding to different values of j.. the second column tells us that 1 1. Thus.289)
The ket on the lefthand side is a j. 0 + 0. whereas those on the righthand side are j. j2 . −1/2 +  − 1/2. and one involving j = 1 states.
Consider the simplest nontrivial system. Can we ﬁnd the approximate eigenvalues and eigenstates of the modiﬁed Hamiltonian. is also timeindependent. (6.1) Here. by performing some sort of perturbation analysis about the eigenvalues and eigenstates of the original Hamiltonian. H1 can. It is usually assumed that the unperturbed Hamiltonian.2 The twostate system Let us begin by considering timeindependent perturbation theory. H0 is a simple Hamiltonian for which we know the exact eigenvalues and eigenstates. in which there are only two independent eigenkets of the unperturbed Hamiltonian. which is very common. has no explicit dependence on time. in which the modiﬁcation to the Hamiltonian.
6.1 Introduction We have developed techniques by which the general energy eigenvalue problem can be reduced to a set of coupled partial differential equations involving various wavefunctions. These are denoted H0 1 = E1 1 . H0 . H1 introduces some interesting additional physics into the problem. we need to develop some techniques for ﬁnding approximate solutions to otherwise intractable problems. H0 + H1 . Clearly.2)
. the number of such problems which yield exactly soluble equations is comparatively small. H1 . However. Unfortunately. be regarded as being small compared to H0 . The Hamiltonian of a system is written H = H 0 + H1 . in some sense (which we shall specify more exactly later on).6 APPROXIMATION METHODS
6
Approximation methods
6. Consider the following problem. H 0 ? Let us investigate. but it is sufﬁciently complicated that when we add it to H0 we can no longer ﬁnd the exact energy eigenvalues and eigenstates.
120
(6.
so that e11 = e22 = 0. 2H1 2 . Since.
(6.7) (6. the eigenkets of H 0 form a complete set.5)
Rightmultiplication of Eq. 1H1 2 . (6. (6.9)
In the special (but common) case of a perturbing Hamiltonian whose diagonal matrix elements (in the unperturbed eigenstates) are zero.2 The twostate system
6 APPROXIMATION METHODS
H0 2
= E2 2 .4) by 1 and 2 yields two coupled equations. (6. by deﬁnition. (6.6)
e11 = e22 = e12 =
1H1 1 . are known.8) (6. an Hermitian operator.12)
. (6.11) 2 Let us expand in the supposedly small parameter = e12  .6. (6. The lengths of these eigenkets are both normalized to unity. Let us now try to solve the modiﬁed energy eigenvalue problem (H0 + H1 )E = E E .4)
In fact.
0 = . we can solve this problem exactly.3)
It is assumed that these states. its two eigenkets are orthonormal and form a complete set.10)
the solution of Eq. 0
(6. and their associated eigenvalues. we can write E = 1E 1 + 2E 2 . Since H0 is.
(6.6) (obtained by setting the determinant of the matrix equal to zero) is (E1 + E2 ) ± (E1 − E2 )2 + 4 e12 2 E= . E1 − E2 
121
(6. which can be written in matrix form:
E1 − E + e11 e12 ∗ e12 E2 − E + e22
1E 2E
Here.
(6. It is easily demonstrated that the modiﬁed eigenkets take the form 1 2 = 1 +
∗ e12 2 + · · · .15)
We obtain
E2 = E 2 −
Note that H1 causes the upper eigenvalue to rise. H0 .16) (6.
122
(6.18)
In other words. Note that the series expansion in Eq. E1 − E 2
(6. are denoted H0 n = En n . and the lower eigenvalue to fall.3 Nondegenerate perturbation theory Let us now generalize our perturbation analysis to deal with systems possessing more than two energy eigenstates. E1 − E 2 e12 = 2 − 1 + · · · . when we say that H1 needs to be small compared to H0 . the modiﬁed energy eigenstates consist of one of the unperturbed eigenstates with a slight admixture of the other.13) only converges if 2   < 1.
6. (6.6.13) 2 2 The above expression yields the modiﬁcations to the energy eigenvalues due to the perturbing Hamiltonian: E E1 e12 2 = E1 + + ···.17)
Thus. 2 (6.3 Nondegenerate perturbation theory
6 APPROXIMATION METHODS
1 1 (E1 + E2 ) ± (E1 − E2 )(1 + 2 2 + · · ·). E1 − E 2 (6. This suggests that the condition for the validity of the perturbation expansion is e12  < E1 − E2  .19)
. E1 − E 2 e12 2 + ···. what we really mean is that the above inequality needs to be satisﬁed.14) (6. The energy eigenstates of the unperturbed Hamiltonian.
neglecting terms which are O( 2 ) according to our expansion scheme.
(6.26) and nE mE = 1. We assume that emk  ∼ O( ). for which E = En + O( ). and rightmultiplying by m. we obtain (Em + emm − E) mE +
k=m
emk kE = 0. where emm  ∼ O( ).27) (6. Substituting the above equation into Eq. Em − E k for all m = k.
(6.22)
where emk = mH1 k .28)
for m = n. (6. (6.21)
where the summation is from k = 1 to N.3 Nondegenerate perturbation theory
6 APPROXIMATION METHODS
where n runs from 1 to N. (6.29)
. Suppose that we write out Eq.20)
We can express E as a linear superposition of the unperturbed energy eigenkets. We also assume that (6. We ﬁnd that (Em − En ) mE + emn
123
0.22) for m = n. and have their lengths normalized to unity.
(6. Let us search for a modiﬁed version of the nth unperturbed energy eigenstate. (6. E =
k
kE k . Em (6.25)
for all m. (6. Let us now develop our perturbation expansion. ∼ O( ). The eigenkets n are orthonormal. form a complete set.20).6.23)
1 is our expansion parameter.24) (6. Let us now try to solve the energy eigenvalue problem for the perturbed Hamiltonian: (H0 + H1 )E = E E .
Ek − E n k=n Thus. Em − E n En − E m
giving
(6. The unperturbed Hamiltonian. evaluated for m = n. (6. spherically symmetric shell)] is subjected to a uniform electric ﬁeld in the positive zdirection. H1 = e E z.34)
Thus.
124
(6. we obtain mE − enk 2 (En + enn − E) − = 0. or an alkali metal atom (which possesses one valance electron orbiting outside a closed.22). the modiﬁed eigenkets remain orthonormal and properly normalized to O( 2 ). The Hamiltonian of the system can be split into two parts. p2 + V(r). and neglecting O( 3 ) terms. either a hydrogen atom.e. E − Ek k=n n
(6. (6.33)
(6. En = En + enn + E − Ek k=n n and a eigenket n = n + Note that mn = δmn +
∗ emn enm + + O( 2 ) = δmn + O( 2 ).36)
.35) H0 = 2 me and the perturbing Hamiltonian. (6.4 The quadratic Stark effect
6 APPROXIMATION METHODS
emn ..31)
(6.30) Em − E n Substituting the above expression into Eq.
6. the modiﬁed nth energy eigenstate possesses an eigenvalue enk 2 + O( 3 ).6.32)
ekn k + O( 2 ).4 The quadratic Stark effect Suppose that a oneelectron atom [i.
mzn . m  n. z] = 0. m is. It is necessary to deal with this case separately. l. 6. l . the change in this energy level induced by a small electric ﬁeld is given by ∆Enlm = e E n. l. Enlm − En l m n . An energy eigenket of the unperturbed Hamiltonian is characterized by three quantum numbers—the radial quantum number n. mzn . This implies that the system possesses no degenerate energy eigenvalues. mzn. It is h clear. (6. mzn. 5. m 2 +e E . from the above relation.6. it follows that [Lz . because the perturbation theory presented in Sect. m = 0.3 breaks down for degenerate unperturbed energy levels. an eigenstate of Lz with eigenvalue m ¯ .m =n. Let us now determine the selection rule for l. l .32). According to Eq. l . l. m[Lz . Thus. since Lz = x p y − y p x . l . We have [L2 . l. z]n . z]
125
(6.40)
. (6. by deﬁnition.4 The quadratic Stark effect
6 APPROXIMATION METHODS
It is assumed that the unperturbed energy eigenvalues and eigenstates are completely known. and let its energy level be Enlm . that the matrix element n. l. l. so we can ignore the spin degrees of freedom of the system. giving (m − m ) n.l . z] = [Lx2 . Let us denote such a ket n. m = 0.39) (6.m
2 2
(6. due to the special properties of a pure Coulomb potential. This is termed the selection rule for the quantum number m.41) since n. n.6). m is zero unless m = m. l.l. The electron spin is irrelevant in this problem (since the spin operators all commute with H1 ). z] + [Ly2 .38) (6.37)
Now. and the two angular quantum numbers l and m (see Sect. This is not true for the n = 1 energy levels of the hydrogen atom. m . l.
50)
. h Thus. l. z] + [Ly .1)–(5. m = 0. the above expression expands to give L4 z − 2 L2 z L2 + z L4 − 2 ¯ 2 (L2 z + z L2 ) = 0.6). (5. y] = 2 i ¯ (Lx z − x Lz ).47) (6.3) that Lx x + Ly y + Lz z = 0. h Equation (6. x] = 2 i ¯ (y Lz − Ly z). z] + [Lx . h This reduces to [L2 .44)
(6. h h where use has been made of Eqs. z]Lx + Ly [Ly . [L2 .42)
(6.48)
(6. Ly x − Lx y + i ¯ z h h = 2 i ¯ Ly [L2 . we obtain [L2 . Hence. [L2 .49) implies that n. h Finally. [L2 . mL4 z − 2 L2 z L2 + z L4 − 2 ¯ 2 (L2 z + z L2 )n .45)
(6. z]Ly = i ¯ (−Lx y − y Lx + Ly x + x Ly ) h = 2 i ¯ (Ly x − Lx y + i ¯ z) h h = 2 i ¯ (Ly x − y Lx ) = 2 i ¯ (x Ly − Lx y). it is clear from Eqs. h [L2 . h h = −4 ¯ 2 Ly (y Lz − Ly z) + 4 ¯ 2 Lx (Lx z − x Lz ) h h −2 ¯ 2 (L2 z − z L2 ). [L2 . (5.49)
(6.6. x] − Lx [L2 .46)
(6.43) (6. l . z]] = 2 ¯ 2 (L2 z + z L2 ). [L2 . h
126
(6. z]] = −¯ 2 4 (Lx x + Ly y + Lz z)Lz − 4 (Lx2 + Ly2 + Lz2 ) z h +2 (L2 z − z L2 ) . z] . y] + i ¯ [L2 . z]] = 2 i ¯ L2 .4 The quadratic Stark effect
6 APPROXIMATION METHODS
= Lx [Lx . Similarly.1)–(5. However.
l.52) = 0. θ . (6. mzn . 0.37) which vary linearly with the electric ﬁeldstrength vanish by symmetry. The polarizability of this state is given by α = 2e
2 n>1
 1.54)
Note that all of the terms in Eq. the matrix element n. l . however.. Enlm − En l m l =l±1
(6.55)
Consider the ground state of a hydrogen atom. l. (Recall. In conclusion. θ . Recall. This is the selection rule for the quantum number l. ϕ ) dV . mzn .53) nlm
where ψnlm (r ) = r n. l . m is zero unless l = l ± 1.6. l . 5. (6. It follows from Eq. l. according to the selection rules. l. the matrix element n. (6. (6. m = ψ∗ (r . l.3): i.51)
According to the above formula. l . Only those terms which vary quadratically with the ﬁeldstrength survive.53) that the matrix element vanishes by symmetry when l = l = 0. l. m . ϕ ) r cos θ ψn m l (r . Application of the selection rules to Eq.37) yields ∆Enlm = e2 E2
n
 n. mzn . (6.56)
. 1. This matrix element can be written n. En00 − E100
127
(6.e. mzn . l. m vanishes unless l = l = 0 or l = l ± 1. m = 0. that we cannot address the n > 1 excited states because they are degenerate. that the wavefunction of an l = 0 state is spherically symmetric (see Sect. mzn .4 The quadratic Stark effect
6 APPROXIMATION METHODS
This expression yields l2 (l + 1)2 − 2 l (l + 1) l (l + 1) + l 2 (l + 1)2 −2 l (l + 1) − 2 l (l + 1)] n. 0zn. (6. ψn00 (r ) = ψn00 (r ). mzn . The polarizability of an atom is deﬁned in terms of the energyshift of the atomic state as follows: 1 ∆E = − α E2 . 0 2 . 2 (6. l . l . and our theory cannot handle this at present). m which reduces to (l + l + 2) (l + l ) (l − l + 1) (l − l − 1) n. m 2 .
without recourse to perturbation theory. 0.
n .62) 1. l z1. where 4π 0 ¯ 2 h = 5. 0.6. We can write En00 − E100 ≥ E200 − E100 Thus. 0 . Thus.3 × 10−11 meters a0 = 2 µe is the Bohr radius.64) 4π 0 a03 = 4.
(6. 0. by solving Schr¨dinger’s equation exactly in parabolic coordinates.5 4π 0 a03 .60)
1.63)
9 (6. 0 2 . 0 (6.
(6.59)
16 4π 3
0 a0 n>1
 1. (5. o α=
128
. we conclude that α< The true result is 16 4π 3
3 0 a0
5. 0zn. 0zn. 0 = a02 . 0zn .58)
3 e2 = . 1.4 The quadratic Stark effect
6 APPROXIMATION METHODS
Here. α< However.61)
=
where we have made use of the fact that the wavefunctions of a hydrogen atom form a complete set.120)] e2 En00 = − (6. The sum in the above expression can be evaluated approximately by noting that [see Eq. 0z2 1.3 4π
3 0 a0 . 0z2 1. 0.m
(6. 0.  1. 0. 1. m . l .57) 8π 0 a0 n2 for a hydrogen atom. 4 8π 0 a0
(6. It is easily demonstrated from the actual form of the ground state wavefunction that (6. 2 It is actually possible to obtain this answer. m 1. 0 2 =
n>1 n . 0. we have made use of the fact that En10 = En00 for a hydrogen atom.l . 0.
respectively. l + where en l nl = n ..69)
.33) yields Enl and n. so that each modiﬁed eigenstate n. H0 . we can choose L such that the quantum numbers n and l uniquely specify each eigenstate. Suppose that for each value of n there are Nn different values of l: i. = En + enlnl + E − En n . It is always possible to ﬁnd a sufﬁcient number of operators which commute with the Hamiltonian in order to ensure that the Lnl are all different.l n en l nl n . possess degenerate energy levels. we expect the perturbation to split the degeneracy of the energy levels. Let us denote this operator (or group of operators) L. L does not commute with the perturbing Hamiltonian. It is always possible to represent degenerate energy eigenstates as the simultaneous eigenstates of the Hamiltonian and some other Hermitian operator (or group of operators). l acquires a unique energy eigenvalue Enl . and n and l.l =n. Here.67)
(6. In general. l = Lnl n. the nth energy eigenstate is Nn fold degenerate. l . In other words.6.e. Let us naively attempt to use the standard perturbation theory of Sect. l = n. In this situation. (6.5 Degenerate perturbation theory
6 APPROXIMATION METHODS
6. 6.66) where [H0 .5 Degenerate perturbation theory Let us now consider systems in which the eigenstates of the unperturbed Hamiltonian. and L n.
129
(6. l H1 n. A direct generalization of Eqs.3 to evaluate the modiﬁed energy eigenstates and energy levels.68)
(6. (6.32) and (6.l n
(6. l . E − En n . This implies that the modiﬁed energy eigenstates are not eigenstates of L. l = En n. H1 . L] = 0. l + O( 2 ).65)
en l nl 2 + O( 3 ). We can write H0 n.l =n. the En and the Lnl are real numbers which depend on the quantum numbers n. l .
71)
Note that these new states are also degenerate energy eigenstates of the unperturbed Hamiltonian corresponding to the eigenvalue En . l(1) n. instead of the old ones. of the perturbing Hamiltonian between distinct.67) and (6. In other words.68) directly. l(1) are chosen in such a manner that they are eigenstates of the perturbing Hamiltonian. The problem terms are those involving unperturbed eigenstates labeled by the same value of n.. if n. unperturbed energy eigenstates corresponding to the eigenvalue En were zero. Eq. and have unit lengths. H1 n. (6. that this problem would not exist if the matrix elements. k . however. It follows that (6.67) and (6. then all of the singular terms in Eqs.5 Degenerate perturbation theory
6 APPROXIMATION METHODS
It is fairly obvious that the summations in Eqs. Let us deﬁne Nn new states which are linear combinations of the Nn original degenerate eigenstates corresponding to the eigenvalue En :
Nn
(6. l(1) are also chosen so that they are orthonormal. kn. Note. since all of the singular terms vanish. (6.70)
n. but different values of l: i. l H1 n. then we can employ Eqs.67) and (6. The n. These terms give rise to singular factors 1/(En − En ) in the summations. l(1) . Thus.e.
130
. (6. l (1) H1 n. (6. l(1) = λnl δll .68) would vanish.
(6. (6. H 1 . those states whose unperturbed energies are En .72) The n. In general. enl nl .68) are not wellbehaved if the nth energy level is degenerate.70) is satisﬁed. if we use the new eigenstates. l(1) = λnl n. Thus. l
(1)
=
k=1
n. degenerate. Fortunately.70) is not satisﬁed. (6.6. l = λnl δll . we can always redeﬁne the unperturbed energy eigenstates belonging to the eigenvalue En in such a manner that Eq.73) n. The only remaining difﬁculty is to determine the new eigenstates in terms of the original ones.
l
en l nl n . En − E n
(6.e.68) yield Enl = En + λnl +
n =n. l(1) +
n =n. l(1) . degenerate. l(1) = n. (6. since the summations are over n = n: i. l n. l
l =1
n. l n. (xnl )k = n.5 Degenerate perturbation theory
6 APPROXIMATION METHODS
Now
Nn
n.77) Provided that the determinant of U is nonzero. where the elements of the Nn × Nn Hermitian matrix U are Ujk = n.
l=1
(6.6.. l = 1.79)
and n. Eqs. l H1 n.67) and (6.75)
This can be written more transparently as U x = λ x. they speciﬁcally exclude the problematic.. l(1) = λnl n.78) for k = 1 to Nn . l n.76).
131
.76)
en l nl 2 + O( 3 ).76) can always be solved to give Nn eigenvalues λnl (for l = 1 to Nn ). l + O( 2 ). with Nn corresponding eigenvectors xnl .72) can be transformed into a straightforward matrix eigenvalue equation:
Nn
n. (6. In our new scheme. the operator eigenvalue equation (6. k . Eq.e. (6. jH1 n. Using this completeness relation.74)
where 1 denotes the identity operator in the subspace of all unperturbed energy eigenkets corresponding to the eigenvalue En .80)
There are no singular terms in these expressions.
(6. (6.l
(6. Note that the ﬁrstorder energy shifts are equivalent to the eigenvalues of the matrix equation (6. kn. unperturbed energy eigenstates corresponding to the eigenvalue En . l(1) . En − E n
(6. The eigenvectors specify the weights of the new eigenstates in terms of the original eigenstates: i.
the perturbing Hamiltonian is H1 = e E z. 1. Here.84)
. For instance.6. that the eigenvalues of U are λ1 = 3 e a0 E. 0 0 0 0
0 0 0 0
0 0 0 0
where the rows and columns correspond to the 2. 0 = 3 a0 . 0z2. 0 0 0
2. 1. 1). (6. 0. 0 = 2. 1 . 1. 0 . There is a single l = 0 state.85) x1 = 0 0
132
. which tell us that the matrix element of z between two hydrogen atom states is zero unless the states possess the same m quantum number. (6. −1 states.6 The linear Stark effect Let us examine the effect of an electric ﬁeld on the excited energy levels of a hydrogen atom. (6. 2. λ2 = −3 e a0 E.6 The linear Stark effect
6 APPROXIMATION METHODS
6.4.
e E
U=
0 2.82) where U is the array of the matrix elements of H1 between the degenerate 2s and 2p states. The corresponding eigenvectors are √ 1/√2 1/ 2 . 1. that 2. 0. 0. 1. Thus. and λ4 = 0. 0z2. and 2. by inspection.83)
It can be seen. 0. respectively. consider the n = 2 states. 0 . and three l = 1 states (with m = −1. 6. usually referred to as 2p. λ3 = 0. 2. It is easily demonstrated. All of these states possess the same energy. we have to solve the matrix eigenvalue equation U x = λ x. 1. (6. usually referred to as 2s. from the exact forms of the 2s and 2p wavefunctions. 0z2. 0. and l quantum numbers which differ by unity. 0. As in Sect. 1. 0z2. we have made use of the selection rules.
(6.81)
In order to apply perturbation theory. E200 = −e2 /(32π 0 a0 ).
96)
Thus. 0 √ .92)
In the absence of an electric ﬁeld. E 200 . 2 = 2.
.88)
It follows from Sect. = 2. by an amount 3 e a0 E in the presence of an electric ﬁeld. ∆E4 = 0.87)
x4 =
(6. 1.
133
. respectively. 1.6 The linear Stark effect
6 APPROXIMATION METHODS
x2 =
√ 1/ √ 2 −1/ 2 0 0 0 0 1 0 0 0 0 1
. (6. 1.5 that the simultaneous eigenstates of the unperturbed Hamiltonian and the perturbing Hamiltonian take the form 1 2 3 4 = 2. (6.95) (6. 0 √ = . the energies of states 1 and 2 are shifted upwards and downwards. all of these states possess the same energy.94) (6.90) (6. 1 . 1. States 1 and 2 are orthogonal linear combinations of the original 2s and 2p(m = 0) states. ∆E3 = 0. 2 2. 0.93) (6. −1 .89) (6. 0 + 2.86)
x3 =
(6. 0 − 2.6. The ﬁrstorder energy shifts induced by an electric ﬁeld are given by ∆E1 = +3 e a0 E. 6.
(6. . ∆E2 = −3 e a0 E. 0.91) (6.
Note that the linear Stark effect depends crucially on the degeneracy of the 2s and 2p states. an electron possesses a spin magnetic moment [see Eq. alkali metal atoms do not exhibit the linear Stark effect. e (6.97)
Now.170)] µ=− eS .6. higher l states have higher energies. Of course.e. Thus. and. 6. (5.7 Fine structure Let us now consider the energy levels of hydrogenlike atoms (i. only applies to a hydrogen atom. The outermost electron moves in a spherically symmetric potential V(r) due to the nuclear charge and the charges of the other electrons (which occupy spherically symmetric closed shells). This splits the degeneracy of states characterized by the same value of n.
6. This degeneracy is a special property of a pure Coulomb potential.99)
134
. so this is a much larger effect that the quadratic effect described in Sect. but different values of l. a charge moving in an electric ﬁeld also experiences an effective magnetic ﬁeld B = −v × E. me
However. (6. This electron experiences an electric ﬁeld E= V . The energies of states 3 and 4 (which are equivalent to the original 2p(m = 1) and 2p(m = −1) states. respectively) are not affected to ﬁrstorder. therefore. In fact. Let us examine a phenomenon known as ﬁne structure. The shielding effect of the inner electrons causes V(r) to depart from the pure Coulomb form. which is due to interaction between the spin and orbital angular momenta of the outermost electron. alkali metal atoms) in more detail.4.98) (6. to secondorder the energies of these states are shifted by an amount which depends on the square of the electric ﬁeldstrength.7 Fine structure
6 APPROXIMATION METHODS
Note that the energy shifts are linear in the electric ﬁeldstrength..
) Moreover. S2 .7 Fine structure
6 APPROXIMATION METHODS
We. Now. When the above expression is compared to the observed spinorbit interaction. Let us now apply perturbation theory to a hydrogenlike atom. different l. and Jz . and p2 + V(r) H0 = 2 me (6. or the simultaneous eigenstates of H0 . L·S = 2
135
(6. We know. therefore. where J = L + S is the total angular momentum.. is proportional to L·S. respectively. but different values of ml . There is a classical explanation for this. Lz and Sz . Although the departure of V(r) from a pure 1/r form splits the degeneracy of same n. it is found to be too large by a factor of two. L2 . where J 2 − L2 − S 2 . ms .e. which we need not go into. (Here. ms = ±1/2). each state is doubly degenerate due to the two possible orientations of the electron spin (i. ml .6. 6. and mj are the quantum numbers corresponding to Lz . the perturbing Hamiltonian.6. due to spin precession. are still degenerate. = me2 r dr
(6. S 2 . L2 .101)
as the unperturbed Hamiltonian. that the application of perturbation theory to a degenerate system is greatly simpliﬁed if the basis eigenstates of the unperturbed Hamiltonian are also eigenstates of the perturbing Hamiltonian. expect a spinorbit contribution to the Hamiltonian of the form HLS = −µ·B 1 r dV eS ·v× = − me e r dr 1 dV L·S. The correct quantum mechanical explanation requires a relativistically covariant treatment of electron dynamics (this is achieved using the socalled Dirac equation). We can adopt the simultaneous eigenstates of H0 . J 2 . we are still dealing with a highly degenerate system. HLS . using H LS as the perturbation (with HLS taking one half of the value given above). from Sect. We have two choices for the energy eigenstates of H0 . those states characterized by the same values of n and l. states. with the addition of spin degrees of freedom. Sz .102)
. Thus.100)
where L = me r × v is the orbital angular momentum. and Jz .
276). and Jz .282). m1 = m − 1/2. We have also made use of the fact that both the j. L2 . j = l + 1/2. mj kets. rather than those of the ﬁrst group. m = cos α m − 1/2.106) which reduces to m − 1/2.103) (6. j2 = 1/2. J 2 . 1/2l + 1/2. and Jz ) does. m l − 1/2. Thus. the allowed values of the total angular momentum are j = l + 1/2 and j = l − 1/2. (6. 1/2 + sin α m + 1/2. −1/2 . (5. where the ClebschGordon coefﬁcient is written in ml . mj and ml . Lz and Sz ) does not commute with HLS . mj form.105)
. We can write l + 1/2. (6. ms kets (the j1 . Eq. for the sake of clarity). S2 . Let us now employ the recursion relation for ClebschGordon coefﬁcients. According to Eq. S 2 . 1/2l + 1/2. 1/2l + 1/2. and have unit lengths. m + 1 . (6. ms j. with j1 = l. 1/2l + 1/2. m . This is equivalent to ﬁnding the eigenstates of the total angular momentum resulting from the addition of two angular momenta: j1 = l. We obtain (l + 1/2) (l + 3/2) − m (m + 1) m − 1/2.107) l + m + 3/2
136
(6. j2 labels have been dropped. = − sin α m − 1/2. L2 . m2 = 1/2 (lower sign). whereas those on the righthand side are ml . S 2 . L·S is just a combination of operators appearing in the second group. and j2 = s = 1/2. J 2 . m = l (l + 1) − (m − 1/2) (m + 1/2) m + 1/2.6. it is advantageous to work in terms of the eigenstates of the second group of operators. We now need to determine cos α = m − 1/2. We now need to ﬁnd the simultaneous eigenstates of H0 . whereas the second group (H0 . the kets on the lefthand side are j. L2 . 1/2 + cos α m + 1/2.104)
Here. In fact. ms kets are orthonormal. −1/2 . m = l + m + 1/2 m + 1/2. 1/2l + 1/2. m + 1 . We have made use of the fact that the ClebschGordon coefﬁcients are automatically zero unless mj = ml + ms .7 Fine structure
6 APPROXIMATION METHODS
It is fairly obvious that the ﬁrst group of operators (H0 . (5.
2l + 1
137
(6. It follows from Eq. 1/2 must be equal to the j. (6. By convention. sin2 α = 1 − l + m + 1/2 l − m + 1/2 = . l and 1/2. m = + l + m + 1/2 m − 1/2.110)
We now need to determine the sign of sin α. Thus. 1/2l + 1/2. 2l + 1 (6. The corresponding value of mj is l + 1/2. m = Now. −1/2 . respectively. this factor is taken to be unity. m = l + m + 1/2 l + m + 3/2 l + m + 3/2 l + m + 5/2 × m + 3/2. A careful examination of the recursion relation. but not when j = l − 1/2.6. 1/2l + 1/2. This value is possible when j = l + 1/2. 1/2l + 1/2. ms ket l. l + 1/2 . Thus. shows that the plus sign is appropriate. l + 1/2 = 1.113)
. 1/2l + 1/2. 2l + 1 (6. mj ket l + 1/2. l + 1/2 . 1/2l + 1/2. Eq.112) l + m + 1/2 .109) that cos α = m − 1/2. 2l + 1 2l + 1 (6. the ml . up to an arbitrary phasefactor.109)
Consider the situation in which ml and m both take their maximum values. For instance. (5. l. m − 1/2.111) (6. m = l + m + 1/2 l. Thus.282). m + 2 . m − 1/2.108) This procedure can be continued until ml attains its maximum possible value. (6. giving l.7 Fine structure
6 APPROXIMATION METHODS
We can use this formula to successively increase the value of ml . 1/2l + 1/2. l + 1/2. 1/2 2l + 1 l − m + 1/2 m + 1/2.
e. which is the quantum number associated with Jz . For a given choice of l. the quantum number j (i. m
= − +
l − m + 1/2 m − 1/2. Thus.117)
(6. ¯2 h L·S j = l ± 1/2. A general wavefunction for an energy eigenstate in a hydrogenlike atom is written The radial part of the wavefunction. = 2 (l + 1) ¯ 2 h l − 1/2. ϕ) χ+ 2l + 1 l m + 1/2 m+1/2 Yl (θ. m . (6.116)
The l ± 1/2. Rnl (r). just as the spherical harmonics are eigenfunctions of the orbital angular momentum. The wavefunction is also labeled by m. ϕ) χ− 2l + 1
m−1/2 . ϕ)
(6. 2 giving L·S l + 1/2. m .115)
ψnlm± = Rnl (r) Y j=l±1/2. m L·S l − 1/2. (6. according to Eq. = − 2
138
(θ.119)
. m kets are eigenstates of L·S. (6.. −1/2 .6. m l¯2 h l + 1/2.
These functions are eigenfunctions of the total angular momentum for spin onehalf particles.7 Fine structure
6 APPROXIMATION METHODS
l − 1/2.102).118) (6. 1/2 2l + 1
l + m + 1/2 m + 1/2. ϕ) 1 ± l ± m + 1/2 Yl = √ m+1/2 2l + 1 l m + 1/2 Yl (θ. the quantum number associated with J 2 ) can take the values l ± 1/2. m . depends on the radial quantum number n and the angular quantum number l.m . mj = m = [j (j + 1) − l (l + 1) − 3/4] j.
(6.m
= ± +
l ± m + 1/2 m−1/2 Yl (θ.114) 2l + 1 It is convenient to deﬁne so called spinangular functions using the Pauli twocomponent formalism: Yl
j=l±1/2.
r dr r dr Equations (6.121)
l−1/2. the ground state is written (1s)2 (2s)2 (2p)6 (3s).121) yield ∆Enlm+ ∆Enlm− where 1 dV l ¯ 2 h 1 .123) (6. (6. To ﬁrstorder. 2 me2 r dr 2 (6. We are interested in the excitation of the eleventh electron from 3s to some higher energy state.116).100) (remember the factor of two). 2
(6.6. (6.m †
) L·S Y ) L·S Y
l+1/2.123)–(6.120) (6.m †
l−1/2. (6. and (6.124)
1 dV 1 dV = (Rnl )∗ Rnl r2 dr. Let us now apply degenerate perturbation theory to evaluate the shift in energy of a state whose wavefunction is ψnlm± due to the spinorbit Hamiltonian HLS . the energyshift is given by ∆Enlm± = (ψnlm± )† HLS ψnlm± dV.m
where the integrals are over all solid angle.126)
The inner ten electrons effectively form a spherically symmetric electron cloud.m
l¯2 h . In chemist’s notation. Equations (6. This state has a higher energy than 3s due to the deviations of the potential from the pure
139
.120)–(6. = + 2 me2 r dr 2 1 1 dV (l + 1) ¯ 2 h = − .7 Fine structure
6 APPROXIMATION METHODS
It follows that (Y (Y
l+1/2.
(6.124) are known as Lande’s interval rule.122)
where the integral is over all space.125)
Let us now apply the above result to the case of a sodium atom. dΩ = 2 (l + 1) ¯ 2 h dΩ = − . The closest (in energy) unoccupied state is 3p.
the relativistic mass correction of the electron) which is the same order of magnitude as spinorbit coupling. where the subscript refers to the value of j. The change in energy of the outermost electron is HB = −µ·B. (6.125) is positive.130)
Suppose that the energyshifts induced by the magnetic ﬁeld are much smaller than those induced by spinorbit interaction. In the absence of spinorbit interaction.
6. The fact that there are two slightly different 3p energy levels (note that spinorbit coupling does not split the 3s energy levels) means that the sodium D line actually consists of two very closely spaced spectroscopic lines. since we have neglected an effect (namely.127)
is the ﬁne structure constant.124) are not entirely correct. Note that Eqs.8 The Zeeman effect
6 APPROXIMATION METHODS
Coulomb form. because the radial integral (6. The spinorbit interaction breaks the degeneracy of these states. The wellknown sodium D line is associated with transitions between the 3p and 3s states. where µ=− (6. HB = eB (Lz + 2 Sz ).129) 2 me is its magnetic moment. including both the spin and orbital contributions. The four (3p)3/2 states lie at a slightly higher energy level than the two (3p)1/2 states. In this situation. Thus. The splitting of the (3p) energy levels of the sodium atom can be observed using a spectroscope. where 1 e2 = α= 2 0 h c 137 (6. It is easily demonstrated that the ratio of the typical spacing of Balmer lines to the splitting brought about by spinorbit interaction is about 1 : α2 .8 The Zeeman effect Consider a hydrogenlike atom placed in a uniform zdirected magnetic ﬁeld. there are six degenerate 3p states.6. The modiﬁed states are labeled (3p)1/2 and (3p)3/2 . 2 me (6.123)–(6. we can treat H B as
140
.128)
e (L + 2 S) (6.
From standard perturbation theory.6. mSz l ± 1/2. a magnetic ﬁeld splits the (3p)3/2
141
. m ) . 1/2 2l + 1 l m + 1/2 m + 1/2. = ± 2l + 1 m + 1/2)] (6. (6.133)
Now. m . these states are the simultaneous eigenstates of J 2 and Jz .136). (6.114).8 The Zeeman effect
6 APPROXIMATION METHODS
Since
a small perturbation acting on the eigenstates of H0 + HLS . (6. the ﬁrstorder energyshift in the presence of a magnetic ﬁeld is ∆Enlm± = l ± 1/2.134)
Thus. we obtain Lande’s formula for the energyshift induced by a weak magnetic ﬁeld: e¯ B h 1 ∆Enlm± = m 1± .132)
we ﬁnd that ∆Enlm± = eB (m ¯ + l ± 1/2. We have already seen that the nonCoulomb potential splits the degeneracy of the 3s and 3p states. h 2 me (6. m = ¯ h [(l ± m + 1/2) − (l 2 (2 l + 1) m¯ h . from Eqs. 2l + 1 (6. Of course. with four j = 3/2 states lying at a slightly higher energy than two j = 1/2 states. mSz l ± 1/2. The spinorbit interaction splits the six 3p states into two groups. mHB l ± 1/2. According to Eq. where j = l ± 1/2.131) Lz + 2 S z = Jz + Sz . labeled by the quantum numbers j and m.113)–(6.136) 2 me 2l + 1 Let us apply this theory to the sodium atom. the latter states acquiring a higher energy. −1/2 . Let us consider one of these states. l ± 1/2. (6.135) l ± m + 1/2 m − 1/2. (6. m = ± + It follows that l ± 1/2.
(6. states with different values of ml + 2 ms acquire different energies. HB . S 2 . the six 3p states form two groups of four and two states. Lz . Let us apply this result to a sodium atom. in this situation. ml . Thus. but different values of ml and ms . l.6. ml . S 2 . but does not commute with L2 . l. the energy of each state becomes dependent on the quantum number m. depending on the values
142
. each state acquiring a different energy. which measures the projection of the total angular momentum along the zaxis. In the absence of a magnetic ﬁeld. In fact. the energy eigenstates of a hydrogenlike atom are approximate eigenstates of the spin and orbital angular momenta. our energy eigenkets are written n. H LS . it is evident from Eq. Note that the magnetic Hamiltonian. H0 . Thus. are degenerate. l. ms e¯ B h (ml + 2 ms ). In the PaschenBack limit we can think of the spinorbit Hamiltonian. Suppose that we increase the strength of the magnetic ﬁeld. and ms . in an intense magnetic ﬁeld. This is called the PaschenBack limit. J 2 . as a small interaction term operating on the eigenstates of H0 + HB . causes states with different values of the quantum numbers n and l to have different energies. but are not eigenstates of the total angular momentum.137)
Thus. Clearly. The unperturbed Hamiltonian.136) that these states are split by a lesser amount than the j = 3/2 states. The shift in energy due to the magnetic ﬁeld is simply ∆Enlml ms = n. However. Let us consider the extreme limit in which the energyshift due to the magnetic ﬁeld greatly exceeds that induced by spinorbit effects. We can label each state by the quantum numbers n (the energy quantum number). ml . In fact. States with higher m values have higher energies. it does not make much sense to think of HB as a small interaction term operating on the eigenstates of H0 + HLS . A magnetic ﬁeld also splits the (3p)1/2 doublet of states. so that the energyshift due to the magnetic ﬁeld becomes comparable to the energyshift induced by spinorbit interaction. ms HB n. Sz . ms . Jz . l. states with the same value of n and l.8 The Zeeman effect
6 APPROXIMATION METHODS
quadruplet of states. ml . this intermediate case is very difﬁcult to analyze. However. = 2 me
(6. commutes with L2 .
143) BPB ∼ α2 0 h a0
143
(6.138) (6. and a state with ml + 2 ms = −2. 1/2). In the presence of an intense magnetic ﬁeld. ¯ 2 ml ms 1 dV h ∆Enlml ms = . in order of decreasing energy. (0. L·S = Lz Sz + (L+ S− + L− S+ )/2 (6. 1/2). respectively. h The energyshift induced by the spinorbit Hamiltonian is given by ∆Enl ml ms = n. −1/2). The net result is that the ﬁve groups of states are no longer equally spaced in energy.6. the energy of the doublet. The sort of magnetic ﬁeldstrength needed to get into the PaschenBach limit is given by e me 25 tesla. In the presence of an intense magnetic ﬁeld the 3p states are split into ﬁve groups. the energy difference between adjacent groups being e ¯ B/2 me . ml . ms ) quantum numbers (1. There is a state with ml + 2 ms = 2. where HLS = Now. and (−1. does not affect the next lowest energy state. l. but does not split. (0.139)
1 1 dV L·S. −1/2). does not affect the next highest energy state. and increases the energy of the lowest energy state. a state with ml + 2 ms = −1. (6. Thus. The spinorbit term increases the energy of the highest energy state. (1. ms HLS n. two states with ml + 2 ms = 0.140) (6.142) 2 me2 r dr Let us apply the above result to a sodium atom.141) = ¯ 2 ml ms . decreases. the 3p states are split into ﬁve groups with (ml . h since L± = S ± = 0 for expectation values taken between the simultaneous eigenkets of Lz and Sz .8 The Zeeman effect
6 APPROXIMATION METHODS
of their total angular momentum. These groups are equally spaced in energy. or (−1. −1/2). 1/2). (6. a state with ml + 2 ms = 1. ms . ml . 2 me2 r dr
. l.
6.9 Timedependent perturbation theory
6 APPROXIMATION METHODS
Obviuously, this is an extremely large ﬁeldstrength.
6.9 Timedependent perturbation theory Suppose that the Hamiltonian of the system under consideration can be written H = H0 + H1 (t), (6.144)
where H0 does not contain time explicitly, and H1 is a small timedependent perturbation. It is assumed that we are able to calculate the eigenkets of the unperturbed Hamiltonian: H0 n = En n . (6.145) We know that if the system is in one of the eigenstates of H0 then, in the absence of the external perturbation, it remains in this state for ever. However, the presence of a small timedependent perturbation can, in principle, give rise to a ﬁnite probability that a system initially in some eigenstate i of the unperturbed Hamiltonian is found in some other eigenstate at a subsequent time (since i is no longer an exact eigenstate of the total Hamiltonian), In other words, a timedependent perturbation causes the system to make transitions between its unperturbed energy eigenstates. Let us investigate this effect. Suppose that at t = t0 the state of the system is represented by A =
n
cn n ,
(6.146)
where the cn are complex numbers. Thus, the initial state is some linear superposition of the unperturbed energy eigenstates. In the absence of the timedependent perturbation, the time evolution of the system is given by A, t0 , t =
n
cn exp([−i En (t − t0 )/¯ ] n . h
(6.147)
Now, the probability of ﬁnding the system in state n at time t is Pn (t) = cn exp[−i En (t − t0 )/¯ ]2 = cn 2 = Pn (t0 ). h
144
(6.148)
6.9 Timedependent perturbation theory
6 APPROXIMATION METHODS
Clearly, with H1 = 0, the probability of ﬁnding the system in state n at time t is exactly the same as the probability of ﬁnding the system in this state at the initial time t0 . However, with H1 = 0, we expect Pn (t) to vary with time. Thus, we can write A, t0 , t = cn (t) exp[−i En (t − t0 )/¯ ] n , h (6.149)
n
where Pn (t) = cn (t) . Here, we have carefully separated the fast phase oscillation of the eigenkets, which depends on the unperturbed Hamiltonian, from the slow variation of the amplitudes cn (t), which depends entirely on the perturbation (i.e., cn is constant if H1 = 0). Note that in Eq. (6.149) the eigenkets n are timeindependent (they are actually the eigenkets of H0 evaluated at the time t0 ). Schr¨dinger’s time evolution equation yields o ∂ A, t0 , t = H A, t0 , t = (H0 + H1 ) A, t0 , t . ∂t It follows from Eq. (6.149) that i¯ h (H0 + H1 )A, t0 , t =
m
2
(6.150)
cm (t) exp[−i Em (t − t0 )/¯ ] (Em + H1 ) m . h
(6.151)
We also have ∂ i ¯ A, t0 , t = h ∂t
i¯ h
m
dcm + cm (t) Em exp[−i Em (t − t0 )/¯ ] m , h dt
(6.152)
where use has been made of the timeindependence of the kets m . According to Eq. (6.150), we can equate the righthand sides of the previous two equations to obtain dcm exp[−i Em (t − t0 )/¯ ]m = h cm (t) exp[−i Em (t − t0 )/¯ ] H1 m . h i¯ h dt m m (6.153) Leftmultiplication by n yields i¯ h where Hnm (t) = nH1 (t)m ,
145
dcn = dt
Hnm (t) exp[i ωnm (t − t0 )] cm (t),
m
(6.154)
(6.155)
6.10 The twostate system
6 APPROXIMATION METHODS
En − E m . (6.156) ¯ h Here, we have made use of the standard orthonormality result, nm = δ nm . Suppose that there are N linearly independent eigenkets of the unperturbed Hamiltonian. According to Eq. (6.154), the time variation of the coefﬁcients cn , which specify the probability of ﬁnding the system in state n at time t, is determined by N coupled ﬁrstorder differential equations. Note that Eq. (6.154) is exact—we have made no approximations at this stage. Unfortunately, we cannot generally ﬁnd exact solutions to this equation, so we have to obtain approximate solutions via suitable expansions in small quantities. However, for the particularly simple case of a twostate system (i.e., N = 2), it is actually possible to solve Eq. (6.154) without approximation. This solution is of enormous practical importance. ωnm =
and
6.10 The twostate system Consider a system in which the timeindependent Hamiltonian possesses two eigenstates, denoted H0 1 H0 2 = E1 1 , = E2 2 . (6.157) (6.158)
Suppose, for the sake of simplicity, that the diagonal matrix elements of the interaction Hamiltonian, H1 , are zero: 1H1 1 = 2H1 2 = 0. (6.159)
The offdiagonal matrix elements are assumed to oscillate sinusoidally at some frequency ω: 1H1 2 = 2H1 1 ∗ = γ exp(i ω t), (6.160) where γ and ω are real. Note that it is only the offdiagonal matrix elements which give rise to the effect which we are interested in—namely, transitions between states 1 and 2.
146
162)
where ω21 = (E2 − E1 )/¯ . and assuming that t0 = 0.6.165)
Now.161) and h (6.161) (6. Let us look for a solution in which the system is certain to be in state 1 at time t = 0.10 The twostate system
6 APPROXIMATION METHODS
For a twostate system. the probability of ﬁnding the system in state 1 at time t is simply P 1 (t) = c1 2 . It follows that γ2 /¯ 2 h P2 (t) = 2 2 γ /¯ + (ω − ω21 )2 /4 h
147
. dt dc2 = γ exp[−i (ω − ω21 ) t ] c1 . dt2 dt ¯ h (6. our boundary conditions are c1 (0) = 1 and c2 (0) = 0.164)
γ2 /¯ 2 + (ω − ω21 )2 /4 t . (6. Equations (6. h
(6. i¯ h dt i¯ h (6. It is easily demonstrated that the appropriate solutions are c2 (t) = −i γ/¯ h γ2 /¯ 2 + (ω − ω21 )2 /4 h × sin exp[−i (ω − ω21 ) t/2] (6. we can use Eq. Eq. h γ2 /¯ 2 + (ω − ω21 )2 /4 t h exp[ i (ω − ω21 ) t/2]
c1 (t) = exp[ i (ω − ω21 ) t/2] cos − i (ω − ω21 )/2 γ2 /¯ 2 + (ω − ω21 )2 /4 h
× sin
γ2 /¯ 2 + (ω − ω21 )2 /4 t .162) can be combined to give a secondorder differential equation for the time variation of the amplitude c2 : dc2 γ2 d2 c 2 + i (ω − ω21 ) + 2 c2 = 0. Likewise.162) to obtain the amplitude c1 . (6.154) reduces to dc1 = γ exp[+i (ω − ω21 ) t ] c2 .163)
Once we have solved for c2 . Thus. the probability of ﬁnding the system in state 2 at time t is P 2 (t) = c2 2 .
In fact. we obtain a resonance curve whose maximum (unity) lies at the resonance.167)
P1 (t) = 1 − P2 (t). The absorptionemission cycle also take place away from the resonance. nor is the minimum value of P1 (t) zero.169)
According to the above result. when the oscillation frequency of the perturbation.e.
Equation (6. we ﬁnd that P1 (t) = cos2 (γ t/¯ ). the narrower the resonance. the smaller γ becomes).
148
. the timedependent perturbation is only effective at causing transitions between states 1 and 2 if its frequency of oscillation lies in the approximate range ω21 ± 2 γ/¯ . the weaker the h perturbation (i. the amplitude of oscillation of the coefﬁcient c2 is reduced. ω. matches the frequency ω21 . After a further time interval h π ¯ /2 γ it is certain to be in state 1. This means that the maximum value of P2 (t) is no longer unity. h (6. Clearly. However. This implies that the system alternatively absorbs and emits energy from the source of the perturbation. the system periodically h ﬂipﬂops between states 1 and 2 under the inﬂuence of the timedependent perturbation. Thus. h P2 (t) = sin2 (γ t/¯ ).168) (6. and so on.. At resonance. and whose fullwidth halfmaximum (in frequency) is 4 γ/¯ . when ω = ω21 . Thus.6. After a time interval π ¯ /2 γ it is certain to be in state 2. ω. h
(6. This result is known as Rabi’s formula. the system starts off at t = 0 in state 1.166) exhibits all the features of a classic resonance. if the applied frequency differs from the resonant frequency by h substantially more than 2 γ/¯ then the probability of the system jumping from h state 1 to state 2 is very small.10 The twostate system
6 APPROXIMATION METHODS
× sin2
γ2 /¯ 2 + (ω − ω21 )2 /4 t .166) (6. if we plot the maximum value of P2 (t) as a function of the applied frequency. In other words.
The rotating magnetic ﬁeld usually represents the magnetic component of an electromagnetic wave propagating along the zaxis.g. me (6.6.174)
The timedependent Hamiltonian can be written H1 = e B1 exp( i ωt) S− + exp(−i ωt) S+ . denoted + and − . The Hamiltonian is written H = −µ·B = H0 + H1 . and then subjected to a small timedependent magnetic ﬁeld rotating in the xy plane.171) (6.173)
The eigenstates of the unperturbed Hamiltonian are the ‘spin up’ and ‘spin down’ states. respectively.170)
where B0 and B1 are constants.175)
where S+ and S− are the conventional raising and lowering operators for the spin angular momentum.172)
e B1 (cos ωt Sx + sin ωt Sy ) .177)
=
e ¯ B1 h exp( i ωt). Thus. me
(6.. H0 ± = ± e ¯ B0 h ± . 2 me (6. with B1 B0 . 2 me
149
. Thus. where H0 = and H1 = e B0 Sz . and −H1 + = +H1 −
∗
(6. (6. It follows that +H1 + = −H1 − = 0. ^ ^ z B = B0 ^ + B1 (cos ωt x + sin ωt y). In this system. a bound electron) placed in a uniform zdirected magnetic ﬁeld.11 Spin magnetic resonance
6 APPROXIMATION METHODS
6.11 Spin magnetic resonance Consider a spin onehalf system (e.176) (6. 2 me (6. the electric component of the wave has no effect.
the expectation values of Sx and Sy oscillate because of the spin precession.180) (6.12 The Dyson series Let us now try to ﬁnd approximate solutions of Eq. (6. ω21 → me e ¯ B1 h γ → . in addition to the spin precession. provided that we make the identiﬁcations 1 2 → − . t) A . e B0 . t).e. it is possible to calculate the magnetic moment.12 The Dyson series
6 APPROXIMATION METHODS
It can be seen that this system is exactly the same as the twostate system discussed in the previous section. and other spin onehalf particles. It is convenient to work in terms of the time evolution operator. In the absence of the perturbation. + − . t0 . 2 me
6. according to the analysis of the previous section. and subjecting them to an oscillating magnetic ﬁeld whose frequency is gradually scanned. ω21 . t = U(t0 .179) (6. We also know that if the oscillation frequency of the applied ﬁeld is very different from the resonant frequency then there is virtually zero probability of the ﬁeld triggering a spinﬂop. Experimentalist are able to measure the magnetic moments of electrons. The width of the resonance (in frequency) is determined by the strength of the oscillating magnetic perturbation. the frequency at which the particles absorb energy from the oscillating ﬁeld). but the expectation value of Sz remains invariant. U(t 0 . By determining the resonant frequency (i. is simply the spin precession frequency for an electron in a uniform magnetic ﬁeld of strength B0 . If we now apply a magnetic perturbation rotating at the resonant frequency then. (6.178) (6. which is deﬁned A. to a high degree of accuracy by placing the particles in a magnetic ﬁeld.154) for a general system.
→ + .182)
150
.181)
The resonant frequency..6. the system undergoes a succession of spinﬂops. (6.
∂t (6. h It is readily demonstrated that UI satisﬁes the differential equation i¯ h where HI (t0 . (6. t = cm (t) exp[−i Em (t − t0 )/¯ ] m .184)
In the absence of the external perturbation. t) = exp[−i H0 (t − t0 )/¯ ].6.183)
(6.189) (6. t0 .187) (6. we would expect UI to contain all of the information regarding transitions between different eigenstates of H 0 caused by the perturbation. h (6. t) = (H0 + H1 ) U(t0 . (6. (6. t0 . t) = exp[+i H0 (t − t0 )/¯ ] H1 exp[−i H0 (t − t0 )/¯ ]. A.186)
Note that UI speciﬁes that component of the time evolution operator which is due to the timedependent perturbation. t) = exp[−i H0 (t − t0 )/¯ ] UI (t0 . h (6.188) ∂UI (t0 . t) UI (t0 . t). t0 ) = 1.12 The Dyson series
6 APPROXIMATION METHODS
Here. i. t0 ) = 1. the time evolution operator reduces to U(t0 . t is the state ket of the system at time t. The subsequent evolution of the state ket is given by Eq.149). t) = HI (t0 .185) Let us switch on the perturbation and look for a solution of the form U(t0 .190)
m
151
. Suppose that the system starts off at time t0 in the eigenstate i of the unperturbed Hamiltonian. t). h h subject to the boundary condition UI (t0 . given that the state ket at the initial time t0 is A . It is easily seen that the time evolution operator satisﬁes the differential equation ∂U(t0 . ∂t subject to the boundary condition i¯ h U(t0 . Thus. t).
is simply Pi→n (t0 .192) and (6. Thus. (6.197)
.187). Note that the differential equation (6. n n n (6. t ) HI (t0 . given that it is deﬁnitely in state i at time t 0 . (6.191)
HI (t0 . (6. t = exp[−i H0 (t − t0 )/¯ ] UI (t0 . t0 . t ) dt + · · · . i UI (t0 . etc.
(6. t) i 1− ¯ h i 1− ¯ h −i + ¯ h i HI (t0 . t) i .192) where use has been made of nm = δnm . the probability that the system is found in state n at time t. we also have i. t) =  nUI (t0 . are equivalent to the following integral equation. t)i 2 .
t0
(6.12 The Dyson series
6 APPROXIMATION METHODS
However. t ) dt . It follows from Eqs.6. h It follows that cn (t) = nUI (t0 .193) This quantity is usually termed the transition probability between states i and n . t ) UI (t0 .194)
We can obtain an approximate solution to this equation by iteration: UI (t0 . t)i . t ) 1 − ¯ h t0
t t t t0
HI (t0 . t ) dt
HI (t0 . n
152
(6.196)
where the superscript (1) refers to a ﬁrstorder term in the expansion. Let cn = c(0) + c(1) + c(2) + · · · .195)
This expansion is known as the Dyson series. t ) dt
t0 2 t t
dt
t0 t0
HI (t0 . t) = 1 − ¯ h
t
(6.195) that c(0) (t) = δin . t ) UI (t0 . plus the boundary condition (6.189).
t0 2 m t t
(6. t ) HI (t0 . However. t )i dt .
t0 2 t t
(6. n n n En − E m . ¯ h
(6.205)
According to the above analysis.e.201)
(2) cn (t)
dt
t0 t0
dt exp[ i ωnm (t − t0 )] (6.
These expressions simplify to
(0) cn (t) = δin .12 The Dyson series
6 APPROXIMATION METHODS
c(1) (t) n c(2) (t) n
i = − ¯ h = −i ¯ h
t
nHI (t0 . if the matrix element is zero. there is no chance of a transition between states i and n (i = n) to zerothorder (i.
153
. the transition probability is proportional to the time integral of the matrix element nH1 i . weighted by some oscillatory phasefactor. in the absence of the perturbation). t) = c(0) + c(1) + c(2) + · · · 2 .200)
t
i = − ¯ h = −i ¯ h
exp[ i ωni (t − t0 )] Hni (t ) dt .202)
×Hnm (t ) exp[ i ωmi (t − t0 )] Hmi (t ). then there is no chance of a ﬁrstorder transition between states i and n .199)
dt
t0 t0
nHI (t0 . where ωnm = and Hnm (t) = nH1 (t)m .204)
(6. t )i dt . (1) cn (t)
(6.198) (6.. The transition probability between states i and n is simply Pi→n (t0 . to secondorder. To ﬁrstorder.6. a transition between states i and n is possible even when the matrix element nH1 i is zero. Thus.203) (6.
It is a good approximation to say that sinc(x) is small except when x < π. ∼ h
154
where
(6.208)
giving (En − Ei ) t 4 Hni 2 (1) 2 sin2 . but is generally a function of the position.209) Pi→n (t) cn  = En − Ei 2 2¯ h for i = n. (6.211) x The sinc function is highly oscillatory.206)
where H1 is timeindependent.207)
(6. (6. The transition probability between states i and n can be written Hni 2 t2 (En − Ei ) t Pi→n (t) = . c(0) (t) = δin . n c(1) (t) n
t i = − Hni exp[ i ωni (t − t)] dt ¯ h 0 Hni = [1 − exp( i ωni t)].13 Constant perturbations
6 APPROXIMATION METHODS
6. It follows ∼ that the transition probability.200)–(6.13 Constant perturbations Consider a constant perturbation which is suddenly switched on at time t = 0: H1 (t) = 0 H1 (t) = H1 for t < 0 for t ≥ 0. is is possible to have transitions which do not exactly conserve energy.e. En = Ei ) have an appreciable probability of occurrence. provided that sinc(x) = ∆E ∆t < ¯ . Pi→n . momentum. At ﬁnite t. is small except when 2π ¯ h . En − E i
(6. (6. and spin operators. (6.6.202) (with t0 = 0). According to Eqs. and decays like 1/x at large x. (6..212) En − Ei  < ∼ t Note that in the limit t → ∞ only those transitions which conserve energy (i. sinc2 2¯ h ¯2 h
(6. Suppose that the system is deﬁnitely in state i at time t = 0.210)
sin x .213)
.
t to t + dt.210). (6.6. This uncertainty relation is fundamentally different to the positionmomentum uncertainty relation. h (6.
Ei
(6.. the probability of a transition from the initial state i to any of the continuum of possible ﬁnal states is Pi→ (t) = giving Pi→ (t) = where x = (En − Ei ) t/2 ¯ .
(6. all possessing nearly the same energy as the energy of the initial state i . In practice. E n = Ei ) is Hin 2 t2 Pi→n (t) = . grows linearly with t.218)
. ¯ h
(6. this result is just a manifestation of the wellknown uncertainty relation for energy and time. there is usually a group of ﬁnal states.217) and use has been made of Eq. Note that this probability grows quadratically with time. Clearly. Thus. since in nonrelativistic quantum mechanics position and momentum are operators. The probability of a transition which conserves energy (i. we can take ρ(En ) and Hni 2 out of the integral in the above formula to obtain Pi→[n] (t) = 2π Hni 2 ρ(En ) t ¯ h En
155
Pi→n (t) ρ(En ) dEn .13 Constant perturbations
6 APPROXIMATION METHODS
where ∆E = En − Ei  is change in energy of the system associated with the transition. whereas time is merely a parameter.215)
2t Hni 2 ρ(En ) sinc2 (x) dx. We know that in the limit t → ∞ the function sinc(x) is only nonzero in an inﬁnitesimally narrow range of ﬁnal energies centred on En = Ei . It is helpful to deﬁne the density of states. where the number of ﬁnal states lying in the energy range E to E + dE is given by ρ(E) dE. and ∆t = t is the time elapsed since the perturbation was switched on. since it implies that the probability of a transition occurring in a ﬁxed time interval. despite the fact that H1 is constant for t > 0.216)
. ρ(E).e. It follows that.214) ¯2 h where use has been made of sinc(0) = 1. (6. This result is somewhat surprising. in this limit.
(6. instead of t2 . Fermi’s golden rule is sometimes written wi→[n] = 2π Hni 2 δ(En − E). ¯ h where it is understood that this formula must be integrated with obtain the actual transition rate.e. wi→n = (6.13 Constant perturbations
6 APPROXIMATION METHODS
where Pi→[n] denotes the transition probability between the initial state i and all ﬁnal states n which have approximately the same energy as the initial state. (6. It is convenient to deﬁne the transition rate.220)
Let us now calculate the secondorder term in the Dyson series.219)
−∞
Note that the transition probability. which is simply the transition probability per unit time.222) ρ(En ) dEn to dPi→[n] . wi→[n] = giving 2π Hni 2 ρ(En ) . is now proportional to t.206). Note that the transition rate is constant in time (for t > 0): i. Pi→[n] . Hni 2 is the average of Hni 2 over all ﬁnal states with approximately the same energy as the initial state. In deriving the above formula. the probability of a transition occurring in the time interval t to t + dt is independent of t for ﬁxed dt.6.202) we ﬁnd that
(2) cn (t)
=
−i ¯ h
2
t
t
Hnm Hmi
m 0 t
dt exp( i ωnm t )
0
dt exp( i ωmi t )
i = ¯ h = it ¯ h
m
Hnm Hmi Em − E i
[exp( i ωni t ) − exp( i ωnm t ] ) dt
0
m
Hnm Hmi [exp( i ωni t/2) sinc(ωni t/2) Em − E i
156
. using the constant perturbation (6.221) ¯ h En E i This appealingly simple result is known as Fermi’s golden rule. Thus. (6.. Here. we have made use of the result ∞ sinc2 (x) dx = π. From Eq. dt (6.
a secondorder transition takes place in two steps. This problem can be avoided by gradually turning on the perturbation: i.224) to average to zero (due to the oscillatory phasefactor) during the evaluation of the transition probability. conserves energy. First.223)
+
m
Hnm Hmi exp( i ωim t/2) sinc(ωnm t/2) . with approximately the same energy as the initial state.224)
where use has been made of Eq. by analogy with the previous analysis. The nonenergyconserving transitions are generally termed virtual transitions.225) from Ei − Em to Ei − Em + i ¯ η. (6. Em − E i
Hnm Hmi sinc(ωni t/2) Em − E i
(6. h
157
. cn (t) = c(1) n + c(2) n it exp( i ωni t/2) = ¯ h −
m H ni
(6. Thus. The above formula clearly breaks down if Hnm Hmi = 0 when Em = Ei . (6.225). (6. It follows.208). and for intermediate states. the system makes a nonenergyconserving transition to some intermediate state m . The net transition. that wi→[n] 2π = Hni + ¯ h Hnm Hmi ρ(En ) Em − E i
2
. The fact that Em = Ei causes the last term on the righthand side of Eq. the system makes another nonenergyconserving transition to the ﬁnal state n .13 Constant perturbations
6 APPROXIMATION METHODS
− exp( i ωnm t/2) sinc(ωnm t/2)] . Subsequently. m whose energies differ from that of the initial state.225)
m
where the transition rate is calculated for all ﬁnal states. i .6. (6. The net result is to change the energy denominator in Eq. n . from i to n .e.
En E i
(6. whereas the energy conserving ﬁrstorder transition is termed a real transition. According to Eq.. H1 → exp(η t) H1 (where η is very small).
(6.201) that c(1) n −i = ¯ h
t 0 † Vni exp(i ωt ) + Vni exp(−i ωt ) exp( i ωni t ) dt . provided that ωni = En − E i → ωni ± ω. It follows from Eq. This is usually called a harmonic perturbation.229)
This formula is analogous to Eq. (6. in general.14 Harmonic perturbations
6 APPROXIMATION METHODS
6. H1 (t) = V exp( i ωt) + V † exp(−i ωt).232) corresponds to the second term. Thus. and switch on the harmonic perturbation at t = 0. momentum. h (6.228) (6. ¯ h 0 or En (6. nV † i = iVn ∗ . H0 . a function of position. h (6.232)
Clearly. (6. ¯ h ωni + ω ωni − ω where Vni =
† Vni =
nVi .226)
where V is.14 Harmonic perturbations Consider a perturbation which oscillates sinusoidally in time. and (6. The former term describes a process by which the system gives up energy ¯ ω to the perturbing ﬁeld.
(6.6.227)
1 1 − exp[ i (ωni + ω) t] 1 − exp[ i (ωni − ω) t] † = Vni + Vni . (6.208).231) ωni − ω 0 or En Ei + ¯ ω. whilst making a h transition to a ﬁnal state whose energy level is less than that of the initial state
158
. (6.230)
Thus. it follows from the previous analysis that the transition probability P i→n (t) = c(1) 2 is only appreciable in the limit t → ∞ if n ωni + ω Ei − ¯ ω.227). and spin operators. Let us initiate the system in the eigenstate i of the unperturbed Hamiltonian. (6.231) corresponds to the ﬁrst term on the righthand side of Eq.
221).15 Absorption and stimulated emission of radiation Let us use some of the results of timedependent perturbation theory to investigate the interaction of an atomic electron with classical (i. nonquantized)
159
. whereas Eq. equals the rate of absorption. The latter term describes h a process by which the system gains energy ¯ ω from the perturbing ﬁeld.233)–(6. h ¯ h 2π † 2 = V  δ(En − Ei − ¯ ω). ¯ h En =Ei −¯ ω h 2π † 2 . whilst h making a transition to a ﬁnal state whose energy level exceeds that of the initial state by ¯ ω.236)
† It is clear from Eqs.229) that Vni 2 = Vni 2 . that of the system plus the perturbing ﬁeld) is conserved. These equations are more usually written wi→n = wi→n 2π Vni 2 δ(En − Ei + ¯ ω).. h ¯ h ni (6.233) (6.234) gives the transition rate for absorption. (6. V  ρ(En ) = ¯ h ni En =Ei +¯ ω h (6.15 Absorption and stimulated emission of radiation
6 APPROXIMATION METHODS
by ¯ ω. the rate of stimulated emission. In both cases. which expresses a fundamental symmetry between absorption and stimulated emission.
6. (6. wi→[n] = wi→[n] 2π Vni 2 ρ(En ) . It follows from Eqs.233) speciﬁes the transition rate for stimulated emission. the total energy h (i.e.e. divided by the density of ﬁnal states for stimulated emission.235) (6. (6. This process is known as absorption. This result.228)(6..6. divided by the density of ﬁnal states for absorption.234)
Equation (6.234) that wi→[n] wn→[i] = .237) ρ(En ) ρ(Ei ) In other words. By analogy with Eq. and is very important in statistical mechanics. is known as detailed balancing. (6. This process is known as stimulated emission. (6.
(6. (6. p·A = A·p. The above equation can be written p2 − e A·p − e p·A + e2 A2 H= + e φ + V0 (r).241) E = − φ− ∂t B = × A. (6. The unperturbed Hamiltonian is p2 + V0 (r). c (6.245) provided that we adopt the gauge p2 e A·p e2 A2 H= − + + e φ + V0 (r). A = 2 A0 cos
160
(6.246)
Suppose that the perturbation corresponds to a monochromatic planewave. (6.238) 2 me The standard classical prescription for obtaining the Hamiltonian of a particle of charge q in the presence of an electromagnetic ﬁeld is H0 = H → H − q φ.6. Thus. 2 me me 2 me ·A = 0. (6.248)
. for which φ = 0. Note that ∂A .239) (6.240)
This prescription also works in quantum mechanics.247) ω n·r − ωt . (6. Hence.15 Absorption and stimulated emission of radiation
6 APPROXIMATION METHODS
electromagnetic radiation.243) 2 me where A and φ are functions of the position operators.244) 2 me Now. p → p + q A. (6.242)
(p − e A)2 H= + e φ + V0 (r).
(6. the Hamiltonian of an atomic electron placed in an electromagnetic ﬁeld is
where A(r) is the vector potential and φ(r) is the scalar potential.
2 me e A·p .249) with H0 = and H1 − p2 + V(r). h 2 ¯ me h (6. by analogy with the previous analysis.251)
where the A2 term.254)
The absorption crosssection is deﬁned as the ratio of the power absorbed by the atom to the incident power per unit area in the electromagnetic ﬁeld. It follows from Eq. (6.6. Note that ·n = 0. (6.236) that the rate of absorption is h wi→n 2π e2 = A0 2  n exp[ i (ω/c) n·r] ·p i 2 δ(En − Ei − ¯ ω). Now the energy density of an electromagnetic ﬁeld is 1 U= 2
2 0 E0
2
B02 + . that the ﬁrst term on the righth hand side of Eq. which is second order in A0 . The Hamiltonian becomes H = H0 + H1 (t).15 Absorption and stimulated emission of radiation
6 APPROXIMATION METHODS
where and n are unit vectors which specify the direction of polarization and the direction of propagation. respectively. has been neglected. (6. me (6. me (6.255)
161
. 2 µ0
(6.252)
This has the same form as Eq. (6.252) describes the absorption of a photon of energy ¯ ω. provided that V =− e A0 ·p exp[−i (ω/c) n·r ] me (6.250)
(6. whereas the second term describes the stimulated emission of a photon of energy ¯ ω.253)
It is clear.226). The perturbing Hamiltonian can be written H1 = − e A0 ·p (exp[ i (ω/c) n·r − i ωt] + exp[−i (ω/c) n·r + i ωt]) .
¯ h
(6. me
(6. The incident power per unit area of the electromagnetic ﬁeld is c U = 2 0 c ω2 A0 2 . cU (6.
(6.261)
me n[r.258) ¯ ω wi→n h . (6.16 The electric dipole approximation In general.257)
6.262)
(6. unity (remember that ω/c = 2π/λ).260)
i¯ p h . transitions between different atomic energy levels is much larger than the typical size of a light atom. Thus.
162
· npi . (6. h 2ωc 0 me (6. respectively. the wavelength of the type of electromagnetic radiation which induces.258). or is emitted during. H0 ]i = i me ωni nri .16 The electric dipole approximation
6 APPROXIMATION METHODS
where E0 and B0 = E0 /c = 2 A0 ω/c are the peak electric and magnetic ﬁeldstrengths.263)
. This approximation is known as the electric dipole approximation. c (6. It follows that n exp[ i (ω/c) n·r] ·p i It is readily demonstrated that [r.6. exp[ i (ω/c) n·r] = 1 + i ω n·r + · · · . H0 ] = so npi = −i Using Eq. σabs = so σabs = π e2  n exp[ i (ω/c) n·r] ·p i 2 δ(En − Ei − ¯ ω).256) Now. we obtain σabs = 4π2 α ωni  n ·ri 2 δ(ω − ωni ).259)
can be approximated by its ﬁrst term.
Thus. l is the quantum number describing the total orbital angular momentum of the electron. but disallows a transition from a 2s to a 1s state.264) (6. Instead. which is due to the interaction between the electron spin and the oscillating magnetic ﬁeld of the incident electromagnetic radiation.265)
∆m = 0. 6.
Here. It is clear. they take place at a far lower rate than transitions which are allowed according to the electric dipole approximation. (6.267)
∆m = ±1. ω. then it exhibits a sharp maximum at ω = ωni = (En − Ei )/¯ .268) (6. We z have already seen.259) yields
163
. Forbidden transitions are not strictly forbidden.4. and m is the quantum number describing the projection of the orbital angular momentum along the zaxis. (6. The latter transition is called a forbidden transition. from Sect. (6.6. After electric dipole transitions.16 The electric dipole approximation
6 APPROXIMATION METHODS
where α = e2 /(2 0 h c) = 1/137 is the ﬁne structure constant.266) (6.269) ∆m = 0. ∆l = ±1.
These are termed the selection rules for electric dipole transitions. so that = ^. It is clear that if the absorption crosssection is regarded as a function of the applied frequency. for generally directed radiation n ·ri is only nonzero if (6. The ﬁrstorder term in Eq. h Suppose that the radiation is polarized in the zdirection. Magnetic dipole transitions are typically about 10 5 times more unlikely than similar electric dipole transitions. the next most likely type of transition is a magnetic dipole transition. that nzi = 0 unless the initial and ﬁnal states satisfy ∆l = ±1. ±1. It is easily demonstrated that nxi and nyi are only nonzero if ∆l = ±1. for instance. that the electric dipole approximation allows a transition from a 2p state to a 1s state.
h ¯2 i[x.270)
Suppose. that the incident radiation is polarized in the xdirection. Eq. In fact. me
2 2
(6.
(6.263) over all possible frequencies of the incident radiation yields σabs (ω) dω =
n
4π2 α ωni  n ·ri 2 . me 2 me ¯ h ωni  nxi 2 = 1. me Thus. ±2. These are typically about 10 8 times more unlikely than electric dipole transitions.
164
.270) reduces to σabs (ω) dω = 2π2 α ¯ h π e2 . (6.271)
(6. Thus. the above formula is exh actly the same as that obtained classically by treating the electron as an oscillator. Magnetic dipole and electric quadrupole transitions satisfy different selection rules than electric dipole transitions: for instance.274)
This is known as the ThomasReicheKuhn sum rule.
n
(6.273)
It follows that
(6. (6. Integrating Eq. for the sake of deﬁniteness. [x. It is easily demonstrated that ¯2 h [x. = me 2 0 me c (6. H0 ] ] = − . the selection rules for electric quadrupole transitions are ∆l = 0. H0 ] ]i = ix H0 + H0 x − 2 x H0 xi = − .275)
Note that ¯ has dropped out of the ﬁnal result.272)
giving 2
n
¯2 h ( ixn Ei nxi − ixn En nxi ) = − . According to this rule.6. transitions which are forbidden as electric dipole transitions may well be allowed as magnetic dipole or electric quadrupole transitions.16 The electric dipole approximation
6 APPROXIMATION METHODS
socalled electric quadrupole transitions. [x.
277)
t
i = − Hni ¯ h
exp[ (η + i ωni )t ] dt
−∞
i exp[ (η + i ωni )t ] = − Hni . Thus.17 Energyshifts and decaywidths
6 APPROXIMATION METHODS
6. For cn=i (t) we have from Eqs. In the remote past. becomes populated as a result of some timedependent perturbation applied to the system. η→0 η + ωni
165
Hni 2 exp(2 η t) 2 . It is convenient to gradually turn on the perturbation from zero at t = −∞. and cn=i (t → −∞) = 0. we want to calculate the time evolution of the coefﬁcient ci (t). exp(η t) → 1. Thus.200)– (6. to ﬁrstorder. In this limit.279)
(6. however. (1) cn (t)
(6.278)
where Hni = nH1 i . (6. H1 (t) = exp(η t) H1 .17 Energyshifts and decaywidths We have examined how a state n . the system is assumed to be in the initial state i . Let us now consider how the initial state becomes depopulated. let us check that our previous Fermi golden rule result still applies when the perturbing potential is turned on slowly. other than the initial state i . (6. dt η2 + ωni ¯2 h Consider the limit η → 0. First.281)
.280)
(6. but η h lim 2 2 = π δ(ωni ) = π ¯ δ(En − Ei ). It follows that. and H1 is a constant. ci (t → −∞) = 1. instead of very suddenly.201) that
(0) cn (t) = 0. ¯ h η + i ωni
(6.276) where η is small and positive. t → −∞. the transition probability from state i to state n is Pi→n (t) = c(1) 2 = n The transition rate is given by 2 Hni 2 η exp(2 η t) dPi→n = wi→n (t) = 2 . Basically.6. ¯ 2 η2 + ωni h
(6.
202).284)
Hmi 
m
2 −∞
dt
−∞
dt
2 m
× exp[ (η + i ωim )t ] exp[ (η + i ωmi )t ].283)
t
i = − Hii ¯ h = −i ¯ h −i ¯ h
2
exp(η t ) dt = −
−∞ t t
i exp(η t) Hii .17 Energyshifts and decaywidths
6 APPROXIMATION METHODS
Thus.282) wi→n = ¯ h It is clear that the deltafunction in the above formula actually represents a function which is highly peaked at some particular energy.6. The width of the peak is determined by how fast the perturbation is switched on. we ˙ ˙ can evaluate this ratio in the limit η → 0. (6.280) yields the standard Fermi golden rule result 2π Hni 2 δ(En − Ei ). where ci ≡ dci /dt. Using Eq. ¯ h η
(6.200)–(6.286). Ei − E m + i ¯ η h m=i
166
(6. to secondorder we have ci (t) exp(η t) −i −i Hii + 1+ ¯ h η ¯ h −i + ¯ h
2
Hii 2
exp(2 η t) 2 η2 (6. We have ci (t) = 1. 2 η (η + i ωmi ) (6. (6.287)
.286)
Hmi 2 exp(2 η t) . (6. Eq. Let us now calculate ci (t) using Eqs. 2 η (Ei − Em + i ¯ η) h m=i
Let us now consider the ratio ci /ci .285)
=
Thus. (6. Hmi 2 exp(2 η t) .
(1) ci (t) (2) ci (t) (0)
(6. We obtain ci ˙ ci
−i −i Hii + ¯ h ¯ h 1−
2
Hii 2 −i + η ¯ h
i Hii ¯ η h
Hmi 2 Ei − E m + i ¯ η h m=i
−i −i Hii + lim η→0 ¯ ¯ h h
Hmi 2 .
x (6. lim where =P (6. It follows that Hmi 2 ∆i = Hii + P − iπ Hmi 2 δ(Ei − Em ). (6. h h (6. t = exp[−i (∆i + Ei ) t/¯ ] i . We can write ci ˙ −i ∆i . = ci ¯ h where Hmi 2 ∆i = Hii + lim η→0 E − Em + i ¯ η h m=i i 1 →0 x + i 1 − i π δ(x). we obtain −i ∆i t ci (t) = exp .296)
.149). h We can rewrite this result as i.290)
> 0. (6. Ei − E m m=i
167
(6.293)
It is clear that the real part of ∆i gives rise to a simple shift in energy of state i . the time evolution of the initial state ket i is given by i.291)
It is convenient to normalize the solution of Eq. h h where Hmi 2 ∆Ei = Re(∆i ) = Hii + P .289)
is a constant. E − Em m=i i m=i (6. i. Note that the righthand side of Eq. whereas the imaginary part of ∆i governs the growth or decay of this state. t = exp[−i (Ei + ∆Ei ) t/¯ ] exp(−Γi t/2¯ ) i .294) (6.288)
(6.292) ¯ h According to Eq.288) so that ci (0) = 1. (6. Thus. According to a wellknown result in pure mathematics.17 Energyshifts and decaywidths
6 APPROXIMATION METHODS
This result is formally correct to secondorder in perturbed quantities.6.295)
(6. and P denotes the principle part.287) is independent of time. (6. t = exp(−i [Ei + Re(∆i ) ] t/¯ ) exp[ Im(∆i ) t/¯ ] i . Thus.
is given by Pi→i (t) = ci 2 = exp(−Γi t/¯ ).297)
Note that the energyshift ∆Ei is the same as that predicted by standard timeindependent perturbation theory. h (6.300)
The quantity ∆i is called the decaywidth of state i . the rate of decay of the initial state is a simple function of the transition rates to the other states. (6. since ci 2 +
m=i
cm 2
(1 − Γi t/¯ ) + h
m=i
wi→m t = 1. It is closely related to the mean lifetime of this state. (6. use has been made of Eq.222). h where Γi = wi→m . Clearly.
(6. we can still represent it as a superposition of stationary states (whose amplitudes simply oscillate in time). the amplitude of state i both oscillates and decays as time progresses. given that it is deﬁnately in state i at time t = 0.17 Energyshifts and decaywidths
6 APPROXIMATION METHODS
and
Γi 2 Im(∆i ) 2π =− = Hmi 2 δ(Ei − Em ). ¯ h τi = .298)
(6. However. exp[−i (Ei + ∆Ei ) t/¯ ] exp(−Γi t/2¯ ) = h h f(E) exp(−i E t/¯ ) dE.302) According to Eq.294).299)
Here. ¯ h m=i (6. (6.303)
168
. state i is not a stationary state in the presence of the timedependent perturbation. Thus.301) Γi where Pi→i = exp(−t/τi ). Clearly. (6. Note that the system conserves probability up to secondorder in perturbed quantities.6. ¯ h ¯ h ¯ m=i h
(6. The probability of observing the system in state i at time t > 0.
(E − [Ei + Re(∆i )])2 + Γi 2 /4 (6.304)
In the absence of the perturbation.17 Energyshifts and decaywidths
6 APPROXIMATION METHODS
where f(E) is the weight of the stationary state with energy E in the superposition.6. f(E)2 is basically a deltafunction centred on the unperturbed energy Ei of state i . Such lines are not as bright as those corresponding to allowed transitions. state i is a stationary state whose energy is completely determined. the larger Γi ).. the energy of the state is smeared over some region of width (in energy) Γi centred around the shifted energy Ei + Re(∆i ). which correspond to fast transitions. This effect is clearly a manifestation of the energytime uncertainty relation ∆E ∆t ∼ ¯ . In the presence of the perturbation. For this reason. The uncertainty in energy of the excited state.e. The fact that the state is no longer stationary (i.. Indeed. due to its propensity to decay.e. gives rise to a slight smearing (in wavelength) of the spectral line associated with the transition. Strong lines.
169
. the energy of state i is shifted by Re(∆i ). it decays in time) implies that its energy cannot be exactly determined. The Fourier inversion theorem yields f(E)2 ∝ 1 . but they are a lot sharper. are smeared out more that weak lines. spectroscopists generally favour forbidden lines for Doppler shift measurements. the more its energy is spread out. The faster the decay of the state (i. One consequence of h this effect is the existence of a natural width of spectral lines associated with the decay of some excited state to the ground state (or any other lower energy state). In other words.
3). possibly. and scattering experiments.7 SCATTERING THEORY
7
Scattering theory
7.2 The LipmannSchwinger equation Consider timeindependent scattering theory. In general. (7. data regarding quantum phenomena has been obtained from two main sources—the study of spectroscopic lines. (7.4)
where ψ is an energy eigenstate of the total Hamiltonian whose wavefunction r ψ is ψ(r ).3) whose wavefunction r φ is φ(r ). corresponding to the o same energy eigenvalue. for which the Hamiltonian of the system is written H = H 0 + H1 . H0 = 2m (7. This state is a planewave state or. (7. We require a solution of Eq. Here.. Schr¨dinger’s equation for the scattering problem is o (H0 + H1 )ψ = E ψ .e.1 Introduction Historically.2)
and H1 represents the nontimevarying source of the scattering. a sphericalwave state.
7. their energy eigenstates are unbound. Let us now examine the quantum theory of scattering. Let φ be an energy eigenket of H0 . (7. both H0 and H0 + H1 have continuous energy spectra: i. We have already developed theories which account for some aspects of the spectra of hydrogenlike atoms. H0 φ = E φ .1) where H0 is the Hamiltonian of a free particle of mass m.
170
. (7.4) which satisﬁes the boundary condition ψ → φ as H1 → 0. φ is a solution of the free particle Schr¨dinger equation. p2 .
and can be inverted using standard Green’s function techniques. The standard prescription is to make the energy eigenvalue E slightly complex. otherwise the above solution is useless.8) ( 2 + k2 ) ψ(r) = 2 rH1 ψ .7.e.10)
. and making use of Eq. Unfortunately. and small. ψ± (r) = φ(r) + r 1 E − H0 ± i r r H1 ψ± d3 r . (7. 2 ¯ h
171
(7.3).7)
Adopting the Schr¨dinger representation. (7. ψ± = φ + 1 H1 ψ± . the operator (E − H0 )−1 is singular: i. Furthermore. r ) r H1 ψ d3 r . we can write the scattering problem o (7. (7.6)
where is real. The physical signiﬁcance of the ± signs will become apparent later on. (7.4) in the form 2m (7. The LipmannSchwinger equation can be converted into an integral equation via left multiplication by r.. We need a prescription for dealing with these inﬁnities. Thus. the desired solution can be written ψ = φ + 1 H1 ψ . it produces inﬁnities when it operates on an eigenstate of H 0 corresponding to the eigenvalue E. and is nonsingular as long as > 0.6) is called the LipmannSchwinger equation.4) by operating on the above equation with E − H0 . ψ(r) = φ(r) + 2m G(r. E − H0 ± i (7. positive.5)
Note that we can recover Eq. Equation (7. Thus. E − H0 (7. ¯ h where ¯ 2 k2 h E= .2 The LipmannSchwinger equation
7 SCATTERING THEORY
Formally.9) 2m This equation is called Helmholtz’s equation. the solution satisﬁes the boundary condition ψ → φ as H1 → 0. Thus.
7. We can write r H1 ψ± = r H1 r r ψ± d3 r (7.
(7. r ) = − Thus.7) and (7.10) satisﬁes the boundary condition ψ → φ as H1 → 0.10) becomes 2 m exp(±i k r − r  ) r H1 ψ d3 r . (7.17)
Suppose that the initial state φ is a planewave with wavevector k (i.2 The LipmannSchwinger equation
7 SCATTERING THEORY
where (
2
+ k2 ) G(r. 4π r − r  (7.16) (7. (7.e. the Green’s function for the Helmholtz problem is given by G(r.7) takes the form 2 m exp(±i k r − r  ) 1 r =− 2 . (7. Eq.. they do. (7. 4π r − r 
(7.13) 2 4π r − r  ¯ h A comparison of Eqs. The associated wavefunction takes the form rk = exp( i k·r) .12)
= V(r ) ψ± (r ). Thus. ψ± (r) = φ(r) − Let us suppose that the scattering Hamiltonian.13) simpliﬁes to ψ± (r) = φ(r) − 2m ¯2 h exp(±i k r − r ) V(r ) ψ± (r ) d3 r . H1 . In fact. (2π)3/2
172
(7.13) suggests that the kernel to Eq.11)
Note that the solution (7. As is wellknown. a stream of particles of deﬁnite momentum p = ¯ k). This implies that r H1 r = V(r) δ(r − r ).14) r E − H0 ± i 4π r − r  ¯ h It is not entirely clear that the ± signs correspond on both sides of this equation.15) exp(±i k r − r  ) . (7. The ket corresponding to h this state is denoted k . is only a function of the position operators. the integral equation (7.18)
. r ) = δ(r − r ). as is easily proved by a more rigorous derivation of this result.
k is the wavevector for particles which possess the same energy as the incoming particles (i. whereas the minus sign corresponds to a wave propagating towards the scattering region. let us adopt the ordering r r .19)
Suppose that the scattering potential V(r) is only nonzero in some relatively localized region centred on the origin (r = 0).7. but propagate from the scattering region to the observation point. (7.20)
r (7. The second term represents a spherical wave centred on the scattering region.17) reduces to ψ(r)± exp( i k·r) m exp(±i k r) exp( i k ·r ) V(r ) ψ± (r ) d3 r . Let us calculate the wavefunction ψ(r) a long way from the scattering region. k = k). In other words. where r − ^·r r (7.. Eq. Let us deﬁne k = k ^. Note that exp(±i k r − r  ) exp(±i k r) exp( i k ·r ). (7. (7.e.21) r is a unit vector which points from the scattering region to the observation point. r (7. The plus sign (on ψ± ) corresponds to a wave propagating away from the scattering region.23)
In the larger limit. (2π)3 (7.
173
.24) − 2 3/2 (2π) r 2π ¯ h
The ﬁrst term on the righthand side is the incident wave. It is easily demonstrated that r − r  to ﬁrstorder in r /r.22) ^= r
Clearly.2 The LipmannSchwinger equation
7 SCATTERING THEORY
The wavefunction is normalized such that kk = = kr rk d3 r exp[−i r·(k − k )] 3 d r = δ(k − k ).
31)
dσ r2 dΩ jscat  dΩ = .32)
. from Sect. the probability ﬂux associated with the incident wavefunction. 4. that the probability ﬂux (i.30)
is jscat Now.
f(k . dΩ jinci 
174
(7.26)
(7. divided by the incident ﬂux of particles. k) . the probability ﬂux associated with the scattered wavefunction. m Thus. Thus. (2π)3/2 r (7.2 The LipmannSchwinger equation
7 SCATTERING THEORY
It is obvious that the former represents the physical solution. ψ(r) = exp( i k·r) + (2π)3/2 r (2π)2 m f(k . k) . ¯2 h
Let us deﬁne the differential crosssection dσ/dΩ as the number of particles per unit time scattered into an element of solid angle dΩ. (7.27)
(7.29) (2π)3 m Likewise. the particle ﬂux) associated with a wavefunction ψ is ¯ h Im(ψ∗ ψ).28)
¯ h k.25)
where
(2π)2 m = − k H1 ψ . Recall. the wavefunction a long way from the scattering region can be written 1 exp( i kr) f(k . k) = − ¯2 h exp(−i k ·r ) V(r ) ψ(r ) d3 r 3/2 (2π) (7.e. j= exp( i k·r) .7. exp( i k r) f(k . k)2 ¯ h k ^.. r = (2π)3 m r2
(7. (2π)3/2 is jinci = (7.
(7.7. k = k). k) − m 2π ¯ 2 h
exp [ i (k − k )·r ] V(r ) d3 r .
175
(7. (7. This is always the case for scattering Hamiltonians of the form shown in Eq. f(k .36)
. V.35)
This is called the Born approximation.3 The Born approximation Equation (7. k) by making the substitution ψ(r) → φ(r) = exp( i k·r) . Thus. we can obtain an expression for f(k . k)2 . (7.34) 2 r 2π ¯ h where φ(r) is the wavefunction of the incident state. Note that the scattered particles possess h the same energy as the incoming particles (i. ψ(r).33) is not particularly useful. In this case. as it stands.e. k) depends on the unknown ket ψ . (2π)3/2 (7. The Born approximation yields f(k . φ(r).
7.. The strength of the sphericalwave emitted at a given point is proportional to the local value of the scattering potential. it is reasonable to suppose that the total wavefunction. does not differ substantially from the incident wavefunction. ψ.15). ψ(r) = φ(r) − Suppose that the scattering is not particularly strong. as well as the local value of the wavefunction.33) dΩ Thus.3 The Born approximation
7 SCATTERING THEORY
giving dσ = f(k . Recall that ψ(r) = rψ is the solution of the integral equation m exp( i k r) exp(−i k ·r ) V(r ) ψ(r ) d3 r . According to the above equation the total wavefunction is a superposition of the incident wavefunction and lots of sphericalwaves emitted from the scattering region. because the quantity f(k . k)2 gives the differential crosssection for particles with incident momentum ¯ k to be scattered into states whose momentum vectors are directed in h a range of solid angles dΩ about ¯ k .
For a spherically symmetric potential. µr (7. (7.7. f(k . (7.3 The Born approximation
7 SCATTERING THEORY
Thus. k) − 2 r V(r ) sin(q r ) dr .
176
(7.43)
q2 = 4 k2 sin2 (θ/2) = 2 k2 (1 − cos θ).41) ¯ µ q2 + µ 2 h since ∞ q exp(−µ r ) sin(q r ) dr = 2 .38) ¯ q 0 h Note that f(k . k) is just a function of q for a spherically symmetric potential. in the Born approximation. (7. [2 k2 (1 − cos θ) + µ2 ]2 ¯2µ h
2
(7.42) µ + q2 0 Thus. f(k . the differential crosssection for scattering by a Yukawa potential is dσ dΩ given that
2 m V0 1 . θ is the angle of scattering.39) − m 2π ¯ 2 h exp( i q r cos θ ) V(r ) r 2 dr sin θ dθ dφ . It follows from Eq. (7.44)
. k) giving 2m ∞ f(k . (7. Consider scattering by a Yukawa potential V(r) = V0 exp(−µ r) . It is easily demonstrated that q ≡ k − k  = 2 k sin(θ/2).37)
where θ is the angle subtended between the vectors k and k .40)
where V0 is a constant and 1/µ measures the “range” of the potential.38) that 2 m V0 1 f(θ) = − 2 . In other words. k) is proportional to the Fourier transform of the scattering potential V(r) with respect to the wavevector q ≡ k − k . Recall that the vectors k and k have the same length by energy conservation. (7.
from Eq.7. It follows. In this limit the Born differential crosssection becomes 2 2 m Z Z e2 dσ 1 . 4 16π 0 E sin (θ/2)
2
(7. so the above equation can be rewritten h dσ dΩ 1 Z Z e2 . (i. that the condition for ψ(r) φ(r) in the vicinity of r = 0 is m 2π ¯ 2 h exp( i k r ) V(r ) d3 r r 1.45) dΩ 4π 0 ¯ 2 h 16 k4 sin4 (θ/2) Recall that ¯ k is equivalent to p.3 The Born approximation
7 SCATTERING THEORY
The Yukawa potential reduces to the familiar Coulomb potential as µ → 0. Equation (7. In the highk limit. ¯ 2 µ2 h (7. Eq.47) yields 2 m V0  1. (7. (7. (7. k we can replace exp( i k r ) by unity. implying that the Born approximation is more accurate at high incident particle energies. giving 2 m V0  ¯ 2 µ2 h 1
(7.46)
where E = p2 /2 m is the kinetic energy of the incident particles.7.50) ¯2 µk h This inequality becomes progressively easier to satisfy as k increases.
177
. The Born approximation is valid provided that ψ(r) is not too different from φ(r) in the scattering region. The condition for the Yukawa potential to develop a bound state is 2 m V0  ≥ 2. provided that V0 /µ → Z Z e2 /4π 0 . if the potential is strong enough to form a bound state then the Born approximation is likely to break down.48)
as the condition for the validity of the Born approximation. At low energies..49)
where V0 is negative. (7. (7.e.47) µ)
Consider the special case of the Yukawa potential.17). Thus.46) is the classical Rutherford scattering crosssection formula.
(2π)3/2 r
(7. Equation (7.54)
What is the most general solution to this equation in spherical polar coordinates which does not depend on the azimuthal angle ϕ? Separation of variables yields ψ(r.4 Partial waves We can assume.
(7. ϕ).38) strongly suggests that for a spherically symmetric scattering potential [i.7.51)
exp( i k z) exp( i k r cos θ) = . 2l + 1
178
(7.53)
depend on the azimuthal angle ϕ.56)
. (2π)3/2 (2π)3/2
(7. the angle subtended between the two wavevectors). 5 via Pl (cos θ) = 4π Yl0 (θ. ϕ) = f(θ).e.. The direction of k is speciﬁed by the polar angle θ (i. points in a different direction.55)
since the Legendre functions Pl (cos θ) form a complete set in θspace. The Legendre functions are related to the spherical harmonics introduced in Sect. It follows that neither the incident wavefunction. without loss of generality.
Outside the range of the scattering potential. θ) =
l
Rl (r) Pl (cos θ).52)
(7. in general.. both φ(r) and ψ(r) satisfy the free space Schr¨dinger equation o (
2
+ k2 ) ψ = 0. 1 exp( i k r) f(θ) ψ(r) = exp( i k r cos θ) + . and an azimuthal angle ϕ about the zaxis.e.4 Partial waves
7 SCATTERING THEORY
7. but. The scattered wavefunction is characterized by a wavevector k which has the same magnitude as k. φ(r) = nor the total wavefunction.
(7. V(r) = V(r)] the scattering amplitude is a function of θ only: f(θ. that the incident wavefunction is characterized by a wavevector k which is aligned parallel to the zaxis.
7.64)
It is wellknown that (−i)l jl (y) = 2
1
exp( i y µ) Pl (µ) dµ.
−1
(7.62)
where the al are constants.
(7. jl (k r).59)
ηl (y) = −y
1 d − y dy
cos y . because they are not wellbehaved as r → 0.58) (7.61)
We can write
exp( i k r cos θ) =
l
al jl (k r) Pl (cos θ). y
Note that spherical Bessel functions are wellbehaved in the limit y → 0 . y
l
(7.57) + 2r 2 dr dr The two independent solutions to this equation are called a spherical Bessel function.63) Pn (µ) Pm (µ) dµ = n + 1/2 −1 so we can invert the above expansion to give
1
al jl (k r) = (l + 1/2)
−1
exp( i k r µ) Pl (µ) dµ. y jl (y) →
(7. It is easily demonstrated that r2 jl (y) = y
l
1 d − y dy
l
l
sin y .54) and (7.55) can be combined to give dRl d2 Rl + [k2 r2 − l (l + 1)]Rl = 0. ηl (k r).65)
179
. The asymptotic behaviour of these functions in the limit y → ∞ is sin(y − l π/2) . (7.60) (7. and a Neumann function. y cos(y − l π/2) ηl (y) → − . 1 δnm . Note there are no Neumann functions in this expansion.
(7.4 Partial waves
7 SCATTERING THEORY
Equations (7. (7. whereas Neumann functions become singular. The Legendre functions are orthonormal.
Equation (7. What is the source of the incoming waves? Obviously. 1965).70) yields ψ(r) 1 (2π)3/2 Cl
l
exp[ i (k r − l π/2 + δl )] − exp[−i (k r − l π/2 + δl )] 2ikr (7. Eq. al = il (2 l + 1). 1. because its region of validity does not include the origin.60)–(7. Thus. · · · [see Abramowitz and Stegun (Dover. The above expression can also be written ψ(r) 1 (2π)3/2 Cl
l
sin(k r − l π/2 + δl ) Pl (cos θ). the total wavefunction reduces to ψ(r) 1 (2π)3/2
l
cos(k r − l π/2) sin(k r − l π/2) A − Bl Pl (cos θ). (7. (7. In the larger limit.
(7. Note that the Neumann functions are allowed to appear in this expansion. The most general solution for the total wavefunction outside the scattering region is 1 [Al jl (k r) + Bl ηl (k r)] Pl (cos θ).67)
The above expression tells us how to decompose a planewave into a series of sphericalwaves (or “partial waves”). l kr kr
(7. 2. they must be part of the larger asymptotic
180
.
which contains both incoming and outgoing sphericalwaves.14]. New York NY.68) ψ(r) = (2π)3/2 l where the Al and Bl are constants. kr
(7.69)
where use has been made of Eqs.71)
× Pl (cos θ).1.70)
where the sine and cosine functions have been combined to give a sine function which is phaseshifted by δl .61).4 Partial waves
7 SCATTERING THEORY
where l = 0.66) giving exp( i k r cos θ) =
l
il (2 l + 1) jl (k r) Pl (cos θ).7. 10. (7.
7. (7. k
(7. Eqs.75)
Clearly. Now. r
(7. In fact.76)
× sin δl sin δl Pl (µ) Pl (µ).52) and (7. it is easily seen that φ(r) 1 (2π)3/2 il (2l + 1)
l
exp[ i (k r − l π/2)] − exp[−i (k r − l π/2)] 2ikr (7. sphericalwaves) is equivalent to determining the phaseshifts δ l .5 The optical theorem The differential scattering crosssection dσ/dΩ is simply the modulus squared of the scattering amplitude f(θ). determining the scattering amplitude f(θ) via a decomposition into partial waves (i. (7.5 The optical theorem
7 SCATTERING THEORY
expansion of the incident wavefunction.7.
181
. Eqs.72) that Cl = (2 l + 1) exp[ i (δl + l π/2)]. The total crosssection is given by σtotal = f(θ)2 dΩ
1
1 = 2 k
dϕ
−1
dµ
l l
(2 l + 1) (2 l + 1) exp[ i (δl − δl ] (7. This implies that the coefﬁcients of the incoming spherical waves in the larger expansions of ψ(r) and φ(r) must be equal..71) and (7.73) yield f(θ) =
∞ l=0
(2 l + 1)
exp( i δl ) sin δl Pl (cos θ). (7. It follows from Eqs.e.74) Thus.72)
× Pl (cos θ) in the larger limit.73)
Note that the righthand side consists only of an outgoing spherical wave.53) give (2π)3/2 [ψ(r) − φ(r)] = exp( i k r) f(θ). (7.71)–(7.
81)
182
.54). It is a reﬂection of the fact that the very existence of scattering requires scattering in the forward (θ = 0) direction in order to interfere with the incident wave. It is usual to write σtotal = where
∞ l=0
σl .79)
4π (2 l + 1) sin2 δl (7. where a is termed the range of the potential. It follows that σtotal = 4π k2 (2 l + 1) sin2 δl .80) 2 k is the lth partial crosssection: i. and thereby reduce the probability current in this direction.6 Determination of phaseshifts Let us now consider how the phaseshifts δl can be evaluated. (7.77)
where use has been made of Eq. The most general solution which is o consistent with no incoming sphericalwaves is 1 ψ(r) = (2π)3/2
∞ l=0
il (2 l + 1) Al (r) Pl (cos θ). the contribution to the total crosssection from the lth partial wave. A comparison of this result with Eq. (7. Note that the maximum value for the lth partial crosssection occurs when the phaseshift δl takes the value π/2. This result is known as the optical theorem.
(7. (7.7.
(7.78) k since Pl (1) = 1.
l
(7.. the wavefunction ψ(r) satisﬁes the freespace Schr¨dinger equation (7. Consider a spherically symmetric potential V(r) which vanishes for r > a.e.63).6 Determination of phaseshifts
7 SCATTERING THEORY
where µ = cos θ.75) yields 4π σtotal = Im [f(0)] . σl =
7. In the region r > a.
(7. the problem of determining the phaseshift δl is equivalent to that of obtaining βl+ . k a ηl (k a) − βl+ ηl (k a)
(7. dr2 r2 ¯ h The boundary condition ul (0) = 0
(7. because its region of validity does not include the origin (where V = 0). etc.7.87) (7. and form the logarithmic derivative βl− = 1 d(ul /r) (ul /r) dr
183
. integrate out to r = a. The most general solution to Schr¨dinger’s equation inside the range of the o potential (r < a) which does not depend on the azimuthal angle ϕ is 1 ψ(r) = (2π)3/2 where Rl (r) = and
∞ l=0
il (2 l + 1) Rl (r) Pl (cos θ).86)
d2 ul 2 2m l (l + 1) + k − 2 V− ul = 0. The above equation can be inverted to give (7.85)
(7.82) Note that Neumann functions are allowed to appear in the above expression. ul (r) .
r=a
(7.6 Determination of phaseshifts
7 SCATTERING THEORY
where Al (r) = exp( i δl ) [ cos δl jl (k r) − sin δl ηl (k r) ] . r
(7. = ka cos δl jl (k a) − sin δl ηl (k a) tan δl = k a jl (k a) − βl+ jl (k a) . The logarithmic derivative of the lth radial wavefunction Al (r) just outside the range of the potential is given by βl+ cos δl jl (k a) − sin δl ηl (k a) .84)
Thus.83)
where jl (x) denotes djl (x)/dx.89)
.88)
ensures that the radial wavefunction is wellbehaved at the origin. We can launch a wellbehaved solution of the above equation from r = 0.
93)
where use has been made of Eqs. and zero for r > a. for which the potential is inﬁnite for r < a. It follows that ψ(r) is zero in the region r < a.7 Hard sphere scattering
7 SCATTERING THEORY
Since ψ(r) and its ﬁrst derivatives are necessarily continuous for physically acceptible wavefunctions. βl− = βl+ = ∞.96)
.92) (7. (7. Consider scattering by a hard sphere. ηl (k a) (7.84) that tan δl = jl (k a) .91) for all l.94)
(7. It follows from Eq. The phaseshift δl is obtainable from Eq.7 Hard sphere scattering Let us test out this scheme using a particularly simple example.7. (7. (7. Thus.92) yields tan δ0 = sin(k a)/k a = − tan k a. which implies that u l = 0 for all l. Equation (7. It follows that (7.58)–(7. 7.59). The swave radial wave function is [cos k a sin k r − sin k a cos k r] kr sin[k (r − a)] = exp(−i k a) . A0 (r) = kr
184
(7. − cos(k a)/ka δ0 = −k a.90)
Consider the l = 0 partial wave. which is usually referred to as the swave.95) kr The corresponding radial wavefunction for the incident wave takes the form A0 (r) = exp(−i k a) sin k r ˜ . (7.84). (7. it follows that βl+ = βl− .
97) (7. (2 l + 1)!! − (2 l − 1)!! . It follows that −(k a)2 l+1 tan δl = . and (7. low energy scattering implies relatively long wavelengths.102) k l=0
185
.98)
where n!! = n (n − 2) (n − 4) · · · 1. In other words. It follows from Eqs. at low energy only swave scattering (i.101) a2 (7. (k r)l+1 (7. In this regime. except that it is phaseshifted by k a. with respect to δ0 .. It follows from Eq..77) that l 4π max σtotal = 2 (2 l + 1) sin2 δl . spherically symmetric scattering) is important.99)
It is clear that we can neglect δl .e. Note that the total crosssection σtotal = dσ dΩ = 4π a2 dΩ (7.33). the spherical Bessel functions and Neumann functions reduce to: jl (k r) ηl (k r) (k r)l . the crosssection for classical particles bouncing off a hard sphere of radius a). Low energy means k a 1.e. so we do not expect to obtain the classical result in this limit. (7. (7.94) that dσ sin2 k a = dΩ k2 for k a 1. with l > 0. At high energies. (2 l + 1) [(2 l − 1)!!]2 (7. (7. Let us consider the low and high energy asymptotic limits of tan δl .75).100)
is four times the geometric crosssection π a2 (i.7. Consider the high energy limit k a 1. all partial waves up to lmax = k a contribute signiﬁcantly to the scattering crosssection. However.7 Hard sphere scattering
7 SCATTERING THEORY
It is clear that the actual l = 0 radial wavefunction is similar to the incident l = 0 wavefunction. (7.
only swave scattering is important.105)
where use has been made of Eqs.e. The outside wavefunction is given by [see Eq. As a speciﬁc example.8 Low energy scattering At low energies (i. in general. which is somewhat surprizing.103) σtotal = 2 k l=0 This is twice the classical result. when 1/k is much larger than the range of the potential) partial waves with l > 0. gives the actual crosssection of 2π a2 . the interference is not completely destructive. The potential is repulsive for V0 > 0.58)–(7. It follows that. (7.104) (7. In fact. there must be scattering in the forward direction (recall the optical theorem) to produce destructive interference with the incident planewave.59). in order to produce a “shadow” behind the sphere. make a negligible contribution to the scattering crosssection. Here.7. and attractive for V0 < 0. when combined with the crosssection for classical reﬂection. let us consider scattering by a ﬁnite potential well. characterized by V = V0 for r < a. ka 2π (2 l + 1) 2π a2 . with a ﬁnite range potential. (7. kr (7. For hard sphere scattering. π a2 .
7. at these energies. However. V0 is a constant.. and V = 0 for r ≥ a. Thus.82)] A0 (r) = exp( i δ0 ) [j0 (k r) cos δ0 − η0 (k r) sin δ0 ] = exp( i δ0 ) sin(k r + δ0 ) . (7. The inside wavefunction follows
186
. and the shadow has a bright spot in the forward direction. it is legitimate to replace sin 2 δl by its average value 1/2. The effective crosssection associated with this bright spot is π a 2 which. since we might expect to obtain the classical result in the short wavelength limit. incident waves with impact parameters less than a must be deﬂected.8 Low energy scattering
7 SCATTERING THEORY
With so many l values contributing.
7. r (7.88). yields tan(k a + δ0 ) = k tan k a k k tanh κ a κ
(7. k
(7. and tan(k a + δ0 ) = for E < V0 .107) E − V0 = 2m Note that Eq. B is a constant.
(7. the righthand side is much less that unity. the depth of the potential well is much larger than the energy of the incik.109)
(7.112)
(7.111)
Consider an attractive potential. r (7.87). ka
187
k tan k a.110) that. V0 − E = 2m Matching A0 (r). unless tan k a dent particles).106) only applies when E > V0 . Suppose that V0  E (i. (7. for which E > V0 .. (7. It follows from Eq.108)
¯ 2 κ2 h . we obtain k a + δ0 This yields δ0 tan k a ka − 1 .e. so that k becomes extremely large.113)
. Here. so replacing the tangent of a small quantity with the quantity itself. (7.110)
for E > V0 . and ¯2k 2 h . We obtain A0 (r) = B sin k r .106)
where use has been made of the boundary condition (7. For E < V0 . we have A0 (r) = B where sinh κ r . and its radial derivative at r = a.8 Low energy scattering
7 SCATTERING THEORY
from Eq. (7.
116)
It follows that the total (swave) scattering crosssection is independent of the energy of the incident particles (provided that this energy is sufﬁciently small). at the value of the incident energy π/2 when k a = π/2. In this case..110) is small. ka 2 m V0  a2 .7. This implies that σtotal = 4π 2 1 sin δ0 = 4π a2 2 2 .9 Resonances There is a signiﬁcant exception to the independence of the crosssection on energy. This is called the RamsauerTownsend effect. so we can no longer assume that the righthand side of Eq. 2 k ka ka= k 2 a2 2 m V0  a2 . (7. (7. It follows that there are certain values of V0 and k which give rise to almost perfect transmission of the incident wave. and the scattering crosssection (7. (7. ¯2 h (7. Note that there are values of k a (e. + ¯2 h
2
(7. can reach the value π/2. and has been observed experimentally.115).110) that k a + δ0 = π/2. But. k a 4. or δ0 (since we are assuming that k a 1). it follows from Eq.115)
so for sufﬁciently small values of k a.114) vanishes. at low incident energies. the crosssection is not exactly zero. which is given by Eq.117)
. because of contributions from l > 0 partial waves.
7.9 Resonances
7 SCATTERING THEORY
According to Eq. these contributions are small. the scattering crosssection is given by σtotal Now tan k a 4π 2 sin δ0 = 4π a2 − 1 . despite the very strong attraction of the potential.114)
(7. As the h incident energy increases. k a. (7.49) at which δ0 → π. k2 k a
188
(7.g. tan k a becomes inﬁnite. In reality. Suppose that the quantity 2 m V0  a2 /¯ 2 is slightly less than π/2. In fact.102).
122)
. the energy of the scattering system is essentially the same as the energy of the bound state. and the subsequent decay of the bound state and release of the particle.119) 2 Let us expand cot δl in the vicinity of the resonant energy: cot δl (E) = cot δl (E0 ) +
d cot δl dE
E=E0
(E − E0 ) + · · ·
(7.7.114) for k a = π/2 (since k a 1). so that π δl (E0 ) = . this sort of resonance scattering is best understood as the capture of an incident particle to form a metastable bound state. the bound state is not stable. The condition π 2 m V0  a2 = 2 2 ¯ h (7. (7. the magnitude of the crosssection is much larger than that given in Eq. since the system has a small positive energy. We have seen that there is a resonant effect when the phaseshift of the swave takes the value π/2. Nevertheless. (7. Suppose that δl attains the value π/2 at the incident energy E0 . dE E=E0 Γ
189
(7. Thus. There is nothing special about the l = 0 partial wave. In this situation. The crosssection for resonance scattering is generally far higher than that for nonresonance scattering.120) (7. for a potential well which satisﬁes the above equation.118)
is equivalent to the condition that a spherical well of depth V0 possesses a bound state at zero energy. The origin of this rather strange behaviour is quite simple. Furthermore. sin δl dE E=E0 Deﬁning
2 dδl (E) = .9 Resonances
7 SCATTERING THEORY
Note that the crosssection now depends on the energy.121)
1 dδl = − 2 (E − E0 ) + · · · . However. an incident particle would like to form a bound state in the potential well. so it is reasonable to assume that there is a similar resonance when the phaseshift of the lth partial wave is π/2.
from Eq.7. (7. (2 l + 1) sin2 δl = 2 (2 l + 1) k2 k 1 + cot2 δl 4π Γ 2 /4 . (2 l + 1) k2 (E − E0 )2 + Γ 2 /4 (7. 6. and lifetime τ = ¯ /Γ (see Sect. of energy E0 .17).124)
Thus. Γ Recall. We can interpret the BreitWigner formula as describing the absorption of an incident particle to form a metastable state.123) cot δl (E) = − (E − E0 ) + · · · .9 Resonances
7 SCATTERING THEORY
we obtain
2 (7. σl (7.125)
This is the famous BreitWigner formula. The variation of the partial crosssection σl with the incident energy has the form of a classical resonance curve. The quantity Γ is the width of the resonance (in energy).80). h
190
. that the contribution of the lth partial wave to the scattering crosssection is σl = 4π 1 4π .