Homeworks PDF

Physics , Condensed Matter
Homework
Due Tuesday, th September
Jacob Lewis Bourjaily
Problem 1
We are asked to study the penetration of normally incident, linearly polarizedwith polarization
parallel to the surfaceelectromagnetic radiation into a conductor using the Drude model. Let the
surface be located at z = 0, with z > 0 vacuum and z 0 be a conductor. We may assume that the
relaxation rate is large relative to the plasma frequency, so p 1, and that the plasma frequency is
large relative to the incident radiation, p 1; we should allow to be arbitrary.
a) Let us first consider the limit of a free electron plasma, where 1/ 0. We are to solve for the
full pattern of the electric field both within and without the conductor, calculate the skin depth, and
determine quantitative skin depth in this approximation for visible light at 61014 Hz incident on copper.
We begin by reminding ourselves of some simple electrodynamics learned by rote long ago
when we took Jackson: for light incident on a surface at z = 0, with outward normal
~n, the conditions to be imposed at the boundary are that the normal components of
B and D and the tangential components of E and H are continuous. If we express
the fields in question as
n
o
n
o
n
o
0
Ein = Re x
Ein ei(kz+t)
Eni = Re x
Eni ei(kzt)
Er = Re x
Er ei(k z+t) ,
(a.1)
where Eni is the reflected wave and the others are self-evident, then making use of
Maxwells equations to relate Bi to Ei then we find these boundary conditionsthe
ones for E and Himply that
(Ein + Eni Er ) ~n = 0
~k Ein ~k Eni k~0 Er ~n = 0
= Er = Ein + Eni ;
(a.2)
= k 0 Er = k (Ein Eni ) .
(a.3)
Now, Maxwells equations give us 2 E = c2 ()E, so k = c in the vacuum and
k 0 = c in the medium. This allows us to solve the boundary conditions above

rather straight-forwardly in no more than a couple of lines of algebra:
1
2
.
Eni = Ein
and
Er = Ein
1+
1+
(a.4)
After the above preliminaries, we are ready to perform the more specific challenges of
the problem. We can easily find the limit of the expression for () predicted by the
Drude model when 1/ 0:
4ine2
,
m(1 i )
4ine2
,
=1+
m (1/ i)
4ne2
1
.
q/ 0
m
For much of the range of light frequency1, this is a negative, real-valued dielectric constant, which means
that light essentially does not penetrate the surface. To see this,
recall that k 0 = , so if is real and negative k 0 is pure imaginary, which means
that the strength of the refracted wave dies exponentially inside the surface. This
exactly follows our intuition about plasmas. The skin depth is given by
() = 1 +
c
1
q
.
4ne2 1
2
m
(a.5)
o o
1The point at which k becomes real for copper (in this approximation, which is crude) is 2.6 1015 Hz.
1
JACOB LEWIS BOURJAILY
Before we move on, we are to calculate the skin depth of copper in this approximation,
using real numbersa headache to most theorists. To do this we need to choose a
consistent set of units. I will use the units for which
e2 = 1.907 1072 m2
me = 6.764 1058 m.
These work out quite well and one finds that

Cu = 19 nm.
(a.6)
b) We are now asked to generalize our work above to the situation where there is scattering in general. We should simplify our expressions as much as possible by keeping only leading terms in /p and
1/(p ). We are to determine the resulting electric fields for the situation of part a above, calculate the
absorption coefficient and plot this as a function of .
Just in case the grader is keeping a tally, please notice that our solution for the full
electric field pattern in part a above did not depend on the assumption that there
was no scattering, so the result applied exactly.
Before we begin, we should comment that we have found nothing slight of horrendous
in this problem. There is little elegance, and in general, everything becomes messy
very fast. Let us just clarify our starting point and our goal: we know that in the
Drude model
ip2
() = 1 +
,
(1 i )
and from our course in electrodynamics so many years ago2 that the absorption
coefficient is given by
4Re { }
T =
2.
|1 + |
Let us now begin. We will make repeated use of the fact that p 1 and p / 1.
The first instance of this appears in the third line, if youre paying attention. To
simplify life a lot, we will define the parameter so that sinh = .
(b.1)
(b.2)
ip2
(1 + i )
,
(1 i ) (1 + i )
p2 2
i
=1
1
,
sinh
1 + sinh2
p2 2
i
,
sinh
cosh2
=1+
p2 2
ei(+) .
cosh sinh
In the last line, we used some hyperbolic trigonometric identities normalizing where
we have defined the phase = Arg {1 i/ sinh } = arctan(1/ ).
Now before
we jump through the last hoops, it is useful to notice right now that
Re { } p , so if we are keeping things to order 1/(p ), then we need only
look at terms in the denominator of the expression
for T that are second order at
least. Indeed, this means we can drop the 1 + 2Re { } bit from the denominator,
simplifying life enormously. Okay, so with that big approximation made clear, we see
directly that
4Re { }
4Re { }
,
T =
2
||
|1 + |
!1
p2 2
4p
1
=
cos
( arctan(1/ ))
,
2
cosh sinh
cosh sinh
=
2It is not really necessary to quote the result, because considering that the power is |E|2 , this expression is fairly
2
obvious from our work in part a.
PHYSICS : CONDENSED MATTER HOMEWORK
Transmission in units of 1
p!
1.5
0.5
0
0
2
!
Figure 1. The absorption coefficient T as a function of as estimated in problem (1.b).
T '
4
p
1 + 2 2 cos
1
( arctan(1/ )) .
2
(b.3)
This is shown in Figure 1.

c) We are to compute the skin depth of copper for = 2.71014 sec using our work above at various
frequencies.
From our work above in part b it we can easily see that (in the approximation used
there)

p
1
Im
=p
sin
( arctan(1/ )) ,
2
1 + 2 2
which allows us to write the skin depth
c
1
.
=
Im { }
Using Mathematica so I wouldnt make any silly mistakes, I found the following:
Cu (60 Hz) = 8.08 mm
Cu (1010 Hz) = 0.625m
Cu (6 1014 Hz) = 18.2 nm.
(c.1)
(c.2)
(c.3)
Problem 2
We are to modify the Sommerfeld theory of electrical and thermal conductivity to incorporate two
disparate types of scattering events: those with a relaxation time of v which are elastic but thermally
randomize the direction of an electrons velocity; and those with a relaxation time of e which fully
equilibrate the electrons energy to thermal equilibrium while leaving the speed and direction of the electron unchanged. This is perfectly valid in the limit of temperatures well below the Fermi temperature,
because in that case virtually all of the effective conduction electrons are on the Fermi surface and have
velocity vF .
a) We are to compute the electrical conductivity in this two-scattering generalization of the Sommerfeld model.
There are various ways we could make this a bit more rigorous, but our intuition strongly
argues that the scatterings which leave the direction of motion unchanged will not
contribute to resistance. Indeed, if one were to follow the same type of analysis we
did in the one-scattering case, we would find the average velocity at a time dt to be
given by
dt
dt
~ + dt h~v (t = 0)i + O(dt2 ),
h~v (dt)i = 1
1
h~v (t = 0)i eE
v
e
e
(a.1)
where the last term is added because with a probability of dte during the interval dt
the electrons can scatter via these inelastic pathways which do not alter the velocity.
A quick glance at the equation above shows that this cancels the resistive force
caused by the e scattering, so there is no change to our derivation of the electrical
conductivity in the original model. Therefore, we see that
ne2 v
.
m
(a.2)
b) We are to compare thermal conductivity in this model with the original Sommerfeld model.
Unfortunately, we will need to work a little less rigourously than we would otherwise
prefer. Most of the results we can more-or-less guess by considering the symmetries
and limits than any solution must have; indeed, it is easy to see that if 1/e 0 the
thermal conductivity will vanish, and similarly if v 0; in the first case there are
too few inelastic scatterings to transport information about temperature gradients,
and in the latter case any thermally interesting transport is washed out by rapid
elastic scattering.
Let us first compute the expected scattering time for the combined, independent scattering processes. This is rather straightforward: notice that the probability for an
electron to survive until a time t without scattering elastically is et/v and the
probability to survive until a time t without scattering thermally is et/e . Because these are independent random variables, the probability to survive to a time
e +v
t without any collision is simply the product, or et e v . For this, the differential
e +v
v t e v
probability of not scattering is ee+
; and from here the evaluation of an
v e
elementary integral shows that the expected time between collisions is
hti =
e v
.
e + v
This of course satisfies our intuition because when one of e or v is small, it will always
dominate: if on process is much more rapid, the other can be effectively ignored.
(a.1)
Now, a correct derivation would begin by using the fact that only the inelastic scatterings
will contribute to the thermal current. One would find something along the lines of
n vF
j=
[(T (x = vF e )) (T (x = vF e ))] ,
2 3
n vF
T
T
=
(T (x = 0)) vF e
(T (x = 0)) vF e
+ ... ,
2 3
T x
T x
v
x T,
= F e n
3
T
v2
= e F cv ,
3
2F
2 k2 T
=
e n B ,
3m
2
F
2
2 e kB
T
=
,
3
m
where we have introduced the parameter which parameterizes our ignorance (not
fundamentally, just the ignorance of the author): represents the fraction of electrons
arriving at x from a given side such that their last scattering was inelastic. A good
guess for would be3
v
?
=
.
v + e
At least it has the right properties and limits. If this were the case, then we would
find
2
1 2 kB
e
=
.
T
3 e2 e + v
3Note added in revision: this is the right answer.
(a.2)
(a.3)

Homework
Due Tuesday, rd October
Problem 1
Consider a trigonal Bravais lattice generated by the primitive vectors ~ai for i = 1, 2, 3 such that
~ai ~aj = a2 cos for i 6= j.
a) We are to determine for what angles this lattice is three-dimensional.
There are many ways by which this answer can be visualized, but to be a bit more
mathematically explicit (and ergo avoiding the necessity of diagrams), we will proceed
differently. If the vectors ~ai are to be taken as a basis in three-dimensions, then the
volume element is given by the square-root of the determinant of the corresponding
metricthe metrics elements are gij = ~ai ~aj . That is,
1
cos cos
2
1
cos (cos 1)2 (cos + ).
(Volume form) = a2 cos
cos cos
1
(a.1)
From the above, it is clear that the space is three-dimensional iff the volume-form is real
and non-vanishing. We therefore see that when cos = 1, the volume vanishes
corresponding to a two-dimensional lattice. Furthermore, we see that because equation (a.1) has a positive coefficient for cos3 and its lowest root at cos = ,
the expression is negative for cos < which amounts to an imaginary volume
element1.
Therefore, a three-dimensional Bravais lattice is obtained only for (0, 2/3).
o
b) Let us show that as varies, the trigonal lattice becomes each of the higher-symmetry cubic
lattices.
There are three values of which give rise to enhanced symmetry. The first, and most
obvious to see, is for = /2 which clearly gives rise to a simple cubic lattice. The
other two cases are a bit more subtle.
Consider the plane spanned by ~a1 and ~a2 . When = /3, we see that a two-dimensional
lattice of equilateral triangles is spanned. Now, because ~a3 lies out of the plane at
an equal angle, the four points ~0, ~a1 , ~a2 , ~a3 form the corners of a regular tetrahedron,
which obviously gives rise to enhanced symmetry. We know that this structurean
equilateral triangle of points with a point on the next layer directly in the center
gives rise to the close-packing of spheres in three-dimensions, so our lattice must
either be face-centred-cubic (fcc) or hexagonal-close-packed (hcp). It is not hard to
see that our Bravais lattice gives only fcc: if a corner of the canonical fcc cube is
taken as the origin, then the vectors ~ai correspond to the points at the centres of the
three faces which are coincident at the corner in question.
The last case of enhanced symmetry arises when {~a1 , ~a2 , ~a3 , (~a1 + ~a2 + ~a3 )} form a
regular tetrahedron. For those of us who loved high-school chemistry, we know
that the internal angle of a regular tetrahedronwhich is the angle between two
hydrogen atoms in CH4 is arccos(1/3) 109.5o . And so our answer is: when
= arccos(1/3) the trigonal Bravais lattice is the body-centred-cubic (bcc) lattice.
Just to motivate the answer for bcc a little better, recall from the textbook that the bcc
Bravais lattice can given as follows: let a corner of the canonical bcc cube be placed
at the origin with the three edges meeting at the corner coincident with the x, y and
z-axes. Then the bcc Bravais lattice can be spanned by ~ai which point toward the
1There are really clear ways of visualizing the unacceptability of cos < 1/2. For example, consider the case where
cos is very near 1: here, we see that this means that ~a2 and ~a3 are both to be nearly anti-coincident with ~a1 , which
means that they cannot be mutually so.
1
centres of the cubes in, e.g., the (+ + +), ( +), and ( +P) quadrants. And the
centre of the cube in the (+ ) quadrant is given by i ~ai . Now, there are a
lot of fancy tricks to determine the internal angle of a tetrahedron; but we shall be
brief, dry and boring and simply compute it directly. In our coordinates, in units
of the lattice spacing a, the vectors ~a1 = 13 (1, 1, 1) and ~a2 = 13 (1, 1, 1). Being
unit vectors, ~a1 ~a2 = 31 (1 2) = 1/3 = cos . Therefore, the angle is given by
arccos(1/3).
c) We are to find the reciprocal lattice of the trigonal lattice and verify the special cases found
above.
We can use the canonical expressions for the reciprocal lattice vectors:
ijk
~bi (~aj ~ak ) ,
~a1 (~a2 ~a3 )
where Einstein summation convention is employed2. At any rate, we find easily that
2
~a1 (~a2 ~a3 ) = a3 sin2 and that this implies |~bi | = a sin
. Using the usual identities
about the inner product of two pairs of cross-products, we see that
2
4
~bi ~bj = 4
a cos2 a4 cos
for i 6= j.
a6 sin4
Therefore, the angles between the reciprocal basis vector ~bi are all equal, and given by
where
~b1 ~b2
cos =
,
|~b1 |2
2 2
4 2
a sin
=
cos cos ,
4
2
4 2
a sin
cos2 cos
,
=
1 cos2
cos
.
1 + cos
For the special values of which correspond to the three cubic lattices, we see
1/2
= 31 . This is the angle which was found to
fcc: = /3 = cos = 1+1/2
generate the bcc lattice.
simple cubic: = /2 = cos = 0 which corresponds to another simple
cubic lattice.
1/3
bcc: cos = 1/3 = cos = 11/3
= 1/2; so = /3, which corresponds to
a fcc lattice.
These results match our understanding of reciprocal these reciprocal lattices.
cos =
(c.1)
(c.2)
(c.3)
2The factor of in the expression is correct: it accounts for the fact that when summing over jk there will be two
contributions.
Problem 2
Consider an ideal, two-dimensional honeycomb lattice of atomsthis could be, for example, graphene.
For specificity, take the honeycomb lattice to be aligned in the xy-plane with the y-axis parallel to one
of the nearest-neighbour atomic spacings. Call the distance between nearest-neighbours d.
a) We are to specify and sketch the reciprocal lattice and state the magnitude of the smallest
reciprocal lattice vector for graphene, where d

= 1.4 A.
The honeycomb lattice can be considered a (2-dimensional) trigonal Bravais lattice with
a basis containing two atoms. For specificity, a little trigonometry tells us that if d
is the spacing between two atoms, then in our chosen orientation of the plane, our
Bravais lattice is generated by
!
1
3
~a1 d 3(1, 0) and ~a2 d 3
,
.
2 2
(a.1)
The Bravais lattice with basis generated by these vectors is illustrated in Figure 1.
Figure 1. This figure shows the original honeycomb lattice, as viewed as a Bravais
lattice of hexagonal cells each containing two atoms, and also the reciprocal lattice of
the Bravais lattice (not to scale, but aligned properly).
To find the corresponding reciprocal lattice, we must satisfy the defining equations ~b1
~a1 = 2 and ~b1 ~a2 = 0, and a similar system for ~b2 . These are easily found by hand,
and it is seen at once that the required reciprocal lattice is generated by
!
4
4
3
1
~b1
and ~b2
,
(0, 1) .
3d
2
2
3d
(a.2)
This clearly generates the same lattice as ~a1 and ~a2 , but rotated by /2. This is
illustrated also in Figure 1.
Using our expression in equation (a.2) above, we see that if d
= 14 nm, then the smallest
reciprocal lattice vector has magnitude 4

2.46
A
.
=
3d
b) Treating the atoms as identical scatterers, we are to determine the intensity of all the Bragg
peaks, normalized so that the strongest peak has unit intensity.
If we let ~r0 = (0, d/2), then the density function (~r) over one cell is given by
(~r) = (~r ~r0 ) + (~r + ~r0 ),
which gives a form factor of
i~q~r0
e
+ ei~q~r0 cos (~q ~r0 ) .
We know that the wave function in ~q space can be expressed as this form factor times
a piece from the Bravais lattice; the Bravais lattice piece gives a multiplicative factor
of N (the number of lattice cells) and enforces that ~q is in the reciprocal latticethat
is, the wave function identically vanishes for ~q not in the reciprocal lattice.
(b.1)
(b.2)
Now, an arbitrary ~q in the reciprocal lattice is given by ~q = n1~b1 +n2~b2 , where n1 , n2 Z

and ~b1 , ~b2 are given in equation (a.2). This shows us immediately that for ~q in the
reciprocal lattice,
~q ~r0 = (2n2 n1 ) .
(b.3)
3
Combining this with our work above, we see that
I |s |2 cos2
(2n2 n1 ) ,
(b.4)
3
and furthermore, the expression has the desired feature of intensitythat the strongest
peak has unit intensityso that we find the normalized intensity to be given by
I(n1 , n2 ) = cos2
(2n2 n1 ) .
(b.5)
3
This is shown in Figure 2.
10
qy
8
6
4
2
0
0
2.5
7.5
10
12.5
15
17.5
qx
Figure 2. This is a plot of the intensity as a function of ~q (in units where the smallest
reciprocal vector is of unit length). Only at discrete values of ~q is there any scattering,
and the intensity is given by the expression (b.5).
c) Consider the cases where the two atoms in the unit cell are distinct, call them A and B, each
with different scattering amplitudes fA and fB . We may assume that they are both real. What condition on the relative scattering amplitudes will cause the intensities of some of the Bragg peaks to vanish?
Because both fA and fB are real, we may without loss of generality suppose that fB =
fA . Just to be exceedingly explicit, we will say that the A atoms are located at
~ + ~r0 and the B atoms are located at R
~ ~r0 where R
~ is the Bravais lattice and
R
~r0 = (0, d/2) as above. With separate scattering sites, we see that the wave function
for scattering goes like3
X
X
~
~
s fA
ei~q(R+~r0 ) + fB
ei~q(R~r0 ) ,
~
R
i~
q ~
r0
fA e
+ e
i~
q~
r0
~
R
Therefore, we find that the intensity evolves like
I |s |2 1 + e2i~q~r0 + e2i~q~r0 + 2 ,
= 1 + 2 cos (2~q ~r0 ) + 2 ,
2
= 4 cos2 (~q ~r0 ) + (1 ) .

3In the second line we make use of the fact that the amplitude is non-vanishing only for ~
q in the reciprocal lattice.
Now, to see if there is any which will cause some of the Bragg peaks to vanish, we
merely need to see if there are any roots to the equation above. It is a simple
quadratic and it is easily reduced to I = 0 iff
p
= 1 2 cos2 (~q ~r0 ) 2 cos (~q ~r0 ) cos2 (~q ~r0 ) 1.
(c.1)
Because we have assumed that both amplitudes are real, R; and because
cos2 () 1, we see that there is a solution iff cos2 (~q ~r0 ) = 1, for which we see
that = 1.
Therefore, if = 1 there is interference causing all the Bragg peaks for which cos2 (~a ~r0 ) =
0; this is satisfied if 2n2
somem Z, which correspond to the mo n1= 3m for
3
mentum transfers ~q = 4
n
3
+
3m
2
3d
2 , 3m .
o o
d) Now let us assume that the atoms A and B are placed randomly on the honeycomb lattice
with an even probability distribution. For this random crystal, we are to determine what fraction of the
total scattering intensity is in the Bragg peaks (i.e. not diffuse scattering).
~ {1, 1}, a function whose value is either +1 or 1 for each
Let us define a map : R
~ It must have the property that | 1 (1)| = | 1 (1)|,
point of the Bravais lattice R.
~ to
which means that its average value over the lattice is zero. We will take (R)
signify the direction of the A atom relative to the center of the cell at the Bravais
~ With this in mind, the atom density functions for A and B are given
lattice point R.
by
X
X
~ + (R)~
~ r0 ,
~ (R)~
~ r0
~r R
~r R
and B (~r) =
A (~r) =
(d.1)
~
R
~
R
where we again use the definition ~r0 = (0, d/2).

Let us again say that fB = fA . We now find that the scattering amplitude goes
like4This is where the equation goes wrong: because of the assumption stated in the
footnote, we are effectively only calculating the Bragg contribution.
X
~
~
~
~
s fA
ei~q(R+(R)~r0 ) +
ei~q(R(R)~r0 ) ,
~
~
R
R
X
~
~
ei(R)~q~r0 +
ei(R)~q~r0 .
~
R
~
R
~ = +1 for half of
Now, when we expand out the two sums we find that because (R)
~
the sites and (R) = 1 for the other half, each sum has an equal number of terms
with positive exponents and negative exponents. Recall that exp{+i}+exp{i} =
2 cos(). Therefore, up to a constant of proportionality, we have
s {cos (~q ~r0 ) + cos (~q ~r0 )} .
(d.2)
This is not right.
4In the second line we again make use of the fact that the amplitude will in general vanish unless q
~ is in the reciprocal
~ = 2 for all R
~ in the Bravais lattice.
lattice, i.e. q~ R
e) Let us consider fully three-dimensional graphite which has a simple hexagonal Bravais lattice.
We are to show that the reciprocal lattice of a simple hexagonal Bravais lattice is also a simple hexagonal
lattice. For graphite, we are to show that the internal symmetries of the basis gives rise to to the vanishing of certain Bragg scattering directions and list which reciprocal lattice vectors show this complete
interference.
First, let us state the vectors which generate the Bravais lattice of graphite. It is not
hard to extend our two-dimensional analysis to three-dimensions:

!
3 3
~a2 d
,d ,0
~a3 (0, 0, ),
~a1 (d 3, 0, 0)
2
2
where is the height between the graphite layers.
Using our general expression for the reciprocal lattice vectors, equation (1.c.2), we can
quite directly compute
~b1 2
~b2 4 (0, 1, 0)
~b3 2 (0, 0, 1) .
3, 1, 0
3d
3d
To see that these generate a simple hexagonal lattice, notice that the angle between
~b1 and ~b2 , called 12 is given by
1
cos 12 = ,
2
which obviously generates a hexagonal lattice in the xy-plane; and because ~b3 is in
the z-direction, we see that the set ~bi generate a simple hexagonal lattice.
If we consider scattering off a perfect graphite lattice, we know that there will be a Bragg
condition forcing the momentum transfers ~q to be elements of the reciprocal lattice:
~q = n1~b1 + n2~b2 + n3~b3 .
Recall that, for a perfect crystal
(e.1)
(e.2)
(e.3)
(e.4)
~
s (~q G)
d~r
(~r)ei~q~r ,
(e.5)
one cell
~ where G
~ denotes the reciprocal lattice. Now, in the basis
and s vanishes for ~q
/ G,
cell, there are four atoms located at
~r1 = (0, 0, 0) ~r2 = (0, d, 0) ~r3 = (0, d, /2) ~r4 = ( 3d/2, d/2, /2).
(e.6)
These give delta-functions in the density, which turn the integral into a discrete sum:
( 4
)
X
s
ei~q~ri .
(e.7)
i=1
Notice that ~r3 = ~r2 + ~r4 ~a2 . Therefore, we see that
2 1 + ei~q~r2 1 + ei~q~r4 = 1 + ei 3 (2n1 n2 ) 1 + ein3 ei 3 (n1 +n2 ) .
(e.8)
Now, the exponential in the first parenthesis cannot be 1 because there are no integer
solutions to the equation 2n2 n1 = 23 . Therefore, the only way for the intensity to
vanish is if the second term in parenthesis vanishes, which requires that simultaneously,
(n1 + n2 ) = 0 mod 3
and
n3 (2Z + 1) .
(e.9)
o o

Homework
Due Tuesday, th October
Problem 1: Electron in a Weak Sinusoidal Potential1
Consider an electron moving in a one-dimensional periodic potential U (r) = V cos (2r/a). We are to
obtain the eigenenergies n (q) and corresponding wavefunctions n,q (r) of the lowest two bands, treating
the potential perturbatively.
a) Away from the edge of the Brillouin zone, there are no degeneracies in the lowest energy band.
Using this fact, we are to compute 1 (q) to order V 2 and the corresponding wavefunctions to order V .
We begin by doing what amounts to Fourier transforming (r) into momentum space,
making use of Blochs theorem to write
X
q (r) =
cqG ei(qG)r ,
(1.a.1)
G
where G represents the reciprocal lattice, which is in this one-dimensional problem

generated simply by b 2
a . Following Ashcroft and Mermin, we will study the
Schrodinger equation in momentum space:
2

X
~
(q G) cqG +
UG G cqG = 0,
(1.a.2)
2m
where Uk are Fourier modes of the potential. In our case, this is extremely easy
to extract: every kindergartener should be able to take the Fourier transform of a
cosine; we find:
V
U+1 = U1 =
and Ui6=1 = 0.
(1.a.3)
2
We are going to be interested in a wave function concentrated well within the first
Brillouin zone, in the limit where there are no nearly degenerate bands. Writing
~2 2
q and inserting our potential Uk into the Schrodinger equation, we see
0q = 2m

V
V
0q cq = (cq+b + cqb )
=
cq =
(cq+b + cqb ) .
(1.a.4)
2
2( 0q )
This expression does not reflect our interest in the first Brillouin zone: it is valid
for all q. Indeed, we see that we can iteratively unfold the equation to obtain an
V
expansion in terms of 2(
0 ):
q
V
(cq+b + cqb ) ,
(1.a.5)
2( 0q )
(
)
V
V
V
=
(cq + cq+2b ) +
(cq + cq2b ) ,
(1.a.6)
2( 0q ) 2( 0q+b )
2( 0qb )
(
) "
#
V 2 cq
1
1
V3
=
+
+
(cq+b + cq+3b ) + (b b) ;
4( 0q ) ( 0q+b ) ( 0qb )
8( 0q )( 0q+b )( 0q+2b )
(1.a.7)
cq =
&tc. Now, because we are expanding in
V
2(0q ) ,
we may safely drop the O(V 3 )
terms. Also, notice that in the curly brackets that we have the expressions 0qb .
Now, because the bands are non-degenerate and we know the band energies are only
affected by terms leading in V 2 , we can are allowed to take ( 0qb ) 7 (0q 0qb )
to this order of approximation. With that in mind, we may divide both sides of the
expression above by cq /( 0q ) obtaining
(
)
V2
1
1
0
( q ) =
+
+ O(V 3 ).
(1.a.8)
4
(0q 0q+b ) (0q 0qb )
1Note added in revision: the solution presented follows Ashcroft and Merminwhich is absolutely horrendous. The
entire first problem can be done in a couple of lines if you read the first few pages of Griffiths Quantum Mechanics chapter
on (time independent) perturbation theory first! Honestly, believe me; learn non-degenerate perturbation theory first (and
see how to apply it in the degenerate case) and the problem will be MUCH easier.
1
It is not altogether delightful, but this expression can of course be quite dramatically
simplified.

V 2 2m 2q + b 2q + b
0
= q +
+ O(V 3 ),
4~2 b
(4q 2 b2 )
(
)
V 2m
1
0
= q + 2
+ O(V 3 ),
2
~
(4q 2 4
)
a2
=
~2 2
V 2m
q +
2m
4~2 q 2
2
a2
+ O(V 3 ).
(1.a.9)
o o
Now we are asked to determine the corrected wave function to leading order in V . To
do this, we start by combining the right hand sides of equations (1.a.5) and (1.a.7):
(
)
V
V 2 cq
1
1
(cq+b + cqb ) =
+
+ O(V 3 ),
2( 0q )
4( 0q ) (0q 0q+b ) (0q 0qb )
(
)
V
1
1
= (cq+b + cqb ) = cq
+
+ O(V 2 ),
2 (0q 0q+b ) (0q 0qb )
= cq
2~2
Vm
q2
2
a2
+ O(V 2 ).
(1.a.10)
Now, remember that equation (1.a.4) allows to write cqb in terms of cq and cq2b , for
example. Using this to rearrange equation (1.a.10), we see
Vm
2
cq+b = cqb + cq
2 + O(V ),
2~2 q 2 a2
V
Vm
2
= 0
(cq + cq2b ) + cq
2 + O(V ),
2
2(q 0qb )
2~ q 2 a2
(
)
Vm
V
= cq
+ O(V 2 ),
2
2(0q 0qb )
2~2 q 2 a2

1
Vm
= cq
1
(aq + ) + O(V 2 );
2
2
2
2
2~ q a2

Vm
aq
cq+b = cq
+ O(V 2 ).
(1.a.11)

1
4~2 q 2 a2
We dont need to reproduce the above steps for cqb : it comes for free once we have
cq+b :
Vm
2
cqb = cq+b + cq
2 + O(V ),
2
2~ q 2 a2

Vm
1
qa
= cq

1
+ O(V 2 );
2
2
2~2 q 2 a2

qa
Vm

cqb = cq
1
+
+ O(V 2 ).
(1.a.12)
2
4~2 q 2 a2
Inserting this in the expansion for q (r), we find directly,
q (r) = cq eiqr + cq+b eiqr eibr + cqb eiqr eibr + O(V 2 ),

(
)
h
Vm
aq irb
qa irb i
iqr
= cq e
1+
1
e + 1+
e
+ O(V 2 );
2
4~2 q 2 a2
(

)

Vm
2r
iqa
2r
iqr
q (r) = cq e
1+
cos
sin
.
2
a
a
2~2 q 2 a2
(1.a.13)
o o
b) At the edge of the Brillouin zone there are degeneracies for small V . We are to work perturbatively near the zone edge to diagonalize the single electron Hamiltonian within the two-state Hilbert
space of the two nearly-degenerate lowest-energy free-electron momentum eigenstates of the same crystal
momentum. Then, we are to add the effects of the higher bands perturbatively. We are to obtain the
eigenenergies of the lowest two energy bands to order V 2 and the wave functions to order V as before.
We are to verify that for small V these results match our work for part (a) when one moves far enough
away from the edge of the Brillouin zone. We are to sketch the dispersions n (q) and determine how
small V must be for this perturbation analysis to be reliable.
We are going to proceed along lines similar to those encountered in part (a). Specifically,
let us start by again by equating the right hand sides of equations (1.a.5) and (1.a.7)
only this time, we will not use the assumption that all the eigenenergies are nondegenerate.
(
)
V2
1
1
0
+
+ O(V 3 ).
(1.b.14)
( q ) =
4
( 0q+b ) ( 0qb )
Now, we are going to consider perturbing the system near the Bragg plane at q = a ;
this will mean that we can consider the term 0q+b where 1/ is at most linear
in V we will justify this and give an explicit expression for later.
Manipulating equation (1.b.14), we see that
!
0qb

V2
0
q ( qb ) =
1+
+ O(V 3 ),
4
!

0qb
V2
V2
2
0
0
0 0
= q + qb +
+ q qb
1
= 0.
4
4
This quadratic is easily solved by calling upon kindergarten identities:

(
!)1/2

2
0qb
1 0
V2
1
V2
0
0
0
0 0
2
=
+ qb +
q + qb +
4q qb + V
1
;
2 q
4
2
4
1
=
2
(
)1/2

2
V2
1
V2
0
0
0
0
2
q + qb +
q qb +
+V
.
4
2
4
(1.b.15)
o o
To evaluate the expression, one can simply insert the equation itself iteratively into
= 0q+b and observe that that it always gives a well-defined expression up to
terms of order O(V 3 )2. This band structure is shown in Figure 1.
We should check that this result makes senseand verify that it agrees with our previous
work once we are far enough away from the Bragg plane. First, notice that at the
Bragg plane, where q = b q = /a, we have

1/2

V2
V
V2
q=
= 0/2 +
1+
,
a
8
2
16 2
~2 2
V
V2
=
+
+ O(V 3 ).
2m a2
2
8
Inserting this into definition of as prescribed, we obtain

~2 2
V
V 2 ma2
q=
=
+ O(V 3 ).
a
2m a2
2
32 2 ~2
(1.b.16)
2The reason for being implicit here is that the two cases we are interestednear and far from the Bragg planegive
different results; but the implicit expression is always correct.
Energy Eigenvalue
40
30
20
10
0
0
0.5
1.5
2
Momentum q
2.5
Figure 1. The second-order band structure for a one-dimensional system in a weak

sinusoidal potential.
Similarly, we can check that equation (1.b.15) gives the right answer when we are far
enough away from the Bragg plane. When we are far from the Bragg plane, then
0q 0qb V 2 so that we may expand
1
=
2
0q
0qb
V2
+
4
(
0q
0qb
V2
+
4
2
1
V2
0q + 0qb + V 0q 0qb

2
4
0 0
2
qb
+V
)1/2
2 1/2

V2
,

2 + 1 +
4 0q 0qb
1/2

V2
4
,

+ O(V )
2 +
2 0q 0qb
2
1
V2
0 + 0 + V 0 0
1
+

q
qb
q
qb
2
4
0q 0qb
2
2
2

1 0
V
V
V
+
=
q + 0qb +
0q 0qb +
+ O(V 4 ) .
2
4
4
0
0
2
qb
Taking the solution corresponding to the lower band3,

(
)
V2
1
1
1
0
1 (q) = q +
+ 0
+
+ O(V 3 ),
4
2
q 0qb
2
(
)
V2
1
1
0
= q +
+ 0
+ O(V 3 ),
4
0q 0q+b
q 0qb
and this we recognize as equation (1.a.8), which implies that this formula (1.b.15)
does indeed agree with our results from part (a).
o o
Our last task is to determine the wave function for electrons at the Bragg plane to first
order in V . We will follow similar lines of thought to those travelled in part (a). Using
the same logic as thereonly this time being careful not to ignore degeneracieswe
can begin our work with the equations
(
)
V
1
1
V
cq+b + cqb = cq
+
+ O(V 2 ) and cq =
(cq+b + cqb ) . (1.b.17)
0
0
2 ( q+b ) ( qb )
2( 0q )
3
The solutions corresponding

to the respective sign the equation (1.b.15) have now switchedthis is simply because
when we extracted
0q 0qb
from the square root, the signs one again become arbitrarily assigned.
This system yields exactly our result in part (a) for the case of cq+b :
(
)
V
1
1
cq+b = cq
+
cbq + O(V 2 ),
2 ( 0q+b ) ( 0qb )
(
)
V
1
1
1
cq+b = cq
+
+ O(V 2 ),
2 ( 0q+b ) ( 0qb ) ( 0qb )
cq+b = cq
cq+b = cq
V
+ O(V 2 ),
2( 0q+b )
V
+ O(V 2 ),
2(0q 0q+b )
= cq
V 2 ma
+ O(V 2 );
4~2 (q + a )

qa
V2

1
+ O(V 2 ).
(1.b.18)
2
4~2 q 2 a2
The story changes, however, for cqb . It is not hard to jump a bit in the calculation and
see
V
cqb = cq
+ O(V 2 ).
(1.b.19)
2( 0qb )
Now, from our calculation of the eigenenergies at the Bragg plane we know that
cq+b = cq
/a 0/ab =
so we see
V
V 2 ma
+ O(V 3 ),
2
16~2 q + a
cqb = cq
2 V2
= cq
cqb = cq
= cq e
iqr
1+
1e
V
16~2 (q+
a)
ibr
8~2
(r) cq e
iqr
2e
i br
2
cos
and
(r) cq e
iqr
2ie
i br
2
r
sin
+i
r
a
+ O(V 2 ),
V
8~2 (q+
a)
+ O(V 2 ),
Vm
q2
qa
2 ma
4~2
so that
+
2 ma

1
Putting all this together, we see

(
q=
(r)
a
2
a2
Vm
q2
(1.b.20)
!

+ O(V 2 ).
(1.b.21)

)

aq ibr eibr
1
e
,
2
2
a2
2~2
Vm
q2

)
aq
2r
eibr
1
sin
i
;
2
a
4
a2
2~2
Vm
q2

)

aq
2r
eibr
1
cos
.
2
a
4
a2
(1.b.22)
(1.b.23)
o o
2
1
0.4
0
-1
0.2
-2
0
-0.4
-0.2
-0.2
0
0.2
-0.4
0.4
0
4
Figure 2. The first Brillouin zone dispersion for a tight-binding model on a twodimensional square lattice.
Problem 2: Tight-Binding Model on a Square Lattice
Consider a tight-binding model on a square, two-dimensional square lattice (lattice spacing a) with
on-site energy 0 and nearest-neighbour hopping matrix element t:
h
io
Xn
H=
0 |rihr| + t |rihr + a
x| + |rihr a
x| + |rihr + a
y| + |rihr a
y| .
r
a) We are to obtain the dispersion relation for this model.
Just for the sake of clearing up notation, our Bravais lattice here will be generated by
~a1 = a(1, 0) and ~a2 = a(0, 1) which has the associated reciprocal lattice generated by
~b1 = 2 (1, 0) and ~b2 = 2 (0, 1). We will write all momenta in terms of the reciprocal
a
a
lattice, so ~q = q1~b1 + q2~b2 . Using Blochs theorem it is quite easy to see that the
Hamiltonian of this system is given by

H = 0 + t ei~q~a1 + ei~q~a1 + ei~q~a2 + ei~q~a2 ,
(2.a.1)
0

i2q1
i2q1
i2q2
i2q2
= +t e
+e
+e
+e
,
(2.a.2)
0

= + 2t (cos (2q1 ) + cos (2q2 )) ;
(2.a.3)
(~q) = 0 + 2t {cos(2q1 ) + cos(2q2 )} .
(2.a.4)
This dispersion relation is shown in the first Brillouin zone in Figure 2.

b-d) Let us sketch the Fermi surface in the first Brillouin zone when the band is less than and more
than half-full, assuming a particle-like band (t < 0). And we are to make an accurate drawing of the
Fermi surface for the case of a precisely half-filled band.
When the Fermi surface is very near the bottom of the band energy, then it is approximately a circle: for qi 1, we can expand the cos(2qi )s to see that (q)
0 + 2t 2t~q 2 + O(~q 3 ), the solution to which is precisely a circle.
As the energy increases, the Fermi surface flattens out along the diagonal directions,
becoming a square when the band is half-filled. When the band is more than halffilled, the square breaks into four disjoint components which encircle the corners of
the Brillouin zone. Expanding cos (2qi ) about qi 21 shows that when the band is
nearly filled, the Fermi surface components do in fact become circles.
These are shown in detail in Figure 3.
#
'
#
'

PHYSICS : #CONDENSED
MATTER
HOMEWORK

'
0.8
0.6
0.4
0.2
0
0
0.2
0.4
0.6
0.8
Figure 3. Several Fermi surfaces observerd for a tight-binding square lattice model.
Dark colouring indicates lower energyFermi surfaces are included for the band both
more than and less than half filled. The half-filled Fermi surface is the clearly visible
s square
s in the
s
s

plot.
# +
#
'
#
'/
+
#
+
# +
#
'
#
'/

s
s
0.4
0.4
0.4
0.2
0.2
0.2
-0.2
-0.2
-0.2
-0.4
-0.4
-0.4
-0.4
-0.2
0.2
0.4
-0.4
-0.2
0.2
0.4
s
-0.4
-0.2
s
0.2
0.4
Figure 4. Fermi surfaces in the tight-binding square-lattice model with t > 0 next-tonearest-neighbour couplings for various values of t /|t|. From left to right: t /|t| = 1/10,
t /|t| = 1/2, and t /|t| = 7/10. Notice the sharp transition at t /|t| = 1/2.
e-f ) We are to add a matrix element t for hopping between next-to-nearest-neighbour sites and
sketch how the Fermi surface of the half-filled band changes for t > 0 and t < 0.
It is simple enough to write down the new dispersion relation coming from the the
Hamiltonian similar to part (a) above. Following that analysis, we find
n

o
H = 0 + 2t cos (2q1 ) + cos (2q2 ) + t ei~q(~a1 +~a2 ) + ei~q(~a1 ~a2 ) + ei~q(~a2 ~a1 ) + ei~q(~a1 +~a2 ) ,
n

o
= 0 + 2t cos (2q1 ) + cos (2q2 ) + t ei2(q1 +q2 ) + ei2(q1 +q2 ) + ei2(q1 q2 ) + ei2(q1 q2 ) ,
n

o
= 0 + 2t cos (2q1 ) + cos (2q2 ) + 2t cos (2(q1 + q2 )) + cos (2(q1 q2 ))
;

(~q) = 0 + 2t cos (2q1 ) + cos (2q2 ) + 2t cos (2(q1 + q2 )) + cos (2(q1 q2 )) .
This modification can have a rather drastic effect on the Fermi surfaceespecially if t /|t|
can be as large as around 12 . Using equation (2.f.5) we have little difficulty plotting
Fermi surfaces for various values of t /t. In Figure 4 we show three qualitatively
different Fermi surfaces for t > 0 for different values of t /|t| and in Figure 5 we
show these for t < 0.
(2.f.5)
# +
#
'
#
'/
+
JACOB
LEWIS
BOURJAILY
s
s
0.4
0.4
0.4
0.2
0.2
0.2
-0.2
-0.2
-0.2
-0.4
-0.4
-0.4
-0.4
-0.2
0.2
0.4
-0.4
-0.2
0.2
0.4
-0.4
# +
#
s
'
-0.2
#
'/
s
0.2
0.4
Figure 5. Fermi surfaces in the tight-binding square-lattice model with t < 0 nextto-nearest-neighbour couplings for various values of t /|t|. From left to right: t /|t| =
1/10, t /|t| = 1/2, and t /|t| = 9/10. Notice the sharp transition at t /|t| = 1/2.
Problem 3: Band Structure of Graphene and Nanotubes
Recall the honeycomb lattice used to describe graphene in homework 2. We are to consider a tightbinding model with a single level per site on a two-dimensional honeycomb lattice with only nearestneighbour hopping with on-site energy 0 and nearest neighbour hopping matrix element t.
a-b) We are to find the energy bands of this model and determine at what momenta the two bands
are degenerate.
Just to get our bearings, let us recall the Bravais and reciprocal lattices of the honeycomb
lattice:
!
3
1
~ = h~a1 , ~a2 i with ~a1 = a 3 (1, 0) and ~a2 = a 3
R
,
,
(3.a.1)
2 2
!
4
3 1
4
~
~
~
~
Q = hb1 , b2 i with b1 =
,
and ~b2 =
(0, 1) .
(3.a.2)
3a
2
2
3a
In this model, the wave function on each Bravais cell contains two linearly independent
parts, coming from the two atoms in each cell; lets call them atoms A and B. The
Hamiltonian of the system can be described by a 2 2 matrix, the diagonal parts
coming from the on-site energy 0 and the off-diagonal parts describing the hopping
matrix elements. The two off-diagonal entries are Hermitian conjugates of each other:
one describe hopping from A B and the other describes hopping from B A.
Because the two processes are conjugate, it is sufficient to describe one.
Let ~q = q1~b1 + q2~b2 where q1 and q2 are not required to be integers. Although it will be
very quickly brushed away, let us say that the vector ~vAB connects the atom at site
A to that at site B. The hopping, or off-diagonal, part of the Hamiltonian is given
n
o
by4
HAB = tei~q~vAB 1 + ei~q~a2 + ei~q(~a2 ~a1 ) ,
(3.a.3)
n
o
t 1 + ei2q2 + ei2(q2 q1 ) ,
(3.a.4)
n
o
= t 1 + 2ei(2q2 q1 ) cos (q1 ) .
(3.a.5)
Let us briefly observe that if HAB were represented as rei , then the solution to the
eigenvalue equation is
0

rei

=
= 0 r.
(3.a.6)
rei 0 = 0
Using this and the work above, we can directly write down the dispersion relation:
p
(~q) = 0 t 1 + 4 cos ( (2q2 q1 )) cos (q1 ) + 4 cos2 (q1 ).
(3.a.7)
o o
4The proportionality is used to ignore a phase factor, which will not affect our analysis of energy eigenvalues.
M
N
r
r
r
r
s
'
M
N
r
r
r
r
s
-0.5
-0.25
0
0.25
0.5
3
0
2
-1
0.4
0.4
1
-2
0.2
-2
-3
0
-0.5
-0.2
0
0.2
-0.5
-0.25
0.5
-0.25
0.25
0
0
0.25
-0.4
0.5
0
5
0.25
-0.25
-0.2
-0.5
0.5
0
5
Figure 6. Energy bands calculated for graphene: the bands are shown separately on
the right and left, and shown together in the middle. Notice there are precisely six
degeneracies located in the firs Brillouin zone.
We see that degeneracy implies that the above discriminant vanishes. This will be the
case if
1
ei(2q2 q1 ) cos (q1 ) = .
(3.a.8)
2
At first glance, there are two possibilities we may try: first, we know that ei(2q2 q1 )
R, so it must be 1. If ei(2q2 q1 ) = 1 then 2q2 q1 = 2n for n Z; the equation above
requires cos(q1 ) = 21 , which means that q1 = 23 or 43 . This gives us an infinite class
of degenerate solutions, and by adding and subtracting reciprocal lattice vectors, we
find six, essentially equivalent degeneracies at the corners of the first Brillouin zone:

1 1
1 1
1 2
2 1
2 1
1 2
,
,
,
,
,
,
,
(3.a.9)
3 3
3 3
3 3
3 3
3 3
3 3
where the components refer to the values of q1 , q2 in ~q = q1~b1 + q2~b2 . Now, because adding and subtracting lattice vectors brought us into the other condition for
degeneracywith 2q2 q1 an odd integerwe know that all of the degeneracies have
been accounted for.
The energy bands are plotted in Figure 6, where the six degenerate points are clearly
visible.
c-d) Describe and sketch the topology for various Fermi surfaces that can occur as the filling of
bands is varied. We should also describe the Fermi surface when the lower band is completely filled.
For very low F , the fermi surface is a circle inscribed within the bowl seen in Figure 6.
As the energy increases, the Fermi surface appears more and more hexagonal until
finally it breaks into six arcsone about each of the corners of the first Brillouin
zone. These six regions shrink as F 0 , when they vanish. This is shown in
Figure 7.
As F grows above 0 , the Fermi surface lies on the upper band and progresses in reverse
of the lower-band: for low energies above 0 , the Fermi surface is composed of six
distinct circular components which grow until they become nearly hexagonal; at
high energies, the Fermi surface again approaches a single circular section. This
progression is also shown in Figure 7.
In the case when the lower band is completely full, the Fermi surface is the union of
the six distinct Dirac points (of course, only two of them are inequivalent).
10
M
N

0.4
0.4
0.2
0.2
-0.2
-0.2
-0.4
-0.4
-0.6
-0.4
-0.2
0.2
0.4
0.6
-0.6
-0.4
r
-0.2
]
^
r
0.2
0.4
0.6
Figure 7. Contours indicating the Fermi surfaces for various F in the lower (left) and
upper (right) bands for the graphene tight-binding model. In both plots the energy is
lower in the darker regionregions in white on the left plot match to the regions in
black on the right plot.
e) If we were to compactify a sheet of graphene in one direction, the result would be a carbon
nanotube. Depending on which Bravais lattice vector is taken as the compactifying direction, there may
or may not be a band gapif there is a gap, then the nanotube is an insulator; if the two bands are
degenerate, then the tube is metallic. We are to determine which types of carbon nanotubes will be
metallic and which would be insulating.
The compactification of graphene into a carbon nanotube can be described as taking
~ r where
a quotient of the Bravais lattice by one of the lattice vectors, written R/~
~r = r1~a1 + r2~a2 where in this case r1 , r2 Z. That is to say, travelling in the
direction ~r brings you around the nanotube and back to where you started: the
~ + ~r are not merely related to those at R,
~ but are actually
atoms at sites related by R
the same atoms. This means that there is no phase factor for travelling any multiple
of times along ~r.
Precisely, this requires that
ei~q~r = 1 ~q.
(3.e.10)
This places a strong constraint on the allowed ~qsindeed, it breaks our continuous
band of allowed values to a discrete set. Whether or not the nanotube will be a metal
or an insulator is completely determined by whether or not this discrete subset of
allowed momenta include the Dirac points explored above.
The condition ei~q~r = ei2(q1 r1 +q2 r2 ) = 1 is that
q1 r1 + q2 r2 Z.
(3.e.11)
Recall that the six Dirac points were located at ~q with components (with respect to
~b1 , ~b2 basis) given in equation (3.a.9). There are two types of points to check: the
first two Dirac points listed in equation (3.a.9) will be present in the nanotube iff
r1
r2
Z
=
(r1 r2 ) 3Z;
(3.e.12)
3
3
the second type of point will be present in the nanotube iff
r1
2r2
+
Z
=
(r1 + 2r2 ) 3Z.
(3.e.13)
3
3
It is not hard to show that these two conditions are in fact equivalent5. Therefore, a
carbon nanotube will be a metal in this model only if
r1 r2 = 3
for some Z.
(3.e.14)
o
5Say r r = 3, then r + 2r = 3 + r + 2r = 3( + r ) 3Z; and conversely, say r + 2r = 3m, then

1
2
1
2
2
2
2
1
2
r1 r2 = 3m 2r2 r2 = 3(m r2 ) 3Z. QED
11
Problem 4: Thermodynamics Near a Dirac Point

Let us return our attention to the tight-binding model of graphene from problem 3. We may for the
sake of convenience take 0 = 0.
a) The points where two bands become degenerate are called Dirac points. We are to determine the
low-temperature behaviour of the specific heat and magnetic spin susceptibility near the Dirac point for
graphene.
The first step in our analysis will be to expand the dispersion relation (3.a.7) found
in problem 3 above near the Dirac points. Now, because the six Dirac points are
obviously translationally related, it is sufficient to consider just one for the moment.
Let us expand equation (3.a.7) about the Dirac point (1/3, 1/3)6:

1/2
3q
q
= t 1 + 4 cos +
+
+
cos
+ 4 cos2
,
3
3
2
3 2
3 2
!
!2 1/2
r
r

2 2
2 2
2 2
1 1 3
q
1 1 3
q
9 q
+ ...
q
+ ... + 4
q
+ ...
,
= t 1 + 4 1 +
4
2 2 2
8
2 2 2
8
!
r

9 2 q 2
3
2 q 2
= t 1 + 2 1 +
+ ...
1
q
+ ... +
4
2
4
!2 1/2
3
q
q
+ ...
,
2
4
2
!
)1/2
r
r

2 q 2
2 q 2
3 2 q 2
9 2 q 2
3
3
= t 1 + 2 1 +
+ ...
1
q
+ ... + 1 2
q
+
+ ...
,
4
2
4
2
2
2
(
!
)1/2
r
r
3
2 q 2
9 2 q 2
3
2 q 2
3 2 q 2
= t 1 + 2 1 +
q +
+
+ ... + 1 2
q
+
+ ...
,
2
4
4
2
2
2

1/2
= t 6 2 q 2 + . . .
;
(
t 6q.
(4.a.1)
This allows us to compute the density of states about a single Dirac point is given
by
Z
d2 q (2)
g() = 2
(0 6q d),
2
(2)
d
q d
;
=
t 2 6
gone
Dirac point ()
6t2 3
gtot () =
t2 3
(4.a.2)
With the density of states, we may compute the total energy7,

Z
u = u0 +
d g()f (),
0
Z
1
2 d
= u0 + 2 3
,
/(k
t 0 e BT ) + 1
3(3) 3 3
= u0 + 2 3 kB
T + O(T 4 ),
2t
where (n) is the Riemann zeta function. Therefore, we see that
cv =
q
6We will expand in
~ = q.
so that |q|
2
7Using Mathematica for the integrals.
3
9(3)kB
T 2 + O(T 3 ).
2
3
2t
(4.a.3)
12
To find the magnetic susceptibility we will begin by referring to the textbook or class
notes wherein it is found that the total magnetization (in the Pauli model) is given
by
Z
M = 2 H d g ()f ().
Evaluating this integral directly, we see

M
kB T log(2)
=
= 2
+ O(T 2 ).
H
t2 3
(4.a.4)
(4.a.5)
b) Consider doping graphene so that F is just above the Dirac point, but by an amount much less
that of T ; we are to again describe the low-temperature approximations of the specific heat and magnetic
spin susceptibility.
I am pretty sure that the picture we are supposed to envision is that we are some F
separated from the Dirac point, yet close enough to it that g() can be still viewed
as a linear function of . If this is the appropriate, then we can take
F
1
,
g() = 2 3
and
f () = ( )/(k T )
B
f
t
e
+1
and integrate above the Fermi surface8. Using a computer algebra package, we find
Z
1
( F ) d
u = u0 + 2 3
,
t F e(F )/(kB T ) + 1
2
3
F kB
3(3)kB
2
T
+
T 3 + O(T 4 ).
12t2
2t2 3
(It is comforting that this reproduces our earlier result for vanishing F .) This allows
us to directly conclude that
3
2
F kB
9(3)kB
T+
T 2 + O(T 3 ).
2
2
3
6t
2t
Now, to find the magnetic susceptibility, we perform the same steps as before and see
Z
1
F
M = 2 H d g()f () = 2 H 2 3 kB T log(2) +
,
t
2
and so
1
eF
= 2 2 3 kB T log(2) +
.
t
2
cv =
8There are alternative ways of looking at this.
(4.b.1)
(4.b.2)
(4.b.3)

Homework
Due Tuesday, th October
Problem 1: Electron in a Two-Dimensional, Weak Sinusoidal Potential

Consider electrons moving in a two-dimensional, weak periodic potential given by

2x
2y
+ cos
,
V (x, y) = U cos
a
a
where U > 0.
a. We are asked to find the two lowest-eigenenergies and eigenstates to first order in U at the edge of
the Brillouin zone, but away from the corners.
To write the Schrodinger equation in momentum space we will need the Fourier modes
of the potential function; the non-vanishing modes are Ubx = Uby = U/2, where
bx = b
qx and by = b
qy where b = 2
a and a is the lattice spacing.
There are four edges of the Brillouin zone, and the essential result will be identical
for all four of them. Without loss of generality, let us consider the Schrodinger
equation for the wave function with momentum on the Bragg plane q = (/a, 0).
~2 2
Letting 0q = 2m
q , we see that 0q = 0qbx on this plane; because this is the only
degeneracybefore considering the effects of the weak potentialwe know that to
leading order in U the Schrodinger equation gives rise to the following system of
equations1

U
0q cq = cqbx ;
2

U
0qbx cqbx = cq .
(1.a.1)
2
This system is obviously solved by
o1/2
1n 0
2
1 0
(q) =
q + qbx
q 0qbx + U 2
.
(1.a.2)
2
2
o o
On the Bragg plane, the two lowest eigenenergies are therefore
~2 2
U
.
(1.a.3)
2
2m a
2
The wave functions are found by considering again equations (1.a.1). On the Bragg
plane, we see that these imply
=
cq = cqbx ,
(1.a.4)
where refers to equation (1.a.2). Up to normalization, this implies that the wave
functions are
x

+ cq eiqr 1 + eibx cos
;
(1.a.5)
a

x
cq eiqr 1 eibx sin
.
(1.a.6)
a
o o
1Here we are using notation which should be common by now: the wave function (r) =
q
1
bG cqG e
i(qG)r .
b. We are to determine the four lowest-energy single-electron eigenstates at the corner of the
Brillouin zone to first order in U and give the corresponding wave functions.
We see at least two equivalent ways of proceeding; because they are of sufficiently different philosophy, we will present both, if briefly. The first, and quickest solution makes
important use of our result in part (a) above. Each two adjacent Bragg planes have a
pair of parabolic energy bands, separated by a gap U . At the corners, pairs of bands
from adjacent planes would meet. From our analysis above and elsewhere, we know
that when this happens in a weak potential U , the apparently degenerate bands will
hybridize. That is, for example, the two = 0q U2 bands from adjacent edges will
hybridize, giving the two lowest states at 1 = 0q U and 2 = 0q . The two upper
bands will hybridize giving rise to 3 = 0q and 4 = 0q + U . Here, q = (/a, /a).
To be a bit more explicitallowing us to determine not just the energies at the corners,
but also nearbywe could have begun with the four nearly degenerate states at the
corners. The Schrodinger equation near the corners, to first order in U , gives rise to
the system of equations

U
U
0q cq =
cqbx + cqby ;
0qbx cqbx =
cq + cqbx by ;
2
2

U
U
0qby cqby =
cq + cqbx by ;
0qbx by cqbx by =
cqbx + cqby . (1.b.7)
2
2
2
1
2
3
4
~
This system is straight-forwardly inverted. In units where 2m
= 1, the eigenenergies
near to the corner are

1/2
q
1
= qx2 +qy2 2(qx +qy )+4 2 +
4U 2 +32 2 qx2 +qy2 2 (qx +qy )+2 2 +4 (U 2 +16 2 (qx )2 ) (U 2 +16 2 (qy )2 )
;
2

1/2
q
2
2
2
2
2
2
2
2 1
2
2
2
2
2
2
= qx +qy 2(qx +qy )+4
4U +32 qx +qy 2 (qx +qy )+2 +4 (U +16 (qx ) ) (U +16 (qy ) )
;
2
1/2

q
1
;
= qx2 +qy2 2(qx +qy )+4 2 +
4U 2 +32 2 qx2 +qy2 2 (qx +qy )+2 2 4 (U 2 +16 2 (qx )2 ) (U 2 +16 2 (qy )2 )
2

1/2
q
1
= qx2 +qy2 2(qx +qy )+4 2
4U 2 +32 2 qx2 +qy2 2 (qx +qy )+2 2 4 (U 2 +16 2 (qx )2 ) (U 2 +16 2 (qy )2 )
.
2
We of course did not need to do the above expansion to note that this analysis agrees
with our previous one for the four energies at the corner :
~2 2
~2 2
~2 2
U,
=
,
=
+ U.
(1.b.8)
2
3
4
m a2
m a2
m a2
Making use of the Schrodinger equation (1.b.7) at the corner, we see that for = 0q U
the solutions are
cq = cqbx by = cqbx and cqbx = cqby .
(1.b.9)
1 =
The two degenerate wave functions do not uniquely solve the Schrodinger equation
(1.b.7)as we should expect. For these two bands, the wave functions satisfy
cq = cqbx by
1
2
3
4
and
cqbx = cqby ,
(1.b.10)
and any wave function can be build out of the two linearly independent relative
pairings cq = cqby . In all, the wave functions (in order of increasing energy) at
the corner are2
n
o n

o
cq eiqr 1 eibx eiby eib(x+y) cos (x + y) cos (x y) ;
(1.b.11)
o n

o
n
cq eiqr 1 eibx + eiby eib(x+y) sin (x + y) + sin (x y) ;
(1.b.12)
n
o n

o
cq eiqr 1 + eibx eiby eib(x+y) sin (x + y) sin (x y) ;
(1.b.13)
n
o n

o
cq eiqr 1 + eibx eiby + eib(x+y) cos (x + y) + cos (x y) .
(1.b.14)
o o
2Just to reiterate: the solutions and could generically be any linear combination of what we have written.
2
3
10
0
10
5
2
-1
-2
-2
-2
0
0
0
-2
2
-3
2
-3
-2
-1
-2
Figure 1. Three different views of the Fermi surfaces for the (metallic) system described in Problem 1. The figures on the left and right-hand sides are show many
equal-energy contours up to the Fermi energy. The middle plot shows just the Fermi
surfaces. (Note that the corners of the figure are on the lower band, n = 1, and the
edges are on the second band.)
c. Assuming that there are two electrons per unit cell, we are to find the lowest value of U for
which the system is a band insulator, and below witch it is a metal. We should make a qualitatively
correct representation of the Fermi surface of the system when it is metallic. changes from being a
band insulator to a metal. For a value of U for which the system describes a metal, we are to make a
qualitatively correct sketch of the locations of the Fermi surfaces in the first Brillouin zone.
Unless we are mistaken, the only point of including the electrons spin is that we know
there are states sufficient to fill the entire lowest energy band: when the system is a
band insulator the entire first band will be filled; when the system is a metal some
of the electrons will pour into the second band.
The system will be a metal only if the minimum energy of the second band is lower than
the highest energy of the first band. This is particularly easy because we know from
part (a) that the global minimum of the second energy is at the midpoint of one of
the first Bragg planes:
U
~2 2
min(2 ) =
+ .
(1.c.15)
2m a2
2
Similarly, we know that the global maximum of the lowest energy band occurs at the
corner of the Brillouin zone (because it increases away from the centre of the Bragg
plane), giving
~2 2
max(1 ) =
U.
(1.c.16)
m a
Therefore, the system will be a band insulator if
min(2 ) max(1 ) > 0
U>
~2 2
.
3ma2
(1.c.17)
The Fermi surfaces for the system when it is a metal are shown in Figure 1.
d. We are to make a qualitatively correct representation of how the zero-temperature spin contribution to the systems magnetic susceptibility varies with U as U passes through the transition
between insulator and conductor. Near-to, but on the insulating side of the transition, we are to describe the lowest-energy electronic excitations above the ground state at and give possible total crystal
momentum for these excitations.
The last part can be done presently. Near the transition, the lowest energy excitations
will bring electrons at the corners of the Brillouin zone to the centres of the Bragg
planes in the second
energy
These
are excitations
with momentum transfer

band.

of q = 2a
, 0 , 0, 2a
, a , 2a
, or 2a
, a . The crystal
momentum
will

then be on the center of the Bragg plane, so q = a , 0 , 0, a , a , 0 , or 0, a .
Problem 2: A Narrow-Gap Semiconductor

Consider two bands of states which overlap but interact only weakly. Let one of them be particle-like
and the other be hole-like. We may assume that they become degenerate and/or hybridize near zero
momentum. The Hamiltonian conserves crystal momentum and has the following matrix elements
~2 k2
~2 k2
k2 ,
h2, k|H|2, ki =
k 2 ,
h1, k|H|2, ki = , (2.a.1)
2me
2mh
me
where we have introduced the constant m
and have chosen appropriate units for the crystal
h
momentum k. In (2.a.1), , me , and mh are all positive and is real.
h1, k|H|1, ki =
a. We are to calculate and describe the dispersions of the resulting band structure after hybridization
for each qualitatively different case which can arise.
We have done these problems enough to know that
(
)1/2

2

1

2

+
,
=0
=
= ( + )

2
2
where in our case = k 2 and = k 2 so that the dispersion is
(
)1/2
2

1 2
k 2 (1 + )
2
=
k (1 ) +
+
.
2
2
(2.a.2)
(2.a.3)
Notice that equation (2.a.3) implies that as long as 6= 0, no matter how small, the
two bands will not intersect.
It is at least intuitively obvious to the author that any generic set of parameters will give
rise to situations where both bands have three stationary points. We can test this
intuition and discover some interesting results by calculating exactly where these stationary points are for each band. Of course we can do these two cases simultaneously
as follows. will have an extremum if
(
)1/2

2
k 2 (1 + ) k(1 + )
k 2 (1 + )
= 0 = k(1 )
where =
+ 2
.
k
2
(2.a.4)
As long as 6= 0 and the other parameters are real, > 0which is all we need for
the moment. The first, obvious extremum is at k = 0, which is uninteresting for the
moment. The other stationary points are then seen to satisfy
(1 + )2
(1 + )
+ (1 ),
1
= k 2
+
;
1+
(1 + )2

1
1
= k 2 =
.
1+
+1
0 = k 2
Indeed, we see that + ( ) will generically have an absolute minimum (maximum)

at k 2 6= 0. However, there can be a conspiracy where the term in curly brackets on
the right vanishes, giving rise to either or + but not bothhaving a third-order
global maximum or minimum at k = 0, respectively.
Let us quickly find the cases when + or does not have three distinct extrema. For
+ , the upper band, , > 0 implies that p
1 < 0 or that me < mh for (2.a.7) to
vanish. Bearing in mind that (k 2 = 0) = 2 /4 + 2 , we have
(1 + ) = (1 ),
= 2 (1 + )2 = (1 )2
+ 2 ;
4

2
(1 )2
2
= 2 =
(1
+
)
.
(1 )2
4
(2.a.5)
(2.a.6)
(2.a.7)
(2.a.8)
(2.a.9)
(2.a.10)

0.25
0.2
0.75
0.8
0.15
0.5
0.1
0.25
0.6
0.05
0.4
0
-0.25
0.2
-0.05
-0.5
0
-1.5
-0.1
-1
-0.5
0.5
1.5
-1
-0.5
0.5
1.2
0.6
0.4
0.8
0.2
0.6
0.4
-0.2
0.2
-0.4
0
-1.5
-1
-0.5
0.5
-0.6
-1
-0.5
0.5
1.5
-0.6
-0.4
-0.2
0.2
0.4
0.6
Figure 2. The qualitatively different possible hybridizations between particle- and

hole-like bands. The possibilities are shown in the order discussed.
Similar algebra in the case of the band shows that we can have a fourth-order
global maximum only when 1 < 0, i.e. that mh < me . Then we have
( + 1) = ( 1),
= 2 ( + 1)2 = ( 1)2
+ 2 ,
4

2
(1 )2
2
2
= =
(1 + )
,
(1 )2
4
exactly as before (only this time we have > 1). Indeed, both cases are rather
contrived.
We have therefore classified the general structure of all possible hybridizations, as illustrated in Figure 2.
(1) Generic situation: when there is no conspiracy in any of the parameters, then
both bands will feature three extremaand global minima of + will not lie
over the global maxima of .
(2) When the upper band, + , has a fourth-order global minimum. Recall that this
only occurs if me < mh and equation (2.a.10) is satisfied. Notice that this gives
rise to two particle-like bands.
(3) When the lower band, , has a fourth-order global maximum. Recall that this
only occurs if mh < me and equation (2.a.13) is satisfied. Notice that this gives
rise to two hole-like bands.
(4) No interaction term: = 0. Here the bands only hybridize in the sense that we
are perfectly free to chose our eigenenergies to be the upper and lower bands.
(5) When = 1 and = 0, the two overlapping bands will separate similar to as
along a Bragg plane. Here both bands stay quadratic.
(2.a.11)
(2.a.12)
(2.a.13)
b. We are to obtain the density of states for = 0 and determine the energies and powers of the
Van Hove singularities.
When = 0, the band energies are simply given by their free-values. Because of this, it
is not difficult to explicitly determine the density of states. Indeed, we find for 1 3
that
k()me
2
k()2
g1 () =
4
,
(2.b.14)
2 k() =
3
~
(2)
2 ~2
me
and because
k() =
2me
,
~
this implies
g1 () =
(2me )3/2
.
2 2 ~3
(2.b.15)
Similarly, we see that

g2 () =
2
k()2
k()mh
4 ~2 k() =
,
3
(2)
2 ~2
(2.b.16)
mh
and because in this case

we see
p
2mh ( )
k() =
,
~
(2mh )3/2
.
2 2 ~3
The Van Hove singularities are located at = 0, = which are both .

g2 () =
(2.b.17)
c. We are to sketch the density of states for each of the qualitative cases studied in problem (a). For
each one, we should comment on the Van Hove singularities and their strengths.
The density of states for each of the five cases discussed in problem (a) are shown in
Figure 3. The Van Hove singularities are obvious by inspection for most of the
plotsapparently the only exception is the left leg of the third plot, which should
show a divergence (the right hand leg does not diverge).
Although it would have been preferable to have labeled the plots explicitly indicating the
singularities and their strengths, we will need to make due with a mere discussion.
Importantly, there are at most four Van Hove singularities which correspond the at
most four crystal momenta giving extrema of . Because there is always a local
extrema at zero crystal momentum, two potential Van Hove singularities are at
+ (0)
and
(0).
(2.c.18)
The other two possible places where Van Hove singularities can arise are the global
extrema of . In problem (a) we calculated the values of k for which would have
an extrema. A bit of algebra allows one to see that the other two possible Van Hove
singularity locations are
1/2
1/2
v
v
u
1 u
u
2
1 u
2

t
t

+
and
.
+
( + 1)2 1 (1)22
( + 1)2 1 (1)22
+ 1
+ 1
4(+1)
4(+1)
(2.c.19)
A quick glance at the plots in Figure 3 shows that the generic case has Van Hove singularities at all four of the possible locations. Plots two and three, when one of the
bands has a fourth-order extrema at the origin show only three of the four possible
singularities (see footnote earlier). The fourth case, where there is no coupling between the bands has only two of the singularities; the last plot, corresponding to the
3Here refers to the particle-like band.

1

1.75
1.2
1.25
1
0.75
g+H/
0.8
g+H/
g+H/
1.4
1.2
1.5
0.6
0.8
0.6
0.5
0.4
0.4
0.25
0.2
0.2
1
0.5
0
H
0.5
0
1
0.5
0
H
0.5
1
0.5
0
H
0.5
0.7
0.6
0.6
0.4
g+H/
g+H/
0.5
0.3
0.5
0.4
0.2
0.1
0.3
0
1
0.5
0
H
0.5
1
0.5
0
H
0.5
Figure 3. The densities of states for the band structures illustrated in Figure 2.
case where the two bands split while maintaining quadratic dependence on k gives
rise to only the two singularities which are seen approaching the gap.
We can actually read off the strength of the singularities with a little background
information. Recall in part (a) above we showed that () was at most a cubic
k2
on general and dimensional grounds, we see
polynomial in k. Because g()
that the worst type of singularity could therefore be 1/k 1/ which is seen
4
as the vertical spikes in plots 1,2,3 and 4 in Figure 3. All
the other Van Hove
singularities come about the usual way, from g() k 2 /k .
d. We are to find the gap energy when me = mh , ( = 1).
We have done enough two-state degeneracy problems to know that the band gap energy
will be 2. To see this, consider zooming in near where the two bands cross; locally,
this is to order simply a two degenerate state problem that we are now so good at.
Simple diagonalizaton gives the result we are now able to guess without diagonalizing
anything.
If it were truly necessary to prove our intuition is correct, recall equation (2.c.19) which
tells us where the global minimum (maximum) of + ( ) is to be found. Using our
assumption that = 1, the expression greatly simplifies and we find
.
2
Putting this in the dispersion relation calculated in equation (2.a.3) immediately
shows that
gap = 2.
(2.d.20)
k2 =
e. In the above analyses, we considered spherically symmetric electron potentials. In any real crystal,
there would be dispersion modifications giving rise to e.g. angular dependencies on the order of k 4 . We
are to describe how including these effects could affect our analysis.
There are two cases when crystal structure could possibly lead to (even dramatic) alterations of our analysis above:
(1) If me , mh are too large: this would soften their intersection (making it more
sensitive to higher-order effects) and also raise the size of k for which become
4We mentioned earlier that although the spike is not seen on the printed plot, it is nevertheless present.
nearly degenerate. In contrast, if me , mh are very small compared to the reciprocal lattice, then their unperturbed intersection will be steepergo most
sensitive to leading effects in kand likely to be closer to small values of k,
where the effects of the zone edges are small.
(2) If is too large: if the two bands are widely separated in energy, then our analysis above would simply have assumed that the two bands hybridize farther out
in k-space. This is only reasonable if the region where the two bands hybridize
is well within the Brillouin zone.
Baring these two caveats, many of our results above will carry over to this case with only
small refinement. Specifically, the general classification of the types of hybridized
bands will still be valid. If one or both of the situations described above apply,
however, there will be much more variety in the types of hybridized band structures
that can arise.
One possibly important change to our analysis would involve our discussion of the density
of levels: k 4 terms will almost certainly add to the number of Van Hove singularities,
for example. When the periodic potential is weak5, for example, (k) vanishes
along Bragg planeswhich will certainly give rise to new Van Hove singularities in
the density of levels (although perhaps not in the energy ranges we are interested in.
Problem 3: Shubnikov-de Haas-van Alphen Oscillations

The low-temperature specific heat of a piece of metal at high magnetic fields is found to be periodic in
inverse field, with period 2 109 Gauss1 . We are asked to interpret this observation using a one-band
free-electron model for the conduction electrons and thereby determine the number density of electrons
and their energy in eV. At low-fields, the oscillations are not visible because of scattering effectsthe
resistivity at low temperature is measure to be 0.1 cm. We are to use this to estimate the mean
free path of conduction electrons in this materials within the Sommerfeld model of conduction and use
this to estimate the minimum field strength necessary to observe oscillations of the specific heat.
Using Onsagers result, we know that

1
2e 1
=
,
H
~ Ae
where in our case (H 1 ) = 2 105 T1 . Ae is the cross-sectional area of the

Fermi surface in a plane normal to the magnetic field axis. If we take the free electron
approximation, then all extremal cross sections through the Fermi surface have area
kF2 a fact which is readily visualized. Inverting this and using real numbers, we
see that the oscillation period observed implies a Fermi momentum of
kF = 1.23 1010 m1 .
(3.b.1)
Using all our favourite results for free electrons, we see that this implies that the density
of conduction electrons is given by
n=
kF3
= 6.33 1028 m3 .
3 2
(3.b.2)
And the Fermi energy is then

~2 kF2
= 5.83 eV.
(3.b.3)
2me
Using some book work in Ashcroft and Mermin, we know how the resistivity is related
to the mean free path and the scattering time. Specifically, we have the mean free
path is given by
50.1 eV
=
92 108 cm
(3.b.4)
F
F =
5Ashcroft and Mermin point out that this is also often true when the potential is not weak: because Bragg planes
are regions of high-symmetry it should not come as a surprise that energy bands are often forced to meet Bragg planes
symmetricallyi.e. smooth crossing in the extended zone scheme.
where is the resistivity in cm. We have

= 8.00 107 m
(3.b.5)
Even at zero temperature, the oscillations will be broadened because of scattering, effectively washing out the signal. To see the effectas discussed in classit the field
H must be such that
eH
~kF
1
=
H
10 Tesla.
(3.b.6)
mc
e
Problem 4: Tight-Binding Resonant Conduction

Consider the one-dimensional tight-binding model with Hamiltonian

X
H = 0 |0ih0| +
tn |nihn + 1| + |n + 1ihn| ,
(4.a.1)
where t1 = t , t0 = t+ and all other tn = t. There is an impurity at site n = 0. We are to obtain

the probability of transmission across the impurity for an incident electron with zero energy. We are to
examine the case when 0 , t , and t+ are all much less than t, showing its resonant form.
In the zero-temperature limit with the Fermi level at zero energy, we are to obtain the current resulting
from a linear potential difference across the impurity. We should estimate the size of the voltage for
which this limit is valid.
Consider the test wave function

X

|i = |0i +
Aeiqm + Reiqm |mi + Seiqm | mi .
(4.a.2)
m>0
Although the notation is a bit compact, this is nothing but an incoming wave with
momentum q 6 toward the origin. The wave function is of course not normalized.
The part of |i proportional to S is nothing more than the transmitted wave, and
the part proportional to R is the reflected wave.
We are seeking a zero-energy eigenstate |i. Acting with the Hamiltonian on our test
function, we see

H|i = 0 + t+ Aeiq + Reiq + t Seiq |0i + t+ + t Aei2q + Rei2q |1i + t + tSei2q | 1i

o
X n
+
t Aeiq(m+1) + Reiq(m+1) + Aeiq(m1) + Reiq(m1) |mi + S eiq(m+1) + eiq(m1) | mi .
m2
(4.a.3)
If |i is an eigenvector of H with eigenvalue 0, then the coefficient of every basis ket

|i must vanish in the expression above. This gives us a large system of constraints.
The constraint coming from the coefficient of | mi in equation (4.a.3)the pieces
proportional to S shows that

eiq(m+1) + eiq(m1) = 0 m
=
eiq + eiq cos(q) = 0.
(4.a.4)
Therefore we see that q = 2 7. To match our sign conventions, this implies that q =
2 . We will save some time analyzing the other constraint equations by automatically
inserting eiq = ei/2 = ei/2 = i as it is encountered. The other constraint
equations then are then
t+ t(A + R) =0,
(4.a.5)
0 + it+ (R A) + it S =0.
(4.a.7)
t tS =0,
(4.a.6)
6The signs are consistent if annoyingthe author did not find time to clean up less-than pedagogical trivialities.
7We can take q to be in the first Brillouin zone without loss of generality.
10
These three equations are linearly independent and allow us to completely solve for
A, R and S in terms of t , t and 0 . Indeed, we have
t2 + t2 it0
t2 t2 + it0
t
and
A= +
and
R= +
.
t
tt+
tt+
This allows us to compute the transmission coefficient8
S=
T =
4t2 t2+
|S|2
n
o.
=
2
|A|2
t2+ + t2 + 20 t2
(4.a.8)
(4.a.9)
In the limit where t and 0 are much less than t, then we see that
n
o

1
2
2 2
2 2
2 2
o
1T = n
t
+
t
+
4t
t
2
+
0
+ ,
t2+ + t2 + 20 t2
o
n
2
t2+ t2 + 20 t2
o,
=n
2
t2+ + t2 + 20 t2
2 2 2

(t+ t )
2
2
t
+ 0
t2
,
= 2 2 2
(t+ +t )
2
t2
+
0
t2
!
!

2
4
2
t2+ t2
t2+ + t2
t2+ + t2
= 1+
1
+
+ ... ,
20 t2
20 t2
40 t4
2
4t2+ t2 t2+ + t2
4t2+ t2
=1 2 2 +
+ ...,
0 t
40 t4
(
)
2
2 2
4
t+ + t2
t+ + t2
4t2+ t2
=1 2 2
1
+
... .
0 t
0 t
0 t
Therefore we see that as t , 0 are taken to be small, T 0.
We are using sloppy notation: if we want to allow t , etc. to be complex, then the terms in the expression for T must
be interpreted as their modulus.

Homework
Due Tuesday, th November
Problem 1: Phonon Spectrum of a Diatomic One-Dimensional Crystal

Consider a one-dimensional, diatomic crystal composed of atoms of mass M1 and M2 , respectively.
We may suppose that the interaction between nearest neighbours is a simple harmonic spring with a
universal spring constant K.
a. We are to determine the full phonon spectrum of this system and sketch the dispersions (q).
Let a denote the lattice spacing of the Bravais lattice, and let us label the displacement
functions u1 and u2 for the atoms with mass M1 and M2 , respectively. Then the
harmonic contribution to the potential is given by
2
2
KX
Uh =
.
(1.a.1)
u1 (na) u2 (na) + u2 (na) u1 ((n + 1)a)
2 n
This immediately implies the following equations of motion:
n
o
Uh
M1 u
1 (na) =
= K 2u1 (na) u2 (na) u2 ((n 1)a) ;
(1.a.2)
u1 (na)
n
o
Uh
M2 u
2 (na) =
= K 2u2 (na) u1 (na) u1 ((n + 1)a) .
(1.a.3)
u2 (na)
We seek phonon solutions to these equations of motion, which have the structure of
plane waves:
u1 (na) = ei(knat)
and
u2 (na) = ei(knat) .
(1.a.4)
Inserting these test functions into the equations of motion and simplifying a bit, we
find the following (independent of n),
M1 2 = 2K K 1 + eika ;
(1.a.5)
2
ika
M2 = 2K K 1 + e
.
(1.a.6)
This is of course equivalent to the eigenvalue equation
M1 2 2K K 1 + eika
= 0,
K 1 + eika
M2 2 2K
(1.a.7)
which only has a solution if the determinant of the operator vanishes. Writing out
the determinant and solving the quadratic equation for 2 , we find that this implies
1/2
(M1 + M2 )2
(M1 + M2 ) 1
K2
4K 2
(1
cos(ka))
,
2 = K
8
M1 M2
2
M12 M22
M1 M2
1/2
M1 + M2
M1 + M2
M1 M2
=K
K
12
(1 cos(ka))
;
M1 M2
M1 M2
(M1 + M2 )2
s
(
)
ka
K
2
2
=
1 14
sin
,
(1.a.8)
(M1 + M2 )
2
M2
where we have introduced the reduced mass: MM11+M
.
2
This phonon dispersion relation is plotted in Figure 1.

0.35
0.3
0.25
Z+k/
0.2
0.15
0.1
0.05
0
Ss6
Ss3
Ss2
k
2Ss3
5Ss6
Figure 1. The phonon dispersion for a diatomic, one-dimensional crystal. The lower
band is represents acoustic mode and the upper band represents optical mode.
b. We are to describe the atomic motion associated with optical and acoustic phonons near the
centre and edge of the first Brillouin zone.
Near the centre of the Brillouin zone k /a, so we may expand sin(ka/2) (ka/2) +
O((ka/2)3 ). Also Taylor expanding the square-root in equation (1.a.8), we obtain
2
2
=
1 1
(ka) . . .
;
(1.b.9)
2(M1 + M2 )
which implies
q
K
2
2(M1 +M2 ) (ka) + O(ka)

q
=
2K(M1 +M2 )
+ O(ka)2
M1 M2
acoustic
optical
(1.b.10)
If we plug this back into the eigenvalue equation (1.a.7), we find for the acoustic mode,
K(2 + ika + O(ka)2 )

K M2 (ka)2
4
2K
=
+ O(ka) ,
2
2 M1 + M2
2K 2 2 (ka)
+ O(ka)4
2
= 1 + O(ak).
Therefore, the acoustic mode is that for which the two types of atoms are oscillating
in phase:
Similarly, looking at the expansion for the optical phonons, we find
K(2 ika + O(ka)2 )

M1 + M2
4
2K 2K
=
+ O(ka) ,
2
M2
2K 2 2 (ka)
+ O(ka)4
2
=
M1
+ O(ak).
M2
This implies that the optical phonons near low crystal momentum (modulo the reciprocal lattice) are excitations where the two types of atoms oscillate in opposite
phase:
Let us now return to equation (1.a.8), only this time keeping track of M1 and M2 . Near
the edge of the Brillouin zone, k = /a , we may expand sin2 (/2 a/2) =
1 (a)2 /4 + O(a)4 . Using this, we have

s
(
)
K(M1 + M2 )
M1 M2
a
2
2
=
1 14
sin
,
M1 M2
(M1 + M2 )2
2
2
s
(
)
K(M1 + M2 )
M1 M2
M1 M2
2
4
=
1 14
+
(a) + O(a) ,
M1 M2
(M1 + M2 )2
(M1 + M2 )2
s
(
)
K(M1 + M2 )
(M1 M2 )2
M1 M2
2
4
1
=
+
(a) + O(a) ,
M1 M2
(M1 + M2 )2
(M1 + M2 )2
s
(
)
K(M1 + M2 )
M1 M2
|M1 M2 |
=
1
1+
(a)2 + O(a)4 .
M1 M2
M1 + M2
(M1 M2 )2
Now, without loss of generality we may suppose that M1 > M2 , in which case this
reduces to
q
K
+ O(ka)2 acoustic
q M1
=
.
(1.b.11)
K
+ O(ka)2
optical
M2
Notice that in this case, the matrix in equation (1.a.7) becomes diagonal, so there is no
constraint on /. Rather, modes with momenta near the edge of the first Brillouin
zone correspond to bulk-modes of lattice of M1 atoms and the lattice of M2 atoms
oscillating independently of each-other.
c. We are to consider the concrete example of a one-dimensional NaCl lattice, for which it is
observed that the highest energy optical phonon is 30 meV. We are to determine the spring constant K
in reasonable atomic-physics units, and determine the minimum energy that an incoming neutron must
posses to excite all phonons at all crystal momenta.
From our work above we know that the highest energy phonon in the spectrum occurs at
zero-crystal momentum, in the optical band1. This was derived explicitly in equation
(1.b.10). Therefore, we know
s that
2
Eph
M1 M2
2K(M1 + M2 )
Ephmax = ~max = ~
=
K = 2 max
.
(1.c.12)
M1 M2
~ 2(M1 + M2 )
Using the fact that max ~ = 30 meV, and plugging in real numbers (with units!), we
see that this gives
2
K = 1.5 eV A .
(1.c.13)
When considering a neutron inelastic scattering process, there are two constraints that
must be satisfied: first, the difference between the initial and final neutron momenta
must be a reciprocal lattice vector; and second, the phonon energy must equal the
difference between the initial and final neutron energies. Specifically, these are2
pi pf = n~
2
a
and
Eph =
(p2i p2f )
.
2mn
(1.c.14)
We can combine these two equations by noting (p2i p2f ) = (pi + pf )(pi pf ) =
(pi + pf )n~ 2
a to arrive at the suggestive pair of equations
pi pf = n
h
a
and
pi + pf =
2Eph mn a
;
nh
(1.c.15)
Eph mn a
h
+ n
where =
and =
.
(1.c.16)
n
h
2a
We must find the n for which pi is minimized for the maximum phonon energy of 30 meV.
Numerically, is about an order of magnitude larger than , so this happens when
n is small; in fact, it is minimized at n = 3, which gives Eneutron = 30.51 meV.
pi =
1It may not be obvious that the global maximum occurs at zero crystal momentum in the optical bandfor the
one-dimensional crystal under considerationbut it turns out to be so.
2The lattice spacing for NaCl crystals is known to be a = 5.64 A

.
Problem 2: Bound States of Phonons Near an Impurity (One-Dimension)

Consider a one-dimensional crystal composed almost entirely of atoms with mass M , but with one
impurity atom of mass M 0 . We may approximate the interactions as nearest-neighbour harmonic potentials with a (universal) spring constant K. We are asked to explore the localized phonon modes that
can possibly arise in this situation.
a. Let us determine the range of M 0 for which localized phonon modes exist.
Our intuition from Quantum Mechanics strongly hints that localized modes exist only if
M 0 < M : this is because the potential energy should be lower there, giving rise to a
potential well in which phonons could be entrapped. Alternatively, when M 0 = M ,
we know there are no localized modes because the situation is identical to a onedimensional crystal; but when M 0 < M wave amplitudes are locally piqued at M 0
so we expect localized modes. Enough intuition, let us show that our intuition isnt
misguided.
The initial set-up is sufficiently similar to the previous problem (and sufficiently canonical) that we may appear brief. Given the Hookes law harmonic potential, we can
immediately write down the equations of motion for the system:
Mu
na = K 2una u(n+1)a u(n1)a
n 6= 0
and
M 0u
0 = K 2u0 ua ua . (2.a.1)
Because we are interested in phonon modes which are localized at n = 0, we will try
the following test functions
un>0 = ean ei(kant) u
n>0 eit
and
un<0 = ean ei(kanit) u

n<0 eit ,
where > 0; we will use our equations of motion (2.a.1) to determine u

0 .
Inserting our test functions into the equations of motion, we find that virtually all of the
system of equations collapses in redundancy, leaving us with
n2
M 2 = K 2 ea eika ea eika ;
n=1
M 2 = K 2 ea eika u
0 ea eika ;
n=0
M 0 2 = K 2
u0 ea eika ea eika ;
n = 1
M 2 = K 2 ea eika u
0 ea eika ;
n 2
M 2 = K 2 ea eika ea eika .
(2.a.2)
(2.a.3)
(2.a.4)
(2.a.5)
(2.a.6)
(2.a.7)
Subtracting equation (2.a.7) from equation (2.a.3) we see that
ea eika eika + ea eika eika = 4i sin(ka) sinh(a) = 0.

This implies that either = 0which would run contrary to our analysis: we are
interested in the case when > 0or that ka = m for some m Z. Actually,
we have no need to keep this level of generality: it is eika that appears throughout
our equations of motion, and this leaves only two possibilities: ei2 = ei2 = 1
or ei = ei = 1; labelling these possibilities as , respectively, equation (2.a.3)
becomes
M 2 = 2K 1 cosh(a) .
(2.a.8)
Now, cosh(a) > 1 if > 1. Because this would lead to negative 2 for the case
abovecorresponding to ka = 2mwe must exclude this as a possibility, leaving
only eika = eika = 1. Which allows us to conclude3
r

K
a
cosh
.
(2.a.9)
=2
M
2
Inserting our expression for 2 into equation (2.a.4) we see that
2(1 + cosh(a)) = 2 + ea + u
0 ea ;
= ea + ea = ea + u
0 ea ,
3We made use of the hyperbolic identity 1 + cosh() = 2 cosh2 (/2).
which of course implies that u

0 = 1. It is not hard to see that the only independent
equation left for us to consider is (2.a.5). Let us see what this equation implies about
M 0 /M .
M 0 2 = K 2 + ea + ea ,

M0
a
2
= 4K
= 2K 1 + ea ,
cosh
M
2
=
M0
1 + ea
,
=
M
2 cosh2 a
2
=e
a/2
2 cosh
2 cosh2
2
a
,
2
a/2
M
e
.
=
(2.a.10)
M
cosh a
2
Now, because the expression on the right hand side of (2.a.10) is strongly bounded
above by 1and is equal to one iff = 0we may conclude that localized phonon
modes exist only if
M0
< 1.
(2.a.11)
M
o o
b. We are to give explicit solutions for the frequency and displacement patterns of this localized mode
and describe what happens as M 0 M .
Perhaps the first thing we should do is revisit equation (2.a.10) and give as a function
0
of M
M . A little bit of manipulation shows that
=
2
,
ea + 1
(2.b.12)
which of course implies that

2
2M M 0
=
.
(2.b.13)
M0
This equation allows us to tidy up much of our previous work. For example, equation
(2.a.9) can be combined with (2.a.10) using our work above,
s
KM
=2
.
(2.b.14)
M 0 (2M M 0 )
ea =
We can put everything together and now give all the functions un at once:
|n|
M0
ei(kant) .
una =
2M M 0
Now, looking at these plane waves (2.b.15) and the dispersion relation (2.b.14), we see
that when M 0 M the suppression term in (2.b.15) becomes 1, and the dispersion
relation matches onto the solution for normal phonons in one-dimensioni.e. there
are no localized modes when M 0 = M .
(2.b.15)
Problem 3: Wigner Crystals in Various Dimensions

At low temperatures and densities, electrons in an ideal semiconductor crystallize into what is
known as a Wigner crystal. This can be modelled as a crystal of electrons in a uniform background
of stationary, neutralizing positive charges, where only the electrons are able to move. The long-range
Coulomb repulsion between the electrons can affect the long-wavelength longitudinal acoustic phonons.
We are to determine the parametric low-q behaviour of this mode (called a plasmon) in one-, two-, and
three-dimensional Wigner crystals.
In order to determine the dispersion of plasmons, we will use the harmonic approximation for the Coulomb interaction potential4; for dimensions higher than one, we will
have need to recall recall some elementary multi-dimensional calculusnamely, the
generalization of Taylors theorem to higher dimensions,
~ (~r) + 1 ~a
~ f (~r) + . . . ,
f (~r ~a) = f (~r) + ~a f
(3.a.1)
2
for any scalar function f (~r). Because the scalar function we are interested in is
radial, it will be helpful to recall the expression for the gradient in polar and spherical
coordinates in two- and three-dimensions respectively.
In one-dimension the situation is rather more elementarythere are at least no angles
to worry us. Let us suppose that at zero temperature, the electrons are spaced a
distance 1 apart (in appropriate units)5so that the electrons are located at r = n
for n Z. Then the equilibrium Coulomb-potential energy at the point r = 0 is
simply
N
X 1
X
1
eq (0) = e2
= 2e2
.
(3.a.2)
|r|
r
r>0
r6=0
It is this potential that we will Taylor-expand for the plasmon perturbations.

Now, let us label the displacement from equilibrium of electron n from the point r as
un . Then the actual potential energy at the site of the electron labeled 0which is
now located at u0 is given by
X 1
1
1
1
(u0 ) = e2
(ur u0 ) 2 + (ur u0 )2 3 . . . .
(3.a.3)
|r|
r
2
|r|
r6=0
Assuming that the values ur = 0 are equilibrium, then the first term in (3.a.3) is a
constant and the second term vanishes. It is the third termthe harmonic term
that we are interested in:
o
e2 X 1 n
2
2
h (u0 ) =
(u
u
)
+
(u
u
)
.
(3.a.4)
r
0
r
0
2 r=1 r3
The equations of motion are relatively simple; for u0 , they give
o
X
h (u0 )
1n
m
u0 =
=e
2u0 (ur + ur ) .
3
u0
r
r=1
(3.a.5)
We are looking for plasmon solutions. These are

ur = ei(krt)
where q is the momentum. Putting this into the equations of motion, we have

X
X
kr
m 2
1
1
2
2
=
2
(1
cos(kr))
=
sin
.
2
3
3
e
r
r
2
r=1
r=1
(3.a.6)
4This is actually a bit more subtle than it may seem at first: the leading order Coulomb interaction potential is a
strong effect. By taking the harmonic approximation, we are implicitly assuming that the leading Coulomb term (which
is highly divergent in all dimensionsat least for infinite crystals) is completely neutralized by the positive charges; or,
put another way, we ignore the machinery by which the electron system finds itself in an equilibrium distribution. Then,
only the electron Coulomb interaction energy is considered as the sub-leading part of the potential. We will find that this
is fine in each dimension under considerationbecause it will dominate cq in the limit of q 0but this was by no
means obvious.
5Because we are only interested in the parametric q-dependence of , there is no reason to bother about units.
In the limit where k 0, this sum is well-approximated as an integral,

Z
sin2 kr
2
2
dr
.
3
r
1
(3.a.7)
Now, there are two obvious regions of interest: first, if r . 2/k, then sin2 (kr/2) can
be well-approximated by its leading Taylor expansion. When r 2/k, the integrand
is rapidly oscillating and can be well-approximated by the average value of sin2 (x)
which is 21 . That is
Z

Z 2/k
1
1
kr
kr
dr 3 sin2
dr 3 sin2
2
+
,
(3.a.8)
r
2
r
2
1
2/k
Z
Z 2/k
k2
1
+
dr
dr 3 ,
(3.a.9)
4r
2r
1
2/k

1
2
1 2
= k log
k2 .
(3.a.10)
4
k
16
Now, the leading term (divergently) dominates as k 0 (and the other terms will
not dominate other ck phonons anyway), so we see that in one-dimension,
Wigner-crystal plasmons have a low-momentum dispersion
p
k log(k).
(3.a.11)
k0
o o
To generalize the work above requires keeping track of coordinate systems, indices, &tc.
In two dimensions, we would like to use polar coordinates for the potential (because
the Coulomb potential is radial), but label the points by their Cartesian coordinates. Specifically, let us denote the equilibrium locations of the electrons as In
two-dimensions, we will use polar coordinates (, ) to expand the potential. Doing
this, we find that6
2
2
2
2 o
e2 X 1 n
h (u~0 ) =
u
u
+
u
u
+
u
u
+
u
u
r
+r
r
r
r
+r
r
r
~
~
~
~
x
y
x
y
x
y
x
y
0
0
0
0
2 r ,r =1 3
x
2
2 o e2 X
2
2 o
e X 1 n
1 n
u
u
+
u
u
+
u
u
,
r
+0
y
r
+0
y
0
x
+r
0
x
r
~
~
~
~
x
x
y
y
0
0
0
0
x
x
y
y
2 r =1 rx3
2 r =1 ry3
2
(3.b.1)
where = (rx2 + ry2 )1/2 . The last two terms in this expression grow only linearly
with N , the number of electrons in one direction, whereas the first term grows like
N 2 ; i.e. both terms in the second line of equation (3.b.1) are of measure zero in
two-dimensions7. Therefore, it is consistent to simply ignore these contributions
when taking the large-N -limitwhen we will replace the sums in equation (3.b.1) by
integrals over the plane.
As before, we are interested in plasmons which are longitudinal plane-waves,
~urx +ry = x
ei(krx t) ,
(3.b.2)
where we have used the longitudinality of the wave, ~k = k

x. Inserting this test
function into equation (3.b.1) we obtain the equations of motion
Z /2 Z
o
X
krx
d
1n
2
2
2
2
d
1 cos (krx ) 8e
sin
.
(3.b.3)
m = 4e
3
2
2
0
1
r ,r =1
x
6The displacement functions u are of course vector quantities. The subscript appearing in the expression is to indicate
~ = a + a
that it is only the -component of the vector-difference that is considered. This comes about because ~a
r

and the second terms contribution vanishes when acting on a radial function.
7This could be a point of confusionif it were not for the fact that the sums along the x- and y-axes gavie no contribution
(are of measure zero in two-dimensions) then we would of course find the one-dimensional result for this one-dimensional
subsystem of the crystal. It can be checked explicitly that this term is subleading in N but the skeptical reader should
also bear in mind that k is bounded below by 1/N .
Noting that rx = sin in polar coordinates and approximating sin2 (x) by its Taylor
expansion for small argument and by its average (which is 1/2) for large argument,
we see that parametrically,
Z
d
k cos
2

d
sin
,
2
2
0
1
(
)
Z 2/(k cos )
Z /2
Z
1
1
d
d k 2 cos2 () +
d
,
2 2/(k cos ) 2
0
1
Z /2
d 3k cos k 2 cos2 ,
Z
/2
= 3k
2
k .
4
The linear term obviously dominates in the limit of k 0, in which we are interested.
Therefore, we see that in two-dimensions the plasmon dispersion is parameterically
given by
k.
(3.b.4)
k0
o o
Lastly, let us turn out attention to the case in three-dimesnions. Like in the twodimensional case, we must be mindful of coordinates. We will again chose to label
the equilibrium positions8 by their Cartesian coordinates ~r = rx x
+ ry y + rz z, but
we will express the potential in spherical coordinates (, , ). In the limit of large
N , the leading contribution to the potential at u~0 9
(u~0 ) =
e2
2
N
X
rx ,ry ,rz
2
2
2
2
1
urx +ry +rz u~0 + urx +ry rz u~0 + urx ry +rz u~0 + urx +ry +rz u~0
3
=1
2
2
2
2
+ urx ry rz u~0 + urx +ry rz u~0 + urx ry +rz u~0 + urx ry rz u~0
.
(3.c.1)
The cpontributions that we are ignoring here are those from electrons in the planes
normal to each of the coordinate axes. As argued before, these contribute nothing
are regions of measure zeroin three-dimensions.
Taking a longitudinal plasmon aligned in the z-direction as our test function,
~urx +ry +rz = zei(krz t)
~k = k
z,
where
(3.c.2)
we find the equations of motion to be

2
m = 8e
N
X
rx ,ry ,rz
o
1n
1
cos
(kr
)
16e2
z
3
=1
/2
1
d d d cos sin2
0
1
0
k cos
2
(3.c.3)
Notice that weve chosen to keep the range of the integration explicit. This will
come in handy later. Now, instead of doing the integration first, notice that that
we can exactly evaluate the angular integrals and greatly simplify the situation.
8We assume here, as before, a cubical lattice. This assumption is probably not accurate physically, but there are reasons
to suspect that the parametric dispersions for small crystal momentum should be independent of the type of lattice.
9As before, the displacement u is obviously a vectorial quantity. However, for the sake of convenienceand using the
~
r
foresight that we will consider plane waves for which u~r only has a component in the radial-directionwe will not write
the vector label over u~r . Lastly, the subscripts that appeared in equation (3.b.1) will be implicit in the expressions to
follow.
Specifically,
2 =
16e2
m
8e2
=
m
/2
d
0
d cos
1
0
N Z k/2
d
1
d
0
1
sin2
k cos
2
2
sin2 () ,
k2
k/2
1
sin()
cos()
,
k2
0
1
Z
4e2 N 1
d
k sin(k) ,
=
2
m 1
k
2 Z Nk
4e
d
=
sin()
;
m k 2
Here, attention to detail has paid off: the minimum (non-vanishing) crystal momentum is k = 1/N so that in the limit of low crystal momentum and infinite Wigner
crystal, kN 1. This means that for the entire range of integration, 1 and we
can effectively approximate the integrand by Taylor-expanding sin(). This gives
Z
2e2 1
3
5
2
=
d
+
... ,
3m k
20
840
!
4e2 X (1)n+1 1 k 2n
,
=
m
(2n + 1)! 2n
n=1
8e2
=
m
Having come this far, we would have really like to have a closed-form expression for
the constant part, but we havent found one. To better than one part in a thousand,
the constant term is
2
983e2
2 e
2
+
k
+ O(k 4 ).
(3.c.4)
733 24 m
3m
To an accuracy of about 2%, the constant term is just
point is that as k 0, constant. That is
e
.
k0 m
e2
m.
At any rate, the important

(3.c.5)
o o

Homework
Due Tuesday, st November
Problem 1: Thermal Expansion of a One-Dimensional Crystal via Anharmonicities

Consider a simple one-dimensional crystal lattice with nearest-neighbour interaction potential
K
(y) = y 2 gy 3 ,
(1.1)
2
where y = xi+1 xi a0 and g is small; a is the lattice spacing. For g = 0 and at zero external pressure
the lattice spacing a = a0 and the probability distribution of y due to zero-point thermal phonons is
Gauian with root-mean-square (T ) at temperature T ; we are to treat (T ) as a known function.
By perturbatively adding the effects of the cubic anharmonicity, we are to determine the equilibrium
lattice spacing a to leading order in g. We should use this to express the thermal expansion coefficient
in terms of the specific heat at g = 0 and other parameters in the problem. We should estimate the
magnitude of g for which this approximation is valid.
A lot of physics intuition leads us to expect that to leading order in g the equilibrium
anharmonic lattice can be viewed as a harmonic system with modified equilibrium
displacements1. The heuristic picture we have in mind is this: if g is turned on
slowly, the lattice spacing may change reaching some new equilibrium value; but the
fluctuations about this new lattice spacing should still be roughly Gauianand the
widths of the distributions shouldnt know anything about g to leading order2.
Because the harmonic system amounts to a collection of N independent oscillators (each
with the same Gauian normal distribution) the entropy of the system is simply the
sum of the entropies of each. This has nothing to do with the lattice spacing, so if we
are allowed to view the perturbed system as identical to the original system with an
expanded lattice spacing, then the entropy should not be changed. We expect this
argument to hold to leading order in g. This implies that the entropy is independent
of g to leading orderand therefore minimization of the free energy is equivalent to
minimization of the energy u.
Using our reasoning above, we see that the expectation value of the total energy per
lattice site u should therefore be given by3
Z
1
(y a)2
K 2
3
hui = u0 +
dy exp
y
gy
+ O(g 2 ),
(1.2)
2 2 (T )
2
(T ) 2
K
K
= u0 + 2 (T ) 3g 2 (T )a + a2 ga3 + O(g 2 ).
2
2
Notice that the expression above makes sense when g = 0: then equation (1.3) reads
K
2
2
hui = u0 + K
2 = u0 + 2 hy i when the displacement is unchanged, i.e. a = 0we
will show presently that g = 0 implies a = 0.
To find the modified lattice spacing for non-vanishing g, we should minimize the total
energy (1.3) with respect to a. This can be done by inspection. We find
Ka 3ga2 3g 2 (T ) = 0,
(1.3)
(1.4)
which implies that

3g 2
3g 2
+ O(g 2 ).
+ a2 =
K
K
To check consistency, we observe that g = 0 implies a vanishes.
a =
(1.5)
1If we think in terms of Feynman diagrams, then it takes one factor of g to communicate the anharmonicity between
neighbouring lattice sites, but at least two powers of g to communicate anharmonicity between two fluctuations; so to
leading order in g, we expect the distributions to be offset, but otherwise unchanged.
2Note added in revisions: this is obvious from the fact that were considering an adiabatic process.
3The higher order terms in equation (1.2) arise from, e.g., non-Gauianity in the structure of the fluctuations past
leading order.
1
The thermal expansion coefficient is then seen to be4

d(a)
1 6g d
1
=
+ O(g 2 ).
(1.6)
a0 + a dT
a0 K dT
Setting g 0 in expression (1.3) we find the original thermal heat capacity to be
2
d
cv = K
.
(1.7)
a0
dT
Therefore, we can write the thermal expansion coefficient as
3g
= 2 cv + O(g 2 ).
(1.8)
K
o o
The leading order approximation in g is valid only when the anharmonic contribution
to the potential is small compared to the harmonic contribution. Inserting our expression for a into equation (1.3) we find that our condition is5
K
3g 2 a + g a3 < ( 2 + a2 );
2
K
3
g a 2 < 2 ;
=
2
2
K2
= g <
;
9g 2
and therefore our approximations are appropriate as long as
K
|g| <
.
3
(1.9)
Problem 2: Electrons in Two-Dimensions

Consider a two-dimensional gas of electrons (confined to, say, the xy-plane) subjected to a uniform
~ = Ex
magnetic field B in the positive z direction and a uniform electric field E
, giving rise to a two6
dimensional potential U (x, y) = eEx. We may assume for the sake of convenience that the system is of
large length L in the y-direction so that we may impose periodic boundary conditions in that direction.
a. We are to find all the single-electron eigenstates and their corresponding eigenenergies.
From our experience working with gauge fields, we know that to upgrade the Schrodinger
~ 7,8,
equation for a free electron to one in a non-trivial gauge potential A = (, A)
q
all we must do is upgrade 7 + i ~ A everywhere. In our setup, the scalar
potential (x, y) = Ex and we will choose the Lorentz gauge for the magnetic field
~ = Bx
so A
y . This means that
= i~t = 1 (i~x )2 + (i~y )2

H
2m
2
e
1
e
2
7 i~ t i (Ex) =
(i~x ) + i~ y + i Bx
,
~
2m
~
= 1 ~2 2 ~2 2 2i~eBxy + e2 B 2 x2 + eEx.
(2.a.1)
i~t = H
x
y
2m
1
4Notice that
= a1 + O(a)and that the part of O(a) is over order g and so can be neglected when multiplying
a0 +a
0
terms of order g.
5A quicker calculation, using equation (1.1), would give g <
K
6
which is a bit stronger than what we obtain abovebut
the difference is not very substantive.

6To avoid confusionwhich is not easy to do in this probleme will always be take to be the absolute value of the
electrons charge. This ensures that the E-field points in the positive x-direction, but it forces us to systematically alter
the equations copied in lecture (where e was often used to denote the charge q).
7We are going to set c = 1 to avoid lots of confusion. If at the end units are desired, there is always a unique way of
adding c to the expressions.
8
However, I am quite sure that the spatial components of A come with a minus sign: this important fact comes about
via the metric. Many textbooks disappointingly do not clarify how all these signs work out. (Indeed, it is a common
practice of field theory textbooks to define the gauge covariant derivative of QED oppositely to all other gauge fields so
that a familiar minus sign is present for the electron.)
iky
We may suppose that is separablespecifically, of the form (x, y) = (x)e
e
. Pe2n
riodic boundary conditions of course require that k kn = L for some nonnegative
integer n9. Inserting this into the Schrodinger equation (2.a.1), we find that
~2 2
~2 2 ~kn eBx
e2 2 2
H(x)
=
x +
kn
+
B x + eEx (x).
e
(2.a.2)
2m
2m
m
2m
To make this conceptually easier, we should try as hard as we can to simplify the
structure. Although not apparently obvious, it may prove useful to define the cyclotron frequency c and magnetic length `B :
eB
~
c
and
`2B
.
(2.a.3)
m
eB
Making use of these constants, we see that equation (2.a.2) becomes
2
2
2
= ~ 2 + mc `4 k 2 m 2 `2 kn x + mc x2 + eEx (x),
H
e
c B
2m x
2 B n
2
"
2 #
E
E
~2 2 mc2
m E
2
2
=
+
x kn `B +
+ eE kn `B
+
(x),
e
2m x
2
Bc
Bc
2 B
"
2 #
~2 2 mc2
m E
2
(x),
e
(2.a.4)
=
+
(x xn ) + eExn +
2m x
2
2 B
where in the last line we have introduced

E
.
Bc
Notice that equation (2.a.4) is the Schrodinger equation for simple harmonic oscillator
with a displaced origin and a lifted energy:
"
2 #
m E
~2 2 mc2
2
+
(x xn ) (x)
e
+ eExn +
(x),
e
(2.a.5)
H(x) =
2m x
2
2 B
2
m E
1
= m + 2 ~c + eExn +
.
(2.a.6)
2 B
xn kn `2B
If we let m (x) denote the canonical simple harmonic oscillator wave function at level
m, then the eigenenergies (2.a.6) correspond to eigenfunctions
m,n (x) = m (x xn )eikn y
(2.a.7)
where m labels the Landau level and n labels the y-momentum.

Now, the m (x xn ) are Hermite polynomials centred at xn . For a sample with a finite
width W in the x
-direction, it should be the case that xn lies within the sample.
This implies that
E
E
< kn `2B < W +
.
(2.a.8)
Bc
Bc
Because this confines kn to a (finite) range of positive values, this agrees with our
choice of signs earlier.
9See the discussion following equation (2.a.7) for an explanation.
b. Using the wavefunctions found above, we are to determine the total current carried by each
state and compare this to the classical result for a particle undergoing cyclotron motion in perpendicular
E and B fields and to the result obtained by using the semiclassical velocity ~v = ~1 dd~k .
It is rather straight-forward to compute the current of the wave functions found above.
Indeed, using equation (2.a.7) we find for the x
-component of the current10
n
o
m,n ,
~jx = evx = e Re m,n
e
= Re m (x xn )eikn y (i~)x m (x xn )eikn y ,
m
=0
because we can choose the simple harmonic oscillator wave functions m (x) to be
real. The y-component of the current is found similarly,
~jy = e Re m (x xn )eikn y (i~y + eBx) m (x xn )eikn y ,

(2.b.9)
m
e
=
(~kn eBx) 2m (x xn )
y.
(2.b.10)
m
Considering the range of kn allowed, this seems to give a current in the +
y -direction,
as we would expect. The minimum value of kn > Em
.
B~
The classical solution to crossed electric and magnetic fields is of course a cycloid. Indeed,
if we consider the trajectory of an individual electron classically, we find it to be
E
x(t) =
(cos(c t) 1), (sin(c t) c t) ,
(2.b.11)
c B
which gives rise to a net current in the positive y-direction
E
~j = ehy(t)i
= y.
(2.b.12)
B
Semiclassically, we take the derivative of the energy (2.a.6) with respect to kn to obtain
~j = (e) 1 ~E = E y.
(2.b.13)
~ eB
B
10Recall that the gauge-covariant momentum operator = i~

~ + e A.
~
~

Homework
Due Tuesday, th December
Problem 1: BCS Mean-Field Theory
The mean-field BCS Hamiltonian is
o
Xn
H BCS =
k (nk, + nk, ) + ck, ck, + ck, ck, .
(1)
When the mean-field ansatz was made to write the Hamiltonian in terms of the gap energy , it was
necessary that obey the consistency-condition
U X
=
hck, ck, i,
(2)
V
k
where U is the effective attraction between unlike spins and V is the sample volume.
We are to determine all of the eigenenergies and eigenstates of this Hamiltonian, and study its
properties at finite temperature. We will do this using the method introduced by Bogoliubov1.
The key insight of Bogoliubov is that although the mean-field Hamiltonian (1) is clearly
not diagonal in the space of states created by the operators ck,/ , it may be diagonal
in a basis of states created by operators related to these via a simple SU2 transformation. Indeed, it may be the case that electrons and holes are the wrong degrees
of freedom to consider; we should at least look to see if the Hamiltonian is simpler
in terms of any collective degrees of freedom.
Along these lines, we define the rotated, collective or Bogoliubov creation and annihilation operators bk,/ and bk,/ given by
bk, = Ak ck, + Bk ck, ;
bk, = Ak ck, Bk ck, .
(3)
It is clear that we desire this to be an SU2 transformation, which implies that |Ak |2 +
|Bk |2 = 1. As an SU2 -related basis of fermionic operators, these Bogoliubov operators
obey the normal anticommutation relations {bj , bk } = {bj , bk } = 0 and {bj , bk } = jk .
To clarify, we merely propose the transformations (3) and hope that an appropriate
choice of Ak and Bk will bring H BCS into diagonal form. Therefore, the first thing
we must do is re-cast the Hamiltonian in terms of Bogoliubonsand to do this, the
first thing we must do is invert the relationship (3).
Using the definitions (3) together with their Hermitian conjugates we obtain the system
ck,
ck,
=
=
Bk
1
Ak bk, Ak ck,
A
B1 bk, + Bk ck,
k
k
and
ck,
ck,
=
=
Ak
1
Bk bk, Bk ck,
B
1
bk, + Ak ck,
A
k
k
(4)
By subtracting the related identities, we find that

ck, = Ak bk, Bk bk, ;
ck, = Bk bk, + Ak bk, .
H BCS
(5)
Now, writing the mean-field Hamiltonian in terms of the Bogoliubov operators, we encounter
(
h
i
X
k Ak bk, Bk bk, Ak bk, Bk bk, + Bk bk, + Ak bk, Bk bk, + Ak bk,
=
k
Ak bk,
Bk bk,
Bk bk,
Ak bk,
Bk bk,
Ak bk,
Ak bk,
Bk bk,
1This is the method discussed the course textbook, Tinkhams Introduction to Superconductivity. Our analysis will
closely follow the discussion in that text.
1
)
,
H BCS
which at first-sight appears horrendous to expand. Using anti-commutation relations

to simplify things a bit,
(
h
i
X
=
k |Ak |2 |Bk |2 bk, bk, + bk, bk, + 2|Bk |2 2Ak Bk bk, bk, 2Ak Bk bk, bk,
k
i
+ Ak Bk bk, bk, + bk, bk, Ak Bk + A2k bk, bk, Bk2 bk, bk,
i
h
2
+ Ak Bk bk, bk, + bk, bk, Ak Bk + A2
k bk, bk, Bk bk, bk,
h
i
2k |Bk |2 2 Re (Ak Bk ) + k |Ak |2 |Bk |2 + 2 Re (Ak Bk ) bk, bk, + bk, bk,
bk, bk,
A2k
Bk2
2k Ak Bk + bk, bk,
A2
k
Bk2
2k Ak Bk
)
. (6)
Now, because we are free to choose Ak and Bk any way wed likeso long as the transformation is SU2 we would obviously like to define them so that the off-diagonal
contributions to H BCS vanishthese are the last two terms in (6). Now, the offdiagonal terms will vanish if the Bogoliubov coefficients are chosen to satisfy
A2k Bk2 2k Ak Bk = 0.
(7)
2
This is a simple quadratic equation, the solution to which is simply
q
Bk
Bk
Ak =
k 2k + ||2
(k Ek ) .
(8)
With this condition, the last line of (6) vanishes. But we actually have a bit more
than that: the additional constraint |Ak |2 + |Bk |2 = 1 allows us to eliminate these
coefficients all-together3. Notice that (8) implies that, after a bit of algebra,
|Bk |2
1
(k Ek )2
1
k
2
2
2
|Ak |2 =
(
E
)
=
(1
|A
|
)
(
E
)
=
=
.
.
.
=
1
. (9)
k
k
k
k
k
||2
||2
||2 + (k Ek )2
2
Ek
And, similarly,
k
1
|Bk |2 =
1
.
(10)
2
Ek
Notice also that Ak Bk is manifestly real and specifically
Re (Ak Bk ) = Ak Bk = |Bk |2 (k Ek ) .
Let us define the quasi-particle (Bogoliubon) number operators n
k,
H BCS
(11)
bk, bk,
and
n
k, bk, bk, . Putting all of this together, we may re-express the BCS Hamiltonian completely in terms of the Bogoliubov operators.
(
)
h
i
X
=
2k |Bk |2 2|Bk |2 (k Ek ) + k |Ak |2 |Bk |2 + 2|Bk |2 (k Ek ) (
nk, + n
k, ) ,
k
X
k
(
2
2|Bk | Ek + k 2|Bk | Ek (
nk, + n
k, ) ,
(
o
k Ek Ek (
nk, + n
k, ) .
The physical requirement that the total energy be bounded below demands that 7+:
otherwise, the creation of Bogoliubons would lower the energy of the system without
bound. Therefore, we have shown that in terms of the Bogoliubon quasi-particles,
o
Xn
k Ek + Ek (
nk, + n
k, ) .
H BCS =
(12)
k
2We have not made use of the freedom to make A realnor will we: it never is necessary.
k
3To be precise, there is still an arbitrary (unphysical) phase between A and B , and there is an insofar unspecified
k
k
sign in Ak . This sign will be determined belowuntil then, however, well keep it unspecified.
Notice that the Hamiltonian is diagonal in terms of the quasi-particlesit manifestly

commutes with the Bogoliubon number operators. Eigenstates are therefore labeled
by their momenta k and quasi-particle numbers: for a given momentum, let |ki be
defined as that which is annihilated by both bk, and bk, ; clearly, |ki has both
quasi-particle numbers 0. Because the quasi particles are fermions, there can be at
most one of each kind at a given momentum. Therefore, the eigenstates of H BCS for
a given k are exactly
bk, |ki,
|ki,
bk, |ki,
bk, bk, |ki,
and
(13)
with eigenenergies, respectively,

k Ek ,
k ,
k ,
and
k + Ek .
(14)
Now, recall that the consistency of the mean-field Hamiltonian requires that
U X
=
hck, ck, i.
V
k
To find the expectation value of the operator ck, ck, at a given momentum k, well
need the partition function
X
Z=
eH = eb(k Ek ) + 2ek + e(k +Ek ) = 2ek (1 + cosh(Ek )) .
(15)
states
Now, we should express the operator (ck, ck, ) in terms of the Bogoliubov operators,
ck, ck, = Bk bk, + Ak bk, Ak bk, Bk bk, ,
2
= Ak Bk (
nk, + n
k, 1) + A2
k bk, bk, Bk bk, bk, .
Because the last two terms do not commute with the Hamiltonian, they will not
contribute anything to the expectation value hck, ck, i. Therefore4,
1 X
Ak Bk h| (
nk, + n
k, 1) eH |i,
hck, ck, i =
Z
|i
n
o
ek
e(k Ek ) + e(k +Ek ) ,
4Ek (1 cosh(Ek ))
sinh(Ek )
=
,
2Ek (1 cosh(Ek ))
tanh (Ek /2)
=
.
2
Ek
So in the large-volume limit, the consistency demands that
p
Z tanh 2 + ||2
Z~c tanh 2 + ||2
k
k
2
2
V
1
V
p
p
=
dk
=
= 0 or
=
dk .
U
2
U
2k + ||2
2k + ||2
=
fermi
surface
(16)
We were asked to argue that for low enough temperature, > 0 is consistent, but for
high enough temperatures only = 0 is possible. We will actually do a bit more and
determine the critical temperate, Tc , above which = 0 is required for consistency.
However, before we do that calculation, let us argue generally to understand the
results qualitatively.
At zero temperature, tanh(Ek /2) 1 so that a non-zero would be determined by
the equation
s
!!
Z~c
dx
||2
2~c
V
p
= log ~c 1 + 1 +
log() log
.
=
U
(~c )2
||
x2 + ||2
(17)
Solving for (T = 0) we find

|(T = 0)| 2~c eV /U .
(18)
4On the borderline of triviality, we recall the identities sinh() = 2 sinh(/2) cosh(/2) and 1 + cosh() = 2 cosh2 (/2).
Therefore, we know that for small enough temperature,

6= 0 is consistent. However,
p
for temperatures greater than about kT > 12 (~c )2 + ||2 , we may Taylor-expand
the integrand of the consistency equation, yielding
Z~c
1
U
3 3
5 5
= dk
Ek
E +
E ... .
V
Ek 2
24 k 240 k
(19)
Parametrically, if we suppose that 6= 0 then the constraint equation becomes a

polynomial linear in and with a leading term of order 3 2 . Regardless of
the details of integration, the general solution to a polynomial equation of the form
c1 2 3 + c2 = c3 has parametrically of the form 1/ 3/2 . But as becomes
small, must therefore grow veryplarge to compensate, immediately in contradiction
with the hypothesis that kT > 21 (~c )2 + ||2 .
Therefore, we know that at sufficiently high temperatures the only consistent meanfield Hamiltonian is one for which = 0. But we have shown also that at zero
temperature, > 0 is consistent and given by the expression above. Let us find the
temperature Tc where first vanishes.
When 0, Ek k so that the consistency equation becomes
V
=
U
cZ
~c /2
tanh(x)dx
= log (2eE c ~c ) ;
x
(20)
the integral was evaluated using a computer algebra package, and E is Leonhard
Eulers constant5. Combining this with the above, we find
kTc = eE (0).
(21)
o o
5For fun, = lim

n
E
1 (n+1)n1+1/n
( n
)
1
(2+n+ n
)
n2
n+1

Homework
Due Thursday, th December
Problem 1: Little-Parks Experiment

Consider a long, thin-walled, hollow cylinder of radius R and thickness d made of a superconductor
subjected to an external magnetic field H which is parallel to the axis of the cylinder. If the wave
function for superconducting electron pairs (r) is taken as the order parameter for a Landau-Ginzburg
theory, the free energy density is then

2

2

~2
~ + 2e A
~ + 1 H
~ B
~ .
i
(1.1)
f = fn + a(T Tc0 )||2 + ||4 +

2
2m
~c
8
Here, we have made the additional assumption that the leading coefficient is a linear function in T
~ =B
~
that vanishes at the zero-field critical temperature Tc0 . To lowest order in the induced current, H
everywhere, and the last term of the free energy may be ignored.
a. We are to find the lowest free-energy state as a function of both T and H, for H such that the
total flux through the interior of the cylinder is no more than a few flux quanta. We are to give the
values of |(r)| and ~vs (r) and calculate the shifted critical temperature as a function of applied field.
Let the cylinder lie with its axis in the z-direction in cylindrical coordinates. We may
write choose our gauge so that the vector potential is
~ = rH (0, 1, 0) ,
A
2
~
which is easily seen to give rise to H = H z.
Before we try to find the minimum of the free energy, it will be helpful to cut away
generality of our analysis for the conveniences offered by the case at hand. Recall
that the field within a thin-walled superconductor is well approximated by London
theory; this is because gradient terms in the magnitude || cost too much free energy
when the field must vanish outside of the thin cylinder. Therefore, we may write
(r) = ei(r) ,
for R.
Of course, must be taken to vanish outside the superconductor, but this will not
really complicate our analysis. Observe that

~
~
(i)
= (r)
,
(1.2)
(1.3)
(1.4)
which encourages us to write the fourth term in (1.1) as

2

2

~2
2e ~
1
~ ~
2e ~
2 2
~ + 2e A
~ = ~ ||2 (r)
~
i
+
A
m
~
v
,
where
~
v
(r)
+
A
.
s
s

2m
~c
2m
~c
2
m
~c
(1.5)
Using the symmetry of the problem it is obvious that there are only currents in
the -direction;
this implies that r () = z () = 01; so is only a function of .
Furthermore, single-valuedness of the wave function requires that
I
1
= 2r = 2n, =
= n, =
() = n.
(1.6)
r
Using this together with the definition
hc
,
2e
and the fact that the flux through the cylinder is = HR2 , we can write

~
vs =
n+
.
mR
0
0
(1.7)
(1.8)
1Actually, symmetry does not exclude currents in the z

-direction: these are excluded because they raise the free energy
unnecessarilyso that the ground state will have no vertical components.

1
vs
FrameTicks
% & (
# &
(
(Automatic, (0, 0)), GridLines ((0), (0))&
&)
0
H in units of !0
&
&)
Figure 1. The velocity operator vs as a function of the applied field H.

)
)*
"
0
H in units of !0
Figure 2. The magnitude of the wave funciton (H).

Putting everything together, we find the free energy of the superconductor to be
1
fs = a(T Tc0 ) 2 + 4 + mvs2 2 .
(1.9)
2
2
Notice that the free energy is naturally lowered by seeking the smallest possible value
of vs . This is done by choosing the integer n so that n 0 is minimized. This is
shown in Figure 1. Minimizing this with respect to the field magnitude , we obtain
the Landau-Ginzburg equation

1
a(T Tc0 ) + 2 + mvs2 = 0.
(1.10)
2

1
1
2
2
=
a(Tc0 T ) mvs .
(1.11)
2
The parametric dependence on H is shown in Figure 2.
If we define Tc (H) to be the field temperature at which 2 vanishes for a given H (which
enters our expression via vs ), we see that
mvs2
.
(1.12)
2a
This was one of the principle experimental results of Little and Parks. The dependence of Tc on H is shown in Figure 3.
Tc (H) = Tc0
Tc0
Tc0
Tc
0
2
0
H in units of !0
Figure 3. The critical temperature as a function of H.
B H!
2
3
H in units of 0
Figure 4. The qualitative structure of B(H).
~
b. We are to obtain the field B(H)
to leading order in the current of the cylinder and describe when
this approximation is valid.
Recall that the current density is

mvs2
2e
2
.
(1.13)
J = 2e vs = vs a(Tc0 T )
2
We can find the (upper bound on) the maximum of J by differentiating with respect
to vs ; we find
2a(Tc0 T )
vs2 =
,
(1.14)
3m
will maximize J. The reason why one could be worried is that vs is of course bounded.
However, the bound on vs extends beyond the vs required to saturate the maximum
of J,
2a(Tc0 T )
vs2
.
(1.15)
m
Anyway, we may use this to find the maximum current,
r
a(Tc0 T ) 2a(Tc0 T )
Jmax = 4e
.
(1.16)
3m
Using the equation above for the current as a function of vs , we can compute the induced
B-field as a function of H. The total field is shown in Figure 4.
Problem 2: Plane Waves on Thin Superconductors

Consider a thin-sheet superconductor with the same Landau-Ginzburg free energy expansion as discussed in problem 1. We are to determine the spatially uniform configurations of (r) which are minima
of the free energy under variation of . We should determine the maximum current that can be carried
by the waveand especially describe the behaviour as Tc0 is approached from below.
Similar to the situation above, we notice that solutions will necessarily have no spatial
variation in magnitudeagain, because this gradient term costs too much in the free
energy. Therefore, we may write (approximate) any solution which minimizes the
Landau Ginzburg free energy as (r) = ei(r) for R. And also like above we
find thatignoring the magnetic fields induced by the supercurrent

1
1
2 =
a(Tc0 T ) mvs2 ,
(2.1)
2
where
~~
((r)).
m
Notice that this time there is no contribution from the vector potential.
A spatially uniform solution must therefore be one such that ~vs (r) = k x for some
constant vector k inside the thin-sheet superconductor and x
is a direction in the
~vs (r)
(2.2)
planewhich we may take as one of the coordinate axes for convenience. Such a
solution will manifestly generate a spatially uniform velocity,
~
x.
~vs = k
m
Recalling some standard notation,
a(Tc0 T )
~2
,
and
2 (T ) =
,
2ma(Tc0 T )
we can rewrite the minimization condition as

~2 k 2
2 = 02 1
= 02 1 2 k 2 .
2ma(Tc0 T )
02 =
Therefore, we see that spatially uniform solutions are of the form

p
(r) = 0 1 2 k 2 eikx .
Notice that this requires that
k<
(2.3)
(2.4)
(2.5)
(2.6)
o o
(2.7)
for a superconducting solution.

This gives us a current of

~
1 2 k2 .
m
The extreme current can be found via differentiation
1
1 3 2 k 2 = 0
=
k= ;
3
which of course implies that
1 2~
Jmax = 02 .
3 3m
Near T = Tc0 from below, we know that
T
1
,
0 1
,
and
(T ) p
T c0
1 T /Tc0
J = 2e 2 vs = 2e02 k
and therefore the maximum current vanishes according to

3/2
T
Jmax 1
.
T c0
(2.8)
(2.9)
(2.10)
(2.11)
(2.12)
o o
Problem 3: Superconducting Spheres and Monopoles

Aluminium is a type-I superconductor with coherence length = 1.5 106 m and a magnetic
penetration length = 2 108 m at low temperature.
a. At zero temperature, a spherical piece of aluminium of radius r is placed in a magnetic field
~ = B(z)
B
z with a small field gradient (r dB/dz) and B < Hc . We are to state Hc (T = 0) and
compute the force on the sphere for the specific case r = 103 m, B = 103 T, and dB/dz = 102 T/m
and compare this to the case if B 7 1 T while keeping everything else the same.
First, we notice that
0
Hc =
= 7.8 103 T = 78 gauss.
(3.1)
2 2(0)(0)
To compute the force of the sphere of superconducting aluminium from the B-field,
we need to find the external field. This is done by first finding the field inside
~ must vanish. Because gradient is small
the superconductor. There, we know that B
compared to the radius of the sphere, we can approximate the solution by considering
the field to be constant throughout the sphere. It is clear that the superconducting
currents on the surface must cancel this field exactly inside. Therefore, we can quickly
discover the required field arrangement if we know of a system which gives rise to a
uniform magnetic inductance within the volume of a sphere.
Now, if we remember our electrodynamics as much as any student studying for Preliminary exams should, then we recall at once that such a field distribution is obtained
by a rotating, insulating sphere of uniform surface charge density. We could derive
this solution but, for the sake of brevity, we will refer the reader to homework solutions prepared many years ago by the author2. Alternatively, we draw the readers
attention to section (5.10) of Jacksons Classical Electrodynamics.
The principle result was that if a sphere has a uniform magnetic field in its interior given
by
20
B z =
M z,
3
then the field outside the sphere is precisely that of a dipole with dipole moment
m
z=
4r3
M z.
3
(3.2)
(3.3)
In our present situation, we need the field within the sphere to exactly cancel that of the
ambient B-field. This tells us that currents which generate a uniform Bin = B(z)
give rise to a perfect dipole field outside the sphere with dipole moment
m
z=
4r3 3
2r3
(B(z)) =
B(z).
3 20
0
(3.4)
Making use again of the fact that the sphere is small compared to the gradient of B, we
will not lose much by considering the interaction between B and the sphere therefore
to be that between a any magnetic inductance and a dipolenamely,
3
~ m
~ = 2r B(z) B(z) .
F~ = (
~ B)
0
z
(3.5)
o o
Putting in the numbers for our problem, we find

F (103 T) = 102 dyne,
and
F (1 T ) = 10 dyne.
(3.6)
b. Consider a Dirac monopole/anti-monopole pair 102 m apart within a much larger piece of superconducting aluminium at zero temperature. The monopole emits magnetic flux hc/e. We are to give a
rough estimate of the force between these two hypothetical particles.
The physical picture to have in mind is the following. If the two monopoles were separated in free-space, their flux lines would be exactly analogous electrostatics. However, magnetic flux by definition cannot exist within a superconducting statethe
superconductor will do all that it can to confine the magnetic flux to a very small
region in the superconductor. Therefore, it is easy to imagine that all of the flux connecting the two monopoles is confined to a narrow sting between the two monopoles.
The width of this string is roughly 2because this is as narrow a region as a Type-I
superconductor can confine a region of non-critical state.
We may approximate the situation as there being a cylindrical band connecting the
two monopoles in which there is confined all of the magnetic flux between them and
completely normal-state Aluminium. In terms of energy costs, the flux lines must
2If the grader truly desires to see this calculation, please see the homework prepared during a course in the fall of
at: http://www.umich.edu/jbourj/jackson/5-13.pdf.
pay a debt of free energy of3

2
1
2hc
2d20
2
Fmag =
d
=
,
8
e2
2 2
where have made use of the flux quantum 0 . Also, we must pay the energy debt
of raising the superconducting minimum to the normal state over the volume of the
flux tube of force. Using some identities from work elsewhere, we have as the leading
term in the Landau Ginzburg free-energy potential,
2e2 Hc2 (T )2 mc2
2 2
=
d
H .
mc2
8e2 2
4 c
Notice that this energy is linear with distance: this is as expected: the energy content
of the flux tube is proportional to its length. We find the force then to be

220
2 2
F =
+ Hc .
2 2
4
This is an attractive force on the order of a dyne.
Fs = d2 02 = d 2
(3.7)
(3.8)
(3.9)
3Im not sure exactly the choice of units here, but if I compte in Tesla-meters2 , then there needs to be a 1/ to
0
0
get the units right. I suspect that

force.
1
8
7 1/0 , but it doesnt really matter: we are only asked to qualitatively describe the

Homeworks PDF

Uploaded by

Copyright:

Available Formats

Homeworks PDF

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Homeworks PDF

Uploaded by

Copyright:

Available Formats

What is the Drude model and how does it describe the penetration of electromagnetic radiation into a conductor?

What is the Drude model and how does it describe the penetration of electromagnetic radiation into a conductor?

How does a superconductor behave in an external magnetic field?

How does a superconductor behave in an external magnetic field?

Physics , Condensed Matter

~k Ein ~k Eni k~0 Er ~n = 0

Now, Maxwells equations give us 2 E = c2 ()E, so k = c in the vacuum and

k 0 = c in the medium. This allows us to solve the boundary conditions above

JACOB LEWIS BOURJAILY

These work out quite well and one finds that

obvious from our work in part a.

PHYSICS : CONDENSED MATTER HOMEWORK

Figure 1. The absorption coefficient T as a function of as estimated in problem (1.b).

This is shown in Figure 1.

Cu (1010 Hz) = 0.625m

Cu (6 1014 Hz) = 18.2 nm.

JACOB LEWIS BOURJAILY

PHYSICS : CONDENSED MATTER HOMEWORK

3Note added in revision: this is the right answer.

Physics , Condensed Matter

JACOB LEWIS BOURJAILY

PHYSICS : CONDENSED MATTER HOMEWORK

reciprocal lattice vector for graphene, where d

reciprocal lattice vector has magnitude 4

JACOB LEWIS BOURJAILY

Now, an arbitrary ~q in the reciprocal lattice is given by ~q = n1~b1 +n2~b2 , where n1 , n2 Z

Therefore, we find that the intensity evolves like

= 4 cos2 (~q ~r0 ) + (1 ) .

PHYSICS : CONDENSED MATTER HOMEWORK

where we again use the definition ~r0 = (0, d/2).

This is not right.

JACOB LEWIS BOURJAILY

Notice that ~r3 = ~r2 + ~r4 ~a2 . Therefore, we see that

2 1 + ei~q~r2 1 + ei~q~r4 = 1 + ei 3 (2n1 n2 ) 1 + ein3 ei 3 (n1 +n2 ) .

Physics , Condensed Matter

where G represents the reciprocal lattice, which is in this one-dimensional problem

&tc. Now, because we are expanding in

we may safely drop the O(V 3 )

JACOB LEWIS BOURJAILY

q (r) = cq eiqr + cq+b eiqr eibr + cqb eiqr eibr + O(V 2 ),

PHYSICS : CONDENSED MATTER HOMEWORK

This quadratic is easily solved by calling upon kindergarten identities:

different results; but the implicit expression is always correct.

JACOB LEWIS BOURJAILY

Figure 1. The second-order band structure for a one-dimensional system in a weak

Taking the solution corresponding to the lower band3,

The solutions corresponding

PHYSICS : CONDENSED MATTER HOMEWORK

Putting all this together, we see

JACOB LEWIS BOURJAILY

a) We are to obtain the dispersion relation for this model.

This dispersion relation is shown in the first Brillouin zone in Figure 2.

PHYSICS : CONDENSED MATTER HOMEWORK

JACOB LEWIS BOURJAILY

5Say r r = 3, then r + 2r = 3 + r + 2r = 3( + r ) 3Z; and conversely, say r + 2r = 3m, then

r1 r2 = 3m 2r2 r2 = 3(m r2 ) 3Z. QED

PHYSICS : CONDENSED MATTER HOMEWORK

Problem 4: Thermodynamics Near a Dirac Point

With the density of states, we may compute the total energy7,

PHYSICS : CONDENSED MATTER HOMEWORK