Debye theory of specific heat

1 The model
A solid is made of electrically neutral atoms arranged on a periodic lattice (mean positions fall on these points). When a solid
absorbs heat, the electrons absorb some of the heat and the remaining is absorbed by the positive charges at their mean lattice
points. The electrons change their distribution from the zero temperature step distribution to the well -known Fermi Dirac
distribution. We have already seen that the energy required to raise the temperature of electrons from T to T + ∆T is given
by 31 D(EF )T ∆T = cV ∆T . Here cV is called specific heat. We want to find a similar answer for the positive charges. The
remaining heat not absorbed by the electrons are absorbed by the positive charges on the lattice. These positive charges vibrate
around their equilibrium positions. Because these positive charges are surrounded by a cloud of negative charges the forces
between neighboring atoms are not Columbic but van der Waals type. This means the potential energy between two neighboring
atoms goes through a minimum as the separation between the atoms is made to deviate from the equilibrium value. This means
we may write,
X p2 X
n
H= + V (Rn − Rn′ ) (1)
n
2M ′
n̸=n

Here the sum over n covers all the lattice points. Firstly the potential energy goes through a minimum. This means,
1 X
V (Rn − Rn′ ) = (Rn − Rn′ ).Vn,n′ .(Rn − Rn′ ) (2)
2 ′
<n,n >

′ ′ ′
where < n, n > means n, n are nearest neighbors. Here Vn,n′ is called a dyad - or a 3 × 3 matrix for each pair n, n . For Rn = 0
to be a minimum, the eigenvalues of Vn,n′ must all be positive. For example consider a vector r = xî + y ĵ + z k̂. This means we
are interested in,    

v11 v12 v13 x
r.V.r = x y z .  v12 v22 v23  .  y  > 0 (3)
v13 v23 v33 z
for each r if and only if all eigenvalues of V are positive (show this). We now want to write this as,
X
H= Ωq a†q aq + E0 (4)
q

where [aq , aq′ ] = 0 and [aq , a†q′ ] = δq,q′ . Also aq is a linear combination of Rn and pn .

1 X 1 X
aq = [aq , pn ] · Xn − [aq , Xn ] · pn (5)
iℏ n iℏ n

and
1 X † 1 X †
a†q = [aq , pn ] · Xn − [a , Xn ] · pn (6)
iℏ n iℏ n q
The inverse of this is, X X
Xn = [Xn , a†q ]aq − [Xn , aq ]a†q (7)
q q

and X X
pn = [pn , a†q ]aq − [pn , aq ]a†q (8)
q q

1
2 The solution
Now on the one hand,
[aq , H] = Ωq aq (9)
but on the other hand,
X [aq , pn ] 1 X 1 X
[aq , H] = · pn + [aq , (Rn − Rn′ )].Vn,n′ .(Rn − Rn′ ) + (Rn − Rn′ ).Vn,n′ .[aq , (Rn − Rn′ )] (10)
n
M 2 ′
2 ′
<n,n > <n,n >

This means,
X [aq , pn ] 1 X 1 X
Ωq aq = · pn + [aq , (Rn − Rn′ )].Vn,n′ .(Rn − Rn′ ) + (Rn − Rn′ ).Vn,n′ .[aq , (Rn − Rn′ )] (11)
M 2 2
n <n,n′ > <n,n′ >

or,
1 X 1 X
Ωq [aq , pm ] = [aq , (Rn − Rn′ )].Vn,n′ .[(Rn − Rn′ ), pm ] + [(Rn − Rn′ ), pm ].Vn,n′ .[aq , (Rn − Rn′ )] (12)
2 2
<n,n′ > <n,n′ >

or,
iℏ X iℏ X
Ωq [aq , pm ] = [aq , (Rm − Rm+δ )].Vm,m+δ . − [aq , (Rm+δ − Rm )].Vm+δ,m . (13)
2 2
δ δ

iℏ X iℏ X
+ .Vm,m+δ .[aq , (Rm − Rm+δ )] − .Vm+δ,m .[aq , (Rm+δ − Rm )] (14)
2 2
δ δ

where δ is the vectorial distance between nearest neighbors. This means,

−iℏ
Ωq [aq , Rm ] = [aq , pm ] (15)
M
For translationally invariant systems, Vn,n′ ≡ Vn−n′ ,0 .

ℏ2 X ℏ2 X
Ω2q [aq , Rm ] = [aq , (2Rm − Rm+δ − Rm−δ )].V−δ,0 + .V−δ,0 .[aq , (2Rm − Rm+δ − Rm−δ )] (16)
2M 2M
δ δ

0
Multiply by eip.Rm and sum over m keeping in mind that R0m is the equilibrium location of the m-th atom whereas Rm is the
deviation from the equilibrium value.
0 ℏ2 X ip.R0m ℏ2 X 0
Ω2q [aq , Rm ]eip.Rm = e [aq , (2Rm − Rm+δ − Rm−δ )].V−δ,0 + .V−δ,0 .[aq , (2Rm − Rm+δ − Rm−δ )]eip.Rm
2M 2M
δ δ
(17)
where, X 0
[aq , Rm ]eip.Rm ≡ [aq , Dp ] (18)
m
or,
ℏ2 X ⃗ ⃗ ℏ2 X ⃗ ⃗
Ω2q [aq , Dp ] = [aq , Dp ].V−δ,0 (2 − eip·δ − e−ip·δ ) + .V−δ,0 .[aq , Dp ](2 − eip·δ − e−ip·δ ) (19)
2M 2M
δ δ

Note also that,

1 X iq.R0m
am = √ e aq (20)
N q
so that,
1 X iq.(R0m −R0 ′ )
[am , a†m′ ] = e m = δm,m′ (21)
N q

Note that due to translational invariance,

[an , Rm ] = [an−m , R0 ] (22)

2
This means,
1 X iq.R0n X 1 X 0 0
[√ e aq , Dp ] = [ √ eiq.Rn−m aq eip.Rm , R0 ] (23)
N q m N q
or,
[aq , Dp ] = N [ap , R0 ] δq,p (24)
Substitute this into Eq.(19) to get,

ℏ2 X ⃗ ⃗ ℏ2 X ⃗ ⃗
Ω2q [aq , R0 ] = [aq , R0 ].V−δ,0 (2 − eiq·δ − e−iq·δ ) + .V−δ,0 .[aq , R0 ] (2 − eiq·δ − e−iq·δ ) (25)
2M 2M
δ δ

Note that Vδ,0 is a symmetric matrix, this means the 9 components of this 3 × 3 matrix may be written as,
X
V−δ,0 = (êi ⊗ êj ) (êi · V−δ,0 · êj ) (26)
i,j=1,2,3

Here (ê1 , ê2 , ê3 ) = (x̂, ŷ, ẑ) are the Cartesian directions. Since (êi · V−δ,0 · êj ) = (êj · V−δ,0 · êi ),

ℏ2 X X ⃗ ⃗
Ω2q [aq , R0 ] = [aq , R0 ]i êj (êi · V−δ,0 · êj )(2 − eiq·δ − e−iq·δ ) (27)
M i,j=1,2,3
δ

where the sum over nearest neighbors is only the linearly independent ones (this means do not include −⃗δ if +δ is already
included). In one dimension this is simply,

ℏ2 1
Ω2q [aq , R0 ] = [aq , R0 ] M ω 2 4 sin( q · ⃗δ)2 (28)
M 2
since in 1D V−δ,0 = M ω 2 . In other words,
1
Ωq = (2ℏω) sin( qa) (29)
2
Now periodic boundary conditions means,
am+N = am (30)
This means the values of q are discrete subject to the condition,
0
eiq.RN = 1 (31)

Also note that there will be some vector G such that ei(q+G).Rm = eiq.Rm so that there is no need to include both aq and aq+G .
1 X iq.R0m 1 X i(q+G).R0m 1 X iq.R0m
am = √ e aq = √ e aq+G = √ e aq+G (32)
N q N q N q

Thus if q is inside the (first) Brillouin zone then q + G is outside it. This makes aq+G ≡ aq so that we need to only consider
vectors inside the first Brillouin zone. In one dimension these ideas are much simpler.

eiqN a = 1 (33)

and
ei(q+G)ma = eiqma (34)
Thus,
2πm
q = (35)
Na
and
2π
G = (36)
a
π
This means −mmax ≤ m ≤ mmax . We then expect −qmax + G = +qmax . This means qmax = a. In one dimension therefore,
the hamiltonian becomes, X
H= Ωq a†q aq + E0 (37)
q

3
The average is,
X Ωq
< H >= + E0 (38)
−qmax <q<qmax
eβΩq − 1

In three dimensions,
ℏ2 X
 
X
2 1 ⃗
Ω2q [aq , R0 ]j = [aq , R0 ]i (êi · V−δ,0 · êj ) 4 sin q·δ (39)
M i=1,2,3
2
δ

The above is an eigenvalue equation for the vector [aq , R0 ]. It is of the form,

M · [aq , R0 ] = 0 (40)

where,
ℏ2 X
 
1 ⃗
Mi,j = Ω2q δi,j − (êi · V−δ,0 · êj ) 4 sin2 q·δ (41)
M 2
δ

The possible values of Ωq are given by setting,

Det(M ) = 0 (42)
k=1,2,3
This in general gives three different answers for Ωq : Ωq,1 , Ωq,2 , Ωq,3 . For each of these we have an eigenvector [aq , R0 ]
For a cubical lattice the eigenvalues are all the same since,

(êi · V−δ,0 · êj ) = M ω 2 δi,j (43)

or,
ℏ2 X
 
2 1 ⃗
Ω2q [aq , R0 ]j = 2
[aq , R0 ]j M ω 4 sin q·δ (44)
M 2
δ

But the eigenvectors are still three linearly independent ones [aq , R0 ]kx , [aq , R0 ]ky , [aq , R0 ]kz ; k = 1, 2, 3. For a cubical lattice
⃗δ = aî, aĵ, ak̂. This means,
       12
2 1 2 1 2 1
Ωq = (2ℏω) sin qx a + sin qy a + sin qz a (45)
2 2 2
This means the annihilation operator acquires an additional “internal” discrete and finite index k = 1, 2, 3.

Ωq a†q,k aq,k + E0
X
H= (46)
q,k=1,2,3

or,
X Ωq
< H >= + E0 (47)
eβΩq −1
−qmax <qx ,qy ,qz <qmax ;k=1,2,3

The components are chosen so as to There are two situations possible now (i) βℏω ≪ 1 (high temperature) or (ii) βℏω ≫ 1 (low
temperature).

High temperature limit: At high temperatures,

X 3Ωq X
<H> ≈ + E0 = 3 T + E0 (48)
(1 + βΩq + ...) − 1
−π π
a <qx ,qy ,qz < a −π π
a <qx ,qy ,qz < a

This means the specific heat is,

d<H>
= 3N = (3N kB = 3R)
Cv = (49)
dT
since kB = 1 in our case. This is the famous Dulong-Petit Law.

Low temperature limit: At low temperatures, βΩq ≫ 1. In this situation the integrand vanishes exponentially. The only
situation where this can be avoided is if we make q very small so that Ωq is also small so that βΩq will not become very large.

4
In the small q limit, Ωq = ℏωaq. Also since the integrand vanishes for large q, no significant errors are introduced by making
the upper limits ∞ instead of πa . Thus in the small temperature limit,
X
<H> ≈ (3ℏωaq) e−βℏωaq + E0 (50)
q

or,
9V
<H> ≈ T 4 + E0 (51)
(πaℏω)3
The specific heat (per unit volume) is,
36
cV ≈ T3 (52)
(πaℏω)3
Thus in three dimensions the lattice specific heat is proportional to cube of temperature at low temperatures. This means if you
want to raise the temperature of a 3D solid from T to T + ∆T , for small T , we have to supply heat of this amount,
1 36
∆Q = D(EF )T ∆T + T 3 ∆T (53)
3 (πaℏω)3

The first term is due to electrons the second due to lattice.