Problem Set 9 - Solutions

PHY 576, Due Friday, April 12, 2013 at 3pm

April 14, 2013
1. Square lattice, free electron energies
(a) Show for a simple square lattice in two dimensions that the kinetic
energy of a free electron at a corner of the first zone is higher than that of
an electron at the midpoint of a side face of the zone by a factor of 2.
Lets do this step by step. The term first zone must be referring to
the first Brillouin zone in ~k-space. The reciprocal lattice of a square space
lattice is also a square lattice. For a real space (i.e., crystal) lattice of
side a, the reciprocal lattice points are separated by steps of 2/a in
the kx and ky directions. The Wigner-Seitz cells in reciprocal space are
squares also. These have boundaries that bisect the segments connecting
reciprocal
points.
The first Brillouin zone is therefore the square
  lattice


a , a a , a .


A corner of the zone is (kx , ky ) = a , a . A free electron with one of
2 2
h
2 ~ 2
these four wavevectors has kinetic energy Ecorner = 2m
k = hma2 . The
faces of the wave zone have one of kx , ky with magnitude a and the other
h
2
component zero, giving Eface = 2m
(/a)2 = Ecorner /2.
(b) What is the corresponding factor for a simple cubic lattice in three
dimensions?
In three dimensions, the reciprocal lattice is a cube and the WignerSeitz cells of reciprocal space are also cubes,
sides 2/a.
 again
 with

 The
corners of the first Brillouin zone
(FBZ) a , a a , a a , a
are (kx , ky , ky ) = a , a , a and so electrons with pseudomomentum (equals real momentum for the free electron) have energy Ecorner =
h
2
3
h2 2
2
2
2
2m (kx + ky + kz ) = 2ma2 . At the face of the FBZ, one of the kx,y,z has
magnitude /a while the other two are zero. This gives Eface = Ecorner /3
in d = 3.
(c) What bearing might the result of (b) have on the conductivity of divalent metals?
The result of (b) is that for the case of free electrons, electrons with
1

pseudomomentum at the face of the FBZ can have an energy of 1/3 of

that of those in the corner. For a divalent metal, the number of electrons
is 2N , where N is the number of ions. This is the same as the number
of states in the first band. If the states in this band had lower energies
than any states in the second or third or higher bands, then the simple
cubic lattice of divalent ions would be an insulator. This is unlikely unless
the lattice potential U (r) is extremely strong, though, as when U = 0,
the lowest states in the second band have the energy of states at the face
of the FBZ: the free second band states have kinetic energy 1/3 of that
of the highest first band states. You have a better chance of having an
insulator if your lattice is fcc, so that the FBZ has a bcc shape, which is
more spherical, i.e., closer to the filling of free electron states.
2. Free electron energies in reduced zone
Consider the free electron energy bands of an fcc crystal lattice in the approximation of an empty lattice (i.e. potential is actually zero), but in the reduced
zone scheme in which all ks are transformed to lie in the first Brillouin zone
for the fcc lattice. Plot roughly in the [111] direction the energies of all bands up
to six times the lowest band energy at the zone boundary at k = (2/2)( 12 , 12 , 12 ).
Let this energy at the zone boundary be the unit of energy for your plot. This
problem shows why band edges need not necessarily be at the zone center. Several of the degeneracies (band crossings) will be removed when account is taken
of the crystal potential.
First note that for a fcc crystal lattice that the reciprocal lattice is bcc and
the first Brilloiun zone (FBZ) is a truncated octahedron. By your texts or standard construction, primitive lattice vectors in reciprocal space are of the form
~ = m1~b1 + m2~b2 + m3~b3 for integer mi and
G


~b1 = 2 X
+ Y + Z
(1)
a


~b2 = 2 +X
Y + Z
(2)
a


~b3 = 2 +X
+ Y Z
(3)
a
Y , Z are the dual vectors in reciprocal space with x
= 1, y X
= 0,
where X,
X
etc.
The [111] direction is given as the set of pseudomomentum vectors ~k whose
energy we are to compute. The energy of each band for free electrons is given
by
2
~ ~k) = h (~k + G)
~ 2.
(4)
E(G,
2m
~ (of small enough magnitude) and for
We are to compute this energy for each G

~k = u
+ Y + Z)
,
(X
(5)
a
2

which gives vectors in reciprocal space along the [111] direction with prefactor
u chosen so that ~k lies within the first Brillouin zone. By either drawing it out

in 3D or by noting that ~b1 + ~b2 + ~b3 = 2

a (X + Y + Z), so that there is a unit
cell in reciprocal space centered at u = 2 in Eqn. (5), we can see that the range
1 u 1 gives ~k that reach faces of the FBZ for Eqn. (5).
~ = 0 and u = 1 gives the energy
Using G
E1 =

3h2 2
2ma2

(6)

for the ends of the [111] vector in the FBZ.

Inserting the expression Eqn. (5) for ~k(u) into the free electron energy Eqn.
~
(4) and plotting vs. E(u, G)/E
1 vs. u, i.e.,
~
E(u, G)
E1

=
=

2
a



(u 2m1 + 2m2 + 2m3 )2 + (u 2m1 + 2m2 + 2m3 )2 + (u 2m1 + 2m2 + 2m3 )2


2
(12 + 12 + 12 )
a


1
(u 2m1 + 2m2 + 2m3 )2 + (u 2m1 + 2m2 + 2m3 )2 + (u 2m1 + 2m2 + 2m3 )2 (7)
3

using the gnuplot file

set yrange [0:6]
set xrange [-1:1]
f(u,a,b,c)=((u-2*(-a+b+c))**2+(u-2*(a-b+c))**2+(u-2*(a+b-c))**2)/3
plot f(x,0,0,0),f(x,1,0,0),f(x,-1,0,0),f(x,1,1,0), \
f(x,-1,-1,0),f(x,1,1,1),f(x,-1,-1,-1)
gives
6
f(x,0,0,0)
f(x,1,0,0)
f(x,-1,0,0)
f(x,1,1,0)
f(x,-1,-1,0)
f(x,1,1,1)
f(x,-1,-1,-1)

0
-1

-0.5

0.5

To see in part how this figure comes about, take a look at two primitive cells
in reciprocal space. The lower truncated octahedron is the FBZ and the red
line in the first FBZ is in the [111] direction. This line starts and terminates on
the center of hexagonal faces. The upper cell is the result of translation by ~b1
~ = ~k + ~b1 .
(barely visible as a faint line) and the black line in its interior is ~k + G
Note that the minimum distance from the origin to this interior line in the second cell is not at either end or the center, as seen by the plot of the normalized

2
k 2 and ~k + ~b1 included:

Here is Maple code for making this figure:

reset: with(plots): with(geom3d): defo := [45, 45]: r:= sqrt(5):
point(s1, -1, -1, -1): point(s2, 1, 1, 1): dsegment(st, [s1, s2]):
point(o, 0, 0, 0): point(c1, -2, 2, 2): point(c2, 2, -2, 2): point(c3, 2, 2, -2):
TruncatedOctahedron(fbz, o, r):
dsegment(g1, [o, c1]): dsegment(g2, [o, c2]): dsegment(g3, [o, c3]):
translation(to1, fbz, g1): translation(to2, fbz, g2): translation(to3, fbz, g3):
translation(st1, st, g1): translation(st2, st, g2): translation(st3, st, g3):
qleft := draw([fbz(transparency = .5), to1(transparency = .5),
st(color = red), st1(color = black), g1], orientation = defo):
f0 := x -> x^2:
f1 := x -> (1/3)*(x-2)^2+(2/3)*(x+2)^2:
qrite := plot([f0(x), f1(x)], x = -1 .. 1, y = 0 .. 6):
A := Array(1 .. 2): A[1] := qleft: A[2] := qrite: display(A):

3. Kronig-Penney model
The Kronig-Penney model is discussed in Chapter 7 of Kittel, and we discussed it briefly in class. The potential is composed of barriers of width b and
height U0 and wells of width a and zero height. Assume the first barrier for
positive x is between a and a + b, and that the energy is E < U0 .
(a) Beginning with Schrodingers equation, prove that inside the wells
1 (x) = Aeix + Beix and inside the barriers it is 2 (x) = Cex +
Dex . What are and ?
Well, Schrodingers eqn. in one dimension is

h2 00
(x) + U (x)(x) = E(x) .
2m

For regions of constant U ,

2 00
h
(x) = (E U )(x) .
2m

Solutions in general for constant U are (just by ordinary differential equations, knowing that there are two independent solutions and checking that
exponentials work)
(x) = exp(ax)
where

2 a2
h
= (E U ) .
2m
For (E U ) > 0, a is imaginary. For (E U ) < 0, a is real. In each case,
there are two roots for a, since there is an a2 in the equation. So for the
regions where U = 0, we have

(x) = Aeix + Beix ,

with real (a = i is imaginary), = 2mE/h, while for the region

where U = U0 > E,
(x) = Cex + Dex ,
p
with = 2m(U0 E)/h. In the forbidden regions, U = U0 > E, the
wave function is exponential (say, decaying from either side of the barrier)
while in the allowed regions is oscillatory.
(b) Write down four equations for the four coefficients (A, B, C, D), using
continuity of and its derivatives at x = 0 and periodicity of and its
derivatives at x = b. Find the matrix Q such that Q [A, B, C, D] = 0.
We will use the Bloch theorem, so that (x + b + a) = eik(b+a) (x) for
some k. We will take
(x)
(x)

= Aeix + Beix
x

= Ce

+ De
5

0 x a (I)

(8)

b x 0 (II)

(9)

Continuity at x = 0 gives
A+B =C +D
or
A+BC D =0 .
The slope at x = 0 from the [0, a] interval is i(A B) and from the
[b, 0] interval is C D, so setting these equal gives
iA iB C + D = 0 .
It gets more complicated matching at x = a, but it is not too bad. Using
form (I),
(a) = Aeia + Beia ,
while using Bloch and (II) gives
(a) = eik(b+a) (b) = eik(b+a) (Ceb + Deb )
so that
eia A + eia B eik(b+a)+b C eik(b+a)b D = 0 .
A similar substitution for the derivative gives
ieia A ieia B beik(b+a)+b C + beik(b+a)b D = 0 .
Rewriting this in matrix form gives

0
A
B 0

Q
C = 0
0
D
where

1
i
Q=
eia
ieia

1
i
eia
ieia

eik(b+a)+b
eik(b+a)+b

.
eik(b+a)b
eik(b+a)b

(c) By setting the determinant of Q = 0 and taking the limit of the delta
function potential, one can show that (P/a)sin(a) + cos(a) = cos(ka)
(You dont have to show this). For the delta-function potential and with
P << 1, find at k = 0 the energy of the lowest energy band.
P

For P  1,  a
sin(a)  1, since |sin(a)/(a)| < 1. For k = 0,

 1, so a  1 and sin(a)/(a)
cos(ka) = 1 and so cos(a) = 1P sin(a)
a
1, with
(a)2
1 cos(a)
P
2
giving

2P

a
and (since E is the kinetic energy where U = 0)
E=

h2 P
2 2
h
.

2m
ma2

(d) For the same problem find the band gap at k = /a.
Now we have to solve
P

sin(a)
+ cos(a) = 1 ,
a

for /a = k. Writing a = + ,
P

sin(a)
sin()
+ cos(a) = P
cos() = 1 .
a
+

For small , this gives to lowest order in P ,

P
2
(1 ) =

2
P
=

(10)

2
2

(11)

which has two roots which are, to lowest order in P or ,

= 0,

2P
.

These are two solutions for at k = /a. This gives a band gap of




2P
E =
(12)
= E = +
a
a
a
"
#
2  
h2
2P
2
=
+

(13)
2m
a
a
a


h2
2P
=
2

(14)
2m
a a
=

2
h2 P
ma2

(15)

4. Square lattice Consider a square lattice in two dimensions with the crystal
potential:
U (x, y) = 4U cos(2x/a)cos(2y/a).

(16)

Apply the central equation to find approximately the energy gap at the corner
point (/a, /a) of the Brillouin zone. In will suffice to solve a 2 2 determinantal equation.
As described in Problem 1 of this set and as you know by now anyway, the
square crystal lattice has a reciprocal lattice that is also square. The first Brillouin zone is a square of side 2/a. To apply the central equation, we should
know the components of the Fourier series for U (x, y). These can be found
~
directly by integration of U (x, y)eik(x,y) or by noting that
U (x, y)

= 4U cos(2x/a) cos(2y/a)

(17)

= U [2 cos(2x/a)] [2 cos(2y/a)]
(18)



= U e2x/a + e2x/a e2y/a + e2y/a
(19)


= U e2x/a+2y/a + e2x/a2y/a + e2x/a+2y/a + e2x/a2y/a
(20)


~
~
~
~
= U eiG++ ~x + eiG+ ~x + eiG+ ~x + eiG ~x
(21)
X
~ x
iG~
=
CG
(22)
~e
~
G

~ = X
Y and C ~ = U for G
~ =G
~ and C ~ = 0 for all other G.
~
where G
G
G
The central equation can be written as

 2 2
X
h k

~k G)
~
~ =0.
U (G)C(
(23)
E C(~k) +
2m
~
G

The approximation that is made here is that only the Fourier coefficients C(~k)
~ are of imfor (nearly) degenerate states that are directly connected by U (G)
portance. (Solving the full central equation will mix in other coefficients, but
only at higher order in U .)
So taking the pseudomomentum ~k = (/a, /a), there is only degenerate
~ . This is ~k 0 = (/a, /a) =
state that is separated by any vector G
~k G
~ ++ . Then the central equation sets two constraints:
 2 2

h k

~ ++ )C(~k G
~ ++ ) = 0
E C(~k) + U (G
(24)
2m
 2 0 2

h (k )

~ )C(~k 0 G
~ ) = 0 .
E C(~k 0 ) + U (G
(25)
2m

Noting (k 0 )2 = k 2 and setting h

2 k 2 2m = , C(~k) = C and C(~k 0 ) = C 0 ,
( E)C U C 0

(26)

U C + ( E)C 0

(27)

and noting that this equation is solvable only for zero determinant of the corresponding matrix,


 E
U 

=0,
 U
E
gives
( E)2 U 2 = 0
finally giving the energies for states with pseudomomentum ~k = (/a, /a) by
the relation
E =U
so that the band gap is ( + U ) ( U ) = 2U .