Chapter 4: Crystal Lattice Dynamics

Debye

December 29, 2001

Contents

1 An Adiabatic Theory of Lattice Vibrations 2

1.1 The Equation of Motion . . . . . . . . . . . . . . . . . . 6
1.2 Example, a Linear Chain . . . . . . . . . . . . . . . . . . 8
1.3 The Constraints of Symmetry . . . . . . . . . . . . . . . 11
1.3.1 Symmetry of the Dispersion . . . . . . . . . . . . 12
1.3.2 Symmetry and the Need for Acoustic modes . . . 15

2 The Counting of Modes 18

2.1 Periodicity and the Quantization of States . . . . . . . . 19
2.2 Translational Invariance: First Brillouin Zone . . . . . . 19
2.3 Point Group Symmetry and Density of States . . . . . . 21

3 Normal Modes and Quantization 21

3.1 Quantization and Second Quantization . . . . . . . . . . 24

1
4 Theory of Neutron Scattering 26
4.1 Classical Theory of Neutron Scattering . . . . . . . . . . 27
4.2 Quantum Theory of Neutron Scattering . . . . . . . . . . 30
4.2.1 The Debye-Waller Factor . . . . . . . . . . . . . . 35
4.2.2 Zero-phonon Elastic Scattering . . . . . . . . . . 36
4.2.3 One-Phonon Inelastic Scattering . . . . . . . . . . 37

2
 A crystal lattice is special due to its long range order. As you ex-
plored in the homework, this yields a sharp diffraction pattern, espe-
cially in 3-d. However, lattice vibrations are important. Among other
things, they contribute to

• the thermal conductivity of insulators is due to dispersive lattice

vibrations, and it can be quite large (in fact, diamond has a ther-
mal conductivity which is about 6 times that of metallic copper).

• in scattering they reduce of the spot intensities, and also allow

for inelastic scattering where the energy of the scatterer (i.e. a
neutron) changes due to the absorption or creation of a phonon in
the target.

• electron-phonon interactions renormalize the properties of elec-

trons (make them heavier).

• superconductivity (conventional) comes from multiple electron-

phonon scattering between time-reversed electrons.

1 An Adiabatic Theory of Lattice Vibrations

At first glance, a theory of lattice vibrations would appear impossibly

daunting. We have N ≈ 1023 ions interacting strongly (with energies of
about (e2 /A)) with N electrons. However, there is a natural expansion
parameter for this problem, which is the ratio of the electronic to the

3
ionic mass:
m
¿1 (1)
M
which allows us to derive an accurate theory.
Due to Newton’s third law, the forces on the ions and electrons are
comparable F ∼ e2 /a2 , where a is the lattice constant. If we imagine
that, at least for small excursions, the forces binding the electrons and
the ions to the lattice may be modeled as harmonic oscillators, then

F ∼ e2 /a2 ∼ mωelectron
2 2
a ∼ M ωion a (2)

This means that

à !1/2
ωion m
∼ ∼ 10−3 to 10−2 (3)
ωelectron M
Which means that the ion is essentially stationary during the period
of the electronic motion. For this reason we may make an adiabatic
approximation:

• we treat the ions as stationary at locations R1 , · · · RN and deter-

mine the electronic ground state energy, E(R1 , · · · RN ). This may
be done using standard ab-initio band structure techniques such
as those used by FJP.

• we then use this as a potential for the ions; i.e.. we recalculate

E as a function of the ionic locations, always assuming that the
electrons remain in their ground state.

4
 n’th unit cell

atom α

rα sα

rn
rnα Rnα = r n + rα + sα
O

Figure 1: Nomenclature for the lattice vibration problem. sn,α is the displacement of
the atom α within the n-th unit cell from its equilibrium position, given by rn,α =
rn + rα , where as usual, rn = n1 a1 + n2 a2 + n3 a3 .

Thus the potential energy for the ions

φ(R1 , · · · RN ) = E(R1 , · · · RN ) + the ion-ion interaction (4)

We will define the zero potential such that when all Rn are at their
equilibrium positions, φ = 0. Then
X Pn2
H= + φ(R1 , · · · RN ) (5)
n 2M
Typical lattice vibrations involve small atomic excursions of the or-
der 0.1A or smaller, thus we may expand about the equilibrium position
of the ions.
∂φ 1 ∂ 2φ
φ({rnαi + snαi }) = φ({rnαi }) + snαi + snαi smβj (6)
∂rnαi 2 ∂rnαi ∂rmβj
The first two terms in the sum are zero; the first by definition, and
the second is zero since it is the first derivative of a potential being

5
evaluated at the equilibrium position. We will define the matrix
∂ 2φ
Φmβj
nαi = (7)
∂rnαi ∂rmβj
From the different conservation laws (related to symmetries) of the
system one may derive some simple relationships for Φ. We will discuss
these in detail later. However, one must be introduced now, that is,

rα sα

Figure 2: Since the coefficients of potential between the atoms linked by the blue lines
(m−n)βj
(or the red lines) must be identical, Φmβj
nαi = Φ0αi .

due to translational invariance.

(m−n)βj ∂ 2φ
Φmβj
nαi = Φ0αi = (8)
∂r0αi ∂r(n−m)βj
ie, it can only depend upon the distance. This is important for the next
subsection.

6
1.1 The Equation of Motion

From the derivative of the potential, we can calculate the force on each
site
∂φ({rmβj + smβj })
Fnαi = − (9)
∂snαi
so that the equation of motion is

−Φmβj
nαi smβj = Mα s̈nαi (10)

If there are N unit cells, each with r atoms, then this gives 3N r equa-
tions of motion. We will take advantage of the periodicity of the lattice
by using Fourier transforms to achieve a significant decoupling of these
equations. Imagine that the coordinate s of each site is decomposed
into its Fourier components. Since the equations are linear, we may just
consider one of these components to derive our equations of motion in
Fourier space
1
snαi = √ uαi (q)ei(q·rn −ωt) (11)
Mα
where the first two terms on the rhs serve as the polarization vector
for the oscillation, uαi (q) is independent of n due to the translational
invariance of the system. In a real system the real s would be composed
of a sum over all q and polarizations. With this substitution, the
equations of motion become
1
ω 2 uαi (q) = q Φmβj
nαi e
iq·(rm −rn )
uβj (q) sum repeated indices .
Mα Mβ
(12)

7
 t=0

t = ∆t

t = 2 ∆t

Figure 3: uαi (q) is independent of n so that a lattice vibration can propagate and
respect the translational invariant of the lattice.

(m−n)βj
Recall that Φmβj
nαi = Φ0αi so that if we identify

βj 1 1
Dαi = q Φmβj
nαi e
iq·(rm −rn )
= q Φpβj
0αi e
iq·(rp )
(13)
Mα Mβ Mα Mβ
where rp = rm − rn , then the equation of motion becomes

βj
ω 2 uαi (q) = Dαi uβj (q) (14)

or
µ ¶
βj βj
Dαi − ω 2 δαi uβj (q) = 0 (15)
³ ´
which only has nontrivial (u 6= 0) solutions if det D(q) − ω 2 I = 0.
For each q there are 3r different solutions (branches) with eigenvalues
ω (n) (q) (or rather ω (n) (q) are the root of the eigenvalues). The de-
pendence of these eigenvalues ω (n) (q) on q is known as the dispersion
relation.

8
 basis
M1 M2 α=1 α=2 f

Figure 4: A linear chain of oscillators composed of a two-element basis with different

masses, M1 and M2 and equal strength springs with spring constant f .

1.2 Example, a Linear Chain

Consider a linear chain of oscillators composed of a two-element ba-

sis with different masses, M1 and M2 and equal strength springs with
spring constant f . It has the potential energy
1 X
φ = f (sn,1 − sn,2 )2 + (sn,2 − sn+1,1 )2 . (16)
2 n
We may suppress the indices i and j, and search for a solution
1
snα = √ uα (q)ei(q·rn −ωt) (17)
Mα
to the equation of motion
1
ω 2 uα (q) = Dαβ uβ (q) where Dαβ = q Φp,β
0α e
iq·(rp )
(18)
Mα Mβ
and,
∂ 2φ
Φm,β
n,α = (19)
∂r0,α ∂r(n−m),β
³ ´
where nontrivial solutions are found by solving det D(q) − ω 2 I = 0.
The potential matrix has the form

Φn,1 n,2
n,1 = Φn,2 = 2f (20)

9
 Φn,2 n,1 n−1,2
n,1 = Φn,2 = Φn,1
n+1,1
= Φn,2 = −f . (21)

This may be Fourier transformed on the space index n by inspection,

so that
1
Dαβ = q Φpβ
0α e
iq·(rp )
Mα Mβ
 ³ ´ 
2f
 M1 − √Mf1 M2 1+e −iqa

=   (22)
 ³ ´ 
− √Mf1 M2 1+e +iqa 2f
M2

Note that the matrix D is hermitian, as it must be to yield real, physi-

cal, eigenvalues ω 2 (however, ω can still be imaginary if ω 2 is negative,
³ ´
indicating an unstable mode). The secular equation det D(q) − ω 2 I =
0 becomes
à !
4 1 21 4f 2
ω − ω 2f + + sin2 (qa/2) = 0 , (23)
M1 M2 M1 M2
with solutions
v
à ! uà !2
1 1 u 1 1 4
ω2 = f + ± ft + − sin2 (qa/2) (24)
M1 M2 M1 M2 M1 M2
This equation simplifies significantly in the q → 0 and q/a → π limits.
µ ¶
1 1
In units where a = 1, and where the reduced mass 1/µ = M1 + M2 ,
v v
u u
fµ u
t
u 2f
t
lim ω− (q) = qa lim ω+ (q) = (25)
q→0 2M1 M2 q→0 µ
and
q q
ω− (q = π/a) = 2f /M2 . ω+ (q = π/a) = 2f /M1 (26)

10
 2.0
optical mode
1.5
acoustic mode ω+
1.0
ω

ω-
0.5 ω ~ ck

0.0
-4 -2 0 2 4
q
Figure 5: Dispersion of the linear chain of oscillators shown in Fig. 4 when M 1 = 1,
M2 = 2 and f = 1. The upper branch ω+ is called the optical and the lower branch is
the acoustic mode.

As a result, the + mode is quite flat; whereas the − mode varies from
zero at the Brillouin zone center q = 0 to a flat value at the edge of the
zone. This behavior is plotted in Fig. 5.
It is also instructive to look at the eigenvectors, since they will tell
us how the atoms vibrate. Let’s look at the optical mode at q = 0,
q
ω+ (0) = 2f /µ. Here,
 
√
 2f /M1 −2f / M1 M2 
D=
 √ 
 . (27)
−2f / M1 M2 2f /M2

11
Eigenvectors are non-trivial solutions to (ω 2 I − D)u = 0, or
  
√
 2f /µ − 2f /M1 2f / M1 M2  u1 
0=
 √ 


 . (28)
2f / M1 M2 2f /µ − 2f /M2 u2
q
with the solution u1 = − M2 /M1 u2 . In terms of the actual displace-
ments Eqs.11 v
u
sn1 u M 2 u1
=t (29)
sn2 M 1 u2
or sn1 /sn2 = −M2 /M1 so that the two atoms in the basis are moving
out of phase with amplitudes of motion inversely proportional to their
masses. These modes are described as optical modes since these atoms,

Figure 6: Optical Mode (bottom) of the linear chain (top).

if oppositely charged, would form an oscillating dipole which would

couple to optical fields with λ ∼ a. Not all optical modes are optically
active.

1.3 The Constraints of Symmetry

We know a great deal about the dispersion of the lattice vibrations

without solving explicitly for them. For example, we know that for each
q, there will be dr modes (where d is the lattice dimension, and r is the

12
number of atoms in the basis). We also expect (and implicitly assumed
above) that the allowed frequencies are real and positive. However,
from simple mathematical identities, the point-group and translational
symmetries of the lattice, and its time-reversal invariance, we can learn
more about the dispersion without solving any particular problem.
The basic symmetries that we will employ are

• The translational invariance of the lattice and reciprocal lattice.

• The point group symmetries of the lattice and reciprocal lattice.

• Time-reversal invariance.

1.3.1 Symmetry of the Dispersion

Complex Properties of the dispersion and Eigenmodes First, from the

symmetry of the second derivative, one may show that ω 2 is real. Recall
³ ´
that the dispersion is determined by the secular equation det D(q) − ω 2 I =
0, so if D is hermitian, then its eigenvalues, ω 2 , must be real.

∗βj 1
Dαi = q Φpβj
0αi e
−iq·(rp )
(30)
Mα Mβ
1 −p,β,j iq·(rp )
= q Φ0,α,i e (31)
Mα Mβ
Then, due to the symmetric properties of the second derivative

∗βj 1 0,α,i 1
Dαi = q Φ−p,β,j eiq·(rp ) = q Φp,α,i
0,β,j e
iq·(rp ) αi
= Dβj (32)
Mα Mβ Mα Mβ

13
Thus, DT ∗ = D† = D so D is hermitian and its eigenvalues ω 2 are
real. This means that either ω are real or they are pure imaginary. We
will assume the former. The latter yields pure exponential growth of
our Fourier solution, indicating an instability of the lattice to a second-
order structural phase transition.
Time-reversal invariance allows us to show related results. We as-
sume a solution of the form
1
snαi = √ uαi (q)ei(q·rn −ωt) (33)
Mα
which is a plane wave. Suppose that the plane wave is moving to the
right so that q = x̂qx , then the plane of stationary phase travels to the
right with
ω
x=t. (34)
qx
Clearly then changing the sign of qx is equivalent to taking t → −t.
If the system is to display proper time-reversal invariance, so that the
plane wave retraces its path under time-reversal, it must have the same
frequency when time, and hence q, is reversed, so

ω(−q) = ω(q) . (35)

αi
Note that this is fully equivalent to the statement that Dβj (q) =
∗αi
Dβj (−q) which is clear from the definition of D.
Now, return to the secular equation, Eq. 15.
µ ¶
βj 2 βj
Dαi (q) −ω (q)δαi ²βj (q) = 0 (36)

14
Lets call the (normalized) eigenvectors of this equation ². They are the
elements of a unitary matrix which diagonalizes D. As a result, they
have orthogonality and completeness relations
X ∗(n) (m)
²α,i (q)²α,i (q) = δm,n orthogonality (37)
α,i
X ∗(n) ∗(n)
²α,i (q)²β,j (q) = δα,β δi,j (38)
n
If we now take the complex conjugate of the secular equation
µ ¶
βj βj
Dαi (−q) − ω 2 (−q)δαi ²∗βj (q) = 0 (39)

Then it must be that

²∗βj (q) ∝ ²βj (−q) . (40)

Since the {²} are normalized the constant of proportionality may be

chosen as one
²∗βj (q) = ²βj (−q) . (41)

Point-Group Symmetry and the Dispersion A point group operation

takes a crystal back to an identical configuration. Both the original
and final lattice must have the same dispersion. Thus, since the recip-
rocal lattice has the same point group as the real lattice, the dispersion
relations have the same point group symmetry as the lattice.
For example, the dispersion must share the periodicity of the Bril-
louin zone. From the definition of D
βj 1
Dαi (q) = q Φpβj
0αi e
iq·(rp )
(42)
Mα Mβ

15
 βj βj
it is easy to see that Dαi (q + G) = Dαi (q) (since G · rp = 2πn, where
n is an integer). I.e., D is periodic in k-space, and so its eigenvalues
(and eigenvectors) must also be periodic.

ω (n) (k + G) = ω (n) (k) (43)

²βj (k + G) = ²βj (k) . (44)

1.3.2 Symmetry and the Need for Acoustic modes

Applying basic symmetries, we can show that an elemental lattice (that

with r = 1) must have an acoustic model. First, look at the transla-

s 11

Figure 7: If each ion is shifted by s1,1,i , then the lattice energy is unchanged.

tional invariance of Φ . Suppose we make an overall shift of the lattice

by an arbitrary displacement sn,α,i for all sites n and elements of the
basis α (i.e. sn,α,i = s1,1,i ). Then, since the interaction is only between

16
ions, the energy of the system should remain unchanged.
1 X m−n,β,j
δE = Φ0,α,i sn,α,i sm,β,j = 0 (45)
2 m,n,α,β,i,j
1 X m−n,β,j
= Φ0,α,i s1,1,i s1,1,j (46)
2 mnα,β,i,j
1X X m−n,β,j
= s1,1,i s1,1,j Φ0,α,i (47)
2 i,j mnα,β

Since we know that s1,1,i is finite, it must be that

X m−n,β,j X
Φ0,α,i = Φp,β,j
0,α,i = 0 (48)
m,n,α,β p,α,β

Now consider a strain on the system Vm,β,j , described by the strain

matrix mα,i
β,j
X
Vm,β,j = mα,i
β,j sm,α,i (49)
α,i

After the stress has been applied, the atoms in the bulk of the sample

Figure 8: After a stress is applied to a lattice, the movement of each ion (strain) is
not only in the direction of the applied stress. The response of the lattice to an applied
stress is described by the strain matrix.

are again in equilibrium (those on the surface are maintained in equi-

librium by the stress), and so the net force must be zero. Looking at

17
the central (n = 0) atom this means that
X
0 = F0,α,i = − Φm,β,j γ,k
0,α,i mβ,j sm,γ,k (50)
m,β,j,γ,k

Since this applies for an arbitrary strain matrix mγ,k

β,j , the coefficients

for each mγ,k

β,j must be zero

X
Φm,β,j
0,α,i sm,γ,k = 0 (51)
m

An alternative way (cf. Callaway) to show this is to recall that the

reflection symmetry of the lattice requires that Φm,β,j
0,α,i be even in m;

whereas, sm,γ,k is odd in m. Thus the sum over all m yields zero.
Now let’s apply these constraints to D for an elemental lattice where
r = 1, and we may suppress the basis indices α.
1 X
Dij (q) = Φp,j
0,i e
iq·(rp )
(52)
M p

For small q we may expand D

à !
1 X 1
Dij (q) = Φp,j
0,i 1 + iq · (rp ) − (q · (rp ))2 + · · · (53)
M p 2
We have shown above that the first two terms in this series are zero.
Thus,
1 X 2
Dij (q) ≈ − Φp,j
0,i (iq · (rp )) (54)
2M p

Thus, the leading order (small q) eigenvalues ω 2 (q) ∼ q 2 . I.e. they are
acoustic modes. We have shown that all elemental lattices must have
acoustic modes for small q.

18
 In fact, one may show that all harmonic lattices in which the energy
is invariant under a rigid translation of the entire lattice must have at
least one acoustic mode. We will not prove this, but rather make a
simple argument. The rigid translation of the lattice corresponds to a
q = 0 translational mode, since no energy is gained by this translation,
it must be that ωs (q = 0) = 0 for the branch s which contains this
mode. The acoustic mode may be obtained by perturbing (in q) around
this point. Physically this mode corresponds to all of the elements of
the basis moving together so as to emulate the motion in the elemental
basis.

2 The Counting of Modes

In the sections to follow, we need to perform sums (integrals) of func-

tions of the dispersion over the crystal momentum states k within the
reciprocal lattice. However, the translational and point group symme-
tries of the crysal, often greatly reduce the set of points we must sum. In
addition, we often approximate very large systems with hypertoroidal
models with periodic boundary conditions. This latter approximation
becomes valid as the system size diverges so that the surface becomes
of zero measure.

19
2.1 Periodicity and the Quantization of States

A consequence of approximating our system as a finite-sized periodic

system is that we now have a discrete sum rather than an integral over
k. Consider a one-dimensional finite system with N atoms and periodic
boundary conditions. We seek solutions to the phonon problem of the
type
sn = ²(q)ei(qrn −ωt) where rn = na (55)

and we require that

sn+N = sn (56)

or
q(n + N )a = qna + 2πm where m is an integer (57)

Then, the allowed values of q = 2πm/N a. This will allow us to convert

the integrals over the Brillouin zone to discrete sums, at least for cubic
systems; however, the method is easily generalized for other Bravais
lattices.

2.2 Translational Invariance: First Brillouin Zone

We can use the translational invariance of the crystal to reduce the

complexity of sums or integrals of functions of the dispersion over the
crystal momentum states. As shown above, translationally invariant
systems have states which are not independent. It is useful then to de-
fine a region of k-space which contains only independent states. Sums

20
 G vector
Bisector

First Brillouin Zone

Figure 9: The First Brillouin Zone. The end points of all vector pairs that satisfy
the Bragg condition k − k0 = Ghkl lie on the perpendicular bisector of Ghkl . The
smallest polyhedron centered at the origin of the reciprocal lattice and enclosed by
perpendicular bisectors of the G’s is called the first Brillouin zone.

over k may then be confined to this region. This region is defined as

the smallest polyhedron centered at the origin of the reciprocal lattice
and enclosed by perpendicular bisectors of the G’s is called the Bril-
louin zone (cf. Fig. 9). Typically, we choose to include only half of the
bounding surface within the first Brillouin zone, so that it can also be
defined as the set of points which contains only independent states.
From the discussion in chapter 3 and in this chapter, it is also clear
that the reciprocal lattice vectors have some interpretation as momen-
tum. For example, the Laue condition requires that the change in
momentum of the scatterer be equal to a reciprocal lattice translation
vector. The end points of all vector pairs that satisfy the Bragg condi-
tion k − k0 = Ghkl lie on the perpendicular bisector of Ghkl . Thus, the
FBZ is also the set of points which cannot satisfy the Bragg condition.

21
2.3 Point Group Symmetry and Density of States

Two other tricks to reduce the complexity of these sums are worth
mentioning here although they are discussed in detail elswhere.
The first is the use of the point group symmetry of the system. It
is clear from their definition in chapter 3, the reciprocal lattice vectors
have the same point group symmetry as the lattice. As we discussed
in chapter 2, the knowledge of the group elements and corresponding
degeneracies may be used to reduce the sums over k to the irreducible
wedge within the the First Brillioun zone. For example, for a cubic
system, this wedge is only 1/23 3! or 1/48th of the the FBZ!
The second is to introduce a phonon density of states to reduce the
multidimensional sum over k to a one-dimensional integral over energy.
This will be discussed in chapter 5.

3 Normal Modes and Quantization

In this section we will derive the equations of motion for the lattice, de-
termine the canonically conjugate variables (the the sense of Lagrangian
mechanics), and use this information to both first and second quantize
the system.
Any lattice displacement may be expressed as a sum over the eigen-

22
vectors of the dynamical matrix D.
1 X
sn,α,i = √ Qs (q, t)²sα,i (q)eiq·rn (58)
Mα N q,s

Recall that ²sα,i (q) are distinguished from usα,i (q) only in that they are
normalized. Also since q + G is equivalent to q, we need sum only over
the first Brillouin zone. Finally we will assume that Qs (q, t) contains
the harmonic time dependence and since sn,α,i is real Q∗s (q) = Qs (−q).
We may rewrite both the kinetic and potential energy of the system
as sums over Q. For example, the kinetic energy of the lattice
1 X
T = Mα (ṡnα,i )2 (59)
2 n,α,i
1 X X
= Q̇r (q)²rα,i (q)eiq·rn Q̇s (k)²sα,i (k)eik·rn (60)
2N n,α,i q,k,r,s
Then as
1 X i(k+q)·r X r ∗s
e n
= δk,−q and ²α,i ²α,i = δrs (61)
N n α,i
the kinetic energy may be reduced to
1 X ¯¯ ¯2
¯
T = ¯Q̇r (q)¯ (62)
2 q,r
The potential energy may be rewritten in a similar fashion
1 X
V = Φm,β,j sn,α,i sm,β,j
2 n,m,α,β,i,j n,α,i
m−n,β,j
1 X Φ0,α,i
= q
2 n,m,α,β,i,j N Mα Mβ
X
Qs (q, t)²sα,i (q)eiq·rn Qr (k, t)²rβ,j (k)eik·rm (63)
q,k,s,r

23
Let rl = rm − rn
l,β,j
1 X Φ0,α,i
V = q
2 n,l,α,β,i,j N Mα Mβ
X
Qs (q, t)²sα,i (q)eiq·rn Qr (k, t)²rβ,j (k)eik·(rl +rn ) (64)
q,k,s,r

and sum over n to obtain the delta function δk,−q so that

1 X 1
V = Qs (−k)²sα,i (−k)Qr (k)²rβ,j (k) q Φl,β,j
0,α,i e
ik·rl
. (65)
2 l,α,β,i,j,s,r Mα Mβ
Note that the sum over l on the last three terms yields D, so that
1 X β,j
V = Dα,i (k)Qs (−k)²sα,i (−k)Qr (k)²rβ,j (k) . (66)
2 l,α,β,i,j,s,r
P βj
Then, since β,j Dαi (k)²rβj (k) = ωr2 (k)²rα,i (k) and ²sα,i (−k) = ²∗s
α,i (k),

1 X r
V = ²α,i (k)²∗s 2 ∗
α,i (k)ωr (k)Qs (k)Qr (k) (67)
2 α,i,k,r,s
P r ∗s
Finally, since α,i ²α,i (k)²α,i (k) = δr,s
1X 2
V = ωs (k) |Qs (k)|2 (68)
2 k,s
Thus we may write the Lagrangian of the ionic system as
1 X µ¯¯ ¯2
¯ 2 2
¶
L=T −V = ¯Q̇s (k)¯ − ωs (k) |Qs (k)| , (69)
2 k,s
where the Qs (k) may be regarded as canonical coordinates, and
∂L
Pr∗ (k) = = Q̇∗s (k) (70)
∂Qr (k)
(no factor of 1/2 since Q∗s (k) = Qs (−k)) are the canonically conjugate
momenta.

24
 The equations of motion are
 
d  ∂L  ∂L
− ∗
or Q̈s (k) + ωs2 (k)Qs (k) = 0 (71)
dt ∂ Q̇s (k)
∗ ∂Qs (k)
for each k, s. These are the equations of motion for 3rN independent
harmonic oscillators. Since going to the Q-coordinates accomplishes the
decoupling of these equations, the {Qs (k)} are referred to as normal
coordinates.

3.1 Quantization and Second Quantization

P.A.M. Dirac laid down the rules of quantization, from Classical Hamilton-
Jacobi classical mechanics to Hamiltonian-based quantum mechanics
following the path (Dirac p.84-89):

1. First, identify the classical canonically conjugate set of variables

{qi , pi }

2. These have Poisson Brackets

à !
X ∂u ∂v ∂u ∂v
{{u, v}} = − (72)
i ∂qi ∂pi ∂pi ∂qi
{{qi , pj }} = δi,j {{pi , pj }} = {{qi , qj }} = 0 (73)

3. Then define the quantum Poisson Bracket (the commutator)

[u, v] = uv − vu = ih̄{{u, v}} (74)

4. In particular, [qi , pj ] = ih̄δi,j , and [qi , qj ] = [pi , pj ] = 0.

25
 Thus, following Dirac, we may now quantize the normal coordinates

[Q∗r (k), Ps (q)] = ih̄δk,q δr,s where the other commutators vanish .
(75)
Furthermore, since we have a system of 3rN uncoupled harmonic os-
cillators we may immediately second quantize by introducing
 
1 q i
as (k) = √  ωs (k)Qs (k) + q Ps (k) (76)
2h̄ ωs (k)
 
1 q i
a†s (k) = √  ωs (k)Q∗s (k) − q Ps∗ (k) , (77)
2h̄ ωs (k)
or
v
u
u h̄ ³ ´
Qs (k) = t as (k) + a†s (−k) (78)
2ωs (k)
v
u
u h̄ωs (k) ³ ´
t
Ps (k) = −i as (k) − a†s (−k) (79)
2
Where

[as (k), a†r (q)] = δr,s δq,k [as (k), ar (q)] = [a†s (k), a†r (q)] = 0 (80)

This transformation {Q, P } → {a, a† } is canonical, since is preserves

the commutator algebra Eq. 75, and the Hamiltonian becomes
à !
X 1
H= h̄ωs (k) a†s (k)as (k) + (81)
k,s 2
which is a sum over 3rN independent quantum oscillators, each one
referred to as a phonon mode!
The number of phonons in state (k, s) is given by the operator

ns (k) = a†s (k)as (k) (82)

26
and a†s (k) and as (k) create and destroy phonons respectively, in the
state (k, s)
q
a†s (k) |ns (k)i = ns (k) + 1 |ns (k) + 1i (83)
q
as (k) |ns (k)i = ns (k) |ns (k) − 1i (84)

If |0i is the normalized state with no phonons present, then the state
with {ns (k)} phonons in each state (k, s) is
 1 
 Y 1  2  Y ³ † ´ns (k)
|{ns (k)}i =   as (k) |0i (85)
k,s ns (k)! k,s

Finally the lattice point displacement

v
u
1 Xu h̄ ³ ´
sn,α,i =√ t as (q) + a†s (−q) ²sα,i (q)eiq·rn (86)
Mα N q,s 2ωs (q)

will be important in the next section, especially with respect to zero-

D E
point motion (i.e. s2 T =0
6= 0).

4 Theory of Neutron Scattering

To “see” the lattice with neutrons, we want their De Broglie wavelength

λ = h/p
0.29A
λneutron = √ E measured in eV (87)
E
to be of the same length as the intersite distance on the lattice. This
means that their kinetic energy E ≈ 12 M v 2 ≈ 0.1eV, or E/kb ≈ 1000K;
i.e. thermal neutrons.

27
 Source of thermal
n λ ≈ | a 1 | or |a 2 |
neutrons a1
a2

Figure 10: Neutron Scattering. The De Broglie wavelength of the neutrons must
be roughly the same size as the lattice constants in order to learn about the lattice
structure and its vibrational modes from the experiment. This dictates the use of
thermal neutrons.

Since the neutron is chargeless, it only interacts with the atomic

nucleus through a short-ranged nuclear interaction (Ignoring any spin-
spin interaction). The range of this interaction is 1 Fermi (10−13 cm.)
or about the radius of the atomic nucleus. Thus

λ ∼ A À range of the interaction ∼ 10−13 cm. (88)

Thus the neutron cannot ”see” the detailed structure of the nucleus,
and so we may approximate the neutron-ion interaction potential as a
contact interaction
X
V (r) = Vn δ(r − rn ) (89)
rn

i.e., we may ignore the angular dependence of the scattering factor f.

4.1 Classical Theory of Neutron Scattering

Due to the importance of lattice vibrations, which are inherently quan-

tum in nature, there is a limit to what we can learn from a classical

28
theory of diffraction. Nevertheless it is useful to compare the classical
result to what we will develop for the quantum problems.
For the classical problem we will assume that the lattice is elemental
(r = 1) and start with a generalization of the formalism developed in
the last chapter
I ∝ |ρ(K, Ω)|2 (90)

where K = k0 − k and Ω = ω0 − ω. Furthermore, we take

X
ρ(r(t)) ∝ δ(r − rn (t)) (91)
n

where
1
rn (t) = rn + sn (t) and sn (t) = √ u(q)ei(q·rn −ω(q)t) (92)
M
describes the harmonic motion of the s-mode with wave-vector q.
XZ
ρ(K, Ω) ∝ dtei[K·(rn +sn (t))−Ωt] . (93)
n

For |K| ∼ 2π/A and sn (t) ¿ A we may expand

XZ
ρ(K, Ω) ∝ dtei[K·(rn )−Ωt] (1 + iK · sn (t) + · · ·) (94)
n

The first term yields a finite contribution only when

K = k0 − k = G and Ω = ω0 − ω = 0 (95)

which are the familiar Bragg conditions for elastic scattering.

The second term, however, yields something new. It only yields a
finite result when

K ± q = k0 − k ± q = G and Ω ± ωs (q) = ω0 − ω ± ωs (q) = 0 (96)

29
When multiplied by h̄ , these can be interpreted as conditions for
the conservation of (crystal) momentum and energy when the scat-
tering event involves the creation (destruction) of a lattice excitation
(phonon). These processes are called Stokes and antistokes processes,
respectively, and are illustrated in Fig. 11.

k = k 0- q , ω = ω 0 - ω q k = k 0+ q , ω = ω 0 + ω q

q ,ω q q ,ω q
n n
k 0, ω0 k 0, ω0
Stokes Process Anti-Stokes Process
(phonon creation) (phonon absorbtion)

Figure 11: Stokes and antistokes processes in inelastic neutron scattering involving
the creation or absorption of a lattice phonon.

Clearly, the anti-Stokes process can only happen at finite temper-

atures where real (as opposed to virtual) phonons are excited. Thus,
our classical formalism does not correctly describe the temperature de-
pendence of the scattering. Several other things are missing, including:

• Security in the validity of the result.

• The effects of zero-point motion.

• Correct temperature dependence.

30
4.2 Quantum Theory of Neutron Scattering

To address these concerns, we will do a fully quantum calculation.

Several useful references for this calculation include

• Ashcroft and Mermin, Appendix N, p. 790)

• Callaway, p. 36–.

• Hook and Hall (for experiment) Ch. 12 p.342-

We will imagine that the scattering shown in Fig. 12 occurs in a box

of volume V . The momentum transfer, from the neutron to the lattice

Initial Final
ψ 0 = 1− e i( k 0•⋅ r - ω 0 t ) ψ f = 1− i( k f •⋅ r - ω f t )
V
√ V e
√
n
k 0 ⋅ω 0
n k ⋅ω

φ0 E0 φf Ef

Figure 12: The initial (left) and final (right) states of the neutron and lattice during
a scattering event. The initial system state is given by Ψ0 = φ0 ψ0 , with energy
²0 = E0 + h̄ω0 where ω0 = k02 /2M . The final system state is given by Ψf = φf ψf ,
with energy ²f = Ef + h̄ωf where ωf = kf2 /2M .

is K = k0 − kf and the energy transfer which is finite for inelastic

scattering is h̄Ω = h̄ (ω0 − ωf ). Again we will take the neutron-lattice

31
interaction to be local
X 1 X Z d3 q X iq·(r−rn )
V (r) = V (r − rn ) = V (q)eiq·(r−rn ) = V0 e (97)
rn N q,n V n

where the locality of the interaction (V (r − rn ) ∝ δ(r − rn )) indicates

that V (q) = V (0) = V0 . Consistent with Aschcroft and Mermin, we
will take
2πh̄2 a 1 XZ
V (r) = d3 qeiq·(r−rn ) (98)
M V rn
2πh̄2 a
where a is the scattering length, and V0 = M is chosen such that the
total cross section σ = 4πa2 .
To formulate our quantum theory, we will use Fermi’s golden rule
for time dependent perturbation theory. (This is fully equivalent to the
lowest-order Born approximation). The probability per unit time for a
neutron to scatter from state k0 to kf is given by
2π X
P = δ(²0 − ²f ) |hΨ0 |V | Ψf i|2 (99)
h̄ f
2π X
= δ(E0 + h̄ω0 − Ef − h̄ωf )
h̄ f
¯ ¯2
¯1 Z ¯
¯
¯
¯V
3
d re i(k−k0 )·r
hφ0 |V (r)| φf i¯¯¯ (100)

If we now substitute in the ion-neutron potential Eq. 98, then the inte-
gral over r will yield a delta function V δ(q + k − k0 ) which allows the
q integral to be evaluated
¯ ¯
(2πh̄)3 X
2
¯X D
¯
¯ ¯
¯ −iK·rn ¯
E ¯2
¯
P =a 2
δ(E0 − Ef + h̄Ω) ¯¯ φ 0 ¯e ¯ φf ¯
¯
(101)
(M V ) f rn

32
 Now, before proceeding to a calculation of the differential cross sec-
dσ
tion dΩdE we must be able to convert this probability (rate) for eigen-
states into a flux of neutrons of energy E and momentum p. A differ-
ential volume element of momentum space d3 p contains V d3 p/(2πh̄)3
neutron states. While this is a natural consequence of the uncertainty
principle, it is useful to show this in a more quantitative sense: Imagine
a cubic volume V = L3 with periodic boundary conditions so that for
any state Ψ in V ,

Ψ(x + L, y, z) = Ψ(x, y, z) (102)

1 P iq·r
If we write Ψ(r) = N qe Ψ(q), then it must be that

qx L = 2πm where m is an integer (103)

with similar relations for the y and z components. So for each volume
³ ´
2π 3
element of q-space L there is one such state. In terms of states p =
h̄q, the volume of a state is (2πh̄/L)3 . Thus d3 p contains V d3 p/(2πh̄)3
states.
The incident neutron flux of states (velocity times density) is
¯ ¯
h̄k0 2 h̄k0 ¯¯ 1 ik0 ·r ¯¯2 h̄k0
j= |Ψ0 | = ¯√ e ¯ = (104)
M M ¯ V ¯ MV
Then since the number of neutrons is conserved
dσ h̄k0 dσ d3 p p2 dpdΩ
j dEdΩ = dEdΩ = P V = PV (105)
dEdΩ M V dEdΩ (2πh̄)3 (2πh̄)3

33
And for thermal (non-relativistic) neutrons E = p2 /2M , so dE =
pdp/M , and
h̄k0 dσ h̄kM dEdΩ
dEdΩ = P V (106)
M V dEdΩ (2πh̄)3
or
dσ k (M V )2
=P . (107)
dEdΩ k0 (2πh̄)3
Substituting in the previous result for P
¯ ¯
dσ k (M V )2 2 (2πh̄)3 X ¯X D
¯
¯ ¯
¯ −iK·rn ¯
E ¯2
¯
= 3
a 2
δ(E0 − Ef + h̄Ω) ¯¯ φ 0 ¯e ¯ φf ¯
¯
dEdΩ k0 (2πh̄) (M V ) f rn
(108)
or
dσ k N a2
= S(K, Ω) (109)
dEdΩ k0 h̄
where
¯ ¯
1 X ¯X D ¯ ¯ E ¯2
¯ ¯ −iK·rn ¯ ¯
S(K, Ω) = δ(E0 − Ef + h̄Ω) ¯¯ φ 0 ¯e ¯ φf ¯
¯
. (110)
N f rn

We may deal with the Dirac delta function by substituting

Z ∞ dt iΩt
δ(Ω) = e . (111)
−∞ 2π

so that
1 XZ ∞ dt i((E0 −Ef )/h̄+Ω)t
S(K, Ω) = e
N f
−∞ 2π
X ¯¯D ¯ ¯
¯ iK·rn ¯
E¯ ¯D
¯¯
¯ ¯
¯ −iK·rm ¯
E¯
¯
¯ φ 0 ¯e ¯ φ f ¯ ¯ φ f ¯e ¯ φ0 ¯ . (112)
rn ,rm

then as
e−iHt/h̄ |φl i = e−iEl t/h̄ |φl i (113)

34
where H is the lattice Hamiltonian, we can write this as
1 XZ ∞ dt iΩt X D ¯¯ iHt/h̄ iK·rn −iHt/h̄ ¯¯ E
S(K, Ω) = e φ 0 ¯e e e ¯ φf
N f
−∞ 2π rn ,rm
D ¯ ¯ E
φf ¯¯e−iK·rm ¯¯ φ0 , (114)

and the argument in the first expectation value is the time-dependent

operator eiK·rn in the Heisenberg representation

eiK·rn (t) = eiHt/h̄ eiK·rn e−iHt/h̄ . (115)

Thus,
1 XZ ∞ dt iΩt X D ¯¯ iK·rn (t) ¯¯ E D ¯¯ −iK·rm ¯¯ E
S(K, Ω) = e φ 0 ¯e ¯ φf φ f ¯e ¯ φ0
N f
−∞ 2π rn ,rm
1 Z ∞ dt X D ¯ ¯ E
= eiΩt φ0 ¯¯eiK·rn (t) e−iK·rm ¯¯ φ0 . (116)
N −∞ 2π rn ,rm

Now since rn (t) = rn + sn (t) (with rn time independent),

1 XZ ∞ dt i(K·(rn −rm )+Ωt) D ¯¯ iK·sn (t) −iK·sm ¯¯ E
S(K, Ω) = e φ 0 ¯e e ¯ φ0 . (117)
N n,m −∞ 2π

This formula is correct at zero temperature. In order to describe

finite T effects (ie., anti-stokes processes involving phonon absorption)
we must introduce a thermal average over all states
X −βE X −βE
hφ0 |A| φ0 i → hAi = l
e hφl |A| φl i / e l
. (118)
l l

With this substitution,

1 XZ ∞ dt i(K·(rn −rm )+Ωt) D iK·sn (t) −iK·sm E
S(K, Ω) = e e e . (119)
N n,m −∞ 2π

and S(K, Ω) is called the dynamical structure factor.

35
4.2.1 The Debye-Waller Factor

To simplify this relation further, recall that the exponentiated operators

within the brackets are linear functions of the creation and annihilation
operators a† and a.
v
u
1 Xu h̄ ³ ´
sn,α (t) = √ t s
²α (q) as (q)(t) + a†s (−q)(t) eiq·rn (120)
Mα N q,s 2ωs (q)

So that, in particular hsn,α,i (t)i = hsn,α,i (0)i = 0. Then let A = iK ·

sn,α,i (t) and B = iK · sm,α,i (0) and suppose that the expectation values
of A and B are small. Then
* +
D
A B
E 1 1
e e = (1 + A + A2 + · · ·)(1 + B + B 2 + · · ·)
*
2 2 +
1 2 1 2
≈ 1 + A + B + AB + A + B + · · ·
2 2
1 D E
≈ 1 + 2AB + A2 + B 2 + · · ·
2
≈
1
e2 h 2AB+A 2
+B 2 i
(121)

This relation is in fact true to all orders, as long as A and B are linear
functions of a† and a . (c.f. Ashcroft and Mermin, p. 792, Callaway pp.
41-48). Thus
D E 2 2
eiK·sn (t) e−iK·sm = e− 2 h(K·sn (t)) i e− 2 h(K·sm ) i ehK·sn (t)K·sm i .
1 1
(122)

Since the Hamiltonian has no time dependence, and the lattice is in-
variant under translations rn
D E 2
eiK·sn (t) e−iK·sm = e−h(K·sn ) i ehK·sn−m (t)K·s0 i , (123)

36
where the first term is called the Debye-Waller factor e−2W .
2
e−2W = e−h(K·sn ) i . (124)

Thus letting l = n − m
Z dt i(K·rl +Ωt) hK·sl (t)K·s0 i
−2W X ∞
S(K, Ω) = e e e . (125)
l
−∞ 2π
Here the Debye-Waller factor contains much of the crucial quantum
physics. It is finite, even at T = 0 due to zero-point fluctuations, and
since hK · sn i2 will increase with temperature, the total strength of the
Bragg peaks will diminish with increasing T . However, as long as a
crystal has long-ranged order, it will remain finite.

4.2.2 Zero-phonon Elastic Scattering

One may disentangle the elastic and inelastic processes by expanding

the exponential in the equation above.
X 1
ehK·sl (t)K·s0 i = (hK · sl (t)K · s0 i)m (126)
m m!
If we approximate the exponential by 1, ie. take only the first, m = 0
term, then
XZ ∞
dt i(K·rl −Ωt)
S0 (K, Ω) = e−2W e . (127)
l −∞ 2π
And we recover the lowest order classical result (modified by the Debye-
Waller factor) which gives us the Bragg conditions that S0 (K, Ω) is only
finite when K = G and Ω = ω0 − ωf = 0.
X
S0 (K, Ω) = e−2W δ(Ω)N δK,G , (128)
G

37
 dσ0 k N a2 −2W X
= e δ(Ω)N δK,G (129)
dEdΩ k0 h̄ G
However, now the scattering intensity is reduced by the Debye-
Waller factor e−2W , which accounts for zero-point motion and thermal
fluctuations.

4.2.3 One-Phonon Inelastic Scattering

When m = 1, then the scattering involves either the absorption or

creation of a phonon. To evaluate
X Z ∞ dt i(K·r +Ωt)
S1 (K, Ω) = e−2W e l
hK · sl (t)K · s0 (0)i . (130)
l −∞ 2π
we need
v
u
1 Xu h̄ ³ ´
sn,α (t) = √ t ²sα (q) as (q, t) + a†s (−q, t) eiq·rn (131)
Mα N q,s 2ωs (q)
in the Heisenberg representation, and therefore we need,

a(q, t) = eiHt/h̄ a(q)e−iHt/h̄

† †
= ei(ω(q)ta (q)a(q)) a(q)e−i(ω(q)ta (q)a(q)
= a(q)ei(ω(q)t(a (q)a(q)−1)) e−i(ω(q)ta (q)a(q)
† †

= a(q)e−iω(q)t (132)

where we have used the fact that (a† a)n a = (a† a)n−1 a† aa = (a† a)n−1 (aa† −
1)a = (a† a)n−1 a(a† a−1) = a(a† a−1)n . Similarly a† (q, t) = a† (q)eiω(q)t .
Thus,
√ iq·r
1 X h̄e n s ³ ´
sn,α (t) = √ q ²α (q) as (q)e−iωs (q)t + a†s (−q)eiωs (q)t
Mα N q,s 2ωs (q)
(133)

38
and
v
u
1 Xu h̄ ³ ´
s0,α (0) = √ t ²rα (p) ar (p) + a†r (−p) (134)
Mα N p,r 2ωr (p)

Recall, we want to evaluate

Z dt i(K·rl +Ωt)
−2W X ∞
S1 (K, Ω) = e e hK · sl (t)K · s0 (0)i . (135)
l
−∞ 2π
Clearly, the only terms which survive in hK · sl (t)K · s0 (0)i are those
with r = s and p = −q. Furthermore, the sum over l yields a delta
P
function N G δK+q,G . Then as ²(G − k) = ²(−k) = ²∗ (k), and ω(G −
k) = ω(k),

−2W
Z ∞dt iΩt X h̄ |K · ²(K)|2
S1 (K, Ω) = e e δK+q,G (136)
−∞ 2π q,G,s 2ωs (q)M
h D E D Ei
e−iωs (q)t as (−K)a†s (−K) + eiωs (q)t a†s (−K)as (−K)

The occupancy of each mode n(q) is given by the Bose factor

1
hn(q)i = (137)
eβω(q) − 1
So, finally
X h̄
S1 (K, Ω) = e−2W |K · ²s (K)|2 (138)
s 2M ωs (K)
[(1 + ns (K))δ(−Ω + ωs (K)) + ns (K)δ(Ω + ωs (K))] .

For the first term, we get a contribution only when Ω − ωs (K) = ω0 −

ωf − ωs (K) = 0; ie., the final energy of the neutron is smaller than the
initial energy. The energy is lost in the creation of a phonon. Note that

39
this can happen at any temperature, since (1 + ns (K)) 6= 0 at any T .
The second term is only finite when Ω + ωs (K) = ω0 − ωf + ωs (K) = 0;
ie., the final energy of the neutron is larger than the initial energy. The
additional energy comes from the absorption of a phonon. Thus phonon
absorption is only allowed at finite temperatures, and in fact, the factor
ns (K) = 0 at zero temperature. These terms correspond to the Stokes
and anti-Stokes processes, respectively, illustrated in Fig. 13.
k = k 0- q , ω = ω 0 - ω q k = k 0+ q , ω = ω 0 + ω q

1 + n (K) n (K)
s s
q ,ω q q ,ω q
n n
k 0, ω0 k 0, ω0
Stokes Process Anti-Stokes Process
(phonon creation) (phonon absorbtion)

Figure 13: Stokes and antistokes processes in inelastic neutron scattering involving
the creation or absorption of a lattice phonon. The antistokes process can only occur
at finite-T, when ns (K) 6= 0.

If we were to continue our expansion of the exponential to larger

values of m, we would find multiple-phonon scattering processes. How-
ever, these terms are usually of minimal contribution to the total cross
section, due to the fact that the average ionic excursion s is small, and
are usually neglected.

40