Lecture 9 — Anharmonic effects in

crystals.
1 Introduction

Most of the effects related to lattice dynamics that you have so far encountered in this course
and in previous courses — lattice specific heat, Debye-Waller factors, thermal diffuse scattering
etc., could be very well interpreted in terms of the harmonic theory of the crystal lattice. Indeed,
deviation from the harmonic expression for the Debye-Waller factors (W ∝ q 2 ) are only really
significant at high temperatures, and so is the anharmonic correction to the expression for the
phonon specific heat. In this lecture, we will focus our attention on two effects that can be
observed at all temperatures, and that can only be explained within the dynamical theory of the
crystal lattice by invoking anharmonic terms in the lattice energy.

Anharmonic terms lead to a lattice potential energy of the form:

∞
eq
X 1 X
U =U + Dµ(n)
1 ...µn
(R1 . . . Rn ) uµ1 (R1 ) . . . uµn (Rn ) (1)
n=2
n! R ...R
1 n

where the term with n = 2 is the familiar harmonic term. It is customary to retain terms up to
n = 4, mainly because the cubic term is unbound from below. In many cases, however, stopping
the series at n = 3 in the small-displacements limit still yields sensible results.

It may come as a surprise to learn that the ubiquitous phenomenon of thermal expansion cannot
be explained within the harmonic theory of the crystal lattice. The reason, as we shall see, is
that the pressure of a gas of harmonic phonons is temperature-independent, since the phonon
frequency does not depend on the amplitude of the oscillations. However, by introducing an
anharmonic term, the phonon gas acquires a finite temperature-dependent pressure, which is
ultimately responsible for the thermal expansion of the crystal.

Another physical effect that requires phonon anharmonicity as a key ingredient is thermal con-
ductivity. In a perfect crystal of an insulators, harmonic phonons would never be scattered, so
such a crystal would have infinite thermal conductivity at all temperatures. Scattering of phonons
from lattice imperfections would produce a finite thermal conductivity, but with a completely
wrong temperature dependence. The only way to explain the data is to admit that phonons can
be scattered by other phonons, and this can only occur by abandoning the harmonic approxi-
mation.

As we shall see, only certain phonon-phonon scattering processes, known as umklapp processes

1
can affect the thermal conductivity. This stems from the fact that real momentum and crystal
momentum are separately conserved quantities. For the phonon gas in an insulator (or by
the phonon and electron gasses in a metal), crystal momentum is conserved to within ~ times a
RL vector. However, the phonon gas can exchange total momentum with the center-of-mass
momentum of the crystal (and with particles, such as neutrons, that are scattered in and out of
the crystal). Umklapp processes, for which the total momentum of the interacting phonons is
not conserved, are the only ones capable to affect the conduction of heat.

As an introduction to the subject of anharmonicity, we will review the important issue of real
momentum and crystal momentum conservation. Our treatment will follow that of Ashcroft and
Mermin, Appendix M [1], but we will use some of the results and notation from the previous
lectures. Thermal expansion and thermal conductivity of insulators are also discussed in details
in Ashcroft and Mermin, Chapter 25 [1].

2 Real momentum and crystal momentum conservation

2.1 Symmetry considerations

In the previous lectures, we have seen that all the translation operators act on normal modes as:

t [u(Ri )] = e−ik·t [u(Ri )] (2)

where t is a symmetry translation and the u(Ri ) will, in general, be complex modes. Eq. 2
will in general be true also for wavefunctions, as they also represent a linear space, provided
that the system as a whole has translational symmetry. We can therefore introduce the quantum
mechanical operator T̂R (R being a real lattice vector), and we can always find a complete set of
wavefunctions so that

T̂R ψ = eK·R ψ (3)

The capital K in eq. 3 is deliberate, as we shall see shortly. Note that T̂R is not Hermitian (it has
complex eigenvalues).

The next step is to look at the symmetry of the Hamiltonian. When we say that “the system
possesses translational symmetry” we may mean two things: that the Hamiltonian containing
coordinates shifted by R is the same operator as the original Hamiltonian or that if ψi is an
eigenstate of Ĥ, T̂R ψi is also an eigenstate with the same eigenvalue. We can convince ourselves

2
that in fact these two statements are identical, provided that we define the “shifted” Hamiltonian
as:

Ĥ(r + R) = T̂R Ĥ T̂R−1 = T̂R Ĥ T̂−R (4)

The translational symmetry condition is thus expressed with the statement that Ĥ commutes with
T̂R for all symmetry translations:

T̂R Ĥ(r)T̂−R = Ĥ(r) (5)

Since H commutes with T̂R , we can always find a set of common eigenvectors for both operators
so that

TR ψi = eiK·R ψi (6)

holds with ψi being eigenvectors of the Hamiltonian.

It is obvious that if ψi is a single-particle Block state, K is the same thing as Bloch wavevector
k we have previously introduced. It is also quite clear that if ψi describes a set of n Bloch-state
non-interacting bosons or fermions

X
K= ki (7)
i

where the ki are the individual Bloch wavevectors of each particle. We can therefore call ~K the
total crystal momentum. This is because we can write explicitly the n-particle wavefunction
as a symmetrised or antisymmetrised product of 1-particle wavefunctions. It can be shown that
eq. 7 represents a completely general result, which is valid even in the interacting case (non-
separable wavefunction) and, in particular, in the case of anharmonic lattice vibrations, where
the phonon number is not a constant of the motion. In fact, one can prove the general equation:

T̂R â†k,j T̂−R = eik·R â†k,j (8)

from which one can prove straightforwardly by induction

Y Y
â†ki ,j |0 >= ei( â†ki ,j |0 >
P
ki )·R
T̂R i (9)
i,j i,j

3
We can also write

1
T̂R = ei ~ K̂·R (10)

where K̂ is the total crystal momentum operator (we have use the hat to indicate that it is an
operator), the definition of which should now be obvious when applied to a generic n-particle
state or linear combination thereof.

2.2 Conservation of the total crystal momentum

2.2.1 Non-interacting particles

As we have just seen, if translational symmetry is present, the Hamiltonian commutes with the
translation operator eiK̂·R for a given set of particles, e.g., electrons and phonons. This means
that eiK·R is a constant of the motion — if it has a definite value in the initial state, it must have
the same value subsequently. Also, since this is true for all lattice translations R, it must be true
that:

~Ki = ~Kf + ~τ (11)

where τ is a Reciprocal Lattice Vector. in other words

Crystal momentum is conserved to within ~ times a RL vector.

If more than one set of Bloch non-interacting particles is present, this statement is true for each
set individually, as the operators K̂ for each set all commute with the Hamiltonian.

2.2.2 Interacting Bloch particles

If an interaction term between different particles is present in the Hamiltonian, the crystal mo-
menta of each set of particles will not, in general, be conserved, but the total crystal momentum
for all particles will be conserved to within ~ times a RL vector. This is because the interac-
tion term will in have the form:

X X
w1 (ri − Rn − u(Rn )) + w2 (ri − rj ) (12)
i,n i,j

4
where the ri ’s are the coordinates of the particles (e.g., electrons) and Rn + u(Rn ) is the position
of the nucleus at the lattice node n (including displacements due to phonons). The first and
second term represent interactions with lattice vibrations and between electronic, respectively.
With periodic boundary conditions, this term is clearly invariant by discrete lattice translations,
for example by:

ri → ri + a1
u(Rn ) → u(Rn−1 ) (13)

We may stop for a second and ponder the significance of eq. 13 — particularly the second line is
a shift by one unit cell of the displacement pattern u(Rn ) due to phonons. The consequence of
this is:

For interacting Bloch particles, the sum of the crystal momenta of the particles and of the
phonons is conserved to within ~ times a RL vector.

2.2.3 Interaction with external particles

In absorption, emission and scattering experiments, there will be one or more particles which
will, in their initial state, final state or both, travel far away from the crystal, so that the interaction
terms with the crystal will be negligible. In these states, the operator K̂ for these particles will
be nothing other than the total real momentum. Therefore, we can conclude that

For a system of Bloch particles interacting with themselves, with phonons and with “ex-
ternal” particles, the sum of the crystal momenta of the Block particles and the phonons
and the real momenta of the external particles is conserved to within ~ times a RL vector.

2.3 Normal and umklapp processes

As we have seen, the total crystal momentum (plus the real momenta of the external particles)
is in general conserved to within ~ times a RL vector. One implication is that we can always
define each of the crystal momenta within the first Brillouin zone (by contrast, the real mo-
menta have an absolute significace to within an overall factor accounting from the centre-of-mass
velocity). Once we have done this we can define:

5
 Normal processes are the ones for which the crystal momentum conservation holds exactly,
i.e., the additional RL vector is zero.
Umklapp processes are the ones for which the crystal momentum conservation holds to
within a finite RL vector times ~.

2.4 Conservation of the total real momentum and energy

As we anticipated, the total real momentum of the crystal plus all the external particles is con-
served, in the absence of external forces. This conservation arises from a different invariance,
this time infinitesimal — the invariance by translation of all the coordinates by an arbitrary (even
infinitesimal) vector r:

ri → ri + r
Rn → Rn + r
u(Rn ) → u(Rn + r) (14)

Again, it is important to recognise the difference between eq. 13 and eq. 14 even when r = R
(a real lattice vector). It is not difficult to see that this difference amounts to a translation of
the center of mass of the crystal by R. Two important consequences of this are (we skip the
detailed demonstrations):

Normal processes conserve real momentum. In an umklapp process, the additive term ~τ
corresponds to real momentum transferred to the center of mass of the crystal.

An interesting implication arises when one considers the possibility of selecting different unit
cells (i.e., different Brillouin zones) in certain crystal systems (triclinic, monoclinic). The possi-
bility arises that a process that is classified as normal in one setting may be umklapp in the other.
This amounts to a different repartition of the total momentum between the phonon system and
the center of mass, and is of no physical consequence, as expected.

2.4.1 Energy conservation

As we have seen, in umklapp processes real momentum is transferred to the crystal as a whole.
However, because of the large mass of the crystal, real momentum transfer to the whole crystal
implies essentially no energy transfer. In other words,

6
 Total energy is conserved in both normal and umklapp processes.

3 Thermal expansion

Let us know examine the phenomenon of thermal expansion, and show that it is related in a fun-
damental way to the anharmonicity of the lattice interaction potential. Let us recall the important
thermodynamic expression for the pressure:

 
∂F
P =− (15)
∂V T

where F is the Helmholtz free energy given by

F = U − TS (16)

When applied to a solid, eq. 15 states that the equilibrium state for a solid (e.g., at ambient
pressure) is reached when the external pressure exactly balances the volume derivative of
the free energy at a given temperature.

The Helmholtz free energy of an insulator has two contributions:

• The equilibrium energy U0 of the crystal lattice, i.e., the energy of the “springs” in their equi-
librium positions. By definition, P0 = ∂U
∂V
0
is temperature independent:

∂P0 ∂ 2 U0
= =0 (17)
∂T ∂V ∂T
• The energy and entropy term for the phonon system. The volume derivative of this part of the
Helmholtz free energy can be considered to be the pressure of the phonon gas Pph .

The calculation of Pph in the general case (including anharmonicity) is not difficult but is rather
lengthy, and is reported in the extended version of these notes. Here, we will assume the result
as given and proceed with the derivation of the linear thermal expansion coefficient:

 
1 ∂V 1 (∂P/∂T )V
α= =− (18)
3V ∂T P 3V (∂P/∂V )T

Eq. 18 is obtained by recognising that T , P and V are linked by the equation of state

7
 f (T, P, V ) = 0 (19)

taking the total derivatives with respect to each variable and solving the resulting determinant
equation.

Using the definition of the bulk modulus (inverse compressibility):

B = −V (∂P/∂V )T (20)

eq. 18 becomes:

   
1 ∂P 1 ∂Pph
α= = (21)
3B ∂T V 3B ∂T V

where we have used eq. 17.

We now write the phonon gas pressure:

" #
∂ 1X X  ∂ωs (k)  1
Pph =− ~ωs (k) + −~ β~ω (k)
(22)
∂V 2 k,s k,s
∂V e s −1

The first term is the volume derivative of the zero-point energy; it is temperature independent
and can be ignored for the calculation of the thermal expansion coefficient. The second term
depends on temperature through the phonon population ns (k) = (exp(β~ωs (k)) − 1)−1 , but is
non-zero only if at least some the phonon frequencies depend on volume. Note that if all the
phonon frequencies are volume-independent, there is an exact cancellation between energy
en entropy terms, so that only the zero-point energy survives.

In analogy with case of the single oscillator, we can conclude that:

If the lattice potential is harmonic, the phonon frequencies are volume-independent, and
the thermal expansion coefficient is zero at all temperatures.

From here onward, we follow closely the treatment by Ashcroft and Mermin. In order to describe
the experimental data, it is therefore necessary to go beyond the harmonic approximation. We
can write

 
1 X ∂~ωs (k) ∂
α= − ns (k) (23)
3B k,s ∂V ∂T

8
Remembering the expression for the specific heat:

X ~ωs (k) ∂
cV = ns (k) (24)
k,s
V ∂T

it is natural to define the contribution to individual phonon modes to the specific heat:

~ωs (k) ∂
cvs (k) = ns (k) (25)
V ∂T

the so-called partial Grüneisen parameter — related to the anharmonicity of individual phonons:

V ∂ωs (k) ∂(ln ωs (k))

and the overall Grüneisen parameter

P
k,s cvs (k)γks
γ= P (27)
k,s cvs (k)

With these definitions, the thermal expansion coefficient is written as

γcv
α= (28)
3B

Note that γ is dimensionless, and for typical materials is positive (springs become stiffer as
the volume is reduced) and is usually of the order of unity, although it can be much larger
and sometimes negative in special cases.

3.1 Example 1: the Debye model

In most material, thermal expansion at low temperatures is dominated by the anharmonicity of

9
In the Debye model, all the phonon frequencies scale linearly with the Debye frequency:

k
ω(k) = ωD (29)
(6π 2 n)1/3

where n is the number of unit cell per unit volume. Therefore

∂(ln ωD )
γD = − (30)
∂(ln V )

and

cvD
γ = γD (31)
cv

where we recall the expression for the Debye specific heat

3 Z ΘD /T
x4 e x

T
cvD = 9nkB dx (32)
ΘD 0 (ex − 1)2

3.2 Example 2: the Einstein model

For some materials, the thermal expansion is certain temperature ranges is dominated by the
anharmonicity of a single optical branch. This may occur near phase transitions, where part
of the optical branch goes soft, or in particular geometries admitting the so-called “rigid unit
modes” — anomalously soft and anharmonic optical phonons. In this, case, one can employ the
Einstein model with:

∂(ln ωE )
γE = − (33)
∂(ln V )

and the Einstein specific heat for a single branch:

10
 X x2 ex
cvE = cvE (k) = nkB (34)
k,s
(ex − 1)2

where x = ΘE /T . Therefore

cvE
γ = γE (35)
cv

Debye and Einstein models can be combined to give a better description of the thermal expansion
in all temperature ranges.

3.3 Thermal expansion in metals

The previous derivation of the thermal expansion coefficient only strictly applied to insulators,
since it only took into account the pressure of the phonon gas. It is natural to extend this to metals
by including the pressure of the electron gas. It is an elementary result of the Sommerfeld theory
of metals that:

2 Uel
Pel =
3 V
 
∂Pel 2 el
= c (36)
∂T V 3 v

The complete expression of the linear thermal expansion coefficient is therefore

 
1 ph 2 el
α= γcv + cv (37)
3B 3

In assessing the relative importance of the two terms, it is important to remember that typically
γ ≈ 1 and that, in the Debye model:

cel
v 5 Θ3D
= Z (38)
cph
v 24π 2 T 2 TF

11
where Z is the nominal valence of the metal. When evaluated numerically, eq. 38 leads to the
conclusion that the pressure of the electron gas contributes significantly to the thermal expansion
only below ∼ 10K. The main difference in the thermal expansion of insulators and metals is
therefore in the low-temperature behaviour: ∝ T 3 for insulators, ∝ T for metals.

4 Lattice thermal conductivity

In this section, we will consider in some detail the phenomenon of thermal conductivity in insula-
tors. As we shall see, the experimental data can only be explained by considering phonon-phonon
scattering, and assuming that umklapp processes alone contribute to the thermal mean free path.
Before discussing the specific subject of phonon thermal conductivity, we will provide a short
introduction to the elementary theory of thermal transport, with an emphasis on the comparison
between different transport phenomena. Although far from being numerically accurate, these
considerations provide good order-of-magnitude assessments and a good way to “think about”
these problems. A complete treatment of transport phenomena (with an emphasis on metals) in
the semiclassical approximation is given in Ashcroft and Mermin [1] chapter 13.

4.1 An elementary guide to thermal transport theory

All transport phenomena (mass transport, charge transport, spin transport, heat transport) have
in common some important features: they are non-equilibrium, steady state phenomena, and
they are characterised by a quantity that is being transported, say Q, and a transport speed v.
One defines a current as j = (Q/V )v were V is the volume. For example, for charge transport
the transported quantity is electric charge, so Q/V = −ene and the transport speed is the drift
velocity vd = −eEτ /m∗ , so that je = e2 ne τ /m∗ E. The conductivity (in this case, the electrical
conductivity) is the ratio of the current and the “driving parameter” (E in this case). For thermal
transport, the transported quantity is heat (or entropy) and the velocity is that of the particles
that transport the entropy. However, as we said right at the beginning, transport is essentially a
non-equilibrium phenomenon: whereas an electron retains its charge after a collision, a particle
releases its entropy when it reaches thermal equilibrium through collisions (not all collisions
are capable of this, as we shall see).

Let us look at the scheme shown in fig. 1 (for complete rigour, we consider 3 domains in a
somewhat pedestrian argument; typical phonon mean free paths at RT are a few tens of nm). The
entropy per unit volume at each point is given by:

Z T
cv
s= dT (39)
0 T

12
 Figure 1: Schematic representation of thermal transport.

The entropy balance at the mid position T2 is given by the entropy flowing in the domain and
released therein minus that flowing out:

1
∆sin = (s1 + s3 )
2
∆sout = s2
1 cv cv
∆stot = (T1 − T3 ) = l ∇T (40)
2T T

where l is the mean free path. The transported quantity is the heat transfer per unit volume

Q
= cv l ∇T (41)
V

while the relevant velocity is the average 1-dimensional speed in the x direction — v/3. Note
that we have made the approximation that the average velocities do not depend on temperature,
which is about (but not exactly) correct for both phonons and electrons (but, of course, would be
very wrong for bosons with a mass). Our discussion is summarised in tab. 1

Table 1: Various parameters relevant to thermal and electrical transport.

Transport Transported quantity Velocity Current Conductivity

Electrical −ene −eEτ /m∗ e2 ne τ /m∗ E σ = e2 ne τ /m∗

1 ph
Phonon Thermal cph
v l ∇T c/3 c l
3 v
c∇T κph = 13 cph
v lc

1 el
Electron Thermal cel
v l ∇T vF /3 c l
3 v
vF ∇T κel = 13 cel
v l vF

13
The mean free path l and the relaxation time τ in eq 1 have in general a complex temperature
dependence, making exact theory of transport a very difficult problem. However, we can write:

l = (ns Σ)−1 (42)

Where ns is the density of scatterers and Σ is the scattering cross section; this simply shifts the
problem to determining the temperature (or energy) dependence of the cross section.

Note that, in the free electron model

π2
   
kB T kB T
cel
v = ne kB = ne kB π 2
(43)
2 EF m∗ vF2

and l = vF τ , so the electronic thermal conductivity becomes:

ne τ π 2 2
el
κ = ∗ kB T (44)
m 3

whence the famous Wiedemann-Franz law:

2
π2

el kB
κ = Tσ (45)
3 e

In other words, the ratio between electrical and electronic thermal conductivities does not depend
on any materials-specific parameter, and should be the same (within the employed approxima-
tions) for all materials.

4.2 Thermal conductivity due to phonons

Keeping in mind the result we obtain for the phonon thermal conductivity in tab. 1, it should now
be easy to predict the temperature dependence of the thermal conductivity due to phonons (i.e., in
insulating materials) within the framework of the harmonic model. In a perfect crystal, harmonic
phonons would propagate without hinderance as free particles; furthermore, in an insulator, we
have removed the possibility for them to scatter off electrons. We would therefore conclude that
phonons can only be scattered by crystal imperfections, i.e., defects and, ultimately, the surfaces
of the crystal themselves. Note that, in both cases, real momentum is transferred between the
phonons and the crystal as a whole. These collisions are therefore efficient in thermalising the
phonon energy distribution and therefore in transferring entropy. On this basis, taking Σ as a
constant in eq. 42, we can as a first approximation write l ∝ 1/nd , nd being the defect density.

14
If nd is very small, l is eventually limited by the crystal size. In either case, the harmonic
approximation predicts that the phonon thermal conductivity should be proportional to
the phonon specific heat, that is, ∝ T 3 at least up to temperatures where the optical phonons
become important. In most materials, the relation κph ∝ T 3 should therefore hold in a wide
domain up to a significant fraction of the Debye temperature.

The experimental situation is, however, rather different. The relation κph ∝ T 3 is indeed obeyed
at very low temperatures (typically to ∼ 10K), but the lattice thermal conductivity drops
rather abruptly above this temperature. This is exemplified in fig. 2.

Figure 2: Thermal conductivity of isotopically pure LiF. The different curves at low temperatures
correspond to different crystal sizes: (A) 7.25 mm, (B) 4.00 mm (C) 2.14 mm and (D) 1.06 mm.
The figure is the same as in Ashcroft and Mermin, an is reproduced from P.D. Thacher, Phys.
Rev. 156, 957 (1967).

The crystals employed in the Thacher experiment (fig. 2) where very pure, so that the scattering
from defects was negligible. Therefore, at low temperatures, we can clearly see the effect of crys-
tal size, which determines the mean free path l . In assessing the data, we have to acknowledge
that, above ∼ 10K something starts scattering the phonons more than the crystal boundaries, and
this can be nothing other than other phonons.

15
However, we cannot simply assume that the mean free path is inversely proportional to the
total density of phonons times the scattering cross section, lph ∝ (Σph nph )−1 , as it would
seem logical. It is easy to see that this would give a completely wrong temperature dependence.
In fact, in the Debye model, the low-temperature phonon density scales like T 3 (like the specific
heat), so we would get cphv l
ph
= const. We should therefore expect the thermal conductivity
to saturate at the point where the phonon mean free path becomes smaller than the crystal
size. One way to reproduce the data is to introduce a low-energy cutoff for the phonons that
contribute to reducing the mean free path. Phonons below this energy cut-off will be ineffective.
This amounts to introduce a “density of effective phonons” (effective in reducing the mean free
path). The total and “effective” phonon densities in the Debye model are:

3 Z ΘD /T
x2

ph T
n = 3n dx x
ΘD 0 e −1
3 Z ΘD /T
x2

T
nph
ef f = 3n dx
ΘD ΘC /T ex − 1
(46)

where ΘC is the cut-off temperature, and, as we shall see, is a significant fraction of room tem-
perature. At low temperatures, x  1 and we can ignore the additive term −1 in the denominator
and the terms with e−ΘD /T as a pre-factor in the integral; we obtain (left as an exercise)

3
Θ2C
  
T
nph −ΘC /T 2
T e−ΘC /T


ef f ≈ 3n e (ΘC /T ) + 2(ΘC /T ) + 2 ≈ 3n (47)
ΘD Θ3D

In other words

With the introduction of a cut-off frequency, the number of “effective” phonons grows
exponentially with temperature in the relevant low-temperature range.

With the correct parameters, it is easy to see that the functional form in eq. 47, combined “in
parallel” with the constant mean free path from defects or boundaries, will produce the observed
peak in the thermal conductivity. In fact, the data in fig. 2 are fitted quantitatively with slightly
more complex but analogous expression.

We still have to justify the introduction of the low-energy cut-off. Before doing that, let us take
a brief look at the high-temperature thermal conductivity.

16
4.2.1 High-temperature phonon thermal conductivity

At high temperatures, the total number of phonons becomes ∝ T , while the specific heat saturates
(Dulong-Petit law). Therefore, one can easily predict that

At high temperatures, the thermal conductivity should decrease as 1/T .

This is in fact only approximately confirmed by experimental data. in fact, κph ∝ 1/T x where
x is usually between one and two. The discrepancy is an indication that not all phonons are
equally effective in reducing the mean free path for entropy transfer.

4.2.2 Umklapp processes and low-energy cut-off

The results of the previous section, and the last general statement in particular, can be understood
by the fact that

Only unklapp processes can bring about the release of entropy and the attainment of
thermal equilibrium.

This can be simply understood by looking again at the diagram in fig. 1. Phonons that transfer en-
tropy from T1 to T2 will carry a net crystal momentum, whereas phonons in thermal equilibrium
P
will have i ki = 0

Since normal processes conserve real momentum exactly, they can never restore a thermal
equilibrium configuration of crystal momenta.

The low-energy cut-off arises naturally from the fact that only unklapp processes can reduce the
mean free path. In fact, simple kinematics accounting for the simultaneous energy and crystal
momentum conservation (the latter to within a non-zero RL vector) imposes that

The crystal momentum of all the phonons involved in an umklapp process must be a sig-
nificant fraction of a non-zero reciprocal lattice vector. This means also that their energy
must be a significant fraction of the Debye energy.

This will be seen better in the following section.

17
Figure 3: Multi-phonon processes enabled by cubic — (a) and (b)— and quartic —(c), (d) and
(f) — terms in the Hamiltonian.

4.2.3 Anharmonicity and phonon “collision” processes

in order to understand the kinematics of normal and umklapp processes, it is useful to think of
them as phonon “collisions”. The justification for this is not complex but goes beyond the scope
of this lecture (for more detail, see Ashcroft and Mermin [1], especially Appendices L,M and O.
In essence, one writes a perturbative anharmonic Hamiltonian in terms of the phonon raising and
lowering operators.

• The harmonic part only contains the phonon counting operators a†s (k)as (k), which is read as
the distruction of a single phonon of branch s and momentum k followed by the creation
of the same phonon.

• The cubic term contains terms of the type a†s1 (k1 )a†s2 (k2 )as3 (k3 ) (process (a) in fig. 3) and
a†s1 (k1 )as2 (k2 )as3 (k3 ), where the three crystal momenta are bound by conservation of en-
ergy and momentum — for instance for (a):

k1 = k2 + k 3 + K
ωs1 (k1 ) = ωs2 (k2 ) + ωs3 (k3 ) (48)

• The quartic term contains terms of the type a†s1 (k1 )a†s2 (k2 )a†s3 (k3 )as4 (k4 ) (process (c)), etc.,
where a conservation law similar to eq. 48 holds for the four crystal momenta.

18
 It is important to understand that the conservation laws as in eq. 48 (particularly for the
cubic term) are very restrictive, and, for acoustic phonons, only allow processes near the
zone boundary .

For example, in 1 dimension and with a linear dispersion ω = ck, the only allowed cubic (a)
process would be

k1 = K/2
k2 = k3 = −K/4 (49)

for which the incoming phonon is exactly at the zone boundary. The deviation from linear dis-
persion and the higher dimensionality lift this strict condition, but 3-phonon acoustic collisions
remain restricted to the outer part of the Brillouin zone. Four-phonon collisions are less restricted,
but are also less likely (at least at low temperatures), and still carry a low-energy cut-off.

References

[1] Neil W. Ashcroft and N. David Mermin, Solid State Physics, HRW International Editions,
CBS Publishing Asia Ltd (1976)

19