PHYS 635 Condensed Matter Physics

Assignment 4 (Nov 9, 2004)

Solutions
1. A&M Problem 12.2. For electrons near a band minimum or maximum, we have
~2
(k k0 )T M1 (k k0 ),
(1.1)
2
where M1 is the reciprocal effective mass tensor. As can be seen from its definition in Eq. (12.29)
in A&M, the reciprocal effective mass tensor is symmetric. Near a band minimum or maximum, the
eigenvalues 1/mi of M1 have the same sign, which means that the surfaces of constant energy as defined
by (1.1) are ellipsoids.
(k) = 0 +

(a) In the absence of a magnetic field we are free to choose the Cartesian axes to coincide with the
principal axes of these ellipsoid. However, when a magnetic field is present, the direction of the magnetic
field will usually be taken as the z-axis, following which the xy-plane will also be fixed. Therefore, for
cyclotron motion, we would in general need to find an area A(, k z ) like the one shown below:
kz

A(, k z)

kz
ky

kx

To go about calculating this area, let us get ready some useful mathematical machinery.
First, consider the integral
$
dx dy dz (z z0 ).

I=

(1.2)

x2 +y2 +z2 R2

By dimensional arguments, [dx] = [dy] = [dz] = length, while [(z z 0 )] = 1/length, the integral has
dimensions of length2 , i.e. it has the dimensions of an area. Going to cylindrical coordinates, we can
write the integral as
Z (R2 z2 )1/2
Z
Z 2

I=
dz
d
d (z z0 ) dz.
(1.3)
0

Performing the integration over z, and noting that this partial integration of the delta function modifies
the limits of the remaining integration over and , we find that
" 2 #(R2 z20 )1/2
Z (R2 z20 )1/2

(1.4)
= (R2 z20 ),
I = 2
d = 2
2
0
0
1

which is none other than the area of the (circular) section of the sphere x 2 + y2 + z2 = R2 at a fixed z = z0 .
Next suppose that the set of unprimed coordinates (x, y, z) are related to another set of primed coordinates (x0 , y0 , z0 ) by a linear transformation r0 = Ar, such that
0

x A xx A xy A xz x A xx x + A xy y + A xz z
y0 = Ayx Ayy Ayz y = Ayx x + Ayy y + Ayz z .
(1.5)
0

z
Azx Azy Azz z
Azx x + Azy y + Azz z

Then the volume element dx0 dy0 dz0 is related to the volume element dx dy dz by
 0

x /x x0 /y x0 /z
dx0 dy0 dz0 =  y0 /x y0 /y y0 /z dx dy dz = |A| dx dy dz,
 0

z /x z0 /y z0 /z 

(1.6)

where |A| is the determinant of the matrix A.

Finally, let us write the equation of a sphere in a vectorial form, i.e. instead of the usual
r 2 = x 2 + y 2 + z2 = R2 ,
we write it as

 x
rT r = x y z y = R2 .

z


(1.7)

(1.8)

Now if the primed coordinates r0 is related to the unprimed coordinates r by the linear transformation
r0 = Ar, and if the singular values of A are all nonzero, and have the same sign, then in general the
primed coordinate axes are rotated and scaled with respect to the unprimed coordinate axes. If we have a
sphere in the unprimed coordinates, under the linear transformation A we would end up with an ellipsoid
in the primed coordinates. Thus
T
(1.9)
r0 T A1 A1 r0 = R2
is the vectorial form of the equation describing an ellipsoid in the primed coordinates.
Comparing (1.9) with
(k k0 )T M1 (k k0 ) = 2( 0 )/~2 ,

(1.10)

we see that the surfaces of constant energy are indeed ellipsoids, as we have stated without proof earlier.
Furthermore, if we think of (k k0 ) as a vector in the primed coordinates, then we may define a vector
q = (q x , qy , qz ) in an unprimed coordinates, such that (k k0 ) = Aq, in which the surfaces of constant
energy are spheres, given by
qT q = 2( 0 )/~2 = 2E/~2 ,
(1.11)
where E =  0 . From (1.10) and (1.9), we see that
T

M1 = A1 A1 ,

(1.12)

M = AAT .

(1.13)

or equivalently,
Now, we are ready to evaluate A(, kz ) as shown in the figure. Our analysis in (1.2) through (1.4) tells
us that this is given by
$
d 3 k0 (kz0 kz ),

A(, kz ) =

k0T M1 k0 2E/~2

(1.14)

where k0 = (k0 k0 ), and we have written the constraint to the volume of an ellipsoid in the vectorial
form given in (1.9).
Next, we change integration variables to the unprimed coordinates q = A 1 (k0 k0 ), so that the
integration volume becomes a sphere. Since the primed integration variable k z0 appears in the deltafunction, we need to write the arguments of (kz0 kz ) in terms of q x , qy and qz . Using (k0 k0 ) = Aq. we
know that
kz0 k0z = Azx q x + Azy qy + Azz qz ,
(1.15)
and so
A(, kz ) =

|A| d 3 q (Azx q x + Azy qy + Azz qz (kz k0z )),

(1.16)

qT q2E/~2

where we have made use of (1.6) relating the volume elements in the primed and unprimed coordinates.
The argument of the delta-function in the above integral looks formidable, but its geometric interpretation is actually quite simple:
Azx q x + Azy qy + Azz qz = (kz k0z )
(1.17)
is the Cartesian equation for a plane in q-space. If we denote the perpendicular distance of this plane
from the origin of q-space as , then clearly the area of the intersection between this plane and the sphere
of radius q = (2E/~2 )1/2 is given by [(2E/~2 ) 2 ]. With this result, we might naively conclude that
#
"
2E
2
(1.18)
A(, kz ) = |A| 2 ,
~
but this is not the case: to obtain the above result, we need the constraining delta-function to have the
form (qn), whereas the delta-function in (1.16) can at most be massaged to a form like ((qn)).
To get the correct answer for A(, kz ), we need to find not just , but also the scaling factor .
To do so, let us assume the form (q n ) for the argument of the delta-function, and compare it
against that in (1.16):
(q n ) = n x q x + ny qy + nz qz = Azx q x + Azy qy + Azz qz (kz k0z ),

(1.19)

which tells us that

n x = Azx ,

ny = Azy ,

nz = Azz ,

Using the fact that n2x + n2y + n2z = 1, we find then that

= (kz k0z ).

= (A2zx + A2zy + A2zz )1/2 .

Finally, using ((x x0 )) = 1 (x x0 ), we can at long last perform the integration to get
"
#
|A| 2( 0 ) (kz k0z )2

.
A(, kz ) =

~2
2

(1.20)

(1.21)

(1.22)

At this point, let us note that we can actually write |A| and in terms of matrix elements of M. First,
we see from M = AAT that
|M| = |A||AT |.
(1.23)

But |AT | = |A|, therefore

|M| = |A|2 ,
3

(1.24)

and we may write |A| = |M|1/2 .

M xx

Myx
Mzx
we find that

Secondly, writing out the matrix elements of M and A explicitly as

M xy M xz A xx A xy A xz A xx Ayx Azx

Myy Myz = Ayx Ayy Ayz A xy Ayy Azy ,
(1.25)

Azx Azy Azz A xz Ayz Azz
Mzy Mzz
Mzz = A2zx + A2zy + A2zz = 2 ,

(1.26)

allowing us to write = Mzz1/2 .

Putting all this together, we find that
|M|
A(, kz ) =
Mzz

!1/2 "

#
2( 0 ) (kz k0z )2

,
~2
Mzz

(1.27)

from which we obtain the cyclotron effective mass as

mc

~2 A
|M|
=
=
2 
Mzz

!1/2

(1.28)

where we have made use of the given information that matrix elements of M do not depend on k (and
hence have no dependence on ). (Shown)
Alternative Method. Alternatively, you can also derive this result by brute force, by first writing out the
dispersion relation
2E
1
1
1
1
1
= M xx
k2x + Myy
ky2 + Mzz1 kz2 + 2M xy
k x ky + 2M xz
k x kz + 2Myz
ky kz
~2

(1.29)

explicitly. For a fixed kz (fixed kz ), this equation can be put into the form
1
1
1
M xx
(k x k1x )2 + Myy
(ky k1y )2 + 2M xy
(k x k1x )(ky k1y ) =

where
k1x =
and

1 1
1 1
M xz
Myy M xy
Myz

1 M 1 M 1 M 1
M xx
yy
xy
xy

kz ,

k1y =

2E
F(kz ),
~2

1 1
1 1
M xx
Myz M xy
M xz

1 M 1 M 1 M 1
M xx
yy
xy
xy

kz ,

1 1
1 1 2
Myz
M xz Myy M xy
k2
1 1
F(kz ) =

z
1
1
M xx Myy M xy M xy
1 1
1 1 2
M xz
M xx Myz M xy
1
k2
Myy 1 1
z
1
1
M xx Myy M xy M xy
1 1
1 1
1 1
1 1
Myz M xy
M xz 2
Myz M xx
M xz Myy M xy
1

k .

2M xy 1 1

z
1 M 1
1 M 1 M 1 M 1
M xx Myy M xy
M
xy
xx
yy
xy
xy
Mzz1 kz2

(1.30)

(1.31)

1
M xx

(1.32)

This is the equation of an ellipse, centered at (k1x , k1y ) and rotated with respect to the k x - and
ky -axes. To find the semi-major and semi-minor axes, we diagonalize the matrix
!
1
1
M xx
M xy
(1.33)
1
1
Myy
M xy
4

to find the eigenvalues as

+ =

1
1
M xx
+ Myy
1
M xx

2
1
+ M xy

2
1
1
M xx + M xy
2



1/2
1  1
1 2
1 1
1 1
M xx + Myy
4 M xx
Myy M xy
M xy
2
1/2

1  1
1 1
1 2
= + ,
+ 4M xy
M xy
M xx Myy
+
2
1/2

1  1
1 1
1 2
+ 4M xy
M xy
M xx Myy
= .

2
+

(1.34)

The semi-major and semi-minor axes of the ellipse are thus

a=

1/2
+

"

2E
F(kz )
~2

#1/2

b=

1/2

"

2E
F(kz )
~2

#1/2

(1.35)

and so the area of the ellipse is

A(, kz ) = ab =

2E/~2 F(kz )
2E/~2 F(kz )
=

,
( + )1/2 ( )1/2
(2 2 )1/2

(1.36)

where
1
1 2

2

M xx + Myy
1  1
1 1
1 1
1
1
1

Myy M xy
M xy .
M xx Myy + 4M xy M xy = M xx
=
2
4
2

(1.37)

Next, we make use of the fact (see for example, in G. Arfken, Mathematical Methods for Physicists )
that
C ji
(1.38)
Mi j = 1 ,
|M |
where C ji is the cofactor of M1 associated with the indices i and j. For i = j = z, we have


1
1 
 M xx
M xy
 M 1 M 1  M 1 M 1 M 1 M 1
xy
yy
xx
yy
xy
xy
Mzz =
=
.
1
1
|M |
|M |
This let us write
1 1
1 1
M xy = Mzz |M1 | =
M xx
Myy M xy

which in turn gives us

|M|
A(, kz ) =
Mzz

!1/2 "

Mzz
,
|M|

#
2E
F(kz ) .
~2

(1.39)

(1.40)

(1.41)

Finally, noting that F(kz ) contains only kz and matrix elements of M1 , which do not depend on ,
we can take the derivative of A(, kz ) with respect to  to find that
A 2 |M|
= 2

~ Mzz
5

!1/2

(1.42)

The cyclotron effective mass is therefore

mc

(b)

|M|
~2 A
=
=
2 
Mzz

!1/2

(Shown)

For the band structure given in (1.1), the density of states is given by
!
Z 3
d k
~2
T 1
 0 (k k0 ) M (k k0 ) .
g() =
43
2

(1.43)

(1.44)

As done in part (a) of the problem, let us go to q-space to get rid of the matrix M 1 . While we are
at it, we might as well scale q further by (2/~2 )1/2 so that the argument of the delta-function becomes
( 0 qT q) = ( 0 q2 ).
As in part (a) of the problem, we pick up a factor of |A| = |M|1/2 with the change from integration
over k to integration over q, as well as a factor of (2/~2 )3/2 from the additional scaling we are introducing
here. We thus have
Z
1
1/2
q2 dq ( 0 q2 )
g() = 2 3 (2|M|)
~
0
#
"

Z
(q  0 ) (q +  0 )
1
1/2
2
(1.45)
= 2 3 (2|M|)
+
q dq

~
 0
 0
0

1
= 2 3 (2|M|)1/2  0 .
~
With this result, the electronic specific heat is therefore given by
cv =

k2 T
2 2
kB T g(F ) = B 3 (2|M|)1/2 (F 0 )1/2 .
3
3~

(1.46)

Now, in the case of free electrons, the density of state is given by

1 2m
g() = 2 2
2 ~

!3/2

 1/2 ,

(1.47)

with the bare mass m of the electron appearing cubed in the expression. If we want to define an effective
mass m so that (1.45) can be written as
1 2m
g() = 2
2 ~2

!3/2

( 0 )1/2 ,

(1.48)

then it is clear that we must have m3 = |M|, i.e. the (specific heat) effective mass is m = |M|1/3 . (Shown)

2. A&M Problem 13.5. For a metal subject simultaneously to a nonzero electric field E , 0 and nonzero
thermal gradient T , 0, the heat production per unit volume dq/dt is given by
dn
dq du
=
,
dt
dt
dt

(2.1)

where u is the energy per unit volume, the chemical potential, and n the number of electrons per unit
volume. The number density n satisfies the continuity equation
dn
= jn ,
dt

(2.2)

where jn is the number current density. The rate of change of the energy density is
du
= j + E j,
dt

(2.3)

where j = ejn is the charge current density, j is the energy current density, and E is the electric field.
Substituting (2.2) and (2.3) into (2.1), we obtain
dq
= j + E j + jn .
dt

(2.4)

To express the rate of heat production dq/dt to the heat current density j q and the charge current
density j, we use Eq. (13.40) in A&M, which tells us that
jq = j jn .

(2.5)

Taking the divergence of this equation, we find that



jq = j jn = j jn jn ,

(2.6)

or, equivalently,

jn = j + jn + jn .

(2.7)

Substituting this into (2.4), we then find

dq
1
= jq jn jn + E j + jn = jq + j + E j = jq + E j,
dt
e

(2.8)

where E = E + 1e is the field associated with the electrochemical potential. (Shown)

Next, we want to write dq/dt in terms of the physically measurable resistivity , thermal conductivity
K and thermoelectric power Q. To do this, let us first write down Eq. (13.45) of A&M,
j = L11 E L12 T,

(2.9)

jq = L21 E L22 T,
and the relations between , K, Q and the tensors Li j ,
1 = = L11 ,


K = L22 L21 L11

Q=L


12


11 1
7

1

L12 ,

(2.10)

When we have cubic symmetry, the tensors Li j are diagonal and of the form
ij

L
0 0

Li j = 0 Li j 0 ,

0 0 Li j

(2.11)

and thus Li j , , K and Q can all be treated as scalars. With this simplification, we have
j = L11 E L12 T,

jq = L21 E L22 T,

(2.12)

and

1
= = L11 ,

L21 L12
(2.13)
K = L22
,
L11
L12
Q = 11 .
L
ij
Furthermore, from the definition of L in Eq. (13.47) in A&M, we see in Eq. (13.51) of A&M that
L21 = T L12 .

(2.14)

With these results at hand, we write E in terms of j and T as

E=

L12
1
j
+
T = j + QT,
L11
L11

(2.15)

and simplify the second term of dq/dt in (2.8) as

E j = j2 + QT j.
To simplify the first term of dq/dt in (2.8), let us substitute E into jq to write
!
L21
L21 L12
L21 L12
q
j = 11 j +
T K +
T = T Qj KT.
L
L11
L11

(2.16)

(2.17)

Taking the divergence of this equation, we then find

jq = QT j + T Q j + T Q j K T K2 T.

(2.18)

For uniform current flow and uniform temperature gradient, we have the conditions j = 0 and 2 T = 0,
which means that
jq = QT j + T Q j K T.
(2.19)
Combining (2.19) and (2.16), we then write
dq
= j2 + K T T Q j.
dt

(2.20)

Now, from the definitions of Li j , we see that K and Q depend on r essentially through the local
chemical potential (r). But as with the case of global thermal equilibrium, where the chemical potential
8

is determined by the particle number constraint to be a function of the temperature T , i.e. = (T ),

we extend this relation to the case of local equilibrium to write (r) = (T (r)). This means then that
K = K((r)) = K(T (r)) and Q = Q((r)) = Q(T (r)), and we can write their gradients as
K =

dK
T,
dT

Q =

dQ
T,
dT

(2.21)

giving us
dq
dK
dQ
= j2 +
(T )2 T
T j.
(Shown)
(2.22)
dt
dT
dT
In the last part of this question, we are asked to compare the numerical values of the coefficient
T (dQ/dT ), obtained using the relaxation time approximation in a quantum-mechanical context, and the
coefficient (ne/m)(d/dT ), obtained using the relaxation time approximation in the classical context
(i.e. the Drude approximation), of T j. For free classical electrons, the Drude relaxation time is
determined from the measured resistivity of the metal as (Eq. (1.7) in A&M)
m
= 2 ,
(2.23)
ne
which means that the coefficient of T j in the Drude approximation is

1 d
,
(2.24)
e dT
where is the mean thermal energy of the electrons. For a free classical electron gas in equilibrium at
temperature T , the average thermal energy ought to be = 32 kB T , which follows from the Equipartition
Theorem. In the presence of a uniform electric field and electron-electron interactions, however, there
is no reason to expect the Equipartition Theorem to be valid. We simply make a leap of faith in the
Drude approximation by assuming that electron-electron interactions results locally in relaxation to a
local thermal distribution characterized by a temperature T (r), which varies from point to point. Within
such an approximation, the average thermal energy (r), which depends on where we are in the metal,
should still be given locally by (r) = 23 kB T (r), and thus
1 d 3 kB
=
.
e dT
2 e
For free quantum-mechanical electrons, we have (Eq. (2.94) in A&M)
!
2 k B k B T
,
Q=
6 e F

(2.25)

(2.26)

and thus the coefficient of T j in the quantum-mechanical relaxation time approximation for free
electrons is
!
dQ 2 kB kB T
T
=
.
(2.27)
dT
6 e F
The ratio of the magnitude of the Thomson effect predicted by the quantum-mechanical relaxation time
approximation as compared to that predicted by the classical Drude approximation is thus
T (dQ/dT )
22 kB T
=
.
(2.28)
(1/e)(d/dT )
3 F
At room temperature, T = 300 K, and using the typical value of  F = 5 eV, we find the numerical value
of this ratio to be on the order of 102 .
. . . /Siew-Ann Cheong
9