1

I.

SYLLABUS AND INTRODUCTION

Let us start with a brief overview of the items that will (hopefully) be covered in this course and to give a guideline
of what we are trying to learn.
Quantum mechanics 660 and 661 are advanced quantum mechanics courses designed for graduate students. The
courses will be treated as a one-year course. It will be assumed that students have already some background in
quantum mechanics (the concepts of waves, quantization, expectation values, etc.). An advised introductory textbook is Introduction to quantum mechanics by Griffiths. A number of things will, however, be repeated albeit in a
more mathematical fashion. Also some basic knowledge in Fourier analysis, differential equations and linear algebra
(calculation of eigenvalues and eigenstates) will be expected. All the information you need to know are in the lecture
notes. However, the lecture notes are relatively concise and further reading is recommended. The book that was
used last year was Liboff, Introductory Quantum Mechanics, fourth edition. However, a lot of other books treat the
material of 660 in a similar fashion, so pick a book that you like best. Some of the material is not covered in Liboff
(the older editions dont have relativistic quantum mechanics and quite a bit of the material of 661 is not in there).
The course is developed from a physicists point of view. Some concepts and mathematical techniques will be introduced along the way (such as matrix techniques) under the assumption that most of you are already somewhat familiar
with them (after all, there are also 3 and 4 hundred level quantum mechanics courses). For a more comprehensive
view of those things, I will refer to the textbook. The subjects that will be covered are
History, Plancks quantum hypothesis, photoelectric effect,specific heat, atomic physics.
The introduction is somewhat different from what is usually given in textbooks. However, since most of you
have already followed a quantum mechanics course and since this is a graduate course, I think it is interesting to
do it this way. Quite some time will be devoted to Plancks hypothesis. The physical language is quite different
from that of the rest of the course (it is more thermodynamics than quantum mechanics). Some of these things
are treated in the exercizes of Chapter 2 of Liboff. Other things will be mentioned more briefly.
Wave mechanics, Schrodinger equation, probability density, free particles, Heisenbergs uncertainty principle.
This section tries to show some of the thinking that led to the Schrodinger equation. As soon as you leave the
idea of a particle as a billiard ball behind you, a number of other things follow directly, such as Heisenbergs
uncertainty principle.
Harmonic oscillator, Schrodinger equation approach, Solution using operators, Operators and wavefunctions,
Correspondence principle.
Although most of you have seen the harmonic oscillator already, it is an important example, since the number
of exactly solvable models in quantum mechanics is limited. Try to pay special attention to the two approaches
to attack this problem. The Schrodinger approach, solving the differential equation for the wavefunction, and
the algebraic or operator approach due to Heisenberg, Jordan, and Born. Although the Schr
odinger approach
is more intuitive, the operator approach will become more important as we go on, since often we do not really
care that much about the wavefunction, but we are more interested in the spectral properties (energies) Of the
problem.
The Hydrogen atom, Legendre polynomials, angular momentum, matrix notation.
Again pay attention to the dual approach to this problem. The Schrodinger approach (solving the wavefunction)
and the operator approach (angular momentum). This involves quite a bit of mathematics (solving differential
equations). Here we will only deal with the potential from the nucleus, leading to the orbitals of the atom.
Other aspects follow in the next section.
Relativistic quantum mechanics, Klein-Gordan equation, Dirac equation, spin-orbit coupling.
This section introduces the Dirac equation and its consequences for the problems that we are interested in:
the introduction of spin, relativistic correction to the Hydrogen atom (spin-orbit coupling, lowest order kinetic
corrections). We could introduce (which would have been simpler) these aspects in a phenomenological or semi
quantum/classical approach, but this is more fun and avoids having to treat relativistic quantum mechanics as
some appendix. Note that we are not really interested in things such as time dilation, length contraction, etc.
Perturbation theory and absorption and emission of photons, time-independent perturbation theory,
time-dependent perturbation theory, interaction of radiation and atom
To be able to describe the emission and absortion of photons by an atom, we discuss perturbation theory. This
will be done in some detail because of its relevance to later problems.
Many-electron atom Paulis exclusion principle, the Helium atom, periodic table, Coulomb interactions,
addition of angular momenta.

FIG. 1: Left: The radiation spectrum of the sun at the top of the atmosphere (yellow) and at sea level (red) compared with
a blackbody spectrum for 5250 C. Right: Max Planck in 1901, around the time he developed the law now named after him.
Source: wikipedia

Having solved the atom for one electron, we would like to fill it with more electrons. First, we need to discuss a
peculiar aspect of electrons (or fermions in general), namely the exclusion principle. We start with the simplest
many-electron atom, Helium. We soon discover that electron-electron interaction have a strong influence and
see that things are not as simple as suggested by the Hydrogen atom solution. We continue with a more general
discussion of Coulomb interaction between electrons and how to couple the electron wavefunction in a smart
way (LS-coupling).
Molecules and solids, Hydrogen molecule, solids, (nearly) free-electron model.
We then continue by taking two atoms amd bringing them close together, thereby forming a molecule. If we
continue adding atoms, we can form a one-dimensional solid, this is known as a tight-binding approximation.
We can also approach it from the other side, by viewing a solid as a box of free electrons. The periodic potential
has a only a small, but important, effect.
Second quantization, the Coulomb gas.
We have a look at the ground-state energy of the electron gas and calculate it in the Hartree-Fock limit. We
also see what the effect are of screening.
II.

A.

OLD QUANTUM MECHANICS

Plancks quantum hypothesis

The quantum theory started with the blackbody radiation. All matter emits electromagnetic radiation when its
temperature is above abolute zero. The radiation results from the conversion of thermal energy into electromagnetic
energy. A black body is a systems that absorbs all incident radiation. On the other hand, this also makes it the
best possible emitter. It is hard to find a perfect black body, but, for example, graphite, is a good approximation.
However, even when a system is not a perfect black body, the basic phenomena remains the same. The sun can, for
example, be described to a good approximation as a black body, see Fig. 1. The spectral distribution has a maximum
and decays for small and large wavelengths. For the sun which has an effective temperature of around 5500 K, the
maximum lies in the visible region of the electromagnetic spectrum and is inherently related to the fact that the sun
is white (and not yellow or even yellowish). Before Max Planck took up the subject there was already a surprising

3
Intensity Harb. un.L
0.25
0.20
0.15
0.10
0.05
E HeVL
0

FIG. 2: The blue line shows the Planck distribution as a function of energy for T = 5523 K. The brown line shows Wiens
distribution curve. The magenta line is the Rayleigh-Jeans law.

amount of understanding of the spectral distribution predominantly due to Wilhelm Wien (1864-1928). For his work,
he received the Nobel prize for Physics in 1911. Using general thermodynamic arguments, Wien showed in 1893 that
the energy dependence is given by

u() = 3 f ( ),
T

(1)

where = 2 and f is a function that only depends on the ratio /T , where T is the temperature. The function
f ( T ) cannot be derived from these considerations, but it can only be a function of /T or T when expressed in
the wavelength . The consequence of that is that the maximum occurs at a particular value of T /. Therefore, if
the temperature changes the maximum occurs at a different wavelength. The human body can also be viewed as a
far-from-perfect black body. The maximum occurs for a wavelength
max,human =

max,Sun TSun
525 109 5500
=
= 9 m,
Thuman
310

(2)

which is in the far infrared. Obviously, we would like to understand that the function f ( T ) is. Wien made an
educated guess that f ( T ) = exp(a/T ), where a is a constant. This dependence was obviously inspired by the
Boltzmann factor and is a surprisingly good guess, see Fig. 2. However, there is something strange about using the
Boltzmann factor. The factor exp(E/kB T ), where kB is Boltzmanns constant, implies that there is some finite
energy associated with exciting electromagnetic radiation. This is in complete contradiction with the classical way of
thinking. The energy density of electro-magnetic radiation is given by 12 0 E2 + 21 0 B2 , and therefore determined by
the magnitudes of the electric and magnetic fields. Since we can take any possible value of these fields, any possible
value of of the energy density is possible.
An attempt to derive a theory for heat radiation was made by Rayleigh and Jeans. We can assume that a black
body is an electromagnetic cavity. Inside the cavity the radiation oscillates between the walls. In the x direction,
the electromagnetic modes are determined by sin kx x which has to be zero at x = 0 and x = L, i.e., the sides of the
cavity. This means that kx L = nx , with nx = 1, 2, 3, . . . . This implies that
(kx2 + ky2 + kz2 )L2 = k 2 L2 =

L
c

2

= (n2x + n2y + n2z ) 2 .

(3)

Note that the modes in the cavity are quantized. However, this is something that is understood from classical
wavemechanics and already occurs when considering standing waves in a string with fixed boundaries. This does not
cause a quantization of the energy. We can compare this with a simple one-dimensional harmonic oscillator. The total
energy of an oscillator is given by 12 m 2 x20 . Even though there is only one mode with frequency , this mode can
have any arbitrary energy, because the amplitude of the oscillation x0 can assume any energy. From the equipartition
theorem in thermodynamics, this energy can be directly related to the temperature, since, in thermal equilibrium,
each independent degree of freedom can be associated with an energy 21 kB T . An oscillator has two degrees of freedom

4
(position and momentum) and therefore has an average energy E = kB T . Note, that the amplitude is not the same.
High-frequency oscillators have a smaller amplitude than low-frequency ones. The argument for electro-magnetic
cavity modes, which also have two degrees of freedom (the electric and magnetic parts), is very similar. So far, there
does not appear to be too much of a problem, until you realize that the number of modes is infinite. If each of the
modes has an energy kB T , the total energy would be infinite as well. Let us have a look at the number of modes in
more detail. The density of modes can be found from Eqn. (3),
 3
2 L2
L
1
L
2
2 2
2
2
=

n
=

2
4n
dn
=

d
=
V
g()d

g()
=
(4)
c2
c
8
c
c3 2
with n2 = n2x + n2y + n2z . The factor 2 comes from the two different polarization vectors of the light. The factor
comes from taking only the positive k values. The Rayleigh-Jeans law can now be derived as

1
8

2
kB T.
(5)
c3 2
Note that the Rayleigh-Jeans equation has a temperature dependence but just increases as a function of , see Fig. 2.
Integrating over would give an infinite internal energy in our cavity. This was later called the ultraviolet catastrophe.
Furthermore, it also deviates from Wiens suggestion at low energies.
This was more or less the state of affairs in 1900, when Max Planck (1858-1947) entered the scene. Planck worked
most of his life and his derivation is based on entropy, effectively obtaining an interpolation between the RayleighJeans law and Wiens expression, which provide the proper limits in the low and high frequency regions, respectively.
Later on, Planck realized that the result can be interpreted as a quantization of the energy in quanta of ~, where, for
clarity we use the conventional notation of Plancks constant. This implies that the energy of an harmonic oscillator
can be given by E = n~, where n = 0, 1, 2, . This looks rather different from the classical result E = 21 m 2 x20 .
There are therefore two aspects in the expression E = n~. First, the energy is directly related to the frequency.
Second, the energy is quantized. Let us consider the former statement first. Let us assume that the energy can be
varied continuously, i.e. E = a, with a 0. Since E is proportional to , there is an equal number of states in
each energy interval of width dE. This allows us to easily calculate the average energy. The probability of finding an
oscillator with a certain energy is determined by the Boltzmann factor exp(E) where we take = 1/kB T . The
average energy can be calculated by integrating the energies multiplied by the probability of finding that energy, and
in the end dividing by the total probability. Let us first determine the total probability:
Z

1
1
dEeE = eE 0 = = kB T
Z=
(6)

0
u() = g()E =

The average energy is now

1
E=
Z

dEEe

1 Z

=
=
Z

 
1
1
= = kB T.

(7)

This is the classical equipartition effect and would directly reproduce the Rayleigh-Jeans result. Therefore, although
E is an interesting finding, it does not solve the ultraviolet catastrophe.
Now let us see what happens if we quantize the energy. In this case, we can no longer integrate over the energies,
but we have to perform a summation. Let us again start with the total probability,
Z=

n=0

The average energy is then

E=

exp(n~) =

1
.
1 exp(~)

1 Z
ln Z

~
~
=
=
ln(1 exp(~)) =
exp(~) =
.
Z

1 exp(~)
exp(~) 1

(8)

(9)

Let us consider the limits of small and large frequencies. For small frequencies, ~ kB T , and we find
E=

~
~
1

= = kB T.
=
exp(~) 1
1 + ~ 1

(10)

Therefore, for small frequencies, the fact that the energy is quantized as n~ is irrelevant and we obtain the classical
equipartition result. For ~ kB T , Plancks formula reduces to
E=

= ~ exp(~),
exp(~) 1

(11)

5
Maximum kinetic energy HeVL
2.0
1.5
1.0
0.5
0.0

EnergyH eVL

-0.5
-1.0
FIG. 3: The maximum kinetic energy of electrons emitted from a solid by electro-magnetic radiation. The work function is
=3 eV. Note that no electrons are observed for photon energies less than 3 eV.

which is equivalent to Wiens suggestion. Note that the average energy E decreases exponentially for larger frequencies.
This effectively solves the ultraviolet catastrophe. The energy distribution is now given by
u() = g()E =

2
~
.
c3 2 exp(~) 1

(12)

Although the result is revolutionary, it took some time to gain momentum. First, Planck did not set out to solve
the ultraviolet catastrophe, but was mainly looking for an effective formula to describe the black-body radiation.
Secondly, there are some subtle arguments about the applicability of the equipartition theorem for a cavity since
thermal equilibrium is not easily established due to the lack of coupling between the different frequency modes.
Thirdly, Planck did not quantize the electromagnetic field, but was more thinking about microscopic dipole oscillators.
B.

Photoelectric effect

The identification of quantization with the electro-magnetic field was done in 1905 by Albert Einstein (1879-1955).
Einstein proposed that the energy can only be transferred in discrete steps of ~. Applying this to a solid gives the
very simple expression
Ekin = ~ .

(13)

The explanation of the photoelectric effect was the primary reason for Einsteins Nobel prize in 1921. Here is the
work function of a solid, i.e., the minimum energy required to remove an electron. The maximum kinetic energy is
plotted in Fig. 13. This is independent of the intensity of the light. This is in complete contradiction with classical
theories where the amplitude of the incoming light would play an important role in determining the kinetic energy

FIG. 4: The specific heat for an Einstein oscillator with an energy = ~. (source wikipedia)

FIG. 5: Arnold Sommerfeld and Niels Bohr, key figures in the establishment of the old quantum mechanics.

of the outgoing electron (in a fashion similar that a greater amplitude of your golf swing gives a larger kinetic energy
to the golf ball). Obviously, the statement of Einstein would have been pretty obvious if experimental data of the
type in Fig. 13 existed. However, Einsteins claims were not experimentally verified until 1914 by Robert Andrews
Millikan (1868-1953), which was partially the reason for his Nobel prize in 1923.
C.

Specific heat

Up to 1910, quantum physics was mainly limited to the blackbody effect and the photoelectric effect. Its impact
on the rest of physics was still rather limited. An important contribution was made by Einstein on the specific heat
of solids. Classically this is expected to follow the Dulong-Petit law, i.e., each atom contributes 3kB T (or 3RT per
mole) to the total energy E, i.e., kB T per degree of freedom (assuming a three-dimensional system). The specific
heat is the quantity that tells how much the total energy of the system changes when the temperature is raised. This
gives a specific heat of
Cv =

E
= 3R.
T

(14)

Einstein recognized that each degree of freedom is governed by the same statistics as the oscillators in the blackbody
radiation problem, giving an average energy of

E=

3N ~
exp k~
1
BT

3R

Cv =

~
kB T

2

exp k~
BT

(exp k~
1)2
BT

(15)

which is plotted in Fig. (4). Note that this reduces to the Dulong-Petit result for high temperatures. This was one of
the first applications of quantum mechanics in solid-state physics and chemistry. This convinced the physical chemist
Walter Nernst of the importance of quantum theory. He was responsible for convincing the Belgian industrialist
Ernest Solvay to organize the famous conferences that bear his name.
D.

Bohr-Sommerfeld model

Additional concepts on quantization were introduced by Niels Bohr (1885-1962) and Arnold Sommerfeld (1868-1951)
in the period 1913-1920. The idea is essentially that classical physics is still obeyed, but that certain motion are not
allowed. Although the theory is now superseded by wave mechanics, it is interesting to have a look into how one tried
to grapple with the idea of quantization for about two decades. The quantization conditions were somewhat heuristic
and based on the notion that the classic action integral can only assume certain value. For example, in one dimension
Z
px dx = nh
(16)

7
Let us consider the case of an harmonic oscillator. The expression for the energy can be rewritten as an ellips in the
x,px plane:
H=
with a =

p2x
1
+ m 2 x2 = E
2m 2

p2x
x2
+ 2 = 1,
2
a
b

(17)

2mE and b = 2E/m 2 . The action integral is the area of the ellips and should be proportional to nh,
Z
2
pdx = ab =
E = nh E = n~,
(18)

giving the right relation between E and .

The most impressive result were however obtained by Bohrs treatment of the hydrogen atom for which he received
the Nobel prize in 1922 (note that this is before the development of the wavefunction formalism of quantum mechanics). Strong evidence of nonclassical behavior was observed in atomic spectra. Although the spectroscopic data
was already available in the late 19th century, it had a strong impact on quantum mechanics from 1910 onwards.
Phenomenologically, it was already established that the wavelengths of the spectral lines could be related to each other
by rational numbers. It was Balmer, a Swiss school teacher with an interest in numerology who noticed that the first
9
four wavelengths could are proportional to each other by 95 , 34 , 25
21 , 8 . After some playing around, he noticed that this
25 36
2
2
2
could also be written as 59 , 16
12 , 21 , 32 or m /(m 2 ), which reproduces the twelve known Hydrogen lines. This was
1885. Much more spectral information was obtained by, for example, Rydberg. After some attempts by Thompson
and others, it was Rutherford who clearly established the model of the atom as a relatively small nucleus with a thin
cloud of electrons turning around it. The stability of the electrons in orbits around the nucleus is in contradiction
with classical electrodynamics, which states that the electron would emit radiation in its orbit and would therefore
and up on the nucleus. This was solved in a some important papers by Bohr in 1913, who wrote down the conditions
for stationary states by relating it to Plancks constant (one of the first uses of quantum mechanics not related to
oscillators).
The quantization condition states
Z
pdq = nh,
(19)
where q is now a generalized coordinate. For an electron in an orbit of radius r, we can write q = r, where is the
angular coordinate.
Z 2
Z 2
prd =
Ld = 2L = nh L = n~,
(20)
0

where the magnitude of the angular momentum is L = |r p|. The equation of motion is then
F =

mv 2
Ze2
= 20
r
r

with

e
e0 =
,
40

(21)

with Z the number of protons in the nucleus. Solving this for v and r gives
vn =

Ze20
n~

and

rn =

n2 ~2
.
Z mne20

(22)

The energy of the state is then

En =

1
Ze20
Z 2 me40
Z2
mvn2
= 2
= 13.6 2 eV.
2
2
rn
n 2~
n

(23)

The emitted or absorbed photon energy is related to the energy of the stationary states
~mn = Em En = R(

1
1
2 ),
n2
m

with

m > n 1,

(24)

The value of 13.6 eV reproduced the experimentally observed Rydberg constant. An impressive result indeed.
Despite some successes of the Bohr-Sommerfeld model, ultimately it turned out to be deficient. First, there was no
standard way of obtaining results, but every problem had to be considered separately. Second, it fails to reproduce
many features, in particular for the model of the hydrogen atom. For example, the lowest has in fact L = 0 and not
L = ~, it cannot describe fine structure in the atomic levels, or the Zeeman effect.

FIG. 6: By assuming that the wavelength of a particle is proportional to its inverse momentum h/p, de Broglie was able to
show that the Bohr quantization condition corresponded to having a standing wave in the orbit of an electron.

FIG. 7: Erwin Schr

odinger developed one of the fundamental equations of quantum mechanics describing matter in terms of
waves.

III.

WAVE MECHANICS

Up to the early 1920s, quantum mechanics had no solid foundations. Applications were based on applying good
physical intuition to the quantization of various problems. This changed when more emphasis was placed on the
dual character of particles and radiation, both particle and wave like. The particle character of radiation was clearly
demonstrated in the photoelectric effect. It was Louis de Broglie (1892-1987) who introduced in his 1924 Ph.D. thesis
the concepts the wave character of particles, identifying the wavelength of a particle with its momentum
=

h
p

or

p = ~k,

(25)

where k = 2/ is the wavenumber. Bohrs quantization of angular momentum then corresponds to L = rp = n~,
giving 2r = n. Implying that an integral number of wavelengths should fit on an orbit, i.e., a standing wave on a
circle. This provides additional understanding of the concept of stationary orbits. If the radius of an electron orbit
would change, this would result in the destructive interference of the wave. Hence, the radius can only assume very
particular values.
The wave character of particles was confirmed in 1927 by Clinton Davisson (1881-1958) and Lester Germer (18961971) at Bell Labs and, independently, by George Paget Thomson (1892-1975, the son of Nobel laureate J. J. Thomson)
at the University of Aberdeen. Essentially, they performed diffraction experiments, not with X-rays, but with an
electron beam. The diffraction pattern could be explained by using the Bragg condition for X-rays but with the de
Broglie wavelength. The 1937 Nobel prize in physics was awarded to Davisson and Thomson for their work on electron
diffraction.

9
A.

Schr
odinger equation

The wave character of the particles as investigated by a number of physicists, most notably Erwin Schr
odinger (18871961). An obvious starting point is the wave differential equation for particles
2 (r, t)

1 2 (r, t)
= 0,
vp2 t2

(26)

where vp is the phase velocity. The nabla operator is Cartesian coordinates is given by
=

2
2
2
+
+
z and 2 =
x
y
+
+
x
y
z
x2
y 2
z 2

(27)

For electro-magnetic waves, this velocity is equal to the speed of light, but obvously this is not correct for particles.
Solutions to this equation are of the form
(r, t) = aei(krt) .

(28)

Using the quantum mechanical identities E = ~ and p = ~k from Planck and de Broglie, respectively, we can also
write the solution as
(r, t) = aei(prEt)/~ .

(29)

Taking the time derivative in the wave equation gives

2

E2
= 0,
~2 vp2

(30)

The unknown quantity in the above equation is the phase velocity vp = /k. We are very used to electromagnetic
waves, where
E = ~

pc = ~kc = ~

c = /k.

(31)

Therefore the phase velocity is equal to the real velocity (the speed of light). This is not true for particles. Here
r
2(E U (r))
1
2
mv = E U (r)

v=
,
(32)
2
m
where U (r) is the potential energy. For the wavevector we have
k=
For the phase velocity we then have
vp =

mv
1p
2m(~ U ).
=
~
~

~
E
= p
.
= p
k
2m(~ U )
2m(E U )

(33)

(34)

This is completely different from the velocity, with vvp = E/m. For electromagnetic waves, we can have comparable
changes in phase velocity when the waves travel through a different medium. In vacuum c2 = 0 0 . In a medium,
this changes to r r 0 0 = n2 /c2 . Moving though a potential landscape is therefore comparable to moving through a
medium with a different refractive index. Inserting (34) in Eq. (30) gives the time-independent Schr
odinger equation.
2 +

2m(E U )
=0
~2

or

~2 2
+ U = E.
2m

(35)

The strange thing is that, whereas in Eq. (30) the energy appears in quadratic form, it now appears linearly in the
equation. Since i~/t = E can put the time-dependent derivative back into the Schrodinger equation giving

~2 2
(r, t)
(r, t) + U (r)(r, t) = i~
2m
t

(36)

10
An alternative way to look at the Schrodinger equation is that the usual scalar quantities are now replaced by
operators:
p i~

and

E i~

.
t

(37)

For particles, we start from the usual expression for the energy
p2
+ U (r) = E.
2m
Inserting the operators and making them work on a wavefunction leads to the Schrodinger equation

~2 2
+ U (r) = i~
.
2m
t
This can be written in a more condensed form by defining the Hamiltonian

~2 2
+ U (r).
2m
The time-independent Schrodinger equation is then simply
H=

H = E.

(38)

(39)

(40)

(41)

For the development of the equation named after him, Schrodinger received the Nobel prize in 1933.
Example: bound states in a delta function
Let us now look at a particular example of a potential. You are probably familiar with a particle in a box, so we will
consider the one-dimensional example of bound states in the presence of a delta function potential V (x) = (x)
with >, where the delta function is defined as

0,
if x 6= 0
(x) =
,
(42)
, if x = 0
where in addition we have
Z

(x)dx = 1.

(43)

The Schrodinger equation can now be written down as

~2 d2
(x)(x) = E.
(44)
2m dx2
In the absence of a potential, we only expect solutions for positive energies, since the kinetic energy is always positive.
The solutions are in then plane waves. For positive energies, we still find plane waves, but the presence of the local
scattering center causes a phase shift in the plane waves at the origin. We will not study this part in this example.
Since the potential is attractive, we also have solutions for negative energies. These are bound states. Let us first
look at the Schr
odinger equation for x 6= 0

~2 d2
= |E|.
2m dx2
Since the energy is negative, the solution are not running waves but exponentials. Using the definition
p
2m|E|
=
,
~
the solutions can be written as

(x) = Aex + Bex .

(45)

(46)

(47)

Neither of them are solutions in the absence of a potential since the potentials blow up at either plus or minus infinity.
However, since there is a potential at x = 0, the solution can be written as

Aex , if x < 0
(x) =
(48)
Bex , if x > 0
So what do we do at the x = 0. To solve this, we have to look at the two boundary conditions

11
is always continuous.
d/dx is continuous except at the points where the potential is infinite.
The first point is very easy and simply implies that A = B. This follows also directly from the symmetry of the
problem, since there is no reason why the left side should not be the mirror image of the right side. The second point
is more tricky, especially since we are dealing with a situation where the potential is infinite. Let us first see why
this is the case. The idea is to integrate the Schrodinger equation around a particular point. Let us focus on x = 0.
Intgrating gives
Z
Z
Z
~2
d2 (x)
dx
+
dx
V
(x)(x)
=
E
dx (x),
(49)

2m
dx2

where we take the limit 0. The first integral is of course d/dx. The integral on the right-hand side goes to zero
if 0. So we find for the change in derivative




Z
d(x)
d(x) 
d(x) 
2m

= 2
dx V (x)(x).
(50)
dx
dx 
dx 
~

The right-hand side goes to zero unless V (x) goes to infinity. For our -function potential we find that the change in
derivative is given by


d(x)
2m
= 2 (0).
(51)

dx
~
evaluting gives

which gives


2m
Aex 0 Aex |0 = 2A = 2 A
~
=

m
.
~2

(52)

(53)

The energies are then given by

E=

~2 2
m2
= 2 .
2m
2~

To find the final result for the wavefunction, we just need to normalize it
Z
Z
|A|2
2
2
dx|(x)| = 2|A|
dxe2x =
=1

(54)

(55)

or
A=

m
.
~

The -function well has, regardless of the strength , one bound state given by



m
m|x|
m2
(x) =
exp
;
E= 2
2
~
~
2~

(56)

(57)

We return to the problem of a bound state when discussing matrix algebra.

B.

Probability density

The discovery of the dual character of particles led to several philosophical problems on how to interpret the waves
that follow from the Schrodinger equation. Max Born (1882-1970) emphasized that we should look at (r) = |(r)|2

12

FIG. 8: Max Born developed the concepts of the interpretation of the wavefunction in terms of probability. An example is the
two-split experiment where quantum mechanics correctly predicts the probability distribution of electrons hitting a wall after
passing through two slits. The right shows the buildup of the image by electrons hitting the detector over time.

as the probability of finding a particle. The Schrodinger equation can be seen as a diffusion equation for probability.
To see this one has to start from the Schrodinger equation in Eqn. (36) and its complex conjugate
i~

~2 2

=
+ U (r) .
t
2m

(58)

Multiplying the equations by and , respectively, and substracting gives

+
=
( 2 2 ).
t
t
2mi

(59)

( )
~
=
( ).
t
2mi

(60)

This can be written as

This can be written as

= J
t

with

J=

~
( ).
2mi

Integrating over the volume gives

Z
Z
Z
Z
Z

dr
=
dr = dr J = Jda =
[ (i~) + (i~ )]da,
t
t
2m S
S

(61)

(62)

where the last term gives the density going in and out of the surface S. Note that if we let the surface
go to infinity and
R

ensuring that 0 when r , then the right-hand side will be zero. This means that t
dr = 0. Therefore
if we had initially normalized the probability, it will remain normalized in time. Three decades after developing the
concepts of probability density, Max Born received the Nobel prize in 1954.
There is a wide variety of interpretations regarding the wavefunction in quantum mechanics. Here, we follow the
practical one, namely that in quantum mechanics there is only the wavefunction and the wavefunction correctly
predicts many experiments. Generally, there is no need to imagine a particle inside the wavefunction. For example,
quantum mechanics correctly predicts the probability distribution of a two-slit experiment see Fig. 8. Since particles
have a wave character, the wavefunction should show destructive and constructive interference, just as an optical wave
passing through two slits as explained by Thomas Young in the 1790s. This was more a thought experiment until it
was actually measured by Clauss Jonsson of the University of T
ubingen in 1961. Even more surprisingly, this even
works with objects such as a C60 (bucky balls) molecule with a diameter of 0.7 nm as was demonstrated in 1999.
This completely shatters our ideas of particles as little balls. Now there is an additional complicating factor of the
measurement. One does not observe the whole electron interference pattern at once. In fact, it requires many electrons

13
for the image to build up, see Fig. 8. However, practically the experiment is one electron passing through two slits
giving rise to an interference pattern. The measurement is the interaction of the electron with the detector which
is a very complicated quantum system consisting of many particles. This causes the collapse of the wavefunction in
a particular position. The problem of measurement and the transition from a wave to an electron hitting a detector
is very difficult. However, often one is only interested in the experiment: one wants to calculate the interference
pattern of one electron passing through a slit. For example, when doing spectroscopy, one calculates a spectrum using
quantum mechanics and generally completely ignores the effect of the measurement.
An additional step to the two-slit experiment is the determination through which slit the electron passes. Here one
sees that the interference pattern disappears and that one obtains two classical probability distributions of electrons
passing through slits like particles, see Fig. 8. Again, the problem arises because of the measurement. In quantum
mechanics, the measurement is never simple, since the measurement is of the same order of magnitude as the
system. One cannot simply separate the system from the measurement. This is different from classical mechanics,
where a measurement is generally an extremely minor perturbance to the system. For example, the measurement of
the speed of a car with the use of radio waves. The situation would change drastically is speed was measured with
cannon balls. For example, the detection of whether an electron passes through a slit can be done by placing an
atom inside each slit. However, is this still a two-slit experiment? From a quantum mechanics point of view, this
is a system of an electron passing through two slits and while passing through the slits it can interact with atoms
positioned in the slits. To really understand what is going on, this is the system you need to calculate. But then it
changes from a one-particle problem (electron through two slits) to a many-particle problem, which is very difficult
to calculate. Since the problem of measurement is highly complicated, we will not discuss it in too much detail
since it just confuses things. The experiment (an electron passing through two slits) is beautiful enough in itself
without further complicating it with the concepts of measurement (and in particular with philosophical questions
about human involvement).
C.

Expectation values

Since particles are no longer little balls, but wavefunctions, it is impossible to say where they are in real space.
However, we can calculate the expectation value of the position. In one dimension, we have
Z
hxi = x|(x)|2 dx,
(63)
since |(x)|2 gives the probability of finding a particle at position x. However, how do we do this for more general
operators. Let us have a look at the momentum, which should be given by
Z
|(x)|2
hxi
=m x
dx.
(64)
hpx i = m
dt
t
Using Eq. (61) in one dimension, we can write the right-hand side as


Z
i~

hpx i =
x

dx.
2
x
x
x

(65)

Using an integration by parts gives

hpx i =

i~
2

Z 

dx,
x
x

(66)

using the fact that x/x = 1 and omit the integrand since the wavefunction () = 0. Now the second term on
the right-hand side is actually equivalent to the first term, which follows from another integration by parts. This gives


Z
Z
Z

hpx i = i~
dx = i~
dx = px dx,
(67)
x
x
where we identify px as the x-component of p = i~. Note that this follows directly from the Schr
odinger equation
which was derived from the wave equation. This justifies the heuristic assumption based on de Broglies principle in
Eq. (37). It is important to realize that the position of px in the integral is important since px is an operator and
the expectation value is given by
therefore affects the wavefunction. For a general operator A,
Z
= dr (r)A(r).

hAi
(68)

14
Another aspect that seems relatively trivial is that the expectation value is a real number. This will become of
importance later on when we start discussing Heisenbergs matrix algebra. Also, the notation for the expectation
value is a shorthand. More, generally, we should write
h|A|i.

hAi

(69)

There are many problems that involve more than one wavefunction, for example the free particle problem in the next
section. You may have notice that on the right-hand side, the appears without the r dependence. This is because
the expectation value does not depend on r, because of the integration over the spatial coordinates. Operators can
also have finite values between different wavefunctions
Z
n i = drn (r)A
n (r).
hn |A|
(70)
n i is called a matrix element. The use of
where the index n labels the different wavefunctions. In this case, hn |A|
matrix element or expectation value for the case n = n depends on the context in which it is used.
Some of you might be bothered that the momentum p is represented by a complex operator. Let us rewrite the
expectation value,


Z

dx,
(71)
hpx i =
i~
x
for the momentum in the x direction by performing an integration by parts


Z 
Z 

hpx i = i~ [ ]
i~
dx =
i~
dx
x
x

(72)

The integrand is zero if the wavefunction properly goes to zero in . Now if is an eigenfunction of px with
eigenvalue q, or px = q then from Eqs. (71) and (X D) it follows that



Z
Z 

hpx i =
i~
dx = q =
i~
dx = q ,
(73)
x
x
and therefore q = q which is only possible if q is real. So that is reassuring that the expectation value of the
momentum operator is a real number.
This can be generalized as follows. The operator A is the Hermitian conjugate of A if
Z
Z

(A ) dr = Adr.
(74)
The operator is Hermitian is A = A. From Eq. (73), we see that the momentum operator satisfies this condition.
Just as for the momentum operator, if is an eigenfunction of A with eigenvalues q, we find that
Z
Z
Adr = q = (A) dr = q
(75)
and therefore that q = q and the eigenvalues q must be real for any Hermitian operator.
It also has important consequences for the eigenstates. Suppose we have two different eigenstates with different
eigenvalues
Aq = qq
If A is Hermitian, we have
Z

q Aq dr

=q

q q dr

and

Aq = q q .

(Aq ) q dr = q

The only way to satisfy this equation if q 6= q is to have

Z
q q dr = 0.

(76)

q q dr.

(77)

(78)

15
This means that the wavefunctions are orthogonal to each other. Unfortunately, this theorem does not help us when
eigenfunctions are degenerate. However, even it that case it is possible to make different eigenfunctions orthogonal to
each other.
For matrix elements between two wavefunctions i and j (that are not necessarily eigenfunctions) for a Hermitian
operator A = A , we find
Z

(79)
Aij = (hi |A|j i) =
dri Aj
Z
Z
=
drj A i = drj Ai
(80)
= hj |A|i i = Aji .

(81)

Note that this implies that Aii = Aii and therefore real. However, matrix elements between different states can be
complex and of the form Aij = a + ib and Aji = a ib where a and b are real numbers.
D.

Free particles

The simplest problem that can be addressed in quantum mechanics are free particles in the absence of extermal
potentials. The equation for a free particle moving in the x-direction is

~2 2 (r, t)
(r, t)
= i~
2
2m x
t

(82)

The solutions are

1
k (x, t) = eikxik t/~
L

with an energy

where L is a normalization length, such that the integral

dimensions giving eigenstates
1
k (r, t) = eikrik t/~
V

k =

~2 k 2
,
2m

(83)

dx = 1. This can be easily extended to three

with an energy

k =

~2 k2
,
2m

(84)

R
where V is now a normalization volume, such that the integral V dr = 1. The running waves obtained from the
differential equation are known as the eigenfunctions. We can also separate the position and time dependence in the
eigenfunctions
k (r, t) = k (r)eik t/~

with

1
k (r) = eikr ,
V

(85)

where the functions k (r) are eigenfunctions of the time-independent Schrodinger equation:

~2 2 (r)
= E(r)
2m x2

(86)

We see that for free particles, the eigenstates are determined by the wavevector k. If we want to maintain running
waves as eigenfunctions (as opposed to particles in a box where we take that the wavefunction is zero at the edges of
the box), then we can apply periodic boundary conditions which state that
eik(x+L) = eikx

eikL = 1

k=

2n
L

where n is an integer.

(87)

Or in three dimensions
k=

2
+ ny y
+ nz z),
(nx x
L

(88)

, y
, and
where x
z are the Cartesian unit vectors. The choice of a rectangular box might appear somewhat arbitrary.
However, we assume here that the normalization volume is large enough, such that the spacing between the levels

16
is unimportant for the problem that we are considering. Effectively, the particle should be free, that is we are not
looking at a problem where a particle is localized inside a box. The energy is now
 2
~2 2
(n2x + n2y + n2z ).
(89)
k =
2m L
We see that the energy and wavefunction are directly related to the quantum numbers nx , ny , and nz . These quantum
numbers are intimately linked to the momentum p = ~k. Therefore, the operator p = i~ is the relevant operator
for the problem of a free particle. The importance of the momentum operator is clear when we operate it on the
eigenfunctions in Eq. (84)


1
pk (r) = i~ eikr = ~kk (r).
(90)
V
The matrix elements of the momentum operator are given by
Z
hk |p|k i =
dr k (r)pk (r)
Z

1
= ~k
drei(kk )r = ~kk,k ,
V

(91)
(92)

where k,k is one for k = k and zero otherwise. The wavevectors k label all the eigenstates. The eigenfunctions are
obviously also eigenstates of the Hamiltonian


~2 2
1 ikr
e
= k k (r).
(93)
Hk (r) =
2m
V
Later, we will find that for other problems, different operators are relevant. For example, for the hydrogen atom,
the relevant operator is the angular momentum. The relevant operators are directly related to the symmetry of the
problem. We are dealing here with free particles. We now want to shift our coordinate system. However, since space
is translationally invariant, nothing should really change by that operation (the situation drastically changes in the
presence of a potential). In classical mechanics, a translation of the coordinates can be easily made by adding a
translation vector to the position, giving r = r + . In quantum mechanices, one is translating wavefunctions and
one needs to find the proper function U that performs this operation. Operating this on the wavefunction should
give
U (r) = (r + ).

(94)

The right-hand side can be expanded as

(r + ) = (r) + +

1
( )2 2 + = exp ( ) (x).
2!

(95)

We would like this transformation to be unitary. Unitary transformations essentially transform the system but
leaves the norms unchanged. Essentially we are going to a new coordinate system. A good example is a rotation
of the Cartesian coordinates system in three-dimensional space. Interesting things can be concluded if the unitary
transformation leaves the system unchanged. This says something about the symmetry of the problem. For example,
if the Hamiltonian does not change under a unitary transformation (H = U HU = H) then the eigenvalues also
remain unchanged. For example, a translation does not do anything with the eigenenergies of a free particle
Z
Z

drk (r)H k (r) =

drk (r)U HU k (r)
(96)
Z
=
drk (r + )Hk (r + ) = k
(97)
Now suppose we can write U = eiL , then it is easy to see that U is unitary

U U = eiL eiL = ei(LL

= 1,

(98)

if L is Hermitian (L = L) (For a translation this means that U translates the system in one direction and U does the
same translation, but now in the opposite direction). If we take the momentum p = i~, then U = exp( ~i p) is

17
unitary, since p is Hermitian. The invariance of the Hamiltonian under a certain unitary transformation has important
consequences, such as the commutation of H and U ,
U H = U U HU = U H.

(99)

This can also be rewritten as

HU U H = 0

[H, U ] = 0,

(100)

using the definition of the commutator of two operators

[A, B] = AB BA

(101)

From this, one can show that an eigenstate n of H with eigenenergy En

H(U n ) = U Hn = U En n = En (U n )

(102)

must also be an eigenstate of U since U n must be proportional to n since it is an eigenstate of H with eigenvalue
En . This directly implies that p commutes with H and that momentum is a good quantum number when a system
is translationally invariant. It also implies that the eigenfunctions of p are also eigenfunctions of H. Defining the
eigenvalue of p as ~k, we obtain the usual plane waves
pk (r) = ~kk (r)

i~k (r) = ~kk (r)

1
k (r) = eikr ,
V

(103)

where V is a normalization volume. So we see that the eigenfunctions of the momentum p are the same as the
eigenfunctions of the Schrodinger equation in the absence of a potential.
E.

Eigenfunctions and basis sets

The momentum p plays a special role when a system has translational symmetry. However, this is no longer the
case when the translational symmetry is broken. For example, when a local scattering potential U (r) is placed in the
system, momentum is no longer a good quantum number. An electron in a particular plane wave can scatter to a
diffent wavefunction. This implies that the matrix element
Z
hk |U |k i = dr k (r)U (r)k (r) 6= 0.
(104)
Although, the plane waves are not eigenfunctions of the system with the local potential, it can still be useful to use
the plane waves as a basis set and express the eigenfunctions in terms of plane waves. Since the plane waves form a
complete basis (there are an infinite number of them), one can express any function in terms of plane waves. This is
the underlying principle of Fourier series. One has to make a clear distinction between a basis set and eigenfunctions.
A basis set is a human choice, a decision to use certain functions to express the eigenfunctions. We shall see that
the eigenfunctions of the hydrogen atom are expressed in terms of Laguerre polynomials and spherical harmonics.
Obviously, the eigenfunctions are a basis set by themselves (again there are an infinite number of them). It is a very
convenient basis set since they are eigenstates of the potential of the nucleus. However, when things are added to the
Hamiltonian (say, an electric field or the spin-orbit coupling), the spherical harmonics are no longer eigenfunctions.
However, they are still a very convenient basis set. The choice of basis set can be very important. For example, trying
to describe the hydrogen atom in terms of plane waves is possible, but very inconvenient. Vice versa expressing plane
waves in terms of spherical harmonics is also possible. You might laugh and think that noone would want to do that,
but this is actually done when one wants to express the interaction of a electromagnetic plane wave with an atom.
To better understand the transitions for absorption and emission, it actually makes sense to express the plane wave
in terms of spherical harmonics and Bessel functions. We will return to the concepts of basis sets and eigenfunctions
when discussing the matrix mechanics developed by Heisenberg (1901-1976), Born, and Jordan (1902-1980).
F.

Wave packet

We found that the eigenfunctions for the Schrodinger equation in the absence of a potential are plane waves. That
means that although they have a very specific momentum, they are completely delocalized in space. This does not

18
necessarily correspond to our idea of a particle. So let us try to localize our particle by putting a Gaussian envelope
of width a around the probability distribution function. So, for t = 0,
1
x2
(r, 0) = p exp( 2 )eik0 x
4a
a 2

1
x2
P (x, 0) = (x, 0)(x, 0) = exp( 2 ).
2a
a 2

(105)

So our particle is nicely localized in space and we have given it the momentum k0 , we think. However, let us look at
the Fourier transform
s
Z
Z
1
1
2a
x2 i(k0 k)x
ikx
= exp(a2 (k k0 )2 ).
b(k) =
dx(x, 0)e
=
dx exp( 2 )e
(106)
4a
a(2)3/4
2
2
We thus find that the distribution is not a delta function in k space at k0 , but a Gaussian with width 1/2a around
k0 . So we have
xp = x~k = a~

1
~
= .
2a
2

(107)

Apparently it is difficult to localize a wavefunction both in momentum and real space. This will be shown more
generally below. This also implies that we cannot say that the particle has a momentum ~k0 . There is a probability
distribution of momenta. The momentum that we measure differs from one experiment to the other. We can, however,
speak of an average momentum. The expectation value of an operator A is defined as
Z

hAi = dx A.
(108)
= 1 P Ai , where Ai is the value of A in the ith experiment. For
So if we could do N experiments, for large N , hAi
i
N
the Gaussian wave packet, we have
Z
x
h
px i = ~ dx (k0 + i 2 ) = ~k0 .
(109)
2a
Let us look briefly at the time development of a Gaussian wave packet (this is treated in more detail in Liboff 6.1).
2 2
k
, it is more convenient to make the time development in k
Since each k value has a different energy E = ~ = ~2m
space and then make a Fourier transform back to real space
Z
1
dkb(k) exp(i(kx t))
(110)
(x, t) =
2
Using Eqn. (106), we find
(x, t) =

1
2

dk

dx (x , 0)eikx exp(i(kx t)) =

dx (x , 0)K(x, x , t),

(111)

with
K(x, x , t) =

1
2

dk exp(i(k(x x ) t))

(112)

describing the time development of the wavefunction. After integration (see Liboff 6.1), we find for the distribution
function at time t
s


1
(x vg t)2
t2 2

P (x, t) = exp
with
a
=
a
1
+
(113)
2a2
k04 a4
a 2
and the group velocity is vg = ~k0 /m. Therefore, the distribution remains Gaussian, but its width, given by a ,
increases in time.

19

FIG. 9: Heisenbergs uncertainty principle has entered the popular imagination, but in the end it is a property of waves and
operators.

G.

Heisenbergs uncertainty principle

We saw above with the Gaussian wavepacket that it is difficult to localize a particle both in real and momentum
space. This is a general aspect of waves and is well known to people familiar with Fourier analysis. A plane wave
(i.e., localized in momentum space) is delocalized
over the entire space, whereas a -function can be represented by
R
an integration over all k values (x) = 12 eikx dx. The bottom line is that it is impossible to localize a particle
without changes the derivatives with respect to x (and hence the momentum) and it is also impossible to reduce the
spread in momentum without delocalizing the wavefunction. This can be proven more generally. We know that

2
Z



I() =
dx x + ~
0
(114)
x 

for any . Thus


2 #




2 2  
I() =
dx |x| + ~(
x + x
)+ ~ 
.
x
x
x 

"

The term with can be evaluated by partial integration

Z
Z
Z

dx(
x + x
) = dxx ( ) = [x ] dx ,
x
x
x

(115)

(116)

where we choose the integration boundaries in such a way that the first term on the right-hand side is zero. Similarly,
we can write for the term with 2

 Z
Z

2
dx
=
dx 2 .
(117)
x x
x
x
This gives a total result
I() =
or

dx x2 ~

dx + 2 (i~)2

dx

2
,
x2

I() = hx2 i ~ + 2 hp2x i 0.

(118)

(119)

Since this quadratic equation in has no solutions, this means that the determinant should be less than zero,
~2 4hx2 ihp2x i 0

hx2 ihp2x i

~2
.
4

(120)

20
If we look at deviations from an average value, we have
p
x = hx2 i hxi2
and

p =

p
hp2x i hpx i2 .

(121)

We can always choose the wavefunction in such a way that hxi = hpi = 0. This gives us the result
xpx

~
,
2

(122)

which is known as the Heisenberg uncertainty principle.

Let us have a look again at the term with which apparently prevents us from making xpx = 0. We can rewrite
this slightly using
Z
Z

dx
x = [ x] dx (x),
(123)
x
x
giving, using Eqn. (116),
~



Z

x = ~ dx = ~.
dx x
x x

(124)

This can also be written as


 


i~
(x) = i~.
x i~
x
x

(125)

xpx px x = i~

(126)

or
or

[x, px ] = i~,

with the commutator defined as [A, B] = AB BA. Note, the intricate relationship between the commutation
relationship and the Heisenberg uncertainty principle. If two operators do not commute we cannot have A and B
zero at the same time.
IV.

MATRIX MECHANICS

Matrix mechanics was the first correct representation of quantum mechanics. It was developed by Werner Heisenberg
together with Max Born and Pacual Jordan. At first, it was not immediately accepted since matrix mechanics were new
concepts in physics at the time. The later introduction of wave mechanics by Schrodinger was generally preferred, since
the concept of waves was much more familiar. However, as we shall see, the two formulations are in fact equivalent.
A.

Matrix mechanics: Two-level system

Let us consider a simple system consisting of only two levels. Examples of such systems are two spin levels (spin up
and spin down), the hydrogen molecule when only the 1s levels are considered, etc. Let us consider an electron hopping
between two levels. Now let us assume we know the wavefunctions 1 (r) and 2 (r) of the two levels. Knowing the
wavefunctions and the Hamiltonian allows us to calculate the energy when an electron is in a particular wavefunction
Z
i = hi |H|i i = dr i (r)Hi (r) for i = 1, 2
(127)
Now if there was no coupling between states 1 and 2, then the wavefunctions 1 and 2 are the eigenfunctions of the
system, with eigenenergies 1 and 2 , respectively. However, if there is a coupling
Z
t = h2 |H|1 i = dr 2 (r)H1 (r)
(128)
Z

= h1 |H|2 i =
dr 1 (r)H2 (r) ,
(129)

21
then this is no longer correct. However, we can use the wavefunctions as a basis to express the eigenstates of the
system. The eigenfunctions can be expressed as a linear combination:
(r) = a1 (r) + b2 (r).

(130)

The Schrodinger equation reads now

H(r) = E(r)

H[a1 (r) + b2 (r)] = E[a1 (r) + b2 (r)].

We can multiply this equation from the left by 1 (r) and integrate over r. This gives
Z
Z
Z
Z

a dr 1 (r)H1 (r) + b dr 1 (r)H2 (r) = Ea dr 1 (r)1 (r) + Eb dr 1 (r)2 (r).

(131)

(132)

Using the orthonormality of the wavefunctions, we find

ah1 |H|1 i + bh1 |H|2 i = Ea.

(133)

ah2 |H|1 i + bh2 |H|2 i = Eb.

(134)

Similarly, for multiplication with 2 (r),

We can combine these two equation by making use of matrices


 
 
h1 |H|1 i h1 |H|2 i
a
a
=E
h2 |H|1 i h2 |H|2 i
b
b
Replacing the matrix elements by the symbols from Eqs. (127) and (129) gives a Hamiltonian

 

H11 H12
1 t
H=
=
,
H21 H22
t 2

(135)

(136)

where we use non-italic letters to indicate matrices and Hij = hi |H|j i. An important thing to not is that the
Hamiltonian is completely independent of the spatial coordinates, since all the r dependence has been integrated out
in deriving the parameters i and t. Therefore, we managed to recast the whole problem into something that only
depends on numbers.
In Eq. (135), the wavefunction is written as a 2 1 column
 
 
 
a
1
0
=a
+b
.
(137)
b
0
1
Again, there is no r dependence. However, this can be easily obtained if we want to since the whole problem is
expressed in the two basis functions 1 (r) and 2 (r). Once we know how much is in each of the basis functions (given
by the coefficients a and b, we can always obtain the spatial dependence of the wavefunction. However, in many
problems we are only interested in the coefficients (or the occupations |a|2 and |b|2 ) and not even in the exact spatial
dependence of the wavefunction. Especially, for complex problems, the spatial dependence of the wavefunction can
sometimes be rather uninformative. Essentially, the wavefunction is vector in a two-dimensional space, expressed in
two basis vectors. To bring out the character of the different components of the basis, we can make the following
identifications
 
 
1
0
|1 i
and |2 i
.
(138)
0
1
The wavefunction is now expressed as
|i = a|1 i + b|2 i.

(139)

We can use this notation also to express other properties of the wavefunction. For example, the normalization of a
wavefunction expressed in two basis functions is given by
Z
h|i = dr (r)(r),
(140)

22
following the notation in Eq. (70). Using Eq. (130), we find
Z
Z
Z
Z
2

h|i = |a|
dr1 (r)1 (r) + |b|
dr2 (r)2 (r) + a b dr1 (r)2 (r) + b a dr2 (r)1 (r),

(141)

where the latter two are zero due to the orthogonality of the basis functions. The first to integral are unity, and we
therefore have
h|i = |a|2 + |b|2 .

(142)

So far, we have not really specificied what h| is. However, let us have a look at what is known in mathematics as
the conjugate transposed or Hermitian conjugate. Given a matrix A, with matrix elements Aij , the matrix elements
of the Hermitian conjugate matrix A are given by
Aij = Aji .

(143)

For the vector h|, we then find

h| = (|i) =

a
b

= (a b ) (a , b ),

(144)

where the comma in the conjugate of a vector is conventional. Note that the conjugate of the vector |i (a 2 1
matrix) is a 1 2 matrix. Let us now look at matrix multiplication. The matrix elements of the product of a matrix
A of dimension m l and a matrix B of dimension l n are
(AB)ij =

l
X

Aik Bkj .

(145)

k=1

Effectively, we are taking the inner product of the rows of A with the columns of B. The resulting matrix is of
dimensions AB is a m n. Interpreting, h|i as a matrix multiplication of h| and |i, we find
 
a
h|i = (a , b )
= |a|2 + |b|2 ,
(146)
b
which is the same result as Eq. (142). The multiplication of a 1 2 matrix and a 2 1 matrix gives a 1 1 matrix,
which is essentially a number. The orthogonality of the basis function is directly satisfied
 
 
1
1
h1 |1 i = (1, 0)
= 1 and
h2 |1 i = (0, 1)
=0
(147)
0
0
B.

Eigenvalues and eigenvectors

Solving the Schrodinger equation now becomes equivalent to finding the solution to the equation


1 t
H|i = E|i

|i = E|i,
t 2

(148)

where for brevity we assume that t is a real number. Thats nice, but what do we do now? How do we find the
eigenvalues E? Let us define the unit matrix


1 0
I2 =
,
(149)
0 1
where the subscript 2 indicates the dimension. Using this we can rewrite Eq. (148) as
(H E I2 )|i = 0,

(150)

where H E I2 is just a new matrix. Now let us assume we can find and inverse of this matrix such that
(H E I2 )1 (H E I2 ) = I2 .

(151)

23
Multiplying Eq. (150) on both sides by the inverse gives
(H E I2 )1 (H E I2 )|i = I2 |i = |i = 0.

(152)

Therefore if there is an inverse for H E I2 then Eq. (150) can only be satisfied by having |i = 0 which is a trivial
solution but not something that is normalized to unity. Now the inverse of any 2 2 matrix can be found




1
d b
a b
A=

A1 =
,
(153)
c d
ad cb c a
unless ad cb = 0. This quantity is also known as the determinant


a b
 = ad bc.
detA = 
c d

This gives a condition for our eigenvalue problem and the eigenenergies can be found from


 1 E

t

 = (1 E)(2 E) t2

t
2 E 
= E 2 (1 + 2 )E + 1 2 t2 = 0.

This quadratic equation is easily solved and gives two eigenvalues

s

2
1 + 2
2 2
E =

+ t2 .
2
2

(154)

(155)
(156)

(157)

We can write the solution of the 2 2 eigenvalue problem in a variety of ways. This is not crucial physics, but it is
convenient for a variety of problems. Let us introduce some shorthand notations
=

1 + 2
2

= 2 1 and =
2
p
2
2
E =
+t

which allows us to write

E = E.

(158)
(159)
(160)

(161)

The eigenvalues for a 2 2 matrix can always be written as

| i = cos |1i sin |2i
|+ i = sin |1i + cos |2i.

(162)
(163)

Let us determine the value of from the equation

t cos + (2 E )( sin ) = t cos [2 ( E)] sin = t cos (E + ) sin = 0,
giving
tan =

t
.
E +

(164)

We can also express this as

tan 2 =

2 tan
=
1 2 tan2

Using that E 2 = 2 + t2 , we obtain

t
2 E+
2t(E + )
2t(E + )
= 2
.

2 =
(E + )2 t2
E + 2E + 2 t2
t
1 E+

tan 2 =

2t(E + )
t
2t
2t
= =
=
2
2E + 2

2 1

(165)

(166)

24
We can use this to obtain
cos 2 =

t2

= .
2
+
E

(167)

Which we can use to express

1 1
+ cos 2 =
2 2
1 1
sin2 =
cos 2 =
2 2

cos2 =


1
1+
2

1
1
2

=
E


=
E



1

1+
2
2 + 4t2


1

1
2
2 + 4t2

(168)

Let us consider some limiting cases. When t = 0, we find that E = 1 and E+ = 2 (where the minus sign indicates
the lowest energy and assuming 1 < 2 . Since = 0, we find that the basis functions are directly the eigenfunctions.
In the limit, 1 = 2 = , the eigenvalues are t and the eigenvectors are 12 (|1 i |2 i).
After we find the eigenvalues and eigenvectors, one might wonder if we could have chosen a different and more
convenient basis set. Obviously it would be nice to work in the basis given by (163), which is the basis of the
eigenvectors. It is often useful to make transformations from one basis set to another. The transformations are given
by the expressions for the eigenfunctions in Eq. (163). This can also be expressed in terms of a matrix


cos sin
U = Ube =
.
(169)
sin cos
The tricky part is making sure that the unitary transformation goes in the right direction. Usually in textbooks, no
subscripts are given, but they aid in seeing between what bases the unitary transformation is. Ube is the unitary
transformation that expresses the eigenvectors (e) in terms of the original basis vectors (b) (you have to read the
indices from right to left, because that is were the matrix multiplications are going). Therefore, applying U = Ube on
the lowest eigenvector | i gives

  

cos sin
1
cos
Ube | i =
=
.
(170)
sin cos
0
sin
The inverse Ueb expresses the basis functions in terms of the eigenfunctions and is given by the Hermitian conjugate


cos sin

U = Ueb =
.
(171)
sin cos
Those somewhat familiar with transformation will have noticed that U = Ueb is a rotation of the coordinate system
by an angle . Its Hermitian conjugate U = Ube is a rotation over . It appears obvious that the two operations
should cancel each other


 
 

cos sin
cos sin
cos2 + sin2
cos sin sin cos
1 0

U U=
=
=
= I2
sin cos
sin cos
0 1
sin cos cos sin
sin2 + cos2
A matrix element between two wavefunctions in the original basis set can be rewritten as
hb |H|b i = hb |Ube Ueb HUbe Ueb |b i = he |Ueb HUbe |e i,

(172)

using that |e i = Ueb |b i, expressing the wavefunction in the basis of eigenfunctions. The Hamiltonian H in this
basis is given by
H Hee = Ueb HUbe = U HU.

(173)

Note that the original Hamiltonian is expressed in the original basis set H Hbb . The evaluation of this matrix is a
bit of a pain


cos2 1 + sin2 2 2 sin cos
sin cos (1 2 ) + t(cos2 sin2 )
U HU =
.
(174)
sin cos (1 2 ) + t(cos2 sin2 )
sin2 1 + cos2 2 2 sin cos
This looks like a horrible expression, however
sin cos (1 2 ) + t(cos2 sin2 ) =

sin 2 + t cos 2 = 0,
2

(175)

25
because of Eq. (166). Furthermore,
1 + 2
1 2

+
cos 1 + sin 2 2 sin cos =
2t
2
2
2
+ 4t2
2

s 

1
2
1 2
4
+ 4t2

(176)

because of Eq. (168). Now since = 2 1 ), this can also be written as

1 + 2
1
2
1
2t
1 + 2
1p 2

+ 4t2 = E .
2t
=
2
2 2 + 4t2
2 2 + 4t2
2
2

(177)

The other diagonal matrix element is given by E+ . The total Hamiltonian in the eigenvector basis is therefore given
by


E 0

H = Hee =
,
(178)
0 E+
as it should be, since H | i = E | i.
In the end, you might wonder how we are ever going to get somewhere like this, since even solving a 2 2 matrix is
quite a bit of work. However, this is because we are doing it analytically, where we can do a 2 2 matrix and maybe
a 3 3. When the parameters take on actual values there are a wide variety of codes available to solve eigenvalue
problems so this is not really a problem. What is a problem is the size. Finding the complete eigenenergies and
eigenvectors can be done (say on your labtop) for matrices with a dimension of the order of a few thousand. This
might seem like a lot, but it is actually for many problems rather small. Using approximate methods, one can deal
with systems with a dimension of several million, but even that is often not sufficient for real problems.
When looking at the basis vectors, we found that their orthogonality is given by hi |j i = ij . However, what
happens when we multiply the vectors the other way around? This gives, for example,




1 0
0 1
|1 ih1 | =
and |1 ih2 | =
,
(179)
0 0
0 0
where the multiplication of a 2 1 and a 1 2 matrix gives a 2 2 matrix. This gives some nice features. For example,
2
X
i=1

|i ihi | =

1 0
0 1

= I2 ,

(180)

this is a completeness relation, since this only works if we sum over all the wavefunctions |i i in the basis. Another
interesting feature is that one can build any possible matrix using all the combinations of |i ihj |, or
H =

2
X

i,j=1

2
X

i,j=1

Hij |i ihj |

(181)

|i ihi |H|j ihj |.

(182)

Operating this on |k i gives

H|k i =

2
X

i,j=1

Hij |i ihj |k i =

2
X
i

|i ihi |H|k i.

(183)

using the orthogonality relation. Therefore the effect on the Hamiltonian is as follows. First, it removes the state |k i.
This state is then scattering into all states |i i with a scattering amplitude given by the matrix element hi |H|k i.
C.

Basis functions and matrix mechanics

In this section, we generalize the matrix algebra introduced in the previous section. Fortunately, most of it consists
of changing the basis from two to any arbitrary number. Matrix algebra depends on choosing a particular basis

26
consisting of wavefunctions i (r). This is often called a Hilbert space. Again, the difference between the basis
functions and eigenfunctions should be stressed. Basis functions are a human choice: a set of wavefunctions that we
feel are good to describe a certain problem. Eigenfunctions are determined by the physics of the problem. We can
approach the same problem with different bases, but in the end the eigenfunctions should be the same. In addition,
wavefunctions that are eigenfunctions in one problem can be basis functions in another. For example, plane waves are
eigenfunctions of free particles. However, as soon as we switch on some sort of potential that depends on the position,
the plane waves are no longer eigenfunctions. We can still use them as a basis. The index/label i can be very general
and can be split into several indices. For example, for a basis set of plane waves, the functions are labelled by the
momentum k. If spin is included in the problem, the label will become i k where =, for up and down spins,
respectively. For atomic orbitals, the index splits into three indices n, l, and m that label the atomic orbitals. There
are very good reasons for choosing indices like that, but in principle, we can label everything with an index i.
A basis has to satisfy, in principle, several conditions. The basis functions have to be orthonormal
hi |j i = ij .

(184)

Some of this is just for convenience and/or mathematical elegance. For example, there is no fundamental problem
having the wavefunctions normalized as hi |i i = ai , where ai is a real number. It is just a pain to keep track of
the norms. Since we have the choice, we might as well take ai 1. Also, people have worked with bases where
hi |j i 6= 0, for example, in applications involving multiple atoms where the atomic orbitals of one atom are not
orthogonal to those of the other atom. Again, this is possible, although it introduces a lot of practical issues.
The second mathematical condition is that the bases are complete, in that we should be able to describe any possible
wavefunction. For example, the basis of plane waves in Eq. (84) is complete if all possible wavevectors k values are
included. Again, this is a mathematical condition that physicists like to break usually for obvious practical reasons.
It is nice to say that your basis has to be complete and all basis functions have to be taken into account, but there
are an infinite number of plane waves and that is just a few too many. Many problems can be solved by restricting
oneself to a limited number of basis functions. Unfortunately, many times this restriction has to be done for purely
mathematical reasons.
In the end, we should be able to express the eigenfunctions in terms of the basis functions. In the case of planes
waves, this gives
n (r) =

N
X

ai i (r),

(185)

i=1

where k (r) is our chosen basis set denoted by quantum numbers i. Our eigenvectors satisfy the Schr
odinger equation
Hn (r) = En n (r).

(186)

with eigenvalues En are indexed by n. Restricting our basis to N functions (for example by taking only N k-values
into account) directly restricts the number of possible eigenfunctions to N . This is directly related to the fact that
going from the our choice of the basis to the eigenfunction basis is a unitary transformation (just as in the previous
example for the two-dimensional problem), which does not change the size of the basis. This might bother you at
first. However, this is not disastrous if we think about it a little longer. For example, for atoms we talk about different
shells. You probably heard of s, p, d, and f shells. However, if we really consider all basis functions, the number of
atomic orbitals would be infinite. However, in practice we are mainly interested in the orbitals that we want to put
electrons in. Since the atomic number of atoms goes to about 100, a same number of orbitals is needed to describe
atoms. Often we can do with even less. For hydrogen (1s1 ) and helium (1s2 ), the presence of 4f orbitals is generally
not relevant for most properties.
The good choice of a basis depends very much on the system that you are studying and its symmetry. For spherical
systems, spherical harmonics are a good starting point. For systems with mainly translational symmetry, plane
wave are an obvious choice. However, sometimes the choice is not that clear. A solid for example has translational
symmetry, but, due to the presence of atoms, many points with spherical symmetry. This has led to the development
of bases where spherical harmonics are centered on the atomic positions, but the regions between the atoms are made
up of plane waves. Although such bases can be convenient, they also lead to some arbitrariness, such as, where does
one switch from atomic regions to plane wave regions? However, such issues are rather advanced and should not
bother us for the moment.
PN
PN
The matrix element between to wavefunctions (r) = i=1 ai i (r) and (r) = i=1 bi i (r) is given by
h|A|i =

N X
N Z
X
j=1 i=1

drbi i (r)Aaj j (r) =

N X
N
X
j=1 i=1

bj Aij ai

(187)

27
where the matrix element is given by
Aij = hi |A|j i =
The result in Eq. (187) can be written as a matrix. The
in matrix form like

A11
A= .
AN 1

dri (r)Ai (r).

(188)

matrix elements between different basis functions can be put

. . A1N
. .
. .
. . AN N

(189)

In expression ((187), we see that the detailed nature of the basis function is no longer relevant, since the entire r
dependence is absorped in the matrix element. The basis function can then be expressed in terms of a vector

0
0

.

.

|i i = 0
,
(190)
1 i

0

.
.
where there is only 1 in the ith row. Its Hermitian conjugate is given by

hi | = (|i i) = (0, 0, . . . , 0, 1, 0, . . . ).

(191)

This notation is known as the bra-ket notation. The inner product of two basis functions is then
hi |j i = ij ,

(192)

where we have assumed that our basis set is limited to n functions. A wavefunction can now be written as

a1
a2

X
.
|i =
ai |i i =
,
.
i
.
an

(193)

where the only thing that is relevant now is the relative contributions of the different basis functions given by the
coefficients ai . The wavefunctions followed the same operations as vectors in an N -dimensional space. For example,
addition is given by

a1
b1
a1 + b 1
a2 b 2 a2 + b 2

X
X
.
. .

|i + |i =
ai |i i +
bi |i i =
(194)
+
=
.
.
. .

i
i
. .

.
an
bn
an + b n
The inner product of two functions becomes

h|i =

N
X

bi ai ,

(195)

i=1

Multiplying the wavefunction |i by the matrix A gives

P
Pi A1i ai
i A2i ai

.
A|i =

P .
i Ani ai

(196)

28
The matrix element between to vectors is
P
Pi A1i ai
i A2i ai

.
h|A|i = (b1 , b2 , . . . , bn )

P .
i Ani ai

bj Aji ai ,
=

ij

(197)

which is the same result as in Eq. (187).

A special operator is the Hamiltonian of the system. The eigenfunctions of the Hamiltonian can be found by solve
the eigenvalue equation
H|n i = E|n i,

(198)

where n labels the eigenfunctions.

1.

Function spaces in infinite dimensions

If you have followed everything so far, thats great. However, some of you might be wondering if we can write
down a direct connection between |i and (r). For completeness, we add this. However, for some of you, this might
be more confusing then helpful. You probably will not use this too often in practice. Let us assign every vector
in real space as a basis vector in an infinite-dimensional space. Let us first do this discretely and later replace the
summations by integrals. The basis vector for vector ri is given by

0
0

.

.

.
(199)
|ri i = 0
1 j

0

.
.

Since the positions in space are independent of each other, we have hrj |rj i = jj . Now we can expand the wavefunction |i in this vector space as

(r1 )
(r2 )

(200)
|i = . ,
(r )

i
.
.
which we can also express this as

|i =

X
i

(ri )|ri i =

dr (r)|ri,

(201)

where in the last step, we have replaced the summation by an integral in the limit that the discretization goes to zero.
If we want to know the value of the wavefunction at a certain rj value, we take the inner product with that basis
vector:
X
hrj |i =
(ri )hrj |ri i = (rj )ij = (rj )
(202)
i

Or if we do it in integral notation
hr|i =

dr (r )hr|r i = (r)(r r ) = (r)

(203)

29
Suppose we now want to calculate a matrix element h |H|i. The expression h |H|i is nice, but in order to
calculate a matrix element, one really needs to now what it looks like in real space. However, in any matrix expression,
we can always squeeze in identity matrices, since a multiplication with an identity matrix does not do anything. The
identity matrix for |ri is given by
Z
dr |rihr| = I ,
(204)
where infinity-dimensional I is the identity matrix (all ones on the diagonal, zeros everywhere else). The matrix
element can now be written as
Z
Z
Z
Z
h |H|i = dr
dr h |r ihr |H|rihr|i = dr
dr (r )hr |H|ri(r).
(205)
Now a potential is only a function of r, so hr |H|ri = H(r)(r r ), and we are left with
Z
h |H|i = dr (r)H(r)(r),

(206)

which is the usual expression for the matrix element.

This might appear somewhat confusing with bases in r of infinite dimensions. However, it might become clearer if
you think of it as a code in your computer. You want to solve a problem using the matrix method. This means you
have to diagonalize a matrix to find the eigenvalues and eigenvectors. However, the matrix consists of numbers that
you need to evaluate. In order to find the matrix element, you really need to know what the basis functions look like
in real space. Therefore, there is a separate routine has the values of the wavefunctions stored. This is essentially the
expression in Eq. (201). You only need that when you actually want to calculate the matrix elements. Sometimes
people skip the calculation of matrix elements all together. For example, the two-dimensional problem that we started
out with can be treated without ever calculating i and t. You can just assume some values. This is then known as
a model Hamiltonian, which are used by a large number of physicist/chemists.
2.

Impurity scattering

In this problem, we want to look at the effect of a local impurity potential on free particles. This is a rather complex
problem that can be treated analytically, although that generally involves complex methods using Greens functions.
We will solve the problem numerically. In the absence of the local impurity the eigenfunctions are plane waves
1
k (r) = eikr ,
V

(207)

with an energy k = ~2 k2 /2m. After introducing the impurity potential, the plane waves are no longer eigenfunctions.
However, since they form a complete set, we can still use them as basis functions. The eigenfunctions in the presence
of an impurity can be expressed in terms of plane waves
X
n (r) =
ak k (r).
(208)
k

In bra-ket notation, we can write the basis |k i. The Hamiltonian for the kinetic energy is then written as
X
H0 =
k |k ihk |,

(209)

which is diagonal in k.
A local potential in real space can be expressed as
HU = U (r),
which is indeed an impurity potential. The matrix elements for such a potential are given by
Z

1
U

hk |HU |k i =
dr eik r U (r)eikr = .
V
V

(210)

(211)

30

Intensity Harb. un.L

Intensity Harb. un.L

0.020
0.015
0.010
0.005

0.05
0.04
0.03
0.02
0.01
E HeVL

0.0 0.2 0.4 0.6 0.8 1.0

0.0 0.2 0.4 0.6 0.8 1.0

E HeVL
Intensity Harb. un.L
0.8

Intensity Harb. un.L

0.6

0.15
0.10
0.05

0.4
0.2

0.0 0.2 0.4 0.6 0.8 1.0

E HeVL
-0.2 0.0 0.2 0.4 0.6 0.8 1.0

E HeVL

P
FIG. 10: The value of |hn |r=0 i|2 where the local state is given |r=0 i = 1N m |m i for scattering of a local attractive
impurity for U = 0, 0.1, 0.2, and 0.6. The number of states that are included in the model is 50.

The Hamiltonian can then be written as

Z
Z
U X
U

|k ihk |,
HU =
dk
dk|k ihk | HU =
V
N

(212)

k,k

where in the last step we have replaced the integration by a summation, where we only take N k-values into account.
This is more appropiate for numerical work since it is impossible to use an infinite basis set. Note that whereas the
kinetic energy is diagonal in the momentum (which is of course a direct result of our choice of plane waves as a basis
set), the local potential couples all the different k values.
To really be able to perform some numerical solution (and in addition simplify our lives somewhat), let us forget
about the parabolic dependence of the energy on k, the density of k values and simply assume that there is only an
energy dependence. These are pretty big approximations, but not directly essential for the physics that we want to
get out. So let us take the kinetic energy term as

H0 =

N
X

m |m ihm | =

m=1

1 0 0
0 2 0

0 0
N

(213)

with
n =

m1
W,
N 1

(214)

where we went from a momentum basis |ki to an basis in energy |ni, where the index n is directly related to the
energy m . Note that the energy values are equidistant and go from 0 to W , where W can be called a bandwidth.

31
The scattering term then becomes

HU =

U
N

N
X

m,m =1

|m ihm | =

U
N

The eigenvalues and eigenvectors satisfy the Schrodinger equation

1 1 1
1 1 1

1 1
1

(H0 + HU )|n i = En |n i.
Obtaining the eigenvalues involves solving the determinant

 1 + U/N
U/N
U/N

2 + U/N U/N
 U/N



 U/N
U/N
N + U/N

(215)

(216)

(217)

This looks pretty horrific, but with the wide availability of codes or commericial software aimed at solving eigenvalue
problems, this problem is acually pretty straightforward (in this case mathematica was used). In the system, there is
one special site, namely that with the impurity. We want to know if we put our system in this local state, how will
it end up in the eigenstates of the system. When dealing with wavevectors k, the local state can be expressed as a
summation over all k values
1 X
|k i.
(218)
|r=0 i =
N k
This is essentially the definition of a -function. In our simplified problem, we take
1 X
|r=0 i =
|m i.
N m

(219)

The weight
wn = |hn |r=0 i|2 .

(220)

The results are given in Fig. 10 for W = 1, N = 50, and U = 0, 0.1, 0.2, and 0.6. It is easy to do calculations
for larger N , but the physics remains the same but the figure becomes worse. Let us first consider the situation for
U = 0. This is basically the absence of an impurity and the basis functions are the eigenfunctions of the system. The
weight is then
1 X
1
wn = |hn |r=0 i|2 = |
hn |m i|2 = ,
N
N m

(221)

which is 1/50 = 0.02 if N = 50. Therefore, a localized state has an equal probability of ending up in any of the
delocalized states. For U = 0.1 and 0.2 the weight increases to lower energies. For U = 0.8, we see that a state
has split off from the other states that has nearly 80% of the weight. For decreasing U even more, the energy becomes
close to U + W/2 and the weight approaches 1. What is the underlying physics of this? Actually, this is not too
hard to understand. The impurity potential is effectively a local attractive potential. If the potential becomes very
attractive, the local state becomes bound to this potential.
V.

HARMONIC OSCILLATOR

The harmonic oscillator played a significant role in the development of quantum mechanics. An harmonic oscillator
is, for example, a particle which is bound in its position by a force proportional with the displacement x of the particle
from its equilibrium position. Classicaly this is easy to solve starting from Newtons equation
m

d2 x
= Kx,
dt2

(222)

32
where K is the spring constant. The oscillation written as A cos t assuming that the oscillation is maximum at t = 0.
Inserting gives
m 2 A cos t = KA cos ,

(223)

p
or = K/m (or K = m 2 ). The velocity is given by v = A sin t. The total energy is therefore
E=

1
1
1
1
1
mv 2 + m 2 x2 = m 2 A2 sin2 t + m 2 A2 cos2 t = m 2 A2
2
2
2
2
2

(224)

Certainly classically there seems to be no indication of quantization. The energy can be varied continuously by
changing the amplitude.
There are two ways to approach this problem in quantum mechanics. This first is writing down the Schr
odinger
equation and solve it, leading to a solution in terms of Hermite polynomials. This solution directly gives you the
probability distribution function as a function of x. The other approach is in terms of effective operators, that tells a
lot about the eigenfunctions and eigenenergies of the system, but not much about the probability distribution function.
However, quite often we are only interested in eigenenergies anyway, since experiments often yield little information
on the probability distribution, but a lot about the spectral, i.e., energy, distribution.
A.

Schr
odinger equation approach

The time-independent Schrodinger equation in case of a harmonic oscillator is

~2 d2
1
+ m 2 x2 = E.
2
2m dx
2

(225)

This can be brought in a somewhat more compact form by introducing the variables = 2E/~ and =
giving

p
m/~x,

d2
+ ( 2 ) = 0.
d 2

(226)

d2
2 = 0.
d 2

(227)

Let us first look at the case for which

Since we are looking at bound particles, the solution has to go to zero when . An approximate solution is
() = exp( 2 /2)
2
d
= e /2 ()
d

2
2
2
d2
= e /2 2 e /2
= 2 e /2 ,
2
d

thereby satisfying Eqn. 227. We now proceed to seek solutions of the form () = H()e
asymptotic behavior. Inserting this into the differential equation gives
d2 H()
dH()
2
+ ( 1)H() = 0.
d 2
d

(228)
2

/2

which have the right

(229)

This can be solved by a series solution.

What are series solutions?
Let us consider a very simple differential equation
dy
= cy.
dx

(230)

Let us assume that the solution can be written as a polynomial

y=

k=0

ak xk .

(231)

33
We can exclude the negative k values since the function does not diverge at x =. The derivative is then given by

X
dy
=
ak kxk1 .
dx

(232)

k=1

In order to compare this with the right-hand side of the differential equation, let us shift the indices by introducing
k = k 1 or k = k + 1.

dy
=
ak +1 (k + 1)xk .
dx

(233)

k =0

Since k is just a summation variable, we can also call the index n again. We can now rewrite our differential equation
as

k=0

((k + 1)ak+1 cak )xk = 0

(234)

The functions xk are independent. This means that we can never find a combination of xk that will be zero for all
values of x. Therefore, the coefficients in front of the xn have to be zero. Therefore,
ak+1 =

c
c c
ck+1
ak =
ak1 =
a0
k+1
k+1k
(k + 1)!

or

ak =

ck
a0
k!

(235)

Our function is now

y=

k=0

a0

1
(cx)n = a0 ecx ,
k!

(236)

where we recognize the series as an expansion of the exponent

X
xk

ex

k!

k=0

(237)

One might also look at this differently and say that ex is the function that is defined by this series and we have now
proven that the derivative of ecx = cex . Of course, we could have solved this problem more simply if we had used
this result directly.
Let us return to the problem of the harmonic oscillator, where for the differential equation we assume that the
solution can be described as a polynomial
H() =

ak k .

(238)

k=0

The derivatives are given by

X
dH
=
ak k k1
d
k=0

k=2

k=0

X
dH
=
ak 2k k
d

(239)

k=0

X
X
d2 H
=
ak k(k 1) k2 =
ak+2 (k + 2)(k + 1) k ,
2
d

(240)

where in the last summation the index has been shifted by 2. Inserting this back into the differential equation gives

k=0

{(k + 1)(k + 2)ak+2 (2k + 1 )ak } k = 0.

(241)

34
Since k are independent functions, the only way to make this zero is by having all the components in front of the k
equal to zero. This leads to the recursion relation
ak+2 =

2k + 1
ak .
(k + 1)(k + 2)

(242)

For large values of the series is dominated by the terms with a large power of k. In the limit of k , this becomes
approximately
2
2 2
2
2
1
ak+2
a0 = 1 a0 .
= ak =
k
kk2k4
2
( 2 k)!

(243)

For the series with even powers of (there is also one with only odd powers), we have in the large limit

X
2
1 2k
= a0 e .
H() = a0
k!

(244)

k=0

This gives as total solution () = H()e /2 = a0 e /2 . Note that this diverges as . We cannot have that,
because this would mean that the particle would spend most of its time in infinity, whereas we would like to have it
close to = 0. Therefore, we cannot have the series continue till infinity. However, there is a way to cut off the series.
If we take = 2n + 1 then the coefficient an+2 will be zero. Note that this implies for the energy
n = 2n + 1 =

2En
~

1
En = (n + )~.
2

(245)

This implies that the energies of the harmonic oscillator are quantized just as Planck originally proposed. There is a
slight difference, since there is what is called the zero-point energy. This is directly related to Heisenbergs uncertainty
principle. Since the oscillation is around zero, we have hxi = hpi = 0. This directly implies that x2 = hx2 i and
p2 = hp2 i. The total energy is therefore
E=

hp2 i 1
p2
1
+ m 2 hx2 i =
+ m 2 x2 .
2m
2
2m
2

(246)

~2
1
+ m 2 x2 .
8mx2
2

(247)

Now since xp ~2 , we have

E

Minimizing this energy with respect to x2 gives

~2
1
+ m 2 = 0
8m(x2 )2
2

(x2 )min =

~
.
2m

(248)

This means that

E

~2 2m 1
~
1
+ m 2
= ~.
8m~
2
2m
2

(249)

This is the mimimum energy an oscillation can have without violating Heisenbergs uncertainty principle. Therefore,
unlike the classical oscillator which can just hang still, in the quantum mechanical case the wavefunction always has
to spread out to avoid a very large kinetic energy. In general, we find that
x2n =

1
~
(n + ).
m
2

Classically, the amplitude of the oscillation is given by A. However, x2 is given by


Z
Z 
A2
A2
1 1
A2
x2 =
dt cos2 t =
dt
+ cos 2t =

2 2
2
0
0

(250)

(251)

The effective amplitude of the oscillation is therefore quantized

A2n = 2x2n =

2~
1
(n + ).
m
2

(252)

35

FIG. 11: The solutions of the Schr

odinger equation for the harmonic oscillator for different values of n. Source: wikipedia.

where / is half an oscillation period. The energy is then

E=

1
1
2~
1
1
m 2 A2 = m 2
(n + ) = (n + )~
2
2
m
2
2

(253)

It can be shown that the function Hn (), which are called Hermite polynomials, can be written as
Hn () = (1)n e

dn 2
(e ).
d n

(254)

The first Hermite poynomials are

H0 ()
H1 ()
H2 ()
H3 ()

=
=
=
=

1
2
4 2 2
8 3 12.

(255)
(256)
(257)
(258)

Note that the relation between the coefficients can be found from the recursion relation in Eqn. (242). For example,
for n = 2, we have = 5 and therefore a2 = (1 5)/(1 2)a0 = 2a0 . For n = 3, we have = 7 and therefore
a3 = (3 7)/(2 3)a1 = 32 a1 . Then there is a constant factor, which ensures that the Hermite polynomial satisfy
the (traditional) normalization relation
Z

2
Hn ()Hm ()e d = 2n n!n,m .
(259)

The total normalized wavefunction is then

2
1
n () = p
Hn ()e /2 .

2n n!

(260)

The solutions for the lowest values of n are shown in Fig. 11.
B.

Solution using operators

An alternative approach to the harmonic oscillator is the use of operators. Classically, the values of x and px are
intimately related. If x = x0 cos t, then px = mx0 sin t, i.e. they differ by a factor and a phase shift. This is also

36
p2

seen from the Hamiltonian, which can be viewed as the values of x and px lying on an ellips, ax2 + xb2 = E. This was
one of the underlying starting points of the old quantum mechanics description of the quantization of the harmonic
oscillator. In general, we are less interested in the phase of the oscillation, which just tells you than something is
swinging. We are more interested in the absolute value. In classical mechanics, we can easily factorize an equation
2
2
i
i

2
of the
form u + v = E, by going to complex coordinates (u iv)(u + iv) = ae ae = a a = |a| = E, with
2
2
a = u + v and tan = v/u. This is roughly the approach that we would like to follow. The classical Hamiltonian
can be factorized as
r
 r

p2x
1
px
px
m
m
2 2
+ m x =
x i
x + i
H=
.
(261)
2m 2
2
2
2m
2m
However, in quantum mechanics, we have to take care of the fact that px and x do not commute
r
 r

m
m
p2
px
p
1
i
1 ~,

x
i
x
+ i
= x + m 2 x
2 (
px x
x
px ) = H
2
2
2m 2
2
2
2m
2m
using Eq. (126). Let us now introduce the operators
!
r
r
r


px
1

m
1
m
~
x
+ i
x+
=
,
+
a=
=
2~
~
m x

2
2
2m~
p
where = m
~ x as defined above. Similarly,
!
r
r
r


m
px
1
m
~
1

a =
x
i
=
x
=

.
2~
~
m x

2
2
2m~

(262)

(263)

(264)

(Textbooks usually put hats on the operators, however this becomes quite cumbersome and it is often clear that we
are dealing with operators. In the majority of scientific papers the hats are simply omitted. We will follow that
practice). The operators follow the following commutation relations


 



+
(265)
[a, a ] = aa a a =
2



1

+
= 1.
(266)
2

The Hamiltonian then obtains the simple form

1
H = (a a + )~.
2

(267)

We want to find the eigenfunctions and eigenvalues of the Hamiltonian. Since the eigenfunctions should have the
property H = E, operating a a on the eigenfunction can only lead to a real number n (although, as we now
already, this later turns our to be an integer), i.e.,
a an = nn .

(268)

What is now the effect of having a a work on an ? Using the commutation relation, we find
a aan = (aa 1)an = a(a a 1)n = a(n 1)n = (n 1)an

(269)

This means that an is proportional to n1 , since the eigenvalue for a a is n 1. Similarly, we can show that
a aa n = (n + 1)a n+1 .

(270)

Thus, we can view a and a as step up and step down operators, respectively. They are also called creation and
annihilation operators, respectively.
So we have already found that the eigenvalues are separated by ~, without solving any differential equation. In
addition, we know that the eigenvalues of H have to be larger than or equal to zero, since H has only x2 and a p2x
operators in it. Thus
1
1
Hn = (a a + )~n = (n + )~n
2
2

with

1
(n + ) 0
2

1
2

(271)

37
So the lowest eigenvector has 21 nmin < 12 . We must have anlow = 0 since we are not allowed to have eigenvalues
with nlow 1. From this it follows that
a anlow = 0

nlow = 0.

(272)

Therefore the eigenvalues for the harmonic oscillator are 21 ~, 32 ~, 52 ~, . . . .

We see that in the operator approach, we can obtain the eigenvalues without solving the differential equation.
Therefore, we do not need to know n () to obtain understanding of our problem. This is comparable to the classical
situation, where it is not necessary to know the complete motion, because everything is determined by the maximum
amplitude x0 . We can therefore introduce new eigenstates |ni, which have the same properties when applying the
operators on them as the wavefunctions n (), although we do not the what the function looks like in real space.
Classically, we can say that we have an oscillation with a maximum oscillation x0 (or its energy 12 m 2 x20 ), the exact
motion is determined by our Hamiltonian (and is a simple cosine). Quantum mechanically, we can say that we have
an eigenstate that is given by its energy n~. Its motion is determined by the Schrodinger equation and is given by
n () although often this knowledge is not essential. Let us now denote the lowest eigenstate in the bra-ket notation
as |0i. Then we can obtain the nth eigenstate by
|
ni = (a )n |0i.

(273)

We put a tilde on this wavefunction to indicate that this state is not normalized. Let us look at some specific examples
for the normalization:
|1i = a |0i

h1|1i = h0|aa |0i = h0|(1 + a a)|0i = 1,

(274)

since a|0i = 0 and since h0|0i = 1. So |1i is also nicely normalized. However, for |2i, we find
|2i = a a |0i

h2|2i = h0|aaa a |0i = h0|a(1 + a a)a |0i = h0|aa + aa aa |0i

= h0|(1 + a a) + (1 + a a)(1 + a a)|0i = 2,

(275)
(276)

So the normalized wavefunction is |2i = 12 a a |0i. If we look at it a bit more carefully, we see that the first
annihilation operator a from the right has to make a 1 with one of the creation operators a . For |ni, there a n
possibilities to do that. The next annihilation operator also has to make a 1, however, now there are only n 1
creation operators left, and so on. Therefore, there are n! different ways of making a non-zero combination. The
normalized wavefunction are therefore given by
1
|ni = (a )n |0i.
n!

(277)

We can use this result to determine the factors for the step operators

1
1
a |ni = (a )n+1 |0i = n + 1 p
(a )n+1 |0i = n + 1|n + 1i
n!
(n + 1)!

and

1
1
1
a|ni = a(a )n |0i = n(a )n1 |0i = n p
(a )n1 |0i = n|n 1i,
n!
n!
(n 1)!

(278)

(279)

again using the fact that there are n possible ways to apply the commutation relation. From this result, we easily
obtain

a a|ni = a n|n 1i = n|ni.

(280)
We can also write x
and px in terms of the creation and annihilation operators
r
r
~
m~

x
=
(a + a)
and
px = i
(a a).
2m
2

(281)

For the expectation values, we find

hn|
x|ni =

dxn (x)xn (x) =

~
hn|(a + a)|ni =
2m

~ 
n + 1hn|n + 1i + nhn|n 1i = 0, (282)
2m

38
where
hn |ni =

dxn (x)n (x) = n,n

(283)

is the inner product between the different wavefunction. This is zero when the quantum numbers are unequal, since
the wavefunctions are orthogonal. Similarly, we find h
px i = 0. This is in agreement with our classical idea that for
an oscillation the average x and px are zero. However, for h
x2 i, we obtain
~
~
hn|(a + a)(a + a)|ni =
hn|(a a + a a + aa + aa)|ni
2m
2m
~
1
~
1
=
hn|(a a + )|ni =
(n + )
m
2
m
2

hn|
x2 |ni =

(284)
(285)

and for h
p2x i, we obtain
hn|
p2x |ni =

m~
m~
1
hn|(a a)(a a)|ni =
hn|(a a aa )|ni = m~(n + ).
2
2
2

(286)

Combining these results gives

xpx =

p
1
1
hx2 ihp2x i = ~(n + ) ~,
2
2

(287)

which satisfies Heisenbergs uncertainty principle.

C.

Operators and wavefunctions

Using the operators it is also possible to obtain the wavefunctions. As noticed above, we should have


1

+
a0 =
0 = 0.

(288)

It is easy to verify that

0 = A0 e

/2

satisfies this equation. The requirement that 0 is normalized to unity gives

Z
Z

2
e d = A20
1=
|0 |2 d = A20

(289)

A0 = 1/4 .

(290)

Having found 0 , it is straightforward to find the other wavefunction with the step operators


2
2
1

1 = a 0 =
e /2 = p 2e /2 .

2
2

(291)

Generally, we have


n
2

n = p

e /2 .

2 n!
1

This leads to an alternative definition of the Hermite polynomial


n
2
2

e /2 = Hn ()e /2 .

(292)

(293)

So we find, again, for the total wavefunction

1
2 /2
n () = p
.
Hn ()e
n
2 n!

(294)

39
In the previous section, we obtained the properties of the creation and annihilation operators by studying the
characteristics of the operators themselves. However, we can also obtain these by operating them on the wavefunction.
Although this is somewhat more complicated, it is instructive to see that we do ge the same results. Let us first look
at yet another way of writing defining the Hermite polynomial. The generating function for the Hermite polynomials
is
2

g(, t) = et

+2t

Hn ()

n=0

tn
.
n!

(295)

This means that is we expand the exponential and collect the terms with the same powers of t then the coefficients in
front of them (which are functions of x) are the Hermite polynomials (try it yourself). From the generating function,
we can derive a number of relationships between Hermite polynomials of different order. These are called recurrence
relations. Differentiating the generating function to gives

X
2
dg
tn
= 2tet +2t =
Hn () .
d
n!
n=1

(296)

Note that H0 = 0. The left-hand side gives

2

2tet

+2t

=2

Hn ()

n=0

X
X
tn
tn+1
tn
=2
Hn 1 ()
=
2nHn1 () .
n!
(n 1)! n=1
n!

(297)

n =1

Therefore, we have
Hn () = 2nHn1 ().

(298)

Another recursion relation can be obtained by differentiating the generating function with respect to t gives

X
2
d t2 +2t
ntn1
(e
) = (2t + 2)et +2t =
Hn ()
.
dt
n!
n=1

(299)

Then by using the result obtained in Eqn. (297) and by shifting the indices of the right-hand side.

(2n)Hn1

n=1

X
X
tn
tn
tn
+ 2
Hn =
Hn+1 ()
n!
n! n=0
n!
n=0

(300)

giving
Hn+1 = 2Hn 2nHn1 .
Not let us apply the annihilation operator on the wavefunction in Eqn. (294)




1
d
1
1
2 /2
an () =
+
n () + ()n () + n
n () =
e
Hn () .
d
(2 n! 1/2 )1/2
2
2

(301)

(302)

Application of the recurrence relation in Eqn. (301) gives

2
1
2n
an () =
e /2 Hn1 ()
n
1/2
1/2
2 (2 n! )

2
1
= n n1
e /2 Hn1 ()
(2
(n 1)! 1/2 )1/2

= nn1 ().

For the creation operator, we find



2
1
1
d
1

a n () =

n () =
e /2 (2Hn () Hn ())
n
1/2
1/2
d
2
2 (2 n! )
2
1
1
=
e /2 (2Hn () 2nHn1 ())
2 (2n n! 1/2 )1/2

(303)
(304)
(305)

(306)
(307)

40

FIG. 12: The classical probability function compared to the probability for a quantum harmonic oscillator for a large value of
n. Source: hyperphysics.

Application of the recurrence relation in Eqn. (298) gives

2
1
e /2 Hn+1 ()
(2n+1 n! 1/2 )1/2

= n + 1n+1 ().

a n () =

(308)
(309)

Note that we obtained the same relationships as in the previous section. However, here we really apply the operator on
the wavefunction, which leads to quite a bit of mathematics. In the previous sections, we made use of our knowledge of
the behavior of the wavefunctions under the operators without even obtaining the x dependence of the wavefunctions.
D.

Correspondence principle

As was shown before, the classical solution of the harmonic oscillator is

x = x0 sin t

and

v = x = x0 cos t,

where x0 is the turning point of the oscillator. We can also write the velocity as
q
v(x) = x20 x2 .

(310)

(311)

To find the classical probability P (x)dx of finding a particle at a certain position x, we have to know the time the
particle spends in a region dx around x. This is simply given by dt = dx/v(x). Now we have to normalize this to half
a period, i.e. /, since we want the particle to pass the interval dx only once. Thus,
P (x)dx =

dx
dx
= p 2
.
v(x)
x0 x2

(312)

This is very close to the quantum-mechanical probability for large n, except for the oscillations. See the figures in
Liboff.
Also for large n the energy should become equivalent to the classical energy, i.e ~ should disappear from the
equations. Classicaly, we have E = 12 m 2 x20 , whereas quantum mechanically, we obtained E = (n + 12 )~. This looks
~
quite different, but in Eqn. (1212), we found that hx2 i = m
(n + 21 ). We can therefore write the energy as
1
En = (n + )~ = m 2 hx2 iQM .
2
The classical expectation value of x2 is
Z
Z
1 T
1 T
1 1
1
hx2 iclass. =
dtx20 sin2 t =
dtx20 ( cos 2t) = x20 .
T 0
T 0
2 2
2

(313)

(314)

Therefore, the classical energy is E = m 2 hx2 iclass. . Thus the energies become equal when the classical and quantum
mechanical expectation values approach each other. This is indeed true for large n.

41
VI.

THE HYDROGEN ATOM

For the Hydrogen atom we are interested in solving a Schrodinger equation of the form

~2 2
+ V (r) = E,
2m

(315)

with V (r) = Ze2 /(40 r) for a Hydrogen-like atom (a single electron, but the nucleus could still have Z protons).
When dealing with atomic physics, the most important operator is angular. Let us first try to understand why this is
so. An atom is spherically symmetric. This means that we can rotate the atom along a certain axis and this should
not directly affect the eigenstates of the system. That also implies that the Hamiltonian is invariant under rotations.
Let us see what happens to a function if we make an infinitesimally small rotation along the z-axis
Rz,d (x, y, z) = (x yd, y + xd, z),

(316)

let us expand this in a Taylor series

Rz,d (x, y, z) = (x, y, z) y

d + x
d =
x
y



 

i
1+ x
y
d (x, y, z) = (1 + Lz d)(x, y, z).(317)
y
x
~

where the angular momentum is given by

Lz = (r p)z = xpy ypx = i~(x

y )
y
x

Now a rotation over an angle can be described by N steps of d = /N

N


i
i
Rz, = lim 1 + Lz
= exp( Lz ).
N
~ N
~

(318)

(319)

Therefore, angular momentum describes a rotation around a certain axis. Obviously, we do not need to make the
+ dy y
+ dz z, would lead to a rotation operator,
rotation around the z-axis. A general rotation over d = dx x
Rd = 1 +

i
i
d L = 1 + (Lx dx + Ly dy + Lz dz )
~
~

(320)

However, we will see later on that the Li do not commute with each other and we can only have eigenfunctions for
one of the components of L. It is customary to choose Lz .
So now let us suppose we make a unitary transformation of our system (the atom) around the z-axis over an angle
.
i
U (r, , ) = Rz, (r, , ) = exp( Lz )(r, , ) = (r, , + ).
~

(321)

Now since a rotation around the z axis is a unitary transformation that does not change the atom, the eigenvalues
for the rotated system, should be identical to those of the original system:
En = hn |H|n i = hn |U HU |n i.

(322)

U H = U U HU = HU

(323)

This directly implies that

H = U HU

[H, U ] = HU U H = 0,

and therefore that U and H commute, which directly implies that Lz and H commute. However, what does that
imply? Suppose we have an operator A that commutes with the Hamiltonian, then AH = HA, therefore
AH|n i = En A|n i = H (A|n i) .

(324)

This means that A|n i can only be const |n i, therefore |n i must also be an eigenfunction of A.
Since we are dealing with spherical symmetry, i.e.,
x = r sin cos
y = r sin sin
z = r cos

(325)
(326)
(327)

42
let us first look at Lz in spherical coordinates. Now since,

x
y
z

=
+
+
= y
+x ,

x y z
x
y

(328)

we can write
Lz = i~

(329)

We know that the wavefunction must be an eigenfunction of Lz ,

Lz = m~,

(330)

where we have taken an eigenvalue of the form m~. Since Lz only depends on , let us try a separation of variables
(r, , ) = R(r)()().

(331)

This implies that

i~

()
= m~()

()
= im()

() = const eim .

(332)

Since we want to be single-valued this implies that

eim(+2) = eim

eim2 = 1

m = 0, 1, 2, ,

(333)

The functions, we want to have orthogonal

Z

eim eim d = 2m,m ,

(334)

giving c = 1/ 2.
However, for the R(r) and (), we need to write down the Hamiltonian in spherical coordinates. From classical
arguments we can get some feeling for what the kinetic energy term will look like. Let us have a look at L2 ,
L2 = r2 p2 sin2 = r2 p2 r2 p2 cos2 = r2 p2 (r p)2

p2
p2
L2
= r +
,
2m
2m 2mr2

(335)

where pr = p r is the momentum in the r-direction. This means that L2 is important in the classical case, can we
expect something similar in the quantum mechanical situation. We note that the components of L commute with the
Hamiltonian. What does this imply for L2 . Since [Li , H] = 0, we can write
Li [Li , H] + [Li , H]Li = L2i H Li HLi + Li HLi HL2i = L2i H HL2i = [L2i , H].

(336)

Thus L2i also commute with the Hamiltonian. A linear combination of L2i should therefore also commute with the
Hamiltonian, which implies that [L2 , H] = 0. Therefore, the eigenfunctions of the Hamiltonian for the atom are
also eigenfunctions of L2 . Let us see if we can find the quantum mechanical equivalent of Eqn. (335). In quantum
mechanics, we are dealing with operators and have to take care that r and p do not commute. First, we have
L2 = |r p|2 = (xpy ypx )2 + c.p.
= xpy xpy + ypx ypx xpy ypx ypx xpy + c.p.

= x2 p2y + y 2 p2x 2xypx py + i~(xpx + ypy ) + c.p.

(337)
(338)
(339)

where use has been made of the commutation relation [x, px ] = i~; c.p. stands for cyclic permutations, i.e., x y z.
On the other hand, we have
r2 p2 (r p)2 = (x2 + y 2 + z 2 )(p2x + p2y + p2z ) (xpx + ypy + zpz )2
=
=

x2 p2x
x2 p2y

+ x2 p2y
+ y 2 p2x

y 2 p2x

+ c.p. (xpx xpx + xpx ypy + ypy xpx + c.p.)

2xypx py + i~xpx + c.p..

(340)
(341)
(342)

43
Collecting the results, we obtain in the quantum mechanical case
L2 = r2 p2 (r p)2 + i~(r p),

(343)

where we obtain an extra term from the fact that r and p do not commute. Let us try to rewrite the result somewhat.
For r p, we have



r p = i~ x
+y
+z
= i~r ,
(344)
x
y
z
r
using the fact that
r=
We can use this to rewrite

p
x2 + y 2 + z 2

r
x2
=x
=
x
x r
r r

)(i~r ) i~(i~r )
r
r
r


2

2
2
2
2
= ~ r
+ 2r
.
= ~ r (r ) ~ r
r r
r
r2
r

(r p)2 i~r p = (i~r

(345)

(346)
(347)

Let us now introduce

pr = i~

1
r = i~(
+ ).
r r
r r

(348)

From this, we obtain that

r2 p2r = ~2 (r2



2
2

2 1
2
2
+
r
+
r
+
1)
=
~
r
+
2r
.
r2
r
r r
r2
r

(349)

This is equivalent to the result obtained above. Therefore, in the quantum mechanical case, we can also write
p2 = p2r +

L2
,
r2

(350)

if we use the definition for pr in Eqn. (348).

Could we have seen more easily to use this definition for pr ? We want to know the component of p in the r direction,
i.e.,
r p. However, this is not very symmetric since we could also write p
r. So let us try a more symmetric form:
1 nr
ro
p+p
.
2 r
r

(351)

f
1 f
1 f

r+
+
,

r
r
r sin

(352)

pr =

From the definition of the gradient in spherical symmetry,

f =

we see that the first term in Eqn. (351) is given by

r

p = i~ .
r
r

(353)

n r
o
1
r
r
p r = i~( ) = i~
+ .
r
r
r
r

(354)

The second term would give

Using that

r
r





r

r
1 2

1
3 1
2
+r 2 =
r+
r = = ,
r
r
r r
r
r
r
r
r

(355)

44
where use has been made of the definition of the divergence in spherical coordinates. Collecting the terms, we obtain
for pr



1
+
,
(356)
pr = i~
r
r
which is identical to Eqn. (348).
Of course, all of this could also have been obtained by starting from the angular momentum in spherical coordinates:


 r




 = i~ 1 +
0
0

L = r p = i~  r
.
(357)

sin

1

 1
r

r sin

Note that the angular momentum is perpendicular to r and has therfore no r component. What is now the angular
momentum in Cartesian coordinates. Therefore, we use
cos cos + y
cos sin z sin
= x
=
cos .

x sin + y

(358)
(359)

This gives
Lx
Ly
Lz




+ cot cos
= i~ sin




= i~ cos
+ cot sin

= i~ .

From what we have seen above, it is obvious to look at L2 = L2x + L2y + L2z ,


1

1 2
L2 = ~2
(sin
)+
.
sin

sin2 2

(360)

(361)

Comparing this with

2 =

1 2
1

1
2
(r
)
+
(sin

)
+
.
2
r2 r
r
r2 sin

r2 sin 2

we note that the angular part can be written as L2 /r2 , and thus


~2 2
~2 1
L2
p2r
L2
2

=
r
+

+
.
2m
2m r2 r
r
2mr2
2m
2mr2

(362)

(363)

Note that we can write the r term in different ways

1 2
1 2
2
2
(r
)=
(r) =
+
.
2
2
r r
r
r r
r2
r r
A.

(364)

Legendre polynomials

Let us try to find the eigenvalues of the eigenvalue equation

L2 = ~2 .

(365)

We assumed already that we could separate the variables in ,

see Eqn. (331). Since L2 only depends on and ,
im
and since we know that the -dependent part is given by e
/ 2, we can write the differential equation for ()
as

1 d
d
m2
(sin
)+
=
sin d
d
sin2

(366)

45
This equation is known as the Legendre equation. This can be put in a different form by substituting x = cos .
This implies that
d
d d
d 1
1 d
=
=
=
dx
d dx
d dx/d
sin d

or

d
d
= sin
.
d
dx

We can write the DE as





1 d
d
m2
d
m2
2 d
(1 x2 )
+ (
)
=
(1

x
)
+ (
) = 0.

2
sin d
dx
1x
dx
dx
1 x2

(367)

(368)

This can also be written as

(1 x2 )

d2
d
m2
2x
+ (
) = 0.
2
dx
dx
1 x2

(369)

Let us first consider m = 0. We will find later on that the other solutions can be related to this solution. As with the
harmonic oscillator, we try to find the solution in form of a series:
(x) = P (x) =

n=0

an xn = a0 + a1 x + a2 x2 + ,

(370)

where the Legendre polynomial is conventionally denoted as P (x). This gives

P (x) =

an xn

(371)

n=0

X
X
dP
n1
2x
= 2x
nan x
= 2
nan xn
dx
n=1
n=1

(1 x2 )

X
X
X
d2 P
2
n2
n
=
(1

x
)
n(n

1)a
x
=
(n
+
1)(n
+
2)a
x

n(n 1)an xn .
n
n+2
dx2
n=2
n=0
n=2

(372)
(373)

Using this we can write

(1 x2 )

dP
d2 P
2x
+ P = a0 + a1 x 2a1 x + 2a2 + 6a3 x
2
dx
dx

X
+
{an 2nan + (n + 2)(n + 1)an+2 n(n 1)an } xn = 0

(374)

n=2

This gives the general equation

( n(n + 1))an + (n + 2)(n + 1)an+2 = 0.

(375)

The other equations a0 + 2a2 = 0 and 1 2a1 + 6a3 = 0 are the same equation for n = 0 and 1, respectively. Thus
we have the recursion relation
an+2 =

n(n + 1)
an .
(n + 2)(n + 1)

(376)

However, just as we found for the harmonic oscillator, this series diverges. Therefore, we have to cut off the series at
a certain value of n. When taking = l(l + 1), the series becomes zero for n > l. Note that this implies
L2 = l(l + 1)~2 .

(377)

A Legendre polynomial of order l is denoted as Pl (x). As for the harmonic oscillator, we have again two series. By
choosing = l(l + 1), we can cut off the series of the even or the odd powers, but not both. Therefore when chosing
to cut off the even powers, we have to choose all the coefficients in front of the odd powers equal to zero, and vice
versa. The normalization of the Legendre polynomial is P (1) = 1. The first terms can be derived rather easily. For
k = 0, we only have a0 , which should be unity, giving P0 = 1. For k = 1, = 2, and we have P1 (x) = a1 x x. For
k = 2, we have the even series, with = k(k + 1) = 6. The recursion relation gives a2 = 6/2a0 = 3a0 , yielding

46
P2 (x) = a0 3a0 x2 . Using P2 (1) = 2a0 = 1 gives a0 = 1/2, which gives P2 (x) = 3/2x2 1/2. For P3 (x), = 12,
giving a3 = (2 12)/(2.3) = 5/3. Thus P3 (x) = a3 x 5/3a3 x3 . The normalization gives P3 (1) = 2/3a3. With
a3 = 3/2, we find P3 (x) = 1/2(5x3 3x). In order to normalize the wavefunctions, we have to integrate over the
Legendre polynomials from 1 to 1 (note, that x comes from cos ). It can be shown that this gives
Z 1
2
m,n ,
(378)
Pn (x)Pm (x)dx =
2n
+1
1
where Legendre polynomials of different order are orthogonal.
Remember that we had put m to zero. Actually, we want to solve the full equation


2 m
dPnm (x)
m2
2 d Pn (x)
2x
+ n(n + 1)
Pnm (x) = 0.
(1 x )
dx2
dx
1 x2

(379)

It can be shown that the associated Legendre polynomials are related to the mth derivative of the Legendre polynomial,
Pnm (x) = (1 x2 )m/2

dm
Pn (x),
dxm

(380)

for positive m, with m l. Note that by solving the differential equation for the associated Legendre polynomials,
we find that |m| l or m = l, l + 1, . . . , l 1, l. The associated Legendre polynomials for negative m are equal
to those for positive m apart from a multiplicative constant. In atomic physics, one often makes use of the spherical
harmonics which are related to the associated Legendre polynomials as
s
2l + 1 (l m)! m
P (cos )eim ,
(381)
Ylm (, ) = (1)m
4 (l + m)! l
where the normalization has been chosen such that
Z 2
Z
d
d sin Yl m (, )Ylm (, ) = l,l m,m
0

(382)

B.

Angular momentum

As was shown before, as a result of the spherical symmetry of the problem Li and L2 commute with the Hamiltonian.
The commutation relationships of angular momentum are given by
[Lx , Ly ] = Lx Ly Ly Lx = (ypz zpy )(zpx xpz ) (zpx xpz )(ypz zpy )
= ypz zpx ypz xpz zpy zpx + zpy xpz zpx ypz + zpx zpy + xpz ypz xpz zpy
= xpy (zpz pz z) ypx (zpz pz z) = i~(xpy ypx ) = i~Lz .

(383)
(384)
(385)

By cyclic permutation, we also obtain

[Ly , Lz ] = i~Lx

and

[Lz , Lx ] = i~Ly ,

(386)

often ~ is taken equal to unity or one assumes that Li is measured in units of ~. Can we make a wavefunction that
has eigenvalues in all operators Lx , Ly , and Lz ? Let us apply this to an eigenfunction which has an eigenvalue mz
iLz = imz = (Lx Ly Ly Lx ) = (Lx my Ly mx ) = ~(mx my my mx ) = 0.

(387)

When mz 6= 0 this implies =0 or that we cannot have eigenfunctions in Lx , Ly , and Lz at the same time. What
about L2 ? Making use of the commutation relations, we find
L2 Lz Lz L2 = L2x Lz Lz L2x + L2y Lz Lz L2y + L2z Lz Lz L2z
= Lx (Lz Lx iLy ) (Lx Lz + iLy )Lx + Ly (Lz Ly + iLx ) (Ly Lz iLx )Ly = 0

(388)

Thus we have eigenvector with eigen values for Lz and L2 . This means we have
Lz bm = mbm L2z bm = Lz mbm = m2 bm

(389)

47
Now we also have
L2 bm = bbm (L2x + L2y )bm = (b m2 )bm .
(390)

Since the operator L2x + L2y is positive, this implies that |m| b. Let us now consider the effect of the operator
Lz (Lx iLy ) = Lx Lz + iLy i(Ly Lz iLx ) = (Lx iLy )(Lz 1).

(391)

Operating this on an eigenfunction gives

Lz (Lx iLy )bm = (Lx iLy )(Lz 1)bm = (Lx iLy )(m 1)bm = (m 1)(Lx iLy )bm .
So (Lx iLy )bm are eigenfunctions of Lz with eigenvalues m 1. Moreover, since Lx and Ly commute with L2 ,
they are also eigenfunctions of L2 . So starting from a particular eigenfunction we can create a set of eigenfunctions
that with eigenvalues
m = , m 2, m 1, m, m + 1, m + 2,

(392)

However, since |m| b, there have to be certain values of m for which

(Lx iLy )bl = 0
(Lx + iLy )bl+ = 0

(393)
(394)
(395)

From the commutation relations we find

(Lx + iLy )(Lx iLy ) = L2x + L2y + Lz
(Lx iLy )(Lx + iLy ) =

L2x

L2y

Lz

(396)
(397)
(398)

These operators will give us eigenvalues

2
(L2x + L2y + Lz )bl = (b l
+ l )bl = 0

(L2x

L2y

Lz )bl+ = (b

2
l+

l+ )bl+ = 0

(399)
(400)
(401)

giving
2
2
(b l
+ l ) = (b l+
l+ ) = 0

(402)

(l+ + l )(l+ l + 1) = 0.

(403)

or

Since l+ l , we find that l = l+ in order to make this equation zero. This directly implies that b = l(l + 1).
Therefore we have the
L2 lm = l(l + 1)lm
Lz lm = mlm ,

(404)
(405)

as we already found from the differential equation. Note that we found the values for m without having to invoke an
argument about the single-valuedness of the function.
As we have already seen that operators can connect different eigenstates
L lm = (Lx iLy )lm = N l,m1 .

(406)

What is now the value of N

|N |2 = hlm |L L |lm i
= hlm |L L |lm i
= hlm |L2x + L2y Lz |lm i

= hlm |l(l + 1) m2 m|lm i

= (l m)(l m + 1)

(407)
(408)
(409)
(410)
(411)

48
giving a value
N =

p
(l m)(l m + 1)ei ,

(412)

by convention we choose = 0.
We have learnt already a lot about the eigenvalue problem of the system without actually having to solve a
differential equation. Just as for the harmonic oscillator, we can also use the operators to determine the wavefunctions.
We know that applying the step-down operator cannot last forever since |m| l, thus
1
(Lx iLy )Yl,l (, ) = (Lx iLy )l,l (, ) eil = 0.
2
Using the expression for Lx and Ly in Eqn.(360), we can show that



+ i cot
.
L = (Lx iLy ) = ei

(413)

(414)

This gives us a differential equation

dl,l
+ l cot l,l = 0.
d

(415)

Let us try as a solution

l,l = c sinl ,

(416)

see Liboff Table 9.1. This gives

cl sinl1 cos + cl

cos
sinl = 0,
sin

(417)

which shows that this is indeed a correct eigenfunction. By applying the step-up operators, all the lm can be
obtained. This is a somewhat tedious process and will be omitted here.
C.

Angular momentum in matrix form

In order to write down a matrix for angular momentum operators, we first need to choose a basis set. Although
we can choose the solution of the Hydrogen atom as a basis set, we have to remember that the angular momentum
operators do not work on the radial part, but only on the spherical harmonics Ylm (, ). In addition, angular
momentum operators do not couple spherical harmonics with each other that have different l values. Therefore, we
can consider the spherical harmonics for one particular l value. Let us begin with the step-operators:
p
L |lmi = ~ (l m)(l m + 1)|l, m 1i.
(418)

In the case that l = 1, we can write down the matrices straightforwardly as

0 1 0
0 0 0

L+ = ~ 2 0 0 1
and
L = ~ 2 1 0 0 .
0 0 0
0 1 0
The functions now become the colums

1
|11i 0 ,
0

where the matrix elements are obtained from

0
|10i 1 ,
0

0
|1, 1i 0 ,
1

(L )m m = hlm |L |lmi

(419)

(420)

(421)

49
Using these matrices, we obtain, for example,

L+ |1, 1i = ~ 2|10i

0 1 0
0
0

~ 2 0 0 1 0 = ~ 2 1
0 0 0
1
0

From the step matrices for the step operators, we can obtain the matrices for Lx and Ly ,

1
1
~ 0 1 0
~ 0 i 0
1 0 1
i 0 i .
Lx = (L+ + L ) =
and
Ly = (L+ L ) =
2
2i
2 0 1 0
2 0 i 0

(422)

(423)

For Lz , we have

1 0 0
Lz = ~ 0 0 0 .
0 0 1

(424)

Note that the matrix elements can also be evaluated from



Z
Z

(Lz )m,m = h1m|Lz |1m i = Y1m i~

Ylm sin dd = m~ Y1m
Ylm sin dd = m~m,m .

Note that the commutation relations,

[Lx , Ly ] = i~Lz .

(425)

can also be derived by simple matrix multiplication.

As in Eqn. (320) we can build a finite rotation over of a function can be built up of infinitesimally small rotations.
For l = 1
Rz (d) = (1 + idLz )
or for d = /N with N ,
Rz () = (1 + i

(426)

Lz )N = exp(iLz ).
N

(427)

The exponential may be expanded in a Taylor series

1
1
exp(iLz ) = 1 + iLz + (iLz )2 + (iLz )3 +
2!
3!

0 0 0
1 0 0
1
1
= 0 1 0 + 0 0 0 {1 2 + 4 }
2!
4!
0 0 0
0 0 1
1
1
+iLz { 3 + 5 }
3!
5!

(428)
(429)
(430)
(431)

(note that the first matrix is 1 L2z ). We have used here that

1 0 0
L2z = ~2 0 0 0
and
0 0 1

L3z = Lz .

(432)

The series may be recognized as the Taylor series for a cosine and sine, respectively, and we can also write the rotation
as

cos + i sin 0
0
e 0 0
= 0 1 0 .
0
1
0
Rz () =
(433)
0
0 cos i sin
0 0 ei
This can be clearly seen as the rotation of a vector over an angle around the z-axis. Thus a rotation of the
function m () gives
1
1
1
Rz ()m () = Rz () eim = eim eim = eim(+) = m ( + )
2
2
2

(434)

50
VII.

LAGUERRE POLYNOMIALS

For the Hydrogen atom, we are left with the Hamiltonian

H=

L2
Ze2
p2r
+

,
2m 2mr2
r

(435)

where, for brevity, e e/ 40 for SI units. The L2 term one sometimes calls the centrifugal term, i.e., it tends to
throw the particle outward, as the centrifugal force in quantum mechanics. This gives us a radial equation




~2 1 d2
~2 l(l + 1) Ze2

r +

+ |E| R = 0,
(436)
2m r dr2
2mr2
r
Introducing the radial function u = rR gives



d2
l(l + 1) 2m Ze2

|E|
u = 0,
2 +
dr
r2
~2
r

(437)

where we have taken E = |E|. Let us try to simplify the equation by introducing
~2 2
and
= 2r
2m

2
Z
Z 2 R0
=
=
,
a0
|E|

|E| =
2

(438)
(439)

with R0 = ~2 /2ma20 the Rydberg constant and a0 = ~2 /me2 the Bohr radius. This gives us the following differential
equation


d2 u l(l + 1)
1

u+

u = 0.
(440)
d2
2
4
Let us take a look at the limiting behavior. When , we have the approximate differential equation
d2 u u
= 0.
d2
4

(441)

u
= Ae/2 + Be/2 .

(442)

This can be easily solved:

However, we do not want our particle to sit in infinity so we want u 0, giving B = 0.

For 0, we obtain
d2 u l(l + 1)

u = 0.
d2
2

(443)

Trying a solution of u = q gives

(q(q 1) l(l + 1))u = 0

q = l

or q = l + 1.

(444)

Thus
u Al + Bl+1 .

(445)

We also do not want our electron to sit at the nucleus, which gives A = 0, or u l+1 . Let us try to find a solution
of the form
u() = e/2 l+1 F ().
/2 l+1

(446)

Writing u = AF with A = e
, the second derivative is given by AF + 2A F + A F , with




1
1 l+1

/2
l+1
/2 l
A () = e

+e
(l + 1) = +
A()
2
2





1 /2 l
1 l + 1 l(l + 1)

/2 1 l+1
/2 l1
A () = e

+2
e
(l + 1) + e
l(l + 1) =

+
A().
4
2
4

(447)
(448)

51
Substituting this into the differential equation and dividing by A() gives




 


1 l+1
1 l + 1 l(l + 1)
l(l + 1)
1

F +2 +
F +

+
F
F+

F = 0.
2

2
2
4

(449)

Cleaning this up and multiplying with gives

F () + (2l + 2 )F () (l + 1 )F = 0.

(450)

Again, we will approach this with a series solution

F () =

c k k .

(451)

k=0

The first and second derivative are given by

F () =

ck kk1

F () =

and

k=1

k=2

ck k(k 1)k2 .

(452)

We would like to make all the powers in the summation equal to k. This can be done by shifting the index. So,
F () =

k=2

ck k(k 1)k1 =

ck+1 (k + 1)kk

(453)

k=1

(note that we can also let the summation start at k = 0). and
(2l + 2)F () = (2l + 2)

ck+1 (k + 1)k .

(454)

k=1

Collecting the term gives

k=0

{ck+1 (k + 1)k + (k + 1)(2l + 2)ck+1 kck (l + 1 )ck } k =

k=0

{(k + 1)(2l + 2 + k)ck+1 (k + l + 1 )ck } k = 0.

(455)
(456)

Since all the k are independent function the only way to make this zero is by having all the coefficients in front of
all powers of k zero. This gives the recursion relation
ck+1 =
In the limit that k , this becomes

(k + l + 1)
ck .
(k + 1)(k + 2l + 2)

ck ck1
c0
ck
ck+1
=
c0 = ,
=
k
k k1
k!

(457)

(458)

giving for F
F =

k=0

c0

k
= c0 e .
k!

(459)

The total solution,

u = e/2 l+1 e = e/2 l+1

(460)

then diverges as . Therefore, just as in the case of the harmonic oscillator and the Legendre polynomials, we
have to end the series at the certain point, so that it remains a simple polynomial. This can be done by taking
= kmax + l + 1,

(461)

52
i.e., since both kmax and l are integers, is integer. Setting n = n, we obtain from Eqn. (439)
2n = n2 =

Z 2 R0
|En |

En = |En | =

Z 2 R0
,
n2

(462)

which is in agreement with Bohrs result. The radial wavefunctions are given by
unl () = e/2 l+1 Fnl () = Anl e/2 l+1

nl1
X

c k k .

(463)

k=0

The function Fnl is the associated Laguerre polynomial, which is usually denoted as L2l+1
nl1 . Note that the maximum
value for l is given by lmax = n 1. We can therefore summarize the quantum number as
n = 1, 2, 3, . . .
l = 0, 1, 2, . . . , n 1
m = l, l + 1, . . . , l

(464)
(465)
(466)

and the total wavefunction of an orbital with a set of quantum numbers nlm is
(r, , ) = Rnl (r)Ylm (, ),

(467)

with an energy
En =
VIII.

Z 2R
mZ 2 e4
=
.
2
n
2~2 n2

(468)

RELATIVISTIC QUANTUM MECHANICS (LIBOFF CHAPTER 15)

Let us have a brief look at relativistic quantum mechanics. We will not deal with relativistic quantum mechanics
in detail. Our main interest is to discover the most important correction to the nonrelativistic limit. We will see
how the spin is introduced in the relativistic equation, how the spin interacts with the magnetic field (obtaining the
gyromagnetic factor g = 2), and obtain the spin-orbit interaction. We all know Einsteins famous formula
m0 c2
E = mc2 = q
,
2
1 vc2

(469)

where m0 is the rest mass of the particle and c the speed of light. We want to transform this into a differential
equation similar to the Schrodinger equation. We are therefore not very happy with the presence of the square root,
so let us start by squaring the equation
m20 c4 (1
m20 c4
E =
=
2
1 vc2
1
2

A.

v2
c2
v2
c2

v2
c2 )

= m20 c4 +

m20 v 2 c2
2 4
2 2
2 . = m0 c + p c
1 vc2

(470)

Klein-Gordon equation

We could start from the previous equation and simply make the substitutions p i~ and E = i~ t
, giving

~2

2
= c2 ~2 2 + m20 c4 .
t2

(471)

or
(2

1 2
m20 c2
)
=
.
c2 t2
~2

(472)

This is known as the Klein-Gordon equation. We see a second derivative to t. This looks more like a wave equation
than a Schrodinger equation, which has a first derivative in t. Unfortunately this equation is not satisfactory as a
relativistic version of the Schrodinger equation. Let us first look at the energies
q
E = p2 c2 + m20 c4 .
(473)

53
We find positive and negative energies. This might look problematic at first but these are actually found in nature.
The electron has a negative energy equivalent in the positron. However, let us try to write the Klein-Gordon equation
as a probability current, as we did for the Schrodinger equation in Eqn. (59). Again, after writing the conjugate
wavefunction and substracting after multiplication with and , respectively, we find


1
2
2

.
(474)
2 2 = 2 2
c
t
t2
Trying to write this as a diffusion equation for probability, cf. Eqn. (61), gives


~
~

( ) =

.
2m0 i
2m0 c2 i t
t
t

(475)

Comparison with Eqn. (61) leads to the identification of the probability density as



i~

.
=
2m0 c2
t
t

(476)

For a stationary state, we can take the time dependence (r, t) = (r) exp(iEt/~), which gives
=

E
( ).
m0 c2

(477)

Note that for particles in the nonrelativistic limit, where E

= m0 c2 this gives the proper probability density = .
However, for relativistic particles, we saw that E could be less than zero, which would lead to negative probability
densities, which are undesirable.
B.

Dirac equation

Since the Klein-Gordon equation is not entirely satisfactory (although it has been used successfully in describing
spinless bosons,-mesons), Dirac set out to find a wave equation that is linear in the derivative to t and not quadratic.
Bringing the p2 term in Eqn. (470) to the other side gives
E2
p2 = m20 c2 .
c2

(478)

We are still stuck with an E 2 and we would prefer to end up with something linear in E, just like the Schr
odinger
equation. We could try to factorize the left-hand side as a2 b2 = (a b)(a + b). However, we cannot take simply
the square root of p2 = p2x + p2y + p2z . We can, however, try to make use of the equation
( a)( b) = a bI + i (a b),

(479)

where the matrix I is given by



1 0
0 1

0 1
1 0

I=

(480)

and ,
=x
x + y
y +
zz = x

+y

0 i
i 0

+
z

1 0
0 1

Using this for the momentum ( p)( p) = p pI = p2 I (note that p p = 0), with


pz
px ipy
p = px x + py y + pz z =
.
px + ipy
pz
would enable us to factorize the equation but would in fact give rise to two equations instead of one



E2
E2
E
E
2
I p I = 2 I ( p)( p) =
I ( p)
I + ( p) = m20 c2 I.
c2
c
c
c

(481)

(482)

(483)

54
Some of you might recognize the i as Pauli spin matrices. What we have effectively done is introducing an extra
degree of freedom, which is the spin of the electron. Although this was first done in the derivation of the relativistic
wave equation, we could have done it, in fact, from the non-relativistic Schrodinger equation, by replacing p2 by
( p)( p).
So far, we have not done any quantum mechanics, so it is time to introduce the operators
E i~

and

p i~.

(484)

It is common in relativity to introduce a four-vector

x = x1 , x2 , x3 , x4 = x, y, z, ict,

(485)

treating space and time on equal footing. The time-dependence then becomes
i~

= ~c
.
t
x4

(486)

Equation (483) now becomes





~
+ i~
~
i~ = m20 c2 ,
x4
x4

(487)

where the unit matrices I have been omitted. Unfortunately, we still do not have a linear differential equation.
For this we have to make another sacrifice. We now have a two-component wavefunction and we have to go to a
four-component wavefunction. We can achieve this by introducing the wavefunctions



~
+ i~ L = m0 c R ,
(488)
x4



~
i~ R = m0 c L ,
(489)
x4
(490)
where L = . By taking the sum and difference of the above equations, we obtain

( R + L ) i~ ( R L ) = m0 c( R + L ),
x4

i~ ( R + L ) ~
( R L ) = mc( R L ).
x4

(491)
(492)

Note that both equations have two spin components. By defining = R L , we can write this equation in matrix
form

 + 
 +

~ x
i~

4
= m0 c
.
(493)

i~ ~ x
4

If we now make the shorthand

=

we can finally write the equation as a linear equation working on the fourvector ,



m0 c
+ 4
+
=0
x4
~

(494)

(495)

or



m0 c

+
= 0,
x
~

(496)

55
where as, using the Einstein notation, the summation over is implicitly assumed. The 4 4 matrices are given by




0 ik
I 0
k =
and
4 =
.
(497)
ik
0
0 I
Often this equation is given in a somewhat different form, which can be obtained by multiplying by ~c4 from the
right side. Using
4

(~c4 ) = i~
x4
t

and

~ck 4 = i~c(ik 4 ) = i~ck

(498)

and
k = ik 4 =

0 k
k 0

(499)

and = 4 , we can explicitly bring out the time dependence

or

H = c (i~) + m0 c2 = i~
t


= ( p)c + m0 c2 .
t
This is known as the Dirac equation. One might wonder whether this has the right probability density.
i~

= ic~( )c + m0 c2 .
t
Multiplying with and substracting the conjugated equation multiplied by gives
i~

( ) + (c ) = 0,
t

(500)

(501)

(502)

(503)

giving a current density of J = c and the proper probability density of = .

We can also write this out in four coupled equations. Rewriting Diracs equation
{E ( p)c m0 c2 } = 0.
Expanding the matrices gives

E m0 c2
0
pz c
(px ipy )c
a1
0
E m0 c2 (px + ipy )c
pz c

a2
a = 0

pz c
(px ipy )c E + m0 c2
0
3
a4
(px + ipy )c
pz c
0
E + m0 c2

(504)

(505)

For a plane wave

= a(p) exp(ip r/~ it),

(506)

the operator pi can be replaced by c-numbers. We now only need to solve the coeffients of a. This can still depend
on p, but not on t and r. These equations have a nonzero solution only if the determinant of the matrix vanishes.
This gives
(E 2 m20 c4 c2 p2 )2 = 0,

(507)

which
p is nothing but the fundamental relativistic energy equation. Solutions for the eigenvectors for E =
c2 p2 + m20 c4 are

1
0
0
1

+ =
+ =
(508)
,
cpz /+
c(px ipy )/+
c(px + ipy )/+
cpz /+

cpz /
c(px ipy )/
cpz /
c(px + ipy )/

=
=
(509)
,
,
1
0
0
1

56
where
= (E m0 c2 ).

(510)

Note that the wavefunction are not normalized. The normalization for, e.g., + is given by
1
Np = q
1+

(511)

c2 p2
2+

Note that in the nonrelativistic limit all the terms with in there vanish and we obtain a two-vector plane wave
for + (with spin up and spin down), and a two-vector plane wave for .
Let us now try to make a connection with the non-relativistic theory. We also want to introduce an electromagnetic
field and a potential. We can rewrite Eqn. (493) as
1
(E V m0 c2 ) +
c
1
= (E V + m0 c2 )
c

(p eA) =

(512)

(p eA) +

(513)
(514)

We can solve these equations for + , giving

(p eA)

c2
(p eA) + = (E V m0 c2 ) + .
E V + m0 c2

In the non-relativistic limit E = Enr + m0 c2

= m0 c2 and v/c 1. We can therefore expand


c2
1
2m0 c2
1
Enr V
=
=
1
+
E V + m0 c2
2m0 2m0 c2 + Enr V
2m0
2m0 c2

(515)

(516)

When keeping only the lowest order this becomes 1/2m0 . This gives for the Hamiltonian
1
( (p eA))2 + = (Enr V ) + .
2m0

(517)

or, using Eqn. (479),



(p eA)2
B
+g
SB+V
2m0
~

+ = Enr + ,

(518)

using the fact that

p (A) + A p = (p A) A p + A p = (p A)

(519)

ie
e~
e~
e
B
(p A) =
( A) =
B=
SB=g
SB
2m0
2m0
2m0
m0
~

(520)

We can now write

e~
with the Bohr magneton B = 2m
and the gyromagnetic factor g = 2. Note that the total interaction with the
0
magnetic field is proportional to (L + gS) B. Obtaining the right value of the gyromagnetic factor was one the
achievements of the relativistic theory.
Note that the other two-component wavefunction

= c

( (p eA)) + ( (p eA)) + v +
=
= ,
E V + m0 c2
2m0 c
2c

(521)

is an order p/mc = v/c smaller than the + component.

We can also keep one term higher in the expansion in Eqn. (516). Transforming it into a decent eigenvalue problem
is somewhat elaborate (see Sakurai) and we will therefore only give the final result (for simplicity we set A = 0)
 2

p
p4
e~
e~2
+V

(E p)
E = Enr ,
(522)
2m0
8m30 c2
4m20 c2
8m20 c2

57
where the electric field comes from pV = i~eU = e~E, where U is the potential. The first two terms are well
known. The first correction terms is the next higher-order correction to the kinetic energy
q
p2
p4

+ .
p2 c2 + m20 c4 m0 c2 =
2m0
8m30 c2

(523)

For a Hydrogen like atom, the next term is

e~
~
(E p) =
2
2
4m0 c
4m20 c2



1 dV
e2

(r p) =
LS
r dr
80 m20 c2 r3

(524)

where the electric field from the nucleus is along the r direction and S = 12 ~. This term is known as the spin-orbit
coupling and will be discussed in more detail below.
The third term is the Darwin term and is related to the negative solution of the Dirac equation. For the Hydrogen
atom, one has

e~2
~2
e 2 ~2
2

E
=

V
=
(r).
2
2
8m0 c2
8m0 c2
8m20 c2

(525)

We like to draw the attention again to the existence of antiparticles in the Dirac theory. One can see this as a
kind of semiconductor picture, where all the states with negative energy are filled. For electrons the negative energy
equivalent is the positron. The energies of the antiparticles start at m0 c2 , and those of the particles at m0 c2 , giving
a gap of 1.02 MeV, see Fig. 15.4 in Liboff. A positron can only be created by making an electron-positron pair across
the gap.
C.

Spin-orbit coupling and relativistic corrections (Liboff 12.2)

The spin-orbit coupling follows directly from the relativistic Dirac equation, but how do we understand this from a
semiclassical point of view? Since the electron is moving through the electric field of the nucleus, it will also experience
an internal magnetic field given by the Biot-Savart law


0 Ze(r v)
Ze
(r mv)
L 1 d Ze2
1 1 dV (r)
B=
=
=
=
L.
(526)
3
2
3
2
4
r
40 mc
r
mec r dr 40 r
mec2 r dr
In a different frame of reference one could see this as the nucleus turning around the electron. The orbiting charged
nucleus gives rise to a current, which leads to a magnetic field interacting with the spin-moment of the electron. The
change in energy for the electron is then given by the scalar product S B. Relativistic kinematic considerations
lead to the introduction of a factor 12 . This gives an energy of


1e 
1 1 dV (r)
1 1 dV (r)
1
S B =
S
L
=
L S = (r)L S.
(527)
2
2 m
mec2 r dr
2m2 c2 r dr
What are the eigenvalues of the spin-orbit coupling for a single electron? Let us first put the parameter = 1. We
would like to couple the orbit l and spin s moments together to a total moment j,
j = l + s.

(528)

j2 = (l + s)2 = l2 + s2 + 2l s,

(529)

The norm of the new moment is

so the eigenvalues are given by

l sjmj =

1 2
1
3
(j l2 s2 )jmj = (j(j + 1) l(l + 1) )~2 jmj .
2
2
4

Since j can only take the values l 21 , we can also write the eigenvalues as

1 2
j =l+
2 l~
hjmj |l s|jmj i =
1
2 (l + 1)~2 j = l

1
2
1
2

(530)

(531)

58
Although this produces the right result, it is a bit based on a trick to obtain the value for l s. When evaluating
matrix elements with a computer, one would often compute matrix elements such as hm sz |L S|msz i (of course, this
depends on the basis set one chooses). To find the matrix elements, it is convenient to rewrite the spin-orbit coupling
as
L S = L x Sx + L y Sy + L z Sz =

1
(L+ S + L S+ ) + Lz Sz ,
2

(532)

since we know the matrix elements for the z-component and the step operators, see Eqns. (406) and (412). Let us
2
consider a p electron (l = 1). For example, the state |1 i has as diagonal element msz = 1 12 ~2 = ~2 . It has no
off-diagonal element, since lowering the orbital momentum by 1 would require raising the spin by 1. This is however
2
impossible since the state already has an up spin. The state |1 i has as diagonal element msz = 1 ( 12 )~2 = ~2 .
This state has an off-diagonal element connecting it to |0 i. The matrix element is given by
p
1
~2 p
h0 | L S+ |1 i =
(l + m)(l m + 1) (s sz )(s + sz + 1)
2
2
r
1
~2 p
1 1
1
~2
(l + 1)(1 1 + 1) ( ( ))( + ( ) + 1) = .
=
2
2
2 2
2
2

The total eigenvalue problem then becomes

1
2

21

2
~

1
2

(533)
(534)

1
2

0
0
1
2

1
2
21

1
2

|1 i

|1 i

|0 i
.

|0 i
| 1 i

| 1 i

(535)

Although a 6 6 matrix, this matrix is easily reduced into 4 smaller matrices. Two are 1 1 giving eigenvalues of
~2 /2. The other two are 2 2 matrices,whose eigenvalues are determined by the determinant


~2 
 ~2 E
2
4
2
4
 2
2  = ( ~ E)(E) ~ = E 2 + ~ E ~ .
(536)


2
~


2
2
2
2
E 
2

The eigenvalues are easily evaluated to give

E1,2

~2
~2
=
4
2




1
1 3
1
~2
2
4
= ~ ( ) =
1 2 .
4
2
4 4
2~

(537)

In total, we find a fourfold degenerate eigenstate with energy ~2 /2 and a twofold degenerate eigenstate with energy
~2 . Since the degeneracy is given by 2j + 1 (since mj = j, . . . , j), this gives j = 23 , 12 , as we would expect. So even
if a computer would not know anything about coupling to jmj and only knows it is working in a basis set |msz i and
the spin-orbit coupling interaction, it still ends up with eigenstates |jmj i. Note also that the center of gravity is zero
since 4 ~2 /2 + 2 (~2 ) = 0.
The eigenstates can also be found. Note that the solutions of the 1 1 matrices are | 32 32 i = |1 21 i and | 32 , 32 i =
| 1, 21 i since m and mz are coupled parallel. For example, writing
1
|j i = a|1 i + b|0 i
2
Substituting E gives


1
1
(1) a + b = 0
2
2

a = 2b

The other state has to be orthoganal to this one giving

r
r
31
1
2
| i=
|1 i +
|0 i.
22
3
3

(538)

11
| i=
22

2
|1 i
3

1
|0 i.
3

(539)

(540)

59
Note that the state which is lowest in energy has more |1 i character since the diagonal element of the state is lower
than that for |0 i.
What is now the value of the parameter ? We will calculate this using perturbation theory. A perturbation means
that there is a small term in the Hamiltonian H = H0 + H . We know how to solve the problem for H0 . However, we
do not know how to solve it for H. Although perturbation theory will be treated later on in more detail, the lowest
order correction to the energy is rather intuitive,
E1 = h0 |H |0 i.

(541)

This means that we calculate the expectation value of the perturbance H for the wavefunction of the unperturbed
state, that follows from H0 |0 i = E0 |0 i. The total energy is then given by E0 + E1 . The assumption is that the
small perturbance does not really change the wavefunction.
For the spin-orbit coupling, the unperturbed Hamiltonian is that of the Hydrogen atom in Eqn. (315), which we
have solved. We want to obtain the expectation value for
1 1 dV
Ze2 1
=
2m2 c2 r dr
2m2 c2 r3

(r) =

(542)

where we have taken e/(40 )1/2 e. Since only depends on r, we can calculated from
Z 2
Ze2
Rnl (r) 2
hinl =
r dr.
2
2
2m c 0
r3

(543)

The expectation value can be straightforwardly evaluated (see the results of problem 10.49 in Liboff). We find
~2
hinl =
2

me4 Z 2
2~2 n2

2

n
.
mc2 (l + 12 )(l + 1)l

(544)

Recognizing that |En | = mZ 2 e4 /(2~2 n2 ), we can rewrite the result as

|En |
n
(Z)2
1
~2 hinl
=
=
,
1
2
2 |En |
mc (l + 2 )(l + 1)l
n (2l + 1)(l + 1)l

(545)

where
=

e2
1
=
~c
137.037

is known as the fine-structure constant. The total energy can now be written as

i
h
2


1
j =l+
|En | 1 (Z)
(Z)2
1
n
(2l+1)(l+1)
h
i
Enlj = |En | 1
=
2
1
|En | 1 + (Z)
n (2l + 1)(j + 21 )
j =l
n
(2l+1)l

(546)

1
2
1
2

(547)

Note that the spin-orbit coupling is not necessarily a small quantity. In atoms with many electrons, the spin-orbit
coupling of a deep-lying level can be several to several tens of electron volts. Compared to the 0.1 eV for shallow
levels, this cannot be treated as fine structure, i.e., a small perturbation.
Relativistic correction In a similar fashion, we can treat the relativistic correction to the kinetic energy from
Eqn. (523),
H =

p4
.
8m3 c2

(548)

The lowest order correction is then given by

E1 = hnl|H |nli =

1
hp4 inl .
8m3 c2

(549)

We do not know how to calculate the expectation value for p4 , but we are more familiar with the p2 term, since
 2

p
+ V |nli = En |nli.
(550)
2m

60
Rearranging gives
p2 |nli = 2m(En V )|nli.

(551)

Using this, we can calculate the perturbation

1
E1 = hnl|H |nli =
hnl|p2 p2 |nli
3 c2
8m
Z
1
1
Rnl [2m(En V )][2m(En V )]Rnl r2 dr =
(E 2 2En hV i + hV i2 ).
=
3
2
8m c 0
2mc2 n
Using the expectation values for r1 and r2 from problem 10.49 in Liboff gives


2 Z 2
8n
E1 = |En |
3 .
4n2
2l + 1

(552)
(553)

(554)

Adding this to Eqn. (547) gives

"

Enlj = |En | 1 +

Z
2n

2 

4n
3
j + 12

#

(555)

This expression gives quite good agreement with the experimentally observed spectra for one-electron atoms. It is
now time to see what happens if we start adding more electrons to the atom.
IX.

PERTURBATION THEORY AND ABSORPTION AND EMISSION OF PHOTONS

After a brief excursion into relativistic quantum mechanics, we take another small detour. A large amount of data
on atoms has been obtained by spectroscopy. We would like to describe the process of absorption and emission of
photon. However, we are missing the necessary mathematical equipment. Usually this is done with perturbation
theory, which was already introduced (in its simplest form) earlier. We will now discuss it in more detail, since we
will need it later on.
A.

Time-independent perturbation theory

We want to know the effect of a (small) perturbation H1 on a Hamiltonian H0 . Often, in quantum mechanics, we
are able to solve the problem for a certain Hamiltonian H0 , but are unable to solve it for the complete Hamiltonian
H0 + H1 . However, when H1 is a relatively small effect, there is no need to solve the entire problem, but we can
calculate the effect of H1 to a certain order. To keep track of the order, we introduce a constant , which goes to
unity when we have sorted out all the different orders. Therefore, we are interested in solving the problem
(H0 + H1 )n = En n .

(556)

We were able to solve the problem H0 n0 = En0 n0 . We would like to know the corrections to the energy and
wavefunction due to H1 :
n = n0 + n = n0 + n1 + 2 n2 +
En = En0 + En = En0 + En1 + 2 En2 + .

(557)
(558)

Equation (556) now becomes

(H0 + H1 )(n0 + n1 + 2 n2 + ) = (En0 + En1 + 2 En2 + )(n0 + n1 + 2 n2 + )

(559)

Keeping the terms up to order 2 gives

H0 0n + (H1 n0 + H0 n1 ) + 2 (H 0 n2 + H1 n1 ) = En0 n0 + (En1 n0 + En0 n1 ) + 2 (En2 n0 + En1 n1 + En0 n2 ). (560)
Of course we can go higher in order but that kind of defeats the purpose of doing perturbation theory. (There is
perturbation theory which goes to infinite order. This is solving the problem exactly but with perturbation methods.

61
This is, however, a different story.) We see that we obtain the unperturbed Schrodinger equation H0 n0 = En0 . To
first order in , we have
H0 n1 + H1 n0 = En0 n1 + En1 n0 .

(561)

Multiplying this by (n0 ) and integrating over the volume gives

hn0 |H0 |n1 i + hn0 |H1 |n0 i = En0 hn0 |n1 i + En1 hn0 |n0 i.

(562)

However hn0 |H0 |n1 i = En0 hn0 |n1 i, and we are therefore left with
En1 = hn0 |H1 |n0 i.

(563)

This is a result that we used already. It says that the first order correction to the energy is the expectation value of
the perturbation in the unperturbed state. This means that one assumes that the perturbation does not really affect
the wavefunction.
The first-order correction to the wavefunction is given by
(H0 En0 )n1 = (H1 En1 )n0 .

(564)

Now since the eigenfunctions of H0 form a complete set, it is always possible to express the perturbed wavefunction
as a linear combination of unperturbed wavefunctions
X
0
n1 =
cnm m
.
(565)
m6=n

Note that we do not need n0 since the total wavefunction (up to first order) is given by n0 + n1 , so n0 is already
included in the wavefunction. Substituting the linear combination into the Schrodinger equation gives
X
0
0
(Em
En0 )cnm m
= (H1 En1 )n0 .
(566)
m6=n

Taking the inner product with l0 gives

X
0
0
(Em
En0 )cnm hl0 |m
i = hl0 |H1 |n0 i + En1 hl0 |n0 i.

(567)

m6=n

If l = n, the left side is zero and we obtain the equation for E1 , when l 6= n
(El0 En0 )cnl = hl0 |H1 |n0 i,

(568)

we obtain the coefficient

hl0 |H1 |n0 i
.
En0 El0

(569)

X h 0 |H1 | 0 i
m
n
0
m
.
0
En0 Em

(570)

cnl =
The perturbed wavefunction is therefore given by
n1 =

m6=n

Notice that this works as long as the eigenstates are nondegenerate, i.e. En 6= Em . Although the energies are quite
good the wavefunctions are usually described less well by perturbation theory.
Example
Calculate the first-order change in energy as a result of the anharmonic potential
H = Kx4 =

K
(a + a )4 ,
4 4

(571)

with
1
x = (a + a ),
2

2 =

m
~

(572)

62
in the state |ni.
Answer Since we are calculating an expectation in state |ni the number of step-up operators has to be equal to the
number of step-down operators, i.e. term such as (a )4 vanish. There are six of those terms, for which we obtain the
matrix elements
hn|(aaa a + a a aa + aa aa + a aa a + aa a a + a aaa )|ni

= n+1 n+2 n+2 n+1+ n n1 n1 n+ n+1 n+1 n+1 n+1

+ n n n n+ n+1 n+1 n n+ n n n+1 n+1
= (n + 1)(n + 2) + n(n 1) + (n + 1)2 + n2 + (n + 1)n + n(n + 1) = 3[2n(n + 1) + 1]

(573)
(574)
(575)
(576)

The total energy is then

En1 =

3K
[2n(n + 1) + 1]
4

(577)

Degenerate perturbation theory

Lowest order perturbation theory works well in the situation where the energies are not degenerate. However,
perturbation theory also works in the case of degenerate energies. The assumption is that only states of the same
energy couple with each other. The splitting between states with different energies is assumed to be significantly
larger than the perturbation. We are then left with a number of states at the same energy. Note that we cannot
simply use lowest order perturbation theory since the wavefunction in Eqn. (570) would go to infinity. Therefore, we
have to solve the problem for our reduced basis set of states at the same energy. This is best illustrated by considering
some examples.
The two-dimensional harmonic oscillator
For a two-dimensional harmonic oscillator, we have the Hamiltonian
H=

K
1 2
(px + p2y ) + (x2 + y 2 ).
2m
2

(578)

This is easily diagonalized by noting that the x and y coordinates do not mix. Thus,
H = (a a + b b + 1)~,

(579)

where x is given in the previous example and y

1
y = (b + b ).
2

(580)

The eigenstates are now products of eigenstates in one direction

mn (x, y) = m (x)n (y)

or

1
|mni =
(a )m (b )n |0i
m!n!

(581)

with energies
Emn = (n + m + 1)~

(582)

Therefore, the energy 2~ is two-fold degenerate: E10 = E01 = 2~.

We would now like to see how the degereneracy of the state with energy 2~ is lifted by a perturbation
H = K xy =

K
K

(a
+
a
)(b
+
b
)
=
(ab + a b + ab + a b ).
2 2
2 2

(583)

Note that the first and last term change the total excited states n + m into n + m 2 and n + m + 2. If we now
restrict ourselves to a basis set of |10i and |01i these terms vanish and we are only left with the a b and ab terms.
This leads to off-diagonal matrix elements. The matrix element coupling |10i and |01i is
T = K h10|xy|01i =

K
K
K

h10|(a
b
+
ab
)|01i
=
h10|a
b|01i
=
.
2 2
2 2
2 2

This leaves us with a simple 2 2 problem



 2~ E

T

 = (2~ E) T 2 = 0

T
2~ E 

E = 2~ T = 2~

(584)

K
2 2

(585)

63
The Stark effect
The Stark effect is the splitting of the levels of the Hydrogen atom under an electric field. This effect is smaller than
the Zeeman splitting resulting from a magnetic field and was only observed in 1913, whereas the Zeeman splitting
was discovered in 1897. Since this energy is small it will only have a significant effect of states that are degenerate in
energy, i.e. with the same quantum number n. The perturbing Hamiltonian is written as
H = eEz z,

(586)

where Ez is the strength of the electric field in the z-direction. Let us now consider the case for n = 2. There a
four degenerate state, (in |nlmi notation): |200i, |211i, |210i, and |21, 1i. What is now the effect of z? In spherical
coordinates z = r cos . This can be written as a spherical harmonic z = rY01 . Note that this is the term with m = 0,
therefore z cannot change the quantum number m (unlike x and y. We will discuss that when treating the interaction
of atoms with radiation). Therefore the matrix elements have the form hnl m |z|nlmi = hnl m|z|nlmim,m . The
diagonal terms are also zero. Note that this would be the expectation value of z in that orbital. For an atom hzi = 0.
In our basis set, this means we only have matrix elements
Z
Z 1
Z 2
ea0 Ez 4
(2 )e d
cos2 d cos
d
(587)
h210|H |200i = h200|H |210i =
32 0
1
0
=

eEz ~2
= 3|e|Ez a0 = T
mZe

(588)

with = r/2a0 . This gives again a 2 2 problem as in the previous problem. The eigenenrgies of the perturbing
Hamiltonian are therefore E = 0, 0, T, T . The eigenvectors are given by |21i = |211i and |2, 1i = |21, 1i, and
|20 i = 12 (|200i |210i) with H |20 i = T |20i. Note that as a result of the Stark effect l is no longer a good
quantum number, but m still is.
Second order energies. Continuing collecting terms, we find for 2
H0 n2 + H1 n1 = En0 n2 + E1 n1 + En2 n0 .

(589)

Again taking the inner product with n0 gives

hn0 |H0 |n2 i + hn0 |H1 |n1 i = En0 hn0 |n2 i + En1 hn0 |n1 i + En2 hn0 |n0 i.

(590)

Since hn0 |H0 |n2 i = En0 hn0 |n2 i, we find

En2 = hn0 |H1 |n1 i En1 hn0 |n1 i.

(591)

But n0 is orthogonal to n1 since

hn0 |n1 i =

m6=n

0
cnm hn0 |m
i = 0.

(592)

So,
En2 = hn0 |H1 |n1 i =

m6=n

0
i=
cnm hn0 |H1 |m

X h 0 |H1 | 0 ih 0 |H1 | 0 i
X |h 0 |H1 | 0 i|2
m
n
n
m
m
n
=
.
0
0
En0 Em
En0 Em

m6=n

(593)

m6=n

Two-level system. One of the simplest, but nevertheless very important, problems is that of a two level system.
Let us take the energies of the two states E1 = 0 and E2 = . This is our unperturbed Hamiltonian H0 . Now let us
introduce a coupling between the two states h2|H1 |1i = T . According to second-order perturbation theory the energy
should be equal to
E12 =

|h2|H1 |1i|2
T2
= .
E1 E2

Similarly, we find that E22 = T 2 /. We can also solve this problem actually by solving the determinant


 0 T 
2
2
2


 T  = E( E) T = E E T = 0.

(594)

(595)

64
This gives
E1,2

1p 2

=
+ 4T 2 =
2
2
2
2

1+

4T 2
2

(1 + 2T ) =
=

2
2
2

(596)

2
T

which is in agreement with perturbation theory in the limit T .

Example
Let us consider a three-level system given by the matrix

0 t T
|1i
H = t 0 T |2i
T T
|3i
In the case that t and T , we can see this as

0 0 0
0
H0 = 0 0 0 ,
Ht = t
0 0
0

T2

(597)

a Hamiltonian plus two different perturbations:

t 0
0
0 T
0 0 ,
0 T .
HT = 0
0 0
T T 0

(598)

So we have three levels of which two a degenerate. So we have two choices: first applying degenerate perturbation
theory and then nondegenerate perturbation theory or the other way around. The former is simpler, so lets do that
first. Degenerate perturbation theory for Ht involves solving a 2 2 eigenvalue problem:


 E t 
1



E = t

| ti = (|1i |2i).
(599)
 t E 
2

Although we can call this degenerate perturbation theory, we can also say we are solving the problem but restricting
ourselves to the basis set containing |1i and |2i. We then can write down the matrix for the new basis set | ti, |ti,
and |3i. For this we need to know the matrix elements

1
2T
.
(600)
h3|HT | ti = (|1i |2i) =
0
2
Let us consider a three-level system given by the matrix

t 0 2T
t
0
H = 0
2T 0

| ti
|ti
|3i

(601)

This matrix reduces to a 2 2 and a 1 1 matric since |ti does not couple to any other states. Solving the 2 2
matrix to lowest order gives and energy shift of 2T 2 /. So we find the following eigenvalues (to lowest order):
0
1
= + t. This
t 2T 2 /, t, + 2T 2 /. Note that we take the difference E30 E1,2
= and not E30 Et
would give a more accurate description but is an order higher in perturbation. This higher-order correction would be
2
t
2T 2 /( + t)
= 2T
(1 )
We could also have started with the perturbation HT . To lowest order, this will gives energies
E3 = E30 +

X |hi|HT |3i|2
2T 2
=

+
,
E30 Ei0

i=1,2

E1 =

T2
,

E2 =

T2
,

(602)

We now need to apply the perturbation Ht . This will couple the perturbed states 1 and 2:
|1 i = |1i +

T
|3i,

|2 i = |2i +

T
|3i.

(603)

T 2 1/2
T 2
Note that it appears that these states are not normalized. However, the normalization is (1 +
)
.
= 1 12
This means that the wavefunction is normalized to order T / the normalization is a higher-order effect. We have to
take some care in calculating the matrix element between |1 i and |2 i,
 2
T
T
T2 T2
T2

h1 |H|2 i = h1|Ht |2i + (h1|HT |3i + h3|HT |2i) +

h3|H0 |3i = t 2
+
= t
.
(604)

65
This leads to the following eigenvalue problem


2 
 T2
 E t T 



2
2
 t T T E 

E=

T2
T2
(t
)=

t
t

2T 2

(605)

giving the same result as above.

B.

Time-dependent perturbation theory

Another important type of perturbations are time-dependent ones. Suppose at a certain point a perturbation H1 (t)
is turned on (say electromagnetic waves) that leads to transition between different states. Thus our total Hamiltonian
is given by
H(t) = H0 + H1 (t).

(606)

The eigenfunctions of the unperturbed Hamiltonian are

|n (t)i = ein t |n i

with

H0 |n i = En |n i = ~n |n i.

The real eigenstates (with perturbation) can be written as a linear combination of the unperturbed states
X
|(t)i =
cn (t)|n (t)i,

(607)

(608)

where the time-dependence of the wavefunction due to the pertubation is put in the coefficients cn (t). This wavefunction is a solution of the Schrodinger equation

X

dcn (t)
{H0 + H1 (t)}|(t)i = i~ |(t)i =
En |n (t)i + i~
|n (t)i .
(609)
t
dt
n
Taking the inner product with hk (t)| gives
i~

dck (t) X
=
hk (t)|H1 (t)|n (t)icn (t).
dt
n

(610)

Now let us suppose that initially our system is in state m. So the unperturbed coefficients are c0m = 1 and c0k = 0 for
k 6= m. To lowest order the coefficients can be written as cn (t) = c0n + c1n (t), which means for k 6= m
i~

dck (t)
dc1 (t) X
= i~ k
hk (t)|H1 (t)|n (t)ic0n = hk (t)|H1 (t)|m (t)i.
=
dt
dt
n

(611)

Integrating gives
ck (t) =

1
i~

hk (t )|H1 (t )|m (t )idt .

Now suppose H1 (t) can be factorized as H1 f (t), then we can write

Z
Z t

1 t
1

ck (t) =
hk (t )|H1 |m (t )if (t )dt = hk |H1 |m i
ei(m k )t f (t )dt .
i~
i~

(612)

(613)

The probability of finding the system in state k is given by |ck |2 .

C.

Transition probabilities

Let us now consider the case of an electromagnetic field that is switched on at t = 0. The perturbation is then

0
t<0
H1 (t) =
,
(614)
2H1 cos t t 0

66
where the factor 2 is introduced to obtain the traditional expression for Fermis Golden Rule in the end. For the
transition from an initial state i to a final state f , we find
Z t

1
cf (t) =
hf |H1 |ii
eif i t (eit + eit )dt
(615)
i~
0
 i(f i )t

1
e
1 ei(f i +)t 1
+
(616)
= hf |H1 |ii
~
f i
f i +
where f i = f i . Using
ei 1 = ei/2 (ei/2 ei/2 ) = 2iei/2 sin(/2),

(617)

we can rewrite the equation as

cf (t) =

 i(f i )t/2

2i
e
sin[(f i )t/2] ei(f i +)t/2 sin[(f i + )t/2]
hf |H1 |ii
+
.
~
f i
f i +

(618)

We can distinguish two cases where cf peaks as a function of . In the first case ~ = ~f i = Ef Ei the final state
energy is higher than the initial state energy. Therefore, the atom absorbs energy and goes into a higher state. In the
second case ~ = ~f i = Ei Ef the initial state has a higher energy. This means that a photon is emitted from
the system. Note that this decay process is stimulated by the presence of radiation of the same frequency.
Let us consider the case of aborption, i.e., f i > 0. We find that the transition probability is given by
Pif = |cf (t)|2 =

4|hf |H1 |ii|2

1
sin2 [ (f i )t],
~2 (f i )2
2

(619)

see Fig. 13.9 in Liboff. The states in the region

|~f i ~| = |Ef (Ei + ~)|2

2~
= E
t

(620)

have greatest probability of being excited. When we study the long-time evolution of the, we can use the approximation
2
sin2 (t/2)
lim
,
t
t 2

(621)

2t
|hf |H1 |ii|2 (f i ).
~2

(622)

dPif
2
=
|hf |H1 |ii|2 (Ef Ei ~),
dt
~

(623)

() =
so that
Pif =
The transition probability per time unit is now
pif =

where the function is now expresses the conservation of energy. This is known as Fermis golden rule.
D.

Interaction of radiation and atom

The Hamiltonian including the electromagnetic field described by the vector potential A is given by
H=

(p eA)2
+ V (r).
2m

(624)

Expanding the kinetic energy gives the interaction terms

H1 =

e
(p A + A p)
2m

and

H1 =

e 2
A .
2m

(625)

The latter term is squared in A. For a nonzero matrix element, the number of photons is conserved. It therefore
scatters the radiation. This is known as Thompson scattering. We shall neglect it in the further discussion.

67
The matrix element is now given by
e
hf |H1 |ii =
2m

drf (p A + A p)i .

(626)

Using that
(A) = A + A.

(627)

If we take the so-called Coulomb gauge, we have A = 0, and

Z
e
drf A pi .
hf |H1 |ii =
m

(628)

We can distinguish two different matrix elements

hf |H1 |ii

hf |H1 |~; ii

or

h~; f |H1 |ii,

corresponding to the absorption and emission of a photon, respectively.

Now suppose we describe the vector potential by plane wave
i
1 X X h
A(r, t) =
ek () ei(krt) + complex conjugate ,
V k =1,2

(629)

(630)

where () denotes the two polarization vectors of the radiation and ek complex coefficients. Using this potential
gives a complicated problem. Often calculation are restricted to the lowest orders of the expansion of the vector
potential in r
eikr = 1 + ik r +

(631)

The first term gives the (electric) dipole approximation. This gives a matrix element proportional to r. This can be
seen as follows. The matrix element is now proportional to
e ()
hf |p|ii
m

(632)

Making use of the commutation relation

[p2 , r] = 2i~p

p=

im
[Hatom , r],
~

(633)

we find
e im ()
ie
hf |(Hatom r rHatom )|ii = (Ef Ei )() hf |r|ii = ie() hf |r|ii,
m ~
~

(634)

which is an expectation value of r, hence the name dipole approximation. Now let us make use of the fact that we
can write the components in spherical coordinates as

1
i
r
r1 = r sin e = 2 (x + iy)
4 1
rq = r
Y (, ) r0 = r cos = z
.
(635)

3 q
r1 = r sin ei = 12 (x iy)

z , = 12 (
i are
Note that the direction of the polarization vectors are + = 12 (
ex + i
ey ), z = e
ex i
ey ), where e
the axis of the system, i.e., the atom in our case. + and are left and right circularly polarized light, respectively;
z is linearly polarized light along the z axis. Obviously, for a free atom, the choice of our axis does not make any
difference for the spectrum that we obtain. The situation changes when we put the atom in a magnetic field. This
will make the spectra strongly depend on the magnetic field. We can now evaluate the matrix elements for atomic
orbitals
Z
Z

1
3

hn l m |rq |nlmi hl m |r|lmihn l |Yq |nli =

dr r Rn l (r)Rnl (r) (Yml ) Yq1 Yml d.
(636)
0

68
The evaluation is rather complicated but leads to some simple selection rules
l = l l = 1
m = m m = 0, 1.

(637)
(638)

The integration over the angular coordinates is related to the vector coupling or Clebsch-Gordan coefficients, that
we
when we start coupling angular momenta. The integration over is straightforward and gives
R will meet again
d exp(i(m + q + m)), leading to the m selection rules. The integration over gives l selection rules. The second
term in Eqn. (631) gives electric quadrupole (proportional to r2 ) and magnetic dipole transitions (proportional to
r p = L). These are small, but certainly not always negligible.
The combination of energy conservation and the selection rules provide a powerful tool to study the electronic states
of atoms but also molecules and solids.

69
X.

MANY-ELECTRON ATOM

Before we can start talking about filling the orbtitals we have to introduce an important concept: Paulis exclusion
principle. Without it, the most obvious thing to do is putting all the electrons in the lowest level, i.e., nl = 10.
Experimentally, this is not observed, but one sees that each orbital from the Hydrogen atom contains only one
electron. This is quite unlike, for example, photons. We can put as many photons in a particular photon state (say,
a mode in an electromagnetic cavity or a plane wave mode) as we want. This corresponds simply to increasing the
intensity of the radiation.
A.

Paulis exclusion principe

In quantum mechanics, particles are indistinguishable. This means that we should be able to interchange two
particles without seeing any effect on the probability function. We can distinguish two fundamentally different types
of particles, namely those with a symmetric and antisymmetric wavefunction. The quantization of the harmonic
oscillator, photon, and a number of fundamental particles have symmetric wavefunctions. These particles are called
bosons. Here we find that a permutation between particle i and j for a N particle wavefunction
Pij (k1 r1 , . . . , ki ri , . . . , kj rJ , . . . , kN rN ) = (k1 r1 , . . . , kj rj , . . . , ki ri , . . . , kN rN ),

(639)

where ki means that a particle is in a certain quantum state ki (for example a plane wave). Note that for the harmonic
oscillator we considered only one type of quantum state, but in principly we can take a set of oscillators. Let us look
at two bosons. The wavefunction is given by
1
S (k1 r1 , k2 r2 ) = (k1 (r1 )k2 (r2 ) + k2 (r1 )k1 (r2 )).
2

(640)

Obviously the interchange of the particles does not affect the wavefunction. However, the other type of particles,
called Fermions, have antisymmetric wavefunctions. These are, e.g., electrons and protons. This gives
Pij (k1 r1 , . . . , ki ri , . . . , kj rj , . . . , kN rN ) = (k1 r1 , . . . , kj rj , . . . , ki ri , . . . , kN rN ).

(641)

Although this looks like a simple phasefactor and would make absolutely no difference in the probability density 2 ,
it has very important consequences. Let us take ki = kj , this would mean that the wavefunction on the right and left
hand side are equal. However, this implies that = , which is only possible if = 0. Therefore, two Fermions
can never be in the same quantum state. A two-fermion state would look like this
1
A (k1 r1 , k2 r2 ) = (k1 (r1 )k2 (r2 ) k1 (r2 )k2 (r1 )).
2

(642)

The symmetric and antisymmetric wavefunctions give two fundamentally different particles. Examples of bosons
are photons, and K mesons. These are particles with integral spin. Also the harmonic oscillator behaves like a
boson-like particle. Particles with half integral spin behave like fermions. Examples are electrons, protons, and
neutrons.
For bosons the general wavefunction is
r
i nki ! X
k1 ,...,kN =
k1 (r1 ) kN (rN ),
(643)
N!
P

where the summations runs over all the possible permutations ofPdifferent indices ki . This gives the normalization
constant. nki indicates how many k-values are ki . Note that the i nki = N . For two particles with S = 0, we have
1
k1 ,k2 = {k1 (r1 )k2 (r2 ) + k2 (r1 )k1 (r2 )} .
2

(644)

For k1 = k2 = k, we find
k1 ,k2 =
using the fact that nk = 2.

2!
k (r1 )k (r2 ) = k (r1 )k (r2 ),
2!

(645)

70
Remember that for the harmonic oscillator we could write the wave function like
1
|ni = (a )N |0i.
N!

(646)

Now forget for the moment the relation of a with x and p, since this result is far more general. We can in principle make
different types of bosons. These operators have some nice characteristics that we would like to see in wavefunctions:
they are indistinguishable. Also they are symmetric. This follows from the more general commutation relations
[ak , ak ] = k,k

[ak , ak ] = 0

and

and

[ak , ak ] = 0

(647)

Thus for k 6= k , we can write

ak ak = ak ak .

(648)

Therefore, interchange of two particles does not change the sign of the wavefunction. Let us look at the two-particle
wavefunctions. If the two particles are in the same state (i.e. have the same quantum number) then the wavefunction
should be similar to that of the harmonic oscillator:
1
|kki = ak ak |0i.
2

(649)

This wavefunction is now properly normalized since

1
1
h0|ak ak ak ak |0i = h0|ak (1 + ak ak )ak |0i
2
2
1
1

= h0|ak ak |0i + h0|ak ak ak ak |0i

2
2
1
1

= h0|(1 + ak ak )|0i + h0|(1 + ak ak )(1 + ak ak )|0i = 1

2
2

hkk|kki =

(650)
(651)
(652)
(653)

since ak |0i = 0. For two different wavevectors, we have

|k1 k2 i = ak1 ak2 |0i.

(654)

This gives the right normalization since

hk1 k2 |k1 k2 i = h0|ak1 ak2 ak2 ak1 |0i = h0|ak1 (1 + ak2 ak2 )ak1 |0i = h0|ak1 ak1 |0i = h0|(1 + ak1 ak1 )|0i = 1

(655)

For wavefunctions there is a dependence on the particle by the coordinate k (ri ). The operators are identical, which
means that we do not have to make all the different permutations. For more than one k value, we can generalize the
wavefunction as
1
|{nki }i = p
i (aki )nki |0i.
i nki

(656)

The situation for electrons (and other spin-1/2 particles) is fact a bit more complicated since the total wavefunction,
i.e., orbital and spin, has to be asymmetric. This means S (r1 , r2 )A (1, 2) or A (r1 , r2 )S (1, 2), where (1, 2) is the
wavefunction for the spin. There are four ways to combine two spins (1) (2), (1) (2), (1) (2), and (1) (2).
(1) (2) and (1) (2) are clearly symmetric spin functions, since interchange of the particles does not change the
sign of the wave function. These are the 11 (1, 2) and 1,1 (1, 2) components of the triplet (S = 1). Since Sz = 1, 0, 1
for S = 1, there should also be a S = 0 component. This can be obtained by using the step operator for the spin
p

S 11 =
(1 + 1)(1 1 + 1)10 = 210 = (1) (2) + (1) (2)
(657)
1

10 = [ (1) (2) + (1) (2)]

(658)
2
where Eqn. (412) has been used. Note that this is also a symmetric wavefunction. The remaining wave function has
to be orthogonal to the previous one
1
00 (1, 2) = [ (1) (2) (1) (2)]
2

(659)

71
which is antisymmetric. Since there is only one component (called a singlet), this has to be S = 0. The four total
wavefunctions for two electrons are therefore 1Sz (r1 , r2 ) = A (r1 , r2 )1Sz (1, 2) with Sz = 1, 0, 1 and 00 (r1 , r2 ) =
S (r1 , r2 )00 (1, 2). The S = 1 and Sz = 1 is given by
1
11 (r1 , r2 ) = A (r1 , r2 )11 (1, 2) = [k1 (r1 )k2 (r2 ) k2 (r1 )k2 (r1 )] (1) (2)
2
1
= [k1 (r1 ) (1)k2 (r2 ) (2) k2 (r1 ) (1)k2 (r1 ) (2)]
2


1  k1 (r1 ) (1) k1 (r2 ) (2) 
= 
2 k2 (r1 ) (1) k2 (r2 ) (2) 

(660)
(661)
(662)

The S = 0 and Sz = 0 wavefunction is given by

1
1
00 (r1 , r2 ) = S (r1 , r2 )00 (1, 2) = [k1 (r1 )k2 (r2 ) + k2 (r1 )k2 (r1 )] [ (1) (2) (1) (2)]
2
2
1
=
[k1 (r1 ) (1)k2 (r2 ) (2) k2 (r1 ) (1)k1 (r2 ) (2)
2
k1 (r1 ) (1)k2 (r2 ) (2) + k2 (r1 ) (1)k1 (r2 ) (2)]
Note that this can also be written as a combination of Slater determinants





1  (r ) (1) k1 (r2 ) (2) 
1  k1 (r1 ) (1) k1 (r2 ) (2) 
1
 k1 1

00 (r1 , r2 ) =
2
2 k2 (r1 ) (1) k2 (r2 ) (2) 
2  k2 (r1 ) (1) k2 (r2 ) (2) 
1
= [k1 ,k2 (r1 , r2 ) k1 ,k2 (r1 , r2 )]
2

(663)
(664)
(665)

(666)
(667)

In fact, we are really entering already the realm of many-body physics (although you might not directly realize the
significance of it). Suppose, k1 k2 ,1Sz and k1 k2 ,00 have different energies, then the energy of the two-particle state
depends strongly on the relation of the two particles in the wavefunction. A lot of theories (band structure calculations
in particular, a very large number of people are working in that area in solid-state physics) do not take into account
that difference. In so-called independent-particle models, 10 and 00 are essentially equivalent. In that case, one
might as well use (1) (2) and (1) (2), which are simpler to handle.
Note that in general we can built up antisymmetric states from Slater determinants


 k1 (r1 )1 (1) k1 (rN )1 (N ) 


1 
.
.

k1 1 ,...,kN N (r1 rN ) = 
(668)

.
.


N! 

(r ) (1) (r ) (N )
kN

kN

where i =, . As a result of the properties of the determinant, having two columns with the same k and implies
that the determinant is zero, just as for the exclusion principle. Interchange of two columns also leads to a change in
sign of the determinant, giving us directly the antisymmetry. This is obvious for the Sz = 1, 1 states of the triplet.
For the Sz = 0 states, one needs a combination of Slater determinants
1
k1 k2 ,00 = (k1 ,k2 k1 ,k2 )
2

and

1
k1 k2 ,10 = (k1 ,k2 + k1 ,k2 ) .
2

(669)

Working with Slater determinants becomes quite elaborate. Often it is much more convenient to work with operators
in a similar way as we saw for bosons. However, unlike bosons that satisfy commutation relations, fermions satisfy
anticommutation relations
{ck , ck } = k,k ,

{ck , ck } = 0,

{ck , ck } = 0,

(670)

where the term in braces is defined as

{A, B} = AB + BA.

(671)

Suppose, we have a vacuum state |0i. We can then put a fermion in there by
ck |0i = |ki,

(672)

72
where |ki indicates that there is one electron in state k and zero in all the others (note that according to Paulis
exclusion principle there can only be zero or one electron in a state).
However, we cannot put two particles in the same state, since
ck ck |0i =

1
{c , c }|0i = 0.
2 k k

(673)

We can put them in different quantum states, thus for k 6= k

ck ck |0i = ck ck |0i.

(674)

Note that the operators directly take care of the antisymmetry of the wavefunctions. Including spin, we can write the
wavefunctions |k1 k2 , SSz i for two particles in states labeled with quantum numbers k1 and k2 with different spin S
and Sz as

1 
|k1 k2 , 11i = ck2 ck1 |0i , |k1 k2 , 10i = ck2 ck1 + ck2 ck1 |0i , |k1 k2 , 1, 1i = ck2 ck1 |0i (675)
2

1 
|k1 k2 , 00i = ck2 ck1 ck2 ck1 |0i.
(676)
2
In general, we can write the many-body Fermion wavefunction as
|{nki }i = i (cki )|0i.

(677)

Note, that the normalization is unity since we can never have more than one fermion in the same quantum state.
B.

The Helium atom

Let us start with the Schrodinger equation for the Helium atom


p21
p22
e2
Z
Z
1
H=
+
+

+
,
2m 2m 40
r1
r2
r12

(678)

takin r12 = |r1 r2 |. Note that this is already a problem which is impossible to solve exactly. Since we increase the
complexity of the system, we simplify the Hamiltonian by leaving out the relativistic terms. Without the last term,
this problem could be solved easily, giving Hydrogen-like solutions. The two electrons would simply go into the 1s
level (nl = 10) giving a ground state singlet
1

10 (r1 , r2 ) = 100 (r1 )100 (r2 )00 ,

(679)

for nlm = 100 and where the spin singlet 00 is given in Eqn. (659). The superscript in 1 10 stands for 2S + 1 = 1,
the subscript for nl = 10. The energy would simply be
E0 = 2 R

22
Z2
=
2

(13.6)
= 8 13.6 = 108.8 eV.
n2
12

(680)

This is quite far of the experimentally observed value of 78.975 eV. Let us therefore try to get an estimate of the
electron-electron interaction using perturbation theory.
Perturbation theory. The lowest order correction is
Z
e2
e2
|100 (r1 )|2 |100 (r2 )|2
1
1
E1 = h 10 |
| 10 i =
dr1 dr2
(681)
40 r12
40
r12




2Z
e2
Z3
1
2Z
=
exp (r1 + r2 ) r12 dr1 d1 r22 dr2 d2 .
(682)
3
40 a0
r12
a0
Let us do the integral over r2 first. Let us fix r1 along the z axis. Then we can write
q
|r1 r2 | = r12 2r1 r2 cos 2 + r22

(683)

73
We can now write
I2 =

e2Zr2 /a0
dr2 = 2
|r1 r2 |

e2Zr2 /a0
p
r22 sin 2 dr2 d2 d2 .
r12 2r1 r2 cos 2 + r22

The integral over 2 is straightforward

q

Z
sin 2
1
2
2
p
r1 2r1 r2 cos 2 + r2
d2 =
r1 r2
r12 2r1 r2 cos 2 + r22
0
0
 2
r2 < r1
1
.
=
(r1 + r2 |r1 r2 |) = r21
r1 r2
r2 r2 > r1
For I2 , we have to separate the integral into two regions
 Z r1



Z
1
a3
Zr1
2Zr2 /a0 2
2Zr2 /a0
2Zr1 /a0
I2 = 4
e
r2 dr2 +
e
r2 dr2 = 3
1e
(1 +
) .
r1 0
Z r1
a0
r1
Putting this back into the equation for E1 gives

Z 
e2 Z 3
Zr1
5 Ze2
2Zr1 /a0
E1 =
1

e
(1
+
)
e2Zr1 /a0 r1 sin 1 dr1 d1 d1 =
.
3
40 a
a0
8 40 a0

(684)

(685)
(686)

(687)

(688)

From this we can calculate the energy

Ze2
E = E0 + E1 =
40 a0



5
Z
= 108.8 + 34 = 74.8 eV.
8

(689)

This is a lot closer to the 79 eV, but we have overshot the ground-state energy somewhat. This is understandable.
We have just taken the unperturbed ground-state wavefunctions to calculate the electron-electron interaction. Since
the electrons repel each other, this is unlikely to be the lowest state. We can gain some energy by increasing the
radial extent of the wavefunction somewhat to try to find a new minimum (obviously, from the point of view of the
potential energy from the nucleus, expanding the wave function is not a good idea). A larger radial extent of the
wavefunction corresponds effectively to a smaller Z of the nucleus. This effect is called screening. An electron feels a
smaller attractive force from the nucleus as a result of the repulsive force of the other electron(s). Let us try to find
an optimal value for Z by means of a variational approach.
Variational approach. Now let us suppose the electron has a wavefunction in which it sees an effective charge of
not Z = 2, but Z < 2. The presence of the other electron makes the charge slightly lower. Our modified wavefunction
then becomes
!3

1
Z
1
1 + r2 )/a0 )00 .
10 (r1 , r2 ) = 100 (r1 )100 (r2 )00 =
exp(Z(r
(690)
a0
Using the above results, we can write down the energy of the Hamiltonian
2
2
2
2
e2
= p1 + p2 Ze Ze +
H
,
2m 2m 40 r1
40 r2
40 r12

(691)

up to first order in the electron-electron interaction


2 
Ze
5
1

h 10 |H| 10 i =
Z
.
40 a0
8
1

(692)

This leaves us still to evaluate the rest of the Hamiltonian

2
= (Z Z) e
H = H H
40

1
1
+
r1
r2

Since this is close to the expectation value of the Coulomb potential of the nucleus, we find


2 Z
1
1

10 |H|1
10 i = (Z Z) e
dr1 dr2 100
(r1 )100
(r2 )
+
100 (r1 )100 (r2 )
E = h1
40
r1
r2
!
2

e
Z
e2
= (Z Z)
2
= 2(Z Z)Z
,
40
a0
40 a0

(693)

(694)
(695)

74
using the fact that h1/ri = Z/a0 . The total energy is then given by




2
e2
5
5
+ E = e
E=E
Z 2 Z 2(Z Z)Z =
Z 2 Z + 2Z Z .
40 a0
8
40 a0
8

We want to minimize this energy with respect to Z,



dE
e2
5
=
2Z + 2Z = 0
40 a0
8
dZ

5
27
Z = Z
=
= 1.6875 . . . ,
16
16

(696)

(697)

with Z = 2 for a Helium atom. Therefore the electron effectively sees a somewhat smaller charge of the nucleus as a
result of the presence of the other electron. The energy of the ground state for Helium is then

2
5
e2
.
E10 = Z
16
40 a20

(698)

For Z = 2, we otbain E10 = 77.38 eV which is quite close to the experimentally observed value of 79 eV. Further
refinements are possible but less didactic.
C.

Periodic table

After the discussion of the Helium atom, we are ready to fill the atoms with electrons. Since we are not quite able
to calculate everything, we are still left with some phenemenology. From the simple Hydrogen, we would expect that
the orbitals with the same quantum number n have the same energy. However, electron-electron interactions severely
change this picture. Orbitals with a small radial extent suffer from larger Coulomb interactions and will therefore be
filled later. A phenemenological scheme of filling the orbitals is
6s

6p

5s

5p

4s

4p

4d

3s

3p

3d

2s

2p

1s

5d

5f
4f

(699)

This is just some scheme to remember it. We have used here the common notation nl, where for l a symbols is used
instead of an integer, i.e., l = 0, 1, 2, 3 = s, p, d, f . Note that there are no g orbitals. The rumor goes that when God
came to the g orbitals, he found them so complicated that he decided to stop there (forgive the political incorrectness).
One starts at the bottom and follows the diagonals, i.e., 1s, 2s, 2p, 3s, 3p, 4s, 3d, . . . . We can put 2(2l + 1) electrons in
an orbital. Note that there are exceptions. First, we observe that the higher l numbers are filled later than lower l. A
physical argument for that is that the orbitals have to be orthogonal to each other, also in radial extent. That means
the Rnl has an extra wiggle when l increases by one (see Fig. 10.16 of Liboff). This pushes the electron density to
larger r and reduces the electron-electron interactions. Even in a solid, the orbitals where a certain l appears for the
first time (3d and 4f ) are more atomic. (the larger Coulomb interaction will also have as a result that the electrons
like to have their spins parallel to reduce the Coulomb interactions, as we will see below. These materials often show
interesting magnetic properties [first row of transition metals for 3d, and rare earths for 4f ]).
We have already treated Hydrogen and Helium. Helium has a filled 1s shell and is the stablest atom, a so-called
noble element.
Because of their wider radial extent, the ns levels (with n > 1) see less of the nucleus and are therefore note very
strongly bound. (See the figure with first ionization energies, Liboff 12.11). For example, for Li, the 1s electrons
screen the nucleus from the 2s electron. Elements with an open s-shell (and no other shells) are unstable. This is
also true for solids. Alkali and also Alkaline metals (the first and second row in the periodic table are very unstable).
For example, Lithium and Sodium react with water, creating H2 gas.

75
Filling the np shell makes the elements stabler. Note the little dip at p4 . Actually this is not a dip but it results
from an increased stability of the p3 configuration making the p4 configuration look less stable. p3 is a half-filled shell,
and all three the electrons have their spin parallel making an extra stable configuration. This will be discussed in
more detail below.
The elements where the p shell is filled are very stable and called noble elements (Ne, Ar, Kr, Xe). They react very
little with other elements and usually appear in the gas phase since they interact only via the weak Van der Waals
forces, making it very difficult to form a solid. Also a filled d shell causes extra stability. That is why we like silver
and gold so much (actually, as you look more closely you see that the d shell borrows an electron from the s shell
to form a nice closed shell. The same you see for Chromium and Molybdenum, but here to get a nice half-filled shell.
Tungsten on the other hand cares less. Here the 6d orbital has a larger radial extent and electron-electron interactions
in the d shell are not strong enough to be able to borrow a s electron.
D.

Coulomb interaction

Now that we have filled the shells, we would like to find out what the state is within a certain open shell. This
is mainly determined by the Coulomb interaction between the electrons. One important problem in physics are the
Coulomb interaction between electrons,
V =

e2
e2
= p 2
,
|r1 r2 |
r1 2r1 r2 cos + r22

(700)

where is the angle between r1 and r2 . We would like to expand the square-root term in a binomial series:
(1 2xt + t2 )1/2 =

(2n)!
0!
2!
4!
(2xt t2 )n = 0
(2xt t2 )0 + 2
(2xt t2 )1 + 4
(2xt t2 )2 +
2n (n!)2
2
2
2
2
2
(0!)
2
(1!)
2
(2!)
n=0

3
1
= 1 + xt + ( x2 )t2 + O(t3 ) +
2
2

(701)

where we have taken x = cos . The term in front of the tn are the Legendre polynomials of order n. This function
is also known as the generating function for the Legendre polynomials,
(1 2xt + t2 )1/2 =

Pn (x)tn .

(702)

n=0

However, we can only use this theorem for |t| < 1. We can achieve this in our potential by defining r> and r< the
greater, respectively, the smaller of r1 and r2 :
e2
V =
r>


1/2

2
k
r<
r<
e2 X r<
Pk (cos ).
1+ 2 2
cos
=
r>
r>
r>
rk
k=0 >

For Hydrogen-like wavefunctions i = Rni li (ri )Yli mi (i , i ), the Coulomb matrix element is given by
Z
Z
(1 2 |V |3 4 ) =
dr1 dr2 1 (r1 )2 (r2 )V 3 (r1 )4 (r2 )

k
e2 X r<
= (1 2 |
Pk (cos )|3 4 )
r>
rk
k=0 >

(Rn1 l1 Rn2 l2 |

k=0

k
e2 r<
|Rn3 l3 Rn4 l4 )(Yl1 m1 Yl2 m2 |Pk (cos )|Yl3 m3 Yl4 m4 )s1 ,s3 s2 ,s4 ,
k+1
r>

(703)

(704)
(705)
(706)

where the delta function imply spin conservation. Here we use ( ) for the matrix element. The different use from
h i will become clear in a minute. We can now use the spherical addition theorem,
l
4 X
Pl (cos ) =
Ylm (1 , 1 )Yl,m (2 , 2 ),
2l + 1
m=l

(707)

76
which connect the Legendre polynomial to the spherical harmonics. We can now write the latter integral as two
integrals over the coordinates of r1 and r2 .
(Yl1 m1 Yl2 m2 |Pk (cos )|Yl3 m3 Yl4 m4 ) =

k
4 X

(Yl1 m1 Yl2 m2 |Ykq (1 1 )Ykq

(2 2 )|Yl3 m3 Yl4 m4 )
2k + 1
q=k

= c (l1 m1 , l3 m3 )m1 m3 ,q ck (l4 m4 , l2 m2 )m2 m4 ,q ,

where ck are simple numbers which can be obtained by evaluating the integral
r
Z 2
Z
4

ck (lm, l m )q,mm =
d
d sin Ylm
Ykq Yl m
2k + 1 0
0

(708)

(709)

where the conservation of momentum q = m m follows from the integration over .

Now suppose we look at one shell and consider diagonal matrix elements only. The radial integral is then always
F k = (Rnl Rnl |

k
e2 r<
|Rnl Rnl ).
k+1
r>

(710)

Now we are dealing with many-body wavefunction where the particles are indistinguishable. So in reality, we have
two different parts in the matrix element
hnlm, nlm |V |nlm, nlm i = (nlm, nlm |V |nlm, nlm ) (nlm, nlm |V |nlm , nlm)
Z
Z
=
dr1 dr2 |nlm (r1 )|2 V |nlm (r1 )|2
Z
Z

dr1 dr2 nlm

(r2 )nlm
(r1 )V nlm (r2 )nlm (r1 )
X

=
ck (lm, lm)ck (lm , lm ) , (ck (lm, lm ))2 F k ,

(711)
(712)
(713)
(714)

where we have introduced the spin =, of the electron. The ( ) now indicates the evaluation of the matrix
element in which the first wavefunction is that of particle 1 and the second that of particle 2. The second term is the
exchange term. Note that the electrons can only be exchanged if they have equal spins.
When we have more than one shell, we can have to different radial matrix elements
F k = (Rnl Rn l |

k
e2 r<
|Rnl Rn l )
k+1
r>

and

Gk = (Rnl Rn l |

k
e2 r<
|Rn l Rnl ).
k+1
r>

(715)

The matrix element now becomes

hnlm, n l m |V |nlm, n l m i = (nlm, n l m |V |nlm, n l m ) (nlm, n l m |V |n l m , nlm)
X
=
ck (lm, lm)ck (l m , l m )F k , (ck (lm, l m ))2 Gk .
(716)
k

Now let us look at the simplest example: two electrons in an s orbital. We have to know now the coefficients
ck (00, 00) = k,0 since lm = 00. The only non zero k turns out to be k = 0. We can see that by noting that the
maximum k one can make by adding l and l is 2l = 0. If the two electron are in the same s orbital our wavefunction
is |n00 n00 i. The matrix element is
0
0
hn00 , n00 |V |n00 , n00 i = c0 (00, 00)2 Fn0,n0
= Fn0,n0
,

(717)

k
0
where we have used the notation Fnl,n
is the roughly 34 eV that we calculated for the He atom in the
l . Where F
previous section. Things become more interesting when we put the electrons in different shells. Again we can take
the spins antiparallel, this gives a very similar result
0
0
hn00 , n 00 |V |n00 , n 00 i = c0 (00, 00)2 Fn0,n
0 = Fn0,n 0 .

(718)

However, when we put the spins parallel, we can also have the exchange term
0
0
0
0
0
0
hn00 , n 00 |V |n00 , n 00 i = c0 (00, 00)2 Fnl,n
l c (00, 00)c (00, 00)Gn0,n 0 = Fn0,n 0 Gn0,n 0 .

(719)

77
Since there is only one way to make a total spin of MS = 1 this must be and eigenenergy of the system with two
electrons in two different s orbitals. In this case this is the triplet with SMS = 11. Similarly, we obtain the term with
both spins antiparallel. The situation for MS = 0 is more complicated. Above we found one term, however there are
more matrix elements:
0
0
hn00 , n 00 |V |n00 , n 00 i = c0 (00, 00)2 Fn0,n
0 = Fn0,n 0

(720)

and also an off-diagonal term

hn00 , n 00 |V |n00 , n 00 i = (n00 , n 00 |V |n00 , n 00 ) (n00 , n 00 |V |n 00 , n00 ) = G0n0,n 0(721)
Note that the first term on the right-hand side is zero since the integration over the spin-part of the wavefunction is
zero. However, for the exchange term, we have a non-zero matrix element. This leads to a 2 2 matrix:
 0

F
G0
n00 , n 00
,
(722)
0
0
G F
n00 , n 00
leading to eigenvalues F 0 G0 . The value F 0 G0 comes from the MS = 0 term of the triplet. The F 0 + G0 term
has only a MS = 0 component (note that this energy does not appear for MS = 1. Note that the sum of the two
eigenenergies is 2F 0 , which is equal to the sum of the diagonal elements (the trace of the matrix). This trace is a
constant under a transformation to a different basis set. So we would have argues differently. We know that the
trace is 2F 0 . We have found an eigenvalue for a triplet F 0 G0 for MS = 1. However, the triplet should also have a
MS = 0 component. Therefore, the other eigenstate for MS = 0 should be 2F 0 (F 0 G0 ) = F 0 + G0 .
Let us consider the simple example of two electrons in a p shell. The different electron configurations are shown
in Table X D. We use the notation for a configuaration (m m ), where the indicates = 21 . For the Coulomb
interaction L, ML , S, and MS are good quantum numbers. The different terms can be found by looking at the extreme
ML and MS . The configuration (1+ 1 ) belongs to the term with L = 2 and S = 0. This is denoted by 1 D (say
singlet D). The letter indicates the value of L, which corresponds to the notation The different values for l are often
denoted by
L: 0 1 2 3 4
S P D F G

(723)

The superscript is 2S + 1. Note that the 1 D term appears in the configurations with ML = 2, , 2 and MS = 0.
This energy level is (2L + 1)(2S + 1) fold degenerate. The configuration (1+ 0+ ) belongs to a 3 P term (say triplet P ).
Then there is only on term unaccounted for with ML = MS = 0, i.e., a 1 S.
The diagonal matrix elements are given by
X

(724)
ck (lm, lm)ck (l m , l m )F k (ck (lm, l m ))2 Gk ,
k

where for p2 , we have l = l = 1. For 1 D, we find

E(1 D) = E(1+ 1 ) =

(ck (11, 11))2 F k = F 0 +

1 2
F ,
25

(725)

where the values of ck are given in Table XIV A. In the case of 3 P , there is also an exchange contribution
X

E(3 P ) = E(1+ 0+ ) =
ck (11, 11)ck (10, 10)F k (ck (11, 10))2 Gk
k

= F0

2 2
3
5
F (0G0 + G2 ) = F 0 F 2 ,
25
25
25

TABLE I: Different configurations for p2 .

p2
ML = 2
1
0
-1
-2

MS = 1
(1+ 0+ )
(1+ 1+ )
(0+ 1+ )

0
(1+ 1 )
(1+ 0 )(1 0+ )
+
(1 1 )(1 1+ )(0+ 0 )
(0+ 1 )(0 1+ )
(1+ 1 )

-1
(1 0 )
(1 1 )
(0 1 )

(726)

78
TABLE II: Values for ck (lm, l m ) for l = l = 1.
c0 (lm, l m )
1
0
1
0

m
1
0
0
1

m
1
1
0
1

c2 (lm, l m )
1

5
3
5
2
5
56

TABLE III: Different configurations for p3 .

p3
ML = 2
1
0
-1
-2

MS =

3
2

(1+ 0+ 1+ )

1
2
+ +

(1 1 0 )
(1+ 0+ 0 )(1+ 1 1+ )
(1 0+ 1+ )(1+ 0 1+ )(1+ 0+ 1 )
(0+ 0 1+ )(1+ 1+ 1 )
(0+ 1+ 1 )

12
+

3
2

(1 1 0 )
(1 0+ 0 )(1+ 1 1 )
(1 0 1+ )(1 0+ 1 )(1+ 0 1 )
(0+ 0 1 )(1 1+ 1 )
(0 1+ 1 )

(1 0 1 )

where use has been made of the fact that within one shell Gk = F k . The energy of the 1 S term cannot be calculated
this way. However, here we can make use of the fact that the sum of the diagonal energies is unchanged when
transforming from our basis set into the LS states,
X
X
ck (11, 11)ck (1 1, 1 1)F k +
ck (10, 10)F k
(727)
E(1+ 1 ) + E(1 1+ ) + E(0+ 0 ) = 2
k

4
1
= 2(F + F 2 ) + F 0 + F 2
25
25
0

= E(1 S) + E(3 P ) + E(1 D) = E(1 S) +

(728)
2 0
4
F F 2,
25
25

10 2
which gives E(1 S) = F 0 + 25
F . G enerally, one finds that for electrons in one shell the lowest term is the one with
first, the highest spin multiplicity, and, secondly, the highest orbital multiplicity. In the case of p2 , the 3 P term. This
can be understood by noting that for the highest spin electrons have to occupy different orbitals, therefore reducing
the Coulomb interaction. The same applies, in a more complicated manner for the angular momentum.
Thus we find in general that that
2

P 2 P =1,3 D +1,3 P +1,3 S

(729)

giving a total degeneracy of 1 5 + 3 5 + 1 3 + 3 3 + 1 1 + 3 1 = 36, which is equal to 6 6. This would

be the case for two electrons in different p shells, i.e. pp. However, for p2 , the Pauli exclusion principle states that
electrons cannot be in the same orbital. This makes matters a bit more complicated. We can easily see that 3 D is
forbidden, since this would require a configuration (1+ 1+ ). We know that the total number of configurations is 15.
So we still have 10 left. The other terms are 3, 9, 3, and 1 fold degenerate. Apparently the only way to make 10 is
to take the 3 P and the 1 S configurations, being 9 and 1-fold degenerate.
p3 configuration The procedure described above can be extended for other configurations. Let us consider p3 . The
configuration are given in Table XIV A. In total we have
 
654
6
=
= 20
(730)
3
32
configurations. Let us now see what different LS terms are in these 20 configurations. The configuration (1+ 1 0+ )
must belong to an LS term with L = 2 and S = 12 since there are no configurations with higher L and S that could
give rise to this term. This term is denotes as a 2 D. The configuration (1+ 0+ 1+ ) can only come from an LS term
with S = 32 and L = 0, therefore a 4 S. The first ML and MS terms for which we see more than one configuration is
ML = 1 and MS = 12 . The LS states are linear combinations of the configurations (1+ 0+ 0 ) and (1+ 1+ 1 ).
One of these combinations is part of the 2 D term. The other combination belongs to the LS term with L = 1 and
S = 21 , i.e. 2 P . The total degeneracy is 5 2 + 1 4 + 3 2 = 20 so we have found all the LS terms.

79
Let us now calculate the eigenenergies. We have calculated the matrix elements between two particles. Now we
have three particles. However, the Coulomb interaction is still a two-particle interaction so we need to find all the
possible pairs of electrons. Therefore, for the 2 D
h1+ 1 0+ |V |1+ 1 0+ i = h1+ 1 |V |1+ 1 i + h1+ 0+ |V |1+ 0+ i + h1 0+ |V |1 0+ i
1
5
2
6
= F 0 + F 2 + F 0 F 2 + F 0 F 2 = 3F 0 F 2 ,
25
25
25
25

(731)
(732)

where results for the p2 configuration have been used. Note that the 3 in front of the monopole term F 0 just counts
the number of different pairs. For the 4 S, we have
h1+ 0+ 1+ |V |1+ 0+ 1+ i = h1+ 0+ |V |1+ 0+ i + h1+ 1+ |V |1+ 1+ i + h0+ 1+ |V |0+ 1+ i

 

5 2
1 2
6 2
2 2
3 2
0
0
0
0
0
= F F + F + F (0G + G ) + F F (0G + G )
25
25
25
25
25
15
= 3F 0 F 2 .
(733)
25
where we have used Gk = F k for the same shell since nl = n l and therefore the radial wavefunction (which enter in
the evaluation of the radial matrix element) are the same.
For the 2 P term, we have to make use again of the diagonal sum rule (or calculate the off-diagonal elements, but
this is more complicated). We have two configurations with matrix elements
h1+ 0+ 0 |V |1+ 0+ 0 i = h1+ 0+ |V |1+ 0+ i + h1+ 0 |V |1+ 0 i + h0+ 0 |V |0+ 0 i
5
2
4
= F0 F2 + F0 F2 + F0 + F2
25
25
25
3 2
0
= 3F F
25

(734)

h1+ 1 1+ |V |1+ 1 1+ i = h1+ 1 |V |1+ 1 i + h1+ 1+ |V |1+ 1+ i + h1 1+ |V |1 1+ i

1
5
1
= F0 + F2 + F0 F2 + F0 + F2
25
25
25
3
= 3F 0 F 2 .
25

(737)

(735)
(736)

and

(738)
(739)

The fact that the trace is constant in going from our (m
1 m2 m3 ) basis set to the LS basis, implies that the sum over
the diagonal matrix elements is the same, i.e.

2(3F 0

6
3
3 2
F ) = 6F 0 F 2 = E(2 D) + E(2 P ) = 3F 0 F 2 + E(2 P ),
25
25
25

(740)

giving E(2 P ) = 3F 0 .
Therefore, for the p3 configuration, we find that the 4 S state is the lowest eigenstate, in agreement with Hunds rule
that says that S should be maximum. This results directly from the fact that they have to be in different orbitals.
Therefore, we like to put the spins parallel in an orbital.
Note also that this whole process of writing done matrices can be done very well by computers. Numerical evaluations of this type are a substantial part of theoretical physics.
E.

Addition of angular momenta

We often find it useful to add angular momenta. As we saw with the Coulomb interaction, the total L and S
are what they call good quantum numbers, i.e., all eigenfunctions with the same L and S, but with different ML
and MS have the same eigenvalues for the Coulomb interaction. Obviously, we would like to be able to create these
eigenfunction. We know that the angular part of the eigenfunction can be expressed as spherical harmonics Ylm . So
we want to express the wavefunction for L and M (omitting the subscript L)in product of the wavefunctions,
|lml m i = |lmi|l m i,

(741)

80
where the |lmi stands for the spherical harmonic Ylm . We now want to couple these to a new total wavefunction,
which are eigenstates of the operators
L2 = L21 + L22 + 2L1 L2
Lz = L1z + L2z ,

(742)
(743)

where L1 and L2 are angular momentum operators working on particle 1 and 2, respectively. We can therefore express
the wavefunction as
X
|ll LM i =
hlml m |LM i|lmlm i.
(744)
mm

The coefficient hlml m |LM i is called the Clebsch-Gordan coefficient or Wigner coefficient or vector-coupling coeffcient.
And they come in a zoo of different notations
hlml m |LM i,

V (ll L; mm M ),

LM
Clm,l
m ,

(745)

or as a 3j symbol


l l L
m m M

(746)

This would be bearable, if they didnt all have different phase factors, and coeffients 2L + 1. Probably the most
elegant are the 3j symbols, where the l, l , and L are all equivalent. However, the Clebsch-Gordan coefficients have
a longer history and are used very regularly. Analytical expressions for the Clebsch-Gordan coefficients in terms of
ll Lmm M can be derived, but the derivation is rather complex. However, they are strongly related to the integrals
over three spherical harmonics, as we saw in the evaluation of the Coulomb interaction.
The situation simplifies if we look at a particular problem. For two p electrons, we now that the we have an
eigenstate |11i|11i, since the is no other way to reach M = 2. This directly implies that L = 2. So the wavefunction
is |1122i = |11i|11i. For the combined wavefunction we use the notation |ll LM i, for the one-particle wavefunction
we have |lmi. However, for M = 1, we can have |10i|11i and |11i|10i, so what are the eigenstates of L2 and Lz . Here
we can make use of the step-down operator
L = L1 + L2

(747)

where
L |lmi =
Therefore,

p
(l + m)(l m + 1)|l, m 1i.

p
(2 + 2)(2 1 + 1)|1121i = 2|1121i =

= (L1 + L2 )|11i|11i = 2(|10i|11i + |11i|10i).

L |1122i =

(748)

(749)
(750)

This gives
1
|1121i = (|10i|11i + |11i|10i).
(751)
2

21
21
This directly gives us the Clebsch-Gordan coefficients C10,11
= 1/ 2 and C11,10
= 1/ 2. Since wavefunction are
orthonormal, we can directly obtain the other one
1
|1111i = (|10i|11i |11i|10i).
2
This must be the wavefunction for L = 1 and M = 1. Again applying the step-down operator
p

L |1121i =
(2 + 1)(2 1 + 1)|1120i = 6|1120i
1
1
= (L1 + L2 ) (|10i|11i + |11i|10i) = 2 (|1, 1i|11i + 2|10i|10i + |11i|1, 1i).
2
2

(752)

(753)
(754)

81
This gives
1
|1120i = (|1, 1i|11i + 2|10i|10i + |11i|1, 1i).
6

(755)

In the case of L = 1, the double product cancels and we obtain

1
|1110i = (|1 1i|11i |11i|1 1i).
2

(756)

However, we started out with three basis functions |1, 1i|11i, |10i|10i, |11i|1, 1i. So the must be a third coupled
wave function left. Since we did not find this one at higher L, it must be the L = 0 term. This one must be orthogonal
to the other two, by inspection we see that
1
|1100i = (|1, 1i|11i |10i|10i + |11i|1, 1i).
3

(757)

So we found the angular momentum part of our p2 problem. The S, P , and D term. There are also spin terms.
The triplet state has as highest component | 21 12 11i = | 12 21 i| 12 12 i Applying the step down operator one can obtain the
MS = 0 component


11
1
1 1 11
11 1 1
| 10i =
(758)
| , i| i + | i| , i .
22
22 2 2
2 2 2 22
From orthogonality we obtain the singlet term
11
1
| 00i =
22
2



1 1 11
11 1 1
| , i| i | i| , i .
2 2 22
22 2 2

(759)

The third term of the triplet is given by | 21 12 1, 1i = | 12 , 21 i| 12 , 21 i.

XI.

SPIN-ORBIT COUPLING

So again we find that L1 = 1 and L2 = 1 add up to a total L = 2, 1, 0. Note that starting from a certain L value
we can now apply a perturbation without really destroying the symmetry. The simplest perturbation is probably a
magnetic field, which is proportional to B L. If B is along the z-axis. then this becomes Bz M . So for example the
L = 1 state, the M valueswould be no longer degenerate but would split in M = 1, 0, 1. For the Hydrogen atom, we
found that further level splitting is caused by the spin-orbit coupling. Another perturbance is the spin-orbit coupling,
which behaves as L S. This can be written differently as
L S = L x Sx + L y Sy + L z Sz =

1
(L+ S + L S+ ) + Lz Sz .
2

(760)

Now we see that L and S are no longer conserved, but the total L and S is. Every time L goes up by one, the S value
goes down, and vices versa. In the case where the spin-orbit coupling is much smaller than the Coulomb interaction,
we can treat the spin-coupling as a perturbation. This is true for the valence shell. However, for deep lying core
level (high binding energy), the spin-orbit coupling can be much larger than the Coulomb interaction. Let suppose
we can apply perturbation theory. In this case one uses the LS coupled wavefunction as the unperturbed states. One
assumes that the splitting between the energies of the LS states is much larger than the spin-orbit coupling, so that
these states do not couple. In lowest-order perturbation, the spin-orbit coupling is given by
hLS|L S|LSi.

(761)

Now this is not a very correct expression since the ML and MS values are missing. Since we assume that the spin-orbit
coupling is small, L and S are still good quantum numbers. However, ML and MS are not (note that all the ML
and MS states of one LS term are degenerate). However, L and S are not good quantum numbers for the spin-orbit
coupling. The corresponding J = |L S|, . . . , L + S values are. Therefore, we can write the lowest order perturbation
of the spin-orbit coupling as
hLSJMJ |L S|LSJMJ i =

1
[J(J + 1) L(L + 1) S(S + 1)].
2

(762)

82
Let us consider as an example the p2 configuration. The LS terms are 1 D, 3 P , and 1 S. It is common to put the
J values as a subscript, i.e. 1 D2 , 3 P0,1,2 , and 1 S0 . Note that the total degeneracy (the sum of all the 2J + 1 values)
should remain the same, i.e. 5 + 1 + 3 + 5 + 1 = 15. Hunds rule now says that for less than half-filled shells, the J
value with the lowest energy is the one where L and S are coupled to L S. For more than half filled shells, L + S
is the lowest. Therefore, the ground state for p2 is 3 P0 . For p4 the LS term are exactly the same, but we the lowest
state is 3 P2 .
XII.

MOLECULES AND SOLIDS

After spending already a considerable time in the world of atomic physics, we would like to know what happens if
we start brining the atoms together to create molcules and solids (we will not be discussing liquids here). Let us first
have a look at a simple two-atom molecule.
A.

Hydrogen molecule

Let us first write the general Hamiltonian for a molecule as a matrix. Suppose we have an eigenstate |i which we
want to write in a certain basis set (atomic orbitals is a very obvious choice, let us denote these as |n i. Note, that
for a molecule the atomic orbitals can be centered around different nuclei. Let us assume our basis set consists of N
atomic-like orbitals).
X
|i =
an |n i.
(763)
n

Since |i is an eigenstate, we have H|i = E|i. For the n th component, we then find
X
X
hn |H|i = Ehn |i

hn |H|n ian = E
hn |n ian .

(764)

This can also be written as a matrix

a1
S11 S1N
a1
H11 H1N
. .
. .
.
.
=E
.
.
. .
. .
aN
SN 1 SN N
aN
HN 1 HN N

(765)

or

H|i = ES|i,

with Hn n = hn |H|n i. We have also introduced here the overlap integral. Note that on an atom the wavefunction
are orthogonal. However, when bringing two atoms together, their wavefunction do not necessarily have zero overlap.
To take that into account we have Sij = hi |j i, not that Sii = 1. (Later, we will assume, for simplicity S to be
the unit matrix). Now let us consider the simplest case with two atoms where only one orbital is involved (say, for
Hydrogen the 1s orbital). Our Hamiltonian then becomes




T
1 S
|i = E
|i.
(766)
T
S 1
The eigenvalues can be found by solving the determinant


 E T ES 
2
2


 T ES E  = ( E) (T ES) = 0

(767)

This gives two eigenvalues

E = (T ES)

E1,2 =

T
.
1S

(768)

The calculation could be a simplified a lot by taking and S zero, giving two eigenvalues T . For T < 0 this gives
eigenstates |B i = 12 (|11s i + |21s i) for E = |T | and |A i = 12 (|11s i |21s i) for E = |T | (the superscript denotes
the two atoms). These are known as bonding and antibonding wavefunctions. Note that in the bonding combination,
the electron wavefunction spreads over both atoms of the molecule. This relaxation causes a lowering of the kinetic

83
energy. Actually we could have guesses these eigenstates beforehand since |11s i and |21s i have to play equivalent
roles in the wavefunction. There is a symmetry axis at the center of the molecule.
So far we have considered the situation with one electron, say a H+
2 molecule. However, a H2 molecule has two
electrons. If the electrons did not interact with each other, the energy would be simply 2|T |. However, as we saw
in the previous sections, electrons do interact with each other. Now let us take S = 0 and put two electrons in
the orbitals. When taking into account the spin of the electron, we find six different combinations: |B B i,
|A B i, |B A i, |A B i, |B A i, and |A A i at energies 2 2T, 2, 2, 2, 2, 2 + 2T ,
respectively. Let us look at one particular combination
|B B i =

1
|11s 11s i + |11s 21s i + |21s 11s i + |21s 21s i .
2

(769)

We see that the electrons spend just as much time on the same atom as they to on different atoms. This is quite a
bit different from the situation were the two Hydrogen atoms were far apart. In that case the eigenstates would be
all degenerate and given by |11s 21s i with , =, with energy twice the binding energy of the 1s orbital, i.e.
2 = 2 13.6 eV. In the remainder, to simplify the expressions, let us measure the energy with respect to 2.
The wavefunctions with two spins parallel are different. For example,
|B A i = |11s 21s i.

(770)

Here we see that the Hydrogen atoms cannot bind to each other because of Paulis principle. This looks then like the
situation of two separate Hydrogen atoms.
Now in the previous sections, we saw a lot about the Coulomb interaction between electrons. That would lead you
to think that maybe the electrons do not like to be on the same atom (of course, there is also a Coulomb interaction
between the electrons when they are not on the same site, but this is much smaller). Let us suppose that this Coulomb
interaction can be written as
hi1s j1s |

e2
|i j i = U i,j ,
|r1 r2 | 1s 1s

(771)

i.e. it is U when the electrons are on the same atom and zero otherwise. (Note that we do not have to deal with all
different kinds of LS terms, since for s2 there is only one: 1 S). This means that the configuration |B A i is
at energy 2. So the hopping does not enter into the problem. Let us for simplicity take = 0 (we can always shift
our absolute energy scale). The question now rises, in the situation where the spins are opposite, can we still gain
something from the hopping? You might think that there is no reason anymore for the electrons to be at the same
atom because of this Coulomb repulsion which can be quite large. However, this is not true.
To solve this problem let us start again with states of the Hydrogen atom (we can in principle also start from
the bonding and anitbonding states, but it somewhat more cumbersome). Let us simplify the notation somewhat
since we know we are only dealing with Hydrogen atoms. Let us write |ij i = |i1s j1s i. We can now write the
Hamiltonian in matrix form as

U T T 0
|1 1 i
T 0 0 T |1 2 i
H=
.
(772)
T 0 0 T |2 1 i
0 T T U
|2 2 i
We can in principle solve this 4 4 Hamiltonian on the computer, we which give us the eigenvalues.
However, we can get a bit further by making clever combination. Note that atom 1 and 2 should be entirely
equivalent. So maybe we get a bit further by taking linear combinations where this condition is satisfied:
1
|0 i = (|1 2 i |2 1 i)
2
1
|U i = (|1 1 i |2 2 i) ,
2

(773)
(774)

where the diagonal elements are 0 and U , respectively.

The matrix elements for H now greatly simplifies
h0 |H|U i =

1
(h1 2 | h2 1 |) H (|1 1 i |2 2 i) .
2

(775)

84
This matrix elements is h0+ |H|U+ i = 2T and zero otherwise. In our new basis set, we can write our Hamiltonian as

0 2T 0 0
|0+ i
2T U 0 0 |U+ i
H=
.
0 0 0 0 |0 i
0 0 0 U
|U i

(776)

This can easily be solved by noting that it consist of two 1 1 matrices giving eigenvalues 0 and U and a 2 2 matrix
with eigenvalues


 E 2T 
U
1p 2
2
2



U + 16T 2 .
(777)

E =
 2T U E  = E U E 4T
2
2

Note that E is less than zero, implying that in the lowest eigenstate (the ground state) the electrons are something
on the same atom. This costs them the Coulomb energy, but they gain kinetic energy.
It is always instructive to look at certain limits. In the case that U = 0, we find E = 2T . This is what we
already found before by assuming the particles to have no Coulomb interaction. This corresponds two having the
electrons in the bonding (2T ) and in the antibonding (2T ) orbitals. The limit T = 0 gives eigenvalues 0, U . All the
eigenvalues are now 0, 0, 0, 0, U, U . This corresponds exactly to the possibilities of putting the electrons on diffrent
and the same atom. Let us now take U t. We can then expand the eigenenergies with a Taylor series
E =

4T 2
U

and

E+ = U +

4T 2
.
U

(778)

In the case for parallel spins, we found an eigenenergy 0. For antiparallel spins, we find a lowest energy of 2T 2 /U .
Since U is large the electrons spend most of their time at separate atoms. However, the can still lower their energy by
having their spins antiparallel so that they can undertake excursion to the neighboring atom, thereby lowering their
energy.
B.

Other molecules

In the same way as two s orbitals bind with each other, two p orbitals can also form bonding and antibonding
combinations. However, here we have to distinguish two different possibilities. Let us take the z axis to be the axis
of the diatomic molecule. The pz orbitals then have a large overlap with each other, and form so-called bonding
orbitals. The px and py orbitals are farther apart and form molecular orbitals. In both cases, we have bonding
and antibonding orbitals. Note that the energy difference between and is smaller than that between and .
This would all be the case for one electron in an molecular orbital. However, when discussing diatomic molecules one
has to take into account the Coulomb interaction when two electrons are in the same orbital. A different notation
that you often find come from the terms gerade (even) and ungerade (odd). This is defined with respect to the center
of symmetry of the molecule. Therefore, the bonding combination of two s orbitals is gerade and the antibonding
is ungerade. For pz orbitals the situation is the same, but for the bonding and antibonding of the px and py , the
situation is reverse. Experimentally, we find the order of molecular orbitals in diatomic molecules

1xu (2px )

1xg
(2px )

1g (1s) 1s
1u (1s) 2pz
2g (2s) 2s
2u (2s) 3pz
1yu (2py ) 2px,y
3g (2pz ) 3s

1yg
(2py ) 3d

3u (2pz ) 4pz

(779)
(780)
(781)
(782)
(783)
(784)
(785)
(786)
(787)

(note that there is a slight redundancy in the notation here, often the asterisk is left out of the notation). Between
parantheses the atomic orbitals from which the molecular orbitals originate are given. Now this looks a bit different
from what you would expect from the overlap. Since the hopping between the pz orbitals is larger than that between
the px and py orbitals, you would expect the 3g to be lower in energy than the 1xu and 1yu molecular orbitals.

85

However, for 3g2 , both electrons are between the two atoms and have a large Coulomb interaction. For the 1xg
and

1yg molecular orbitals the electrons are farther apart. In addition, if we have only two electrons to put in the 1

orbitals, we can put one in the 1xg

and the other in the 1xu
. This will lead to a further reduction in the Coulomb
interaction. We already considers the situation of H2 , which has a dissociation energy of 4.75 eV (2 bonding electrons).
Let us consider some diatomic molecules for different elements:
1. He2 , (1g )2 (1u )2 . Both 1s orbitals are full, so effectively no bonding occurs.
2. Li2 KK(2g )2 . The two 1s (or K) shells are complete and do not contribute in the bonding. The bonding is
done by the 2s electrons forming a stable species. Dissociation energy 1.14 eV (two bonding electrons).
3. Be2 , KK(2g )2 (2u )2 . Both 2s orbitals are full, so effectively no bonding occurs. The situation is analogous to
He.
4. B2 , KK(2g )2 (2u )2 (1u )2 . We have two degenerate orbitals that contribute to the bonding 1xu and 1yu . In
order to minimize the Coulomb interaction, a configuration of 1xu 1yu is formed (i.e. a triplet is formed, in total
three combinations). Dissociation energy 3.0 eV (two bonding electrons).
5. C2 , KK(2g )2 (2u )2 (1u )4 . With a total of six bonding and two antibonding states, bonding occurs. However,
Carbon forms stabler bondings with Hydrogen, Oxygen, etc. Dissociation energy 6.24 eV (four bonding electrons).
6. N2 , KK(2g )2 (2u )2 (1u )4 (3g )2 . With a total of eight bonding and two antibonding states, a very stable
bonding occurs. N2 has the largest dissociation energy of he molecules in this list. Dissociation energy 9.76 eV (six
bonding electrons).
7. O2 , KK(2g )2 (2u )2 (1u )4 (3g )2 (1g )2 . Eight bonding electrons and four antibonding electrons. Again we
have two electrons in two degenerate 1g orbitals. These will again form a triplet as in the case of B2 . The oxygen
molecule is therefore paramagnetic, which was one of the great triumph of the molecular orbital theory. Dissociation
energy 5.12 eV (four bonding electrons).
8. F2 , KK(2g )2 (2u )2 (1u )4 (3g )2 (1g )4 . Dissociation energy 1.60 eV (two bonding electrons).
9. Ne2 , KK(2g )2 (2u )2 (1u )4 (3g )2 (1g )4 (3u )2 . Unstable since all the 2s and 2p orbitals are full.

C.

Hybrids

The configuration of the Carbon atom is given by (1s)2 (2s)2 (2px )1 (2py )1 . Therefore, Carbon would have two
unpaired electrons which are at an angle of 90 with each other. However, this is not what is observed experimentally
in many Carbon compounds such as methane CH4 . Here Carbon appears to be tetravalent. This can be seen if we
represent the configuration of Carbon somewhat differently (1s)2 (2s)1 (2px )1 (2py )1 (2pz )1 . This treats the s and p
orbitals on a similar footing. Now we can consider the mixing of s and p orbitals.
sp hybrids Let us consider the formation of hybrid functions composed of one s and one pz function. The two hybrid
wavefunction can be expressed as
1 = a1 s + a2 pz

and

2 = b1 s + b2 pz .

The coefficients are determined such that the wavefunctions are orthogonal and normalized
Z
Z
Z
1 1 dr = 1
and
2 2 dr = 1
and
1 2 dr = 0

(788)

(789)

We want that the new wavefunctions contain both an equal amount of s character. Combining all the equations leads
to the solution
1
1 = (s + pz )
2

and

1
2 = (s pz ).
2

(790)

An example of an sp hybrid is acetylene (C2 H2 ). Here the two sp hybrids of the Carbon atoms form a bonding
molecular orbital (if we take the z axis along the molecule, the sp hybrid is formed by the s and pz orbitals. This
leaves the px and py orbitals which are perpendicular to the z axis. These orbitals of the two Carbon atoms form
bonding molecular orbitals. This creates a triple bond.
sp2 hybrids
Following the same procedure as described above, we can find three new wavefunction consisting of s, px , and py

86
orbitals:
1
2
2

1
= s +
3
1
= s
3
1
= s
3

2
p
3 x

1
1
px + py
6
2
1
1
px py .
6
2

(791)
(792)
(793)

Note that every orbital has 1/3 s orbital and 2/3 p orbital character. The first hybrid has on the positive x side both
the positive s orbital and the positive part of the p orbital. There the wave functions add up. On the negative x
side, we have a positive s orbital and the negative lobe of the p orbital. There the wavefunction partially cancel each
other. For the second hydrid, the situation is very similar. Note that the wave function can be written as
r 

1
2
1
1
2 = s +
px +
3py
(794)
3
2
2
3
r
r

1
2
1
2
= s +
cos 120 px + sin 120 py = s +
p .
(795)
3
3 120
3
3

The combination of px and py orbtital is effectively a p orbital at 120 degrees with respect to the x-axis. This sp2
hybrid is therefore equivalent to the first hybrid except it is rotated by 120 around the z-axis.
A typical example of an sp2 hybrid is Ethylene (C2 H4 ). One 2s and two 2p orbitals form three sp2 hybrids at a
120 degree angle with each other. Two of those couple two the Hydrogen atom, the other connects the two Carbon
atoms, forming a planar molecule. The remaining p is perpendicular to the plane of the molecule (the pz orbital).
The two pz orbitals of the Carbon atoms for a bonding with each other.
sp3 hybrids In an sp3 hybrid, all the orbitals assume equivalent roles, this leads to wavefunction of the type
1
(s + px
2
1
= (s + px
2
1
= (s px
2
1
= (s px
2

1 =

+ py + pz )

(796)

py pz )

(797)

+ py pz )

(798)

py + pz ).

(799)

3
4

The orbitals form a tertrahedron and have mutual angles of 109 .28 . One of the simplest examples of an sp3 hybrid
can be found in Methane (CH4 ). The four single bonds form bonding with the Hydrogen s orbitals. A somewhat
more complicated example is Ethane (C2 H6 ). Four sp3 orbitals are formed at each C atom. One of these connects
the two Carbon atoms, the other three connect to the Hydrogen s orbitals.
D.

The H
uckel molecular orbital method

It has long been clear that numerous chemical and spectroscopic proporties of hydrocarbons involve primarily
electrons. Apparently, the energy spearation between the and orbitals is large, so we can consider them spearately.
For ethylene, we find a typical bonding-antibonding solution for the molecular bond. If we take 0 = hpzi |H|pzi i
where i = 1, 2 and T = hpz1 |H|pz2 i, we have the eigenvalue problem


 E0 E

T

=0

E = E0 T.
(800)

T
E0 E 
For Ethylene, there are a total of 12 electrons (8 from the two Carbon atoms, 4 from the Hydrogen atoms). All the
bonding molecular orbitals are occupied. This accounts for ten of the electrons (8 will go into the bonding orbitals
between the Carbon and the Hydrogen, the other two will go into the binding orbital between the two Carbons).
This leaves two electrons, that will go into the bonding molecular orbital. The orbital is empty.
Let us consider a somewhat more complicated molecule, such as butadiene (C4 H6 ), whose chemical formula is
H
H
H
H
|
|
|
|
H C1 = C2 C3 = C4 H

(801)

87
There are a total of 22 electrons. The stongly-bound bonds are: 6 between the Carbon and the Hydrogen, 3 between
the different Carbons. This accounts for (6 + 3) 2 = 18 of the electrons. So the four other electrons go into the
more losely bound molecular orbital states. The electrons can be assumed to be delocalized. Let us take the
T = hpzi |H|pzj i < 0, where i and j are nearest neighbors. The eigenvalue problem can then be written as


 E0 E
T
0
0



T
E0 E
T
0


(802)

.
0
T
E0 E
T




0
0
T
E E
0

The roots can be found to be

E = E0 1.6180T,

E0 0.6180T.

(803)

The four electrons go into the lowest two orbitals, giving a total energy of
E = 2(E0 + 1.6180T ) + 2(E0 + 0.6180T ) = 4E0 + 4.4720T.

(804)

Now let us assume that the orbitals are not delocalized but localized between Carbon atoms 1,2 and 3,4. This
leads to a determinant



 E0 E
T
0
0


T
E0 E
0
0


(805)

.
0
0
E0 E
T




0
0
T
E0 E
This is easily solved since it splits into two 2 2 matrices. giving eigenvalues E = E0 + T, E0 + T, E0 T, E0 T .
The total energy is then E = 4E0 + 4T , which is higher than for the delocalized case.
Benzene The Benzene molecule (C6 H6 ) consists of 6 Carbon atom that form a circle by the sp2 bonds. There are
a total of 6 4 + 6 = 30 electrons. There are 6 bonds of the sp2 hybrid between the different Carbon atoms. There
are another six bonding orbitals between Carbon and Hydrogen. This accounts for (6 + 6) 2 = 24 of the electrons.
The other 6 electrons delocalize in the pz orbitals of the 6 Carbon atoms. The secular matrix is given by



 E0 E
T
0
0
0
T




T
E0 E
T
0
0
0


0
T
E0 E
T
0
0


(806)
.

0
0
T
E0 E
T
0




0
0
0
T
E0 E
T



T
0
0
0
T
E E 
0

The eigenvalues are found to be

E = E0 + 2T, E0 + T, E0 + T, E0 T, E0 T, E0 2T.

(807)

The 6 electrons go into the lowest molecular orbitals giving a total energy of E = 6E0 + 8T .
The wavefunction for the eigenstate with energy E0 + 2T is completely symmetric
1
1 = (p1z + p2z + p3z + p4z + p5z + p6z ),
6

(808)

and contains two electrons. The two wavefunctions at energy E0 + T are

1 1
1
1
1
2 = ( p1z p2z p3z p4z + p5z + p6z ),
2
2
2
3 2

(809)

and
3 =

1
(p1z + p2z p4z p5z ).
2

(810)

These two wavefunctions contain the other four electrons. The wavefunctions for E0 T are
1 1
1
1
1
4 = ( p1z + p2z p3z + p4z + p5z p6z ),
2
2
2
2
3

(811)

88
and
5 =

1
(p1z p2z + p4z p5z ).
2

(812)

Finally the wavefunction at energy E0 2T is given by

1
6 = (p1z p2z + p3z p4z + p5z p6z ).
6

(813)

The last three orbitals are empty. Although at first sight this might seem like some coefficients which just happen to
come out of a calculation, on further inspection you can see that there is some system to the coefficients. A rather
trivial one is
1 X
1 =
pnz .
(814)
6 n
But 6 we can write as
1 X
6 =
pnz cos n.
6 n
Looking more carefulle at the wavefunctions, we see that

1 X
2 =
pnz cos n
3
3 n

(815)

(816)

and

1 X
pnz sin n .
3 =
3
3 n

(817)

The same thing we can do for 4 and 5 , but now using 2/3. In addition, since 2 and 3 are degenerate, we can
make linear combinations of them that are still eigenfunctions

1
1 X
+
2,3
= (2 + i3 ) =
pnz ein 3 ,
(818)
2
6 n
and
1
1 X

+
2,3
= (2 i3 ) =
pnz ein 3 .
2
6 n

(819)

The same can be done for 4 and 5 . So it appears that the eigenstates look like a Fourier transform of the states in
real space
1 X
2
k =
pnz eikn ,
k = 0, , , .
(820)
3
3
6 n
In fact, even the eigenvalues can be written as Ek = E0 + 2T cos k. Is this surprising? Of course not entirely since
we have a repeating function, and we could have expected that making a Fourier transform would yield some results.
This is a very important concept when going to solids.
E.

Solids

Let us try to extend our molecule into a chain [which one could see as a polymer, or a one-dimensional solid, or
even a three-dimensional solid in certain cases (note that, in a cubic solid, the pz orbital has only overlap with a
neigboring pz orbital along the z-axis)]. Our Hamiltonian will then look like (in its simplest form)

0 T
T 0 T

T 0 T
H =
(821)
.

T 0 T

89

FIG. 13: Eigenvalues for a linear chain of atoms. For periodic boundary conditions, eigenvalues always lie on the 2T cos ka
curve. For fixed boundary conditions, eigenvalues are shown for N = 2, 3, 5, 11. The dashed lines are drawn as a guide to the
eye.

The solutions can be easily found by means of some computer program. Let us just plot the eigenvalues equidistant
on a scale from 0 to 1, see Fig. 13 (the same eigenvalues are plotted from -1 to 0). For
two atoms,
we had T .
So we plot T at 0 and T at 1 (note that T < 0). For three atoms, we have eigenvalues 2T, 0, 2T , which we
plot at 0, 12 , 1. Also plotted in Fig. 13 are the eigenvalues for 5 and 11 atoms. However, quite quickly the maximum
eigenvalues approach 2T . Very quickly, the eigenvalues seem to behave a bit like a cosine. Already for 11 atoms,
the behavior is quite close to an inifinite number of atoms. Apparently, we are missing something in our approach by
connecting all the atoms together. It looks like we should be able to solve the problem more easily.
The input that we are missing here is translational symmetry. Suppose the atom are a distance a apart. In a
molecule not all the atoms are equivalent. Although we see from the eigenvalues that if we increase the length of the
molecule the eigenvalues follow more or less the same curve (except that we find more and more points of the curve).
Things become easier when we connect the first and the last atom with each other. This means that if the chain is
N atoms long, then N + 1 is equivalent to 1. So an electron that tries to hop from the N th atom to (N + 1)th atom
ends up at the first atom. This makes all the atoms in the chain equivalent, since on every atom we can go in both
directions. This is called a periodic boundary condition (or Born-Von Karman boundary condition). For a linear
chain we might see this as a kind of a circle, but we can also apply periodic boundary conditions to two and three
dimensional systems, where such pictures are a bit more difficult to imagine. Now are boundary condition becomes
different. We can use a complex Fourier transform, and impose the condition that if we go L further we end up at
the same atom:
eik(x+L) = eikx kL = 2n k =

2n
L

and

n = 0, . . . , N/2

(822)

or
k = 0,

2 4
6
N 2
, , ,...,
L
L
L
2 L

(823)

What we are actually saying is that the ends of the chain or the surface of a solid is not really important for
the bulk properties. For our linear chain, we see that it already does not make much difference whether it is 11 or
inifinity atoms long. In general, we are having of the order of 1023 atoms, so we are probably okay. It has some other
mathematical advantages, since we do not have to take bound solutions (sin kx), but we can take running waves eikx .
A Fourier transform can now be given as
1 X ikxj
1 X ikxj
|ki =
e
|ji
and
|ji =
e
|ki,
(824)
N j
N k
where N is the number of sites (with N ), |ji is the wavefunction at site j, and |ki the wavevector in reciprocal
space. The Hamiltonian can be written as
X
X
H=
|j + ihj + |H|jihj| =
|j + iT hj|,
(825)
j,

j,

90
where = 1 indicate that the electron can hop to a neighboring site. Making the Fourier transform gives
H =

1 XX
T |k ieik xj+ hk|eikxj
N

(826)

j kk

1 XX
T |k ieik (j+)a hk|eikja
N

(827)

j kk

X 1 X
X

ei(kk )ja
eika T |k ihk|,
N

j
k,k

(828)

where xj = ja with a the distance between the particles. The first term gives a delta function
X
X
H=
k,k k |k ihk| =
k |kihk|,
k,k

(829)

where
k =

eika T = 2T cos(ka).

(830)

=1

We find that the Hamiltonian is diagonal in k. Therefore, k is obviously a good quantum number for translational
symmetry. Note, that in principle the matrix does not have to be entirely diagonal. Here we took only one type
of orbital, but in principle there could be different kinds. This approach to a solid is what is usually known as a
tight-binding scheme or Linear Combination of Atomic Orbitals (LCAO). Also it does not necessarily have to be a
circle. It can also be some linear chain of N atoms, where the N th atom is connected to the first. Note that if the
system is large then what happen at the ends of the chain should be unimportant. Note that this solution works
regardless of how many atoms there are. For two atoms, we can also apply periodic boundary conditions (although it
is a bit artificial). From atom two, we should be able to hop to atoms 2 1 = 1, 3. However, since we have periodic
boundary conditions atom 3 is atom 1. This is a bit like have a circle of length 2a with two atoms on it. So in both
cases we hop to atom 1. Instead of a hopping matrix element T , we now have 2T , finding eigenvalues 2T . We

have wavevectors k = 0, 2
L = 0, a . This looks a bit like three wavevectors whereas we only had two sites, but
the maximum wavevectors only count for a half. This gives eigenenergies 2T cos 0 a = 2T and 2T cos a a = 2T .
For N = 3 with periodic boundary conditions, solving a 3 3 matrix would give eigenvalues E = 2T, T, T (do
it yourself). The wavevectors are 0, 2
3a . (We should be a bit careful here with the wavevectors in Eqn. (823).
N/2 is not an integer, so these wavevectors should not be included.) The eigenvalues are now 2T cos 0 = 2T and
2T cos 2
3a a = T . This is in agreement with eigenvalues of the 3 3 matrix. Show yourself that you can also find
the eigenvalues for benzene this way.
We can also write the Eqns. (825) and (829) in a somewhat different way which is more often observed in the
scientific literature. Making use of the operators defined earlier, we can write
X
H=
T cj+ cj .
(831)
j,=1

The cj removes an electron from site j, whereas the cj+ creates and electron at site j + , thereby moving effectively
an electron one site to left or to the right. In k space we can write
X
X
H=
k ck ck =
k nk .
(832)
k

Note that the wavevector k is not changed the operators ck and ck , they just measure the presence of an electron in
the state with energy k .
F.

(Nearly) free-electron model

Usually discussion on solids do not start with the tight-binding model (which consider the electrons to be in states
that resemble closely the orbitals in atoms). Generally one starts at the opposite end. One often considers the solid to
be a box of free electrons. The nuclei are there to provide a positive background and to prevent the electrons from flying
out of the box. (Usually it takes some minimum energy to remove an electron from the solid, see Eqn.(photoelectric)

91
and the discussion on the photoelectric effect). In fact, it is not at all obvious that we can just ignore the potential
landscape in the solid, but it works well for so-called simple metals, such as Be, Mg, Ca, Al, Ga, or Pb. These
are all systems with free s or p electrons. For d and f electrons this approximation works much less (remember our
discussion on the radial extent of the wavefunction for atoms).
Application of period boundary conditions means that an increase in x by the length of the chain should result in
the same function,
eik(x+L) = eikx

k = 0,

2 4
6
N
, , ,...,
.
L
L
L
L

(833)

Note that the maximum k value is NL = a . We now consider a three-dimensional solid of free electrons. (we
assume it is cubic, but that does not really matter since N is so large that k can be generally assumed continous. A
free electron satisfies the Schrodinger equation

~2 2
= E
2m

1
|ki = eikr
V

with

E = k =

~2 k 2
.
2m

(834)

Note that the normalization is chosen such that integration of hk|ki over the volume V = L3 gives unity. The
Hamiltonian can now directly be written as
H=

k ck ck =

k nk

and

k =

~2 k 2
.
2m

(835)

Note the difference between the tight-binding model and the free electron model. In the tight-binding model, we
start from atoms and then let the electrons move between the atoms. In the free-electron model, we start letting the
electrons move, and put the atoms in as a perturbation, as we shall see below. Note that the resemblance between the
tight-binding model and the free-electron model is greatest near k = 0. In that case we can expand the tight-binding
energy as


k 2 a2
+ ,
(836)
2T cos ka = 2T 1 +
a
where the second is comparable to a free electron with an effective mass m
T k 2 a2 =

~2 k 2
2m

m =

~2
.
2a2 T

(837)

Note that if the hopping matrix element T goes to zero, the effective mass goes to infinity.
Now we have to start filling the levels with electrons. This is similar to what we have done for Plancks distribution
law with the major difference that we are now dealing with Fermion instead of bosons in Eqn. (8). For Fermions the
partition function (the sum of all the probabilities) is
Z=

exp(nE E) = 1 + exp(E),

with

= 1/kB T

(838)

nE =0,1

The chance of finding a state occupied is then simply the nonnormalized chance of finding that state occupied divided
by the total chance (Z)
hnE i =

exp(E)
1
=
.
1 + exp(E)
1 + exp(E)

(839)

However, this does not entirely work, since for T 0 we find f (0) = 12 and f (E) = 0 for E > 0, which means that
we could only put one particle in our solid. Therefore, E should be measured with respect to a certain energy known
as the chemical potential , i.e., E = . The chemical potential should be chosen in such a way that the number
of particles is equal to N . In general is temperature dependent.
f () = hn i =

1
1+

e()/kB T

(840)

This is known as the Fermi-Dirac function. In general, the Fermi-Dirac function is one up to close to the chemical
potential, where it is broadened over a region of kB T . The chemical potential for T 0 is also known as the

92
Fermi energy EF (although the distinction between Fermi energy and chemical potential is a bit blurred in everyday
language).
Also we want to know the number of k states. When k becomes large there are many different ways to form a
wavevector with a length between k and k + dk:
2

L
2

3

4 3
V
k = 2 k 3
3
3

(841)

Note that in Eqn. (4) there is a factor 18 , but the density of k states is twice as large in each direction, i.e., 23 = 8
in total. We can now write down the Fermi wave vector and energy by noting that the number of k states should be
equal to the number of electrons Ne :
kF =

3 2 Ne
V

1/3

~2 kF2
~2
EF =
=
2m
2m

and

3 2 Ne
V

2/3

(842)

A very important quantity in solid state physics is the density of states. Note that we can write the number of
particles as
V
Ne = 2
3

2mE
~2

3/2

(843)

The density of states is now defined from the number of states in a region dE,
dNe = (E)dE

(E) =

dNe
V
=
dE
2 2

2m
~2

3/2

E.

(844)

Obviously, free electron are an oversimplification of the wavefunctions in a solid. The real eigenfunctions are a
linear combination of the plane-wave solutions (let us, for simplicity, restrict ourselves to one dimension).
X
|i =
ck eikx .
(845)
k

Therefore,
P we have almost plane waves in a potential landscape, which can also be given in Fourier components as
U (x) = K UK eiKx Now the most important thing to note about the potential is that it is periodic, i.e.,
U (x + a) = U (x)

eiKa = 1

K = 0,

2 4 6
, , ,...,
a
a
a

(846)

i.e., K can only assume, what are called, reciprocal lattice vectors. Reciprocal lattice vectors are found when making
a Fourier transform of the lattice (Some of you might know them from Bragg reflection). Thus we can write
U (x) =

UK eiKx

with

K = 0,

2 4 6
, , ,....
a
a
a

(847)

Now let us try to solve the Schrodinger equation

~2 d2
+ U (x) = E.
2m dx2

(848)

Writing out the components gives

X
X
~2 k 2 X
ck eikx +
UK ck ei(k+K)x = E
ck eikx .
2m
k

K,k

(849)

Making the substitution k + K k in the second term gives the equation for the coefficients (since all the eikx are
independent the coefficient in front of each eikx must be zero)
X
UK ckK = 0,
(850)
(k E)ck +
K

93
with k = ~2 k 2 /2m. It is important to note that a plane wave only couples with other plane waves whose k differ by
a reciprocal lattice vector.
Now let us suppose that we only couple to the plane wave with the smallest reciprocal lattice vector (which is not
a bad assumption in a large number of cases), we are then left with only two coupled equations:
(k E)ck + UK ckK = 0;
(kK E)ckK + UK ck = 0.

(851)
(852)

This can be written as a matrix whose eigenvalues can be found by solving the determinant
r


 k E

1
1
UK
1,2
2.

=0

Ek = (kK + k ) ( (kK k ))2 + UK

 UK kK E 
2
2

(853)

When U is small compared to kK k , the effect of the perturbance will be small. However, when k = 21 K,
kK = k and we find that Ek1,2 = 21 K and U opens up a gap of of 2U .
What is so special about the reciprocal lattice vector 12 K? This is exactly the wavevector for which one would find

a Bragg reflection. Note that the eigenfunction for Ek1,2 are approximately 12 (eikx eikx ) = 2 cos kx, 2 sin kx
with k
= a . Note that the cosine has its extrema (i.e., maxima in the probability distribution function) at the atoms,
whereas the sine is zero. The particle in the cosine wavefunction therefore feels more of the attractive potential of the
atoms and is at U and the sine is at +U .
Note that this can also be understood in a more orbital-like approach. The cosine has its maximum at the atom
and we can therefore look at it as a kind of s orbital. The sine changes sign and is therefore similar to a p orbital.
The results are therefore in agreement with our expectation that a s like wavefunction is lower in energy than a p like
wavefunction.
Since the potential only couples plane waves that differ by a reciprocal lattice vector, the eigenfunctions can be
written as
X
k (x) =
ckK ei(kK)x .
(854)
K

This can be rearranged as

k (x) =

X
K

ckK eiKx

eikx = eikx uk (x).

(855)

Note that the wavefunction now consists of a product of a function with the periodicity of the lattice (uk (x + a) = uk ,
since eiKa = 1) and a plane wave. Generalizing this to three dimensions, it says that we can write the wavefunction
in a periodic potential as
k (x) = eikr uk (r).

(856)

This is known as Blochs theorem.

Also the effect if the periodic potential can be generalized. Note that the potential only couples wavevectors that
differ by a reciprocal lattice vector, therefore
k = k + G.

(857)

Since the energy of the electron is conserved, we must have k = k . We can therefore write
2k G + G2 = 0.

(858)

This is the condition for Bragg reflection. You might not recognize it directly, so let us try to rewrite this condition
for x-rays. We can always replace G by G, since both are reciprocal lattice vectors. This gives us
2k G = G2

2
2
sin = n

2d sin = n.

(859)

In a usual experiment is the angle that the incoming and outgoing wave make with a crystal plane. (Note that
in Fig. 8.15 in Liboff, the angle with the normal is taken, this differs from the usual practice in diffraction). By
satisfying the Bragg condition, the x-ray or electron feels the repeated action of the periodic potential.

94
XIII.

SECOND QUANTIZATION

As you might remember, for the atom we treated the effects of the Coulomb interaction. We would like to do
something similar for the electron gas. However, now we are entering a really complicated world. Electron correlations
in solid are one of the most complicated areas in physics. We will try to do a bit. Before doing this, let us introduce,
in a somewhat hand-waving way, the ideas of second quantization.
Remember that for the harmonic oscillator, we had this nice Hamiltonian
1
H = (a a + )~,
2

(860)

which contains an enormous amount of information without having to worry about Hermite polynomials (life is already
difficult enough). This Hamiltonian applies to a large number of possibible oscillators. It is easy to generalize it to
include more than one type of oscillator. For example,
H=

X
i

1
(ai ai + )~i
2

or

1
H = (aq aq + )~q ,
2

(861)

where we can have oscillators on different sites (which could in principle have different ) or oscillators in momentum
space (for example, creation of photons in plane waves (eikr ) or lattice oscillations [also called phonons (for those of
you that are already know some solid-state physics: in principle we could add another index k, since we have two
types of phonon optical and acoustic)]. We can now proceed in the same as Eqns. (278) and (279) except that we
have to extend our bookkeeping slightly

(862)
ai |n1 n2 ni i = ni + 1|n1 n2 , ni + 1, i

ai |n1 n2 ni i = ni |n1 n2 , ni 1, i,
(863)
where we have raised and lowered the occupation of state i by one (we could do exactly the same thing for any index).
These operators follow the commutation relation
[ai , aj ] = i,j .

(864)

We can do exactly the same thing for Fermions, which we in fact already described a bit in Section X A on Paulis
exclusion principle, see Eqns. (670)-(674). We can use the fermion operators c and c to rewrite the Hamiltonians in
Dirac notation in Eqns. (825) and (829) as
X
X
H0 = T
cj+ cj
and
H0 =
k ck ck ,
(865)
j

respectively. Which looks similar to the Hamiltonian we had for the harmonic oscillator. Let us first look at the
second Hamiltonian. Note that ck ck just measures the number of electrons in state k. The first term is somewhat
more complicated: cj annihililates an electron at site j. If there is no electron, nothing happens (i.e., cj gives zero).
If there is an electron, then the cj+ term creates an electron at site j + , which is a neighboring site since = 1.
Which means that effectively an electron has hopped from site j to site j 1.
We can do the same thing for a two-particle Hamiltonians H1 (r1 , r2 ) such as the Coulomb interaction. Let us write
it down in k space
H1 =

1
2

Vk1 k2 k3 k4 ck4 ck3 ck2 ck1

(866)

k1 k2 k3 k4

with
Vk1 k2 k3 k4 =

dr1

dr2 k4 (r1 )k3 (r2 )H1 (r1 , r2 )k2 (r2 )k1 (r1 ).

(867)

The factor 1/2 is to avoid double counting of the electron interactions. This looks pretty horrible and that is exactly
what many-body interactions are. Now let us assume that the i are plane waves V 1/2 eikr . It is important to
remember that these are the basis functions and not the eigenfunctions. Remember, that for the case of the periodic

95
potential, we choose plane waves, but the resulting Bloch waves (the eigenfunctions) were linear combinations of plane
waves.
Z
Z
1
e2
Vk1 k2 k3 k4 = 2
dr1 dr2 ei(k4 r1 k3 r2 +k2 r2 +k1 r1 )
,
(868)
V
|r1 r2 |
taking e2 /40 e2 . Now let us put R = r1 r2 and r = r2 , this gives
 Z
 Z

e2
1 1
dr ei(k4 k3 +k2 +k1 )r
dR ei(k1 k4 )R .
Vk1 k2 k3 k4 =
V V
R

(869)

The first term between parentheses gives (k1 + k2 k3 k4 ), which means that if we choose k1 = k, k2 = k , and ,
k3 = k q, that k4 = k + q. The second term in parantheses then only depends on k1 k3 = q.
For Vq we obtain
Vq

Z
Z
Z
Z 1
e2 iqR
2
2
iqR cos 1
2
=
dR e
= 2e
dR
dR sin e
= 2e
dRR
d cos eiqR cos
R
R
0
0
0
1
Z
2e2
4e2
dR(ei(q+i)R ei(qi)R ) = 2
,
=
iq 0
q + 2
Z

(870)
(871)

where a has been introduced to let the oscillations in the sine damp out slowly. In the limit 0, we obtain
Vq = 4e2 /q 2 . Later on the will actually take on a physical role.
The Hamiltonian now has the form
H = H0 + H1 =

k ck ck +

1 X Vq
ck+q ck q ck ck .
2
V

(872)

qkk

Sometimes you might see the Coulomb term as a Feynman diagram language (it is a little bit more complicated
than this, but. . . )
k + q@
@

k
Vq .

@
@k

k q

(873)
(874)

Here we see two electrons that have an interaction with each other via the Coulomb potential thereby exchanging
momentum q.
A.

The electron gas

If we forget about the Coulomb interactions, the ground state at T = 0 is the Fermi sphere, i.e. all the independentparticle states (plane wave) up to the Fermi energy are filled:
Y
|F i =
ck ck |0i,
(875)
|k|<kF

where all the plane wavefunction with a k whose norm is less than kF = 2mEF /~ are filled; |0i is the state with no
electrons. We fill both spin states. So far we did not include spin in our Hamiltonian. This can be easily introduced
H = H0 + H1 =

X
k

k ck ck +

1
2

qkk

Vq
c
c
ck ck .
V k+q, k q,

(876)

We would now like to now the ground state energy. Let us first have a look at the kinetic energy H0 . We need to
calculate the expectation value of H0 ,
E0 = hF |H0 |F i =

~2
2m

|k|<kF ,

k2 .

(877)

96
E0 = hF |H0 |F i =

~2
2m

k2 .

(878)

|k|<kF ,

Let us replace the summation by an integral

Z
Z
~2 V
~2 V
~2 V 1 5
3 ~2 kF
3
E0 =
2 3
dk k 2 =
4
dk k 4 =
kF = N
= N EF ,
3
2
2m 8 |k|<kF
2m 4
2m

5
5
2m
5
|k|<kF

(879)

where we have used N = V kF3 /3 2 from Eqn. (841).

Let us now try to calculate the first order correction resulting from H1 . Again we assume that is can be calculated
as an expectation value in the unperturbed ground state:
E1 = hF |H1 |F i =

X X Vq
hF |ck+q, ck q, ck ck |F i.
V

(880)

kk q

Since the electrons are indistinguishable there are two different ways to make pairs. The most obvious one is to
make pairs ck+q, ck and ck q, ck which implies that q = 0. This is called the Hartree contribution. (Note
what happens if we try to make q 6= 0. Then we annihilate a electron (create a hole) with k, which necessarily has
to be below the Fermi surface and we create an electron with k + q, which has to be above the above the Fermi
surface. So, effectively, we have created a Fermi surface with an electron-hole pair. However, this is orthogonal to
the state |F i and we obtain a zero expectation value). Let us consider the contribution of the q = 0 term. Since
Vq = 4e2 /(q 2 + 2 ) this would diverge if we take the limit q 0. However, for a finite , we have
1 V0 X X
1 V0 X X
hF |ck, ck , ck ck |F i =
hF |ck , (k,k , ck ck, )ck |F i
2V
2V

kk

(881)

kk

1 V0 X X
1 V0 2
1 V0 2 2e2 N 2
hF |(nk , nk, nk , )|F i =
(N N )
N =
,
=
2V
2V
2V
V 2

(882)

kk

where nk, = ck, ck gives the occupation of the state with k, . This does not need to surprise us terribly. It is
rather difficult to take of the order of 1023 electrons and put them together with all the negative charge present. So
far we left out the hole positive background of the nuclei making the total system neutral. Suppose we can describe
this background by a uniform density (r) = N/V . We then have two contributions:
Vbackground

1
= e2
2

(r1 )(r2 )

e|r1 r2 |
2e2 N 2
dr1 dr2 =
,
|r1 r2 |
V 2

(883)

e|rri |
4e2 N 2
dr =
,
|r ri |
V 2

(884)

describing the interaction between the nuclei, and

Vebackground = e2

XZ

(r)

describing the interaction of the electrons with the nuclei. Note the absence of the factor 1/2, since we do not have to
correct for double counting the interactions here. The contributions from Eqns. (882), (883), and (884) cancel each
other. From now on, we implicitly assume that the summation excludes q = 0.
However, there is another way of combining the creation and annihilation operators ck+q, ck and ck q, ck
which implies that k = k + q and = (thus only parallel spins). This is called the Hartree-Fock contribution and
is given by
X X Vq

kq

hF |ck+q, ck, ck+q ck |F i =

X X Vq

kq

hF |nk+q, nk, |F i.

(885)

It is also known as the exchange term, since the electron are exchanged (therefore they need to have the same spin).
We want to write this in a somewhat different form. We can write the kinetic energy per particle as
hEk i =

1 X
k hF |nk, |F i.
N
k,

(886)

97
We can rewrite the exchange term as
hEexch i =

1 X
hF |nk, |F iexch
k
N

exch
=
k

with

k,

1 X
Vq hF |nk+q |F i.
V q

(887)

exch
is known as the self energy of the particle. We see that it is negative. This is a result of Paulis principle. Since
k
electrons with the same spin cannot be in the same quantum state, they tend to avoid each other. This effectively
leads to a somewhat lower Coulomb interaction between electrons with parallel spins.
Let us take k = k + q. We can evaluate the self energy term as
exch
k

dk
4e2
e2
=
n
k
(2)3 |k k |2 + 2

e2 kF

kF

da
+
2kk a + 2
0
1
!
2
2
1k
(1 + k)2 +
1k
1+k
e2 kF
1+

F (k),
ln

arctan

arctan
=

4k

(1 k)2 +
Z

dk k

da

k2

k 2

with k = k/kF and = /kF . Thus the energy is given by



~2 kF
1
2
exch
F (k) ,
Ek = k + k =
k +
2m
a0 kF

(888)
(889)

(890)

where a0 = ~2 /me2 is the Bohr radius. Figure 14 shows the energies as a function of k. Unfortunately the results are
physically incorrect for = 0. Suppose that the modified dispersion of the electron could be described by an effective
mass m
Ek = Eshift +

~2 k 2
2m

dEk =

~2 k
dk.
m

Since this also applies for the nonperturbed case (with m = m), we can write for the ratio
!


2
1 + k 
m
exch
e2 m 1 1 + k
k
2 ,
ln 
=1+
=
m
k
2kF k 2
k
1 k

(891)

(892)

which has a logarithmic divergence at the Fermi energy k = k/kF 1. This would have serious consequences at low
temperature, where the Fermi gas would become unstable. For example, the specific heat would diverge as T / ln T ,
which is not observed experimentally. The divergence is a result of the very long range of the Coulomb interaction. If
can be removed by a non zero . In Fig. 14, one sees that, although the divergence is removed, the energies change
relatively little. The effective Coulomb interaction is the e2 (er /r). The limited range of the Coulomb interaction
is known as screening.
B.

Thomas-Fermi screening

How can we obtain such a screening of the Coulomb potential? It is possible to obtain it from the Fermi sphere
and allow the creation of electron-hole pairs (known as the random phase approximation). However, this requires a
knowledge of Greens functions and would go beyond the scope of this course.
The results are very similar to the Thomas-Fermi model. We can understand screening physically by noting that
the electron cloud reacts to the electron and will move as a result of the potential created by the electron. However,
in a system in equilibrium we assume that the chemical potential = EF (r) eV (r) remains constant. Thus the
change in potential has to be compensated by a change in number of electrons
dEF
n eV = 0.
dn

(893)

As we saw before, in a free electron model, the Fermi energy is given by

kF = (3 2 n)1/3 EF =

~2 kF2
~2
=
(3n)2/3 ,
2m
2m

(894)

98

FIG. 14: The energies k and exch

plus the total energy for a0 kF / = 1, where EF = ~2 kF2 /2m. The solid line are the results
k
for = 0, the dashed line for = 0.1.

with the electron density n = N/V . This gives dEF /dn = 23 EF /n. The induced charge is therefore
ind = en =

3 e2 nV
2 EF

(895)

where use has been made of Laplaces equation 2 V = tot /0 . If we write it in Fourier components, we can use
q 2 V = tot /0 , leading to
ind (q) =

3 e2 n
tot (q)
2 EF 0 q 2

(896)

We now like to relate the effect of our applied field (the electron) and the real field
E=

1
(D P)
0

E=

1
( D P)
0

tot = ext + ind

(897)

Which means that the real electric field is the field from the electron minus a polarization field as a result of the
reaction of the electron gas to that field. Now if we assume that the real electric field is directly proportional to the
applied field 0 = D/E, we can write
0 E = D

or

tot = ext = tot ind

(898)

For the dielectric function we then find

(q) = 1
with qT F =

3 e2 n
2 EF 0 .

ind
3 e2 n
qT2 F
=1+
=
1
+
tot
2 EF 0 q 2
q2

(899)

The Coulomb potential in momentum space is then

Vq =

1 e2
1
e2
=
,
q2
0 q 2 + qT2 F

(900)

which shows that our = qT F .

XIV.

LATTICE VIBRATIONS

Let us consider a (linear) solid consisting of equal atoms. The force pulling the atom back into its equilibrium
position is linearly proportional to the distance with the neighboring atom. This leads to the Hamiltonian for the
atoms
X p2
KX
n
H=
+
(xn xn+1 )2 .
(901)
2M
2
n
n