Journal of Non-Crystalline Solids 621 (2023) 122618

Contents lists available at ScienceDirect

Journal of Non-Crystalline Solids

journal homepage: www.elsevier.com/locate/jnoncrysol

Practical approach to thermal conductivity calculations of small SiO2 samples

Mbaye Ndour a , Philippe Jund a , Laurent Chaput b ,∗
a
ICGM, Univ Montpellier, CNRS, ENSCM, 34293 Montpellier, France
b
Université de Lorraine, LEMTA, Centre National de la Recherche Scientifique, Unité Mixte de Recherche 7563, BP 70239, 54506, Vandœuvre Cedex, France

ARTICLE INFO ABSTRACT

Keywords: We compute the thermal conductivity of amorphous SiO2 combining density functional theory calculations
Thermal conductivity and lifetime models. The supercell we use to represent amorphous SiO2 is small, 78 atoms. We show that
Amorphous semiconductors this rudimentary model is nevertheless sufficient to reproduce the temperature dependence of the thermal
conductivity from 0.1 K to 1000 K if we use a lifetime model at low frequencies accounting for Rayleigh
scattering, the absorption of phonons by two levels systems and structural relaxations. The necessary conditions
needed to observe the plateau in the thermal conductivity around 10 K are also discussed according to this
model.

1. Introduction at present beyond the reach of ab initio calculations. In the remaining

part of the paper, the frequencies are angular frequencies, 𝜔, and there
Heat transport in crystals is relatively well understood since the units will be indicated in THz.
work of Peierls [1]. During the last decade it has even become possible
to compute the lattice thermal conductivity from first principles [2].
On the other hand, heat transport in amorphous, disordered, or 2. Thermal conductivity models
more generally non-periodic systems, is still difficult to compute from
first principles. Many compounds of practical interest, such as a-Si, In the harmonic approximation, the vibrational hamiltonian of a
a-SiO2 , . . . , fall however in this category. Moreover, an adequate condensed matter system is the sum of the kinetic energy of the
modeling would also contribute to develop the intuitive understanding atoms and of the Born–Oppenheimer electronic energy, expanded up to
which is needed to fill the gap between solid and biological matter.
second order in the atomic displacements. The eigenstates of the hamil-
The relevant theory behind the modeling of heat transport in non-
tonian, |{𝑛𝑚 }⟩, are labeled using the occupations 𝑛𝑚 of the harmonic
periodic systems started with the work of Kittel [3] whereas additional
mode 𝑚 having a frequency 𝜔𝑚 .
effects where included later in a series of papers [4–6]. Recently, the
Hardy [9] has shown that in this approximation the harmonic part
theory has been clarified and applied to realistic systems in [7,8].
of the heat current operator can be written as
The goal of this paper is not to develop the theory further, but to
evidence the key assumptions used in the theory and the numerical ∑
𝐒= 𝐒𝑛𝑚 𝑎†𝑛 𝑎𝑚 , (1)
modeling. The paper is therefore organized in two sections. In the 𝑛𝑚
first we recall a simple derivation of the thermal conductivity valid in
periodic and non-periodic systems. In the second section we apply this where 𝑎†𝑛 and 𝑎𝑚 are creation and annihilation operators in the har-
model to amorphous SiO2 , denoted hereafter as a-SiO2 . In particular monic modes 𝑛 and 𝑚, and 𝐒𝑛𝑚 matrix elements of the heat current
we show that the coupling of ab initio calculations with a model of operator in between those harmonic modes.
scattering for the very long wavelength phonons allows to obtain the If 𝜌 is the density matrix of the system, the average heat current is
thermal conductivity over a wide temperature range. Especially, we then computed as
discuss the origin of the plateau observed around 10 K. ∑ ∑
⟨𝐒⟩ = Tr(𝜌𝐒) = 𝐒𝑛𝑚 Tr(𝜌𝑎†𝑛 𝑎𝑚 ) = 𝐒𝑛𝑚 ⟨𝑎†𝑛 𝑎𝑚 ⟩. (2)
We believe that the use of such a model is, presently, practically
𝑛𝑚 𝑛𝑚
unavoidable to describe the long wavelength phonons. Indeed, to access
the sub THz range, the supercell used to represent the amorphous In crystals, the harmonic modes can be labeled using a wavevector, 𝐪,
system would need to contain more than 1.6 million atoms which is and a band index, 𝑗, 𝑛 = (𝐪𝑗). Periodicity imposes that ⟨𝑎†𝑖 𝑎𝑗 ⟩ is diagonal

∗ Corresponding author.
E-mail address: laurent.chaput@univ-lorraine.fr (L. Chaput).

https://doi.org/10.1016/j.jnoncrysol.2023.122618
Received 11 June 2023; Received in revised form 29 August 2023; Accepted 2 September 2023
Available online 12 September 2023
0022-3093/© 2023 Elsevier B.V. All rights reserved.
M. Ndour et al. Journal of Non-Crystalline Solids 621 (2023) 122618

in 𝐪, therefore the reciprocal temperature. We obtain (see Eq. 24 in [5] and Eq. (A.12)
∑ in the appendix)
⟨𝐒⟩ = 𝐒𝐪𝑗𝑗 ′ ⟨𝑎†𝐪𝑗 𝑎𝐪𝑗 ′ ⟩. (3)
( 𝑛0𝐪𝑗 ′ − 𝑛0𝐪𝑗 ) 𝐒𝑥𝐪𝑗𝑗 ′ 𝐒𝑦𝐪𝑗 ′ 𝑗
𝐪𝑗𝑗 ′
𝑖𝑉 ∑
𝜅𝑥𝑦 = Re − (5)
⟨𝑎†𝐪𝑗 𝑎𝐪𝑗 ′ ⟩ need to be computed. In the Boltzmann–Peierls approach, the 𝑇 𝐪𝑗𝑗 ′ ℏ(𝜔𝐪𝑗 ′ − 𝜔𝐪𝑗 ) (𝜔𝐪𝑗 ′ − 𝜔𝐪𝑗 ) + 𝑖𝜂
interband contributions 𝑗 ≠ 𝑗 ′ are neglected, and 𝑛𝐪𝑗 = ⟨𝑎†𝐪𝑗 𝑎𝐪𝑗 ⟩ is
with 𝑛0𝐪𝑗 = 1∕(exp 𝛽𝜔𝐪𝑗 − 1) the Bose–Einstein occupation. The result
computed from a balance equation. In the stationary regime we assume
for amorphous systems in [5] is obtained omitting the 𝐪 sum, and the
that the rate at which phonons leave the mode 𝐪𝑗, due to scattering
𝑗 ′ = 𝑗 contribution, as we would obtain considering amorphous systems
with other phonons, is the same than the rate at which they occupy this
as crystals with an infinitely large unit cell.
mode due to their drift along the temperature gradient existing in the
The above result is obtained from a purely harmonic hamiltonian.
system. The argument for neglecting the interband contributions and
However, it can straightforwardly be generalized to include anhar-
using the Boltzmann equation for computing intraband contributions
monicity if we assume the time evolution of the operators in the
is inherited from the kinetic theory of gases, where molecules are seen
Heisenberg picture to be damped at a rate 𝛾𝐪𝑗 ,
as propagating in between sudden scattering events. The picture is
the same for heat transport. If the vibrations are propagating long 𝑎†𝐪𝑗 (𝑡) = 𝑒𝑖(𝜔𝐪𝑗 +𝑖𝛾𝐪𝑗 )𝑡 𝑎†𝐪𝑗 (6)
enough before being scattered, they sample the periodicity of the
−𝑖(𝜔𝐪𝑗 −𝑖𝛾𝐪𝑗 )𝑡
crystal and the wavevector becomes a good quantum number. At that 𝑎𝐪𝑗 (𝑡) = 𝑒 𝑎𝐪𝑗 . (7)
wavevector, the neglect of the interband contribution is then part of
𝛾𝐪𝑗 is due to phonon–phonon interactions, and may be computed from
the semiclassical approximation used to write the Boltzmann–Peierls
perturbation theory [10].
equation. The scattering of the phonons with other phonons is obtained
For each 𝐪 the current operators in Eq. (4) therefore change by an
from higher order terms in the expansion of the Born–Oppenheimer
extra 𝑒−(𝛾𝐪𝑗 +𝛾𝐪𝑗 ′ )𝑡 factor. This means that the final result is obtained from
energy, treated as a perturbation to the hamiltonian.
Eq. (5) simply replacing 𝜂 by 𝛾𝐪𝑗 + 𝛾𝐪𝑗 ′ . We obtain
This scattering picture cannot be used if the vibrations do not
propagate enough before being scattered. This can be the case in ( 𝑛0𝐪𝑗 ′ − 𝑛0𝐪𝑗 )
𝑉 ∑ (𝛾𝐪𝑗 + 𝛾𝐪𝑗 ′ )
disordered compounds, where the mean free path of the vibrations 𝜅𝑥𝑥 = − |𝐒𝑥𝐪𝑗𝑗 ′ |2
𝑇 𝐪𝑗𝑗 ′ ℏ(𝜔𝐪𝑗 ′ − 𝜔𝐪𝑗 ) (𝜔𝐪𝑗 ′ − 𝜔𝐪𝑗 )2 + (𝛾𝐪𝑗 + 𝛾𝐪𝑗 ′ )2
can be of the order of the interatomic distances. Therefore Allen and
Feldman [5] took a different road to study such systems. In particular 𝑖𝑛𝑡𝑟𝑎 𝑖𝑛𝑡𝑒𝑟
≡ 𝜅𝑥𝑥 + 𝜅𝑥𝑥 (8)
they considered systems where the disorder is sufficient for the vibra-
tions not to propagate, but weak enough for them not to be localized. with
( 𝜕𝑛0𝐪𝑗 )
This allows considering an harmonic hamiltonian. To study localized 𝑉 ∑ 1
modes higher order approximations would be necessary since then heat
𝑖𝑛𝑡𝑟𝑎
𝜅𝑥𝑥 = − |𝐒𝑥𝐪𝑗𝑗 |2 (9)
𝑇 𝐪𝑗 𝜕ℏ𝜔𝐪𝑗 2𝛾𝐪𝑗
transport occurs through virtual transitions.
In harmonic crystals, phonons propagate without being scattered and
and therefore the thermal conductivity is infinite. This can be under-
( 𝑛0𝐪𝑗 ′ − 𝑛0𝐪𝑗 )
stood from the formula for the thermal conductivity shown below, 1 ∑ (𝛾𝐪𝑗 + 𝛾𝐪𝑗 ′ )
𝑖𝑛𝑡𝑒𝑟
𝜅𝑥𝑥 = − |𝐒𝑥𝐪𝑗𝑗 ′ |2 .
Eq. (9), or directly from Eq. (3). Indeed, assume the system is prepared 𝑇 𝑉 𝐪𝑗≠𝑗 ′ ℏ(𝜔𝐪𝑗 ′ − 𝜔𝐪𝑗 ) (𝜔𝐪𝑗 ′ − 𝜔𝐪𝑗 )2 + (𝛾𝐪𝑗 + 𝛾𝐪𝑗 ′ )2
with a distribution in which more phonons travel in one direction than
(10)
in the opposite one. Because phonons are not scattered, this distribution
would remain indefinitely in time. For crystals, the diagonal matrix Noticing that the heat current matrix elements are [9]
elements of the heat current are 𝐒𝐪𝑗𝑗 = ℏ𝜔𝐪𝑗 𝐯𝐪𝑗 ∕𝑉 , where 𝐯𝐪𝑗 is the
1 𝜔𝐪𝑗 + 𝜔𝐪𝑗 ′
group velocity and 𝑉 the volume of the crystal. They are in general 𝐒𝐪𝑗𝑗 ′ = ℏ 𝐯𝐪𝑗𝑗 ′ (11)
𝑉 2
different from zero, and therefore we can obtain a finite heat current 𝜕𝐃(𝐪)
1
possibly without a temperature gradient in the system. This means that 𝐯𝐪𝑗𝑗 ′ = √ ⟨𝐞(𝐪𝑗)| |𝐞(𝐪𝑗 ′ )⟩ (12)
2 𝜔𝐪𝑗 𝜔𝐪𝑗 ′ 𝜕𝐪
the thermal conductivity is infinite.
The same reasoning applies for disordered systems. However in this where 𝐃(𝐪) is the dynamical matrix and 𝐞(𝐪𝑗) the polarizations, the
case the thermal conductivity is finite, even for a purely harmonic intraband and interband contributions to the isotropic thermal conduc-
∑
hamiltonian. This can only be true if the diagonal matrix elements of tivity 𝜅 = 1∕3 𝑥 𝜅𝑥𝑥 can be rewritten as
the heat current operator are zero, 𝐒𝑛𝑛 = 0. In Eq. (2) the heat current 1 ∑
is therefore obtained through the non diagonal elements, 𝐒𝑛𝑚 , 𝑛 ≠ 𝑚. 𝜅 𝑖𝑛𝑡𝑟𝑎 = |𝐯 |2 𝐶𝐪𝑗 𝜏𝐪𝑗 (13)
3𝑉 𝐪𝑗 𝐪𝑗
Therefore we end up with a peculiar situation where heat is transported
1 ∑∑
through non conducting states. 𝐒𝑛𝑛 = 0 is in fact the definition of non 𝜅 𝑖𝑛𝑡𝑒𝑟 = |𝐯 ′ |2
3𝑉 𝐪𝑗 𝑗 ′ ≠𝑗 𝐪𝑗𝑗
conducting states.
Crystals with complicated unit cells can be seen as an intermediate [ ( 𝜔𝐪𝑗 + 𝜔𝐪𝑗 ′ )2 ( 𝑛0𝐪𝑗 ′ − 𝑛0𝐪𝑗 )]( 𝛾𝐪𝑗 + 𝛾𝐪𝑗 ′ )
1
ℏ − 2 2
case. In those systems phonons may have a short enough mean free 𝑇 2 ℏ(𝜔𝐪𝑗 ′ − 𝜔𝐪𝑗 ) (𝜔𝐪𝑗 ′ − 𝜔𝐪𝑗 ) + (𝛾𝐪𝑗 + 𝛾𝐪𝑗 ′ )
2𝜋
path 𝛬𝐪𝑗 for the scattering hypothesis (𝛬𝐪𝑗 ≫ |𝐪| ) not to be fulfilled. (14)
But still, in this case, phonon–phonon interactions should be considered
since heat conduction is not entirely off diagonal. To describe such a with the mode heat capacity 𝐶𝐪𝑗 = 𝐶(𝜔𝐪𝑗 ) = 𝜕(ℏ𝜔𝐪𝑗 𝑛0𝐪𝑗 )∕𝜕𝑇
and the
situation, the approach of Allen and Feldmann can be straightforwardly relaxation time 𝜏𝐪𝑗 = 1∕2𝛾𝐪𝑗 .
generalized. In fact it was already considered in [4]. The thermal conductivity is therefore the sum of an intraband
At first consider a crystal in the harmonic approximation. Then the contribution 𝜅 𝑖𝑛𝑡𝑟𝑎 , which is equal to the result obtained from the
Kubo formula for the thermal conductivity, Boltzmann equation in the relaxation time approximation [2], and an
𝛽 +∞
interband contribution, 𝜅 𝑖𝑛𝑡𝑒𝑟 . It has obviously some similarity with the
𝑉 electronic optical conductivity since it involves band to band transitions
𝜅𝑥𝑦 = Re 𝑑𝜆 𝑑𝑡 𝑒−𝜂𝑡 ⟨𝐒𝑥 (−𝑖ℏ𝜆)𝐒𝑦 (𝑡)⟩, (4)
𝑇 ∫0 ∫0 (See for example Eq. 6.5.58 in [11] and Appendix B).
can be evaluated exactly. In the above formula, 𝜂 is a positive infinites- The physics contained in those equations can be understood as fol-
imal, 𝜂 → 0+ , 𝑥 and 𝑦 label the Cartesian directions and 𝛽 = 1∕𝑘𝐵 𝑇 is lows. If the harmonic modes are delocalized, for a temperature gradient

2
M. Ndour et al.
to exist, the wavefunction of the system needs to be wavepacket like,
and therefore be a superposition of the harmonic eigenstates |𝑖⟩, |𝛹 ⟩ =
∑
𝑖 𝑐𝑖 |𝑖⟩ [5]. |𝑖⟩ is usually written in the occupation number formalism,
but such details are not necessary here. Looking at Eq. (2), we see
that such superposition allows 𝑛 to be different from 𝑚 to give a non
zero expectation value. Therefore, if a temperature gradient exists in
a system with delocalized states, heat can be transported through the
non diagonal elements of the heat current operator.
One should not be surprised of this result since the interband optical
conductivity of electrons also has its origin in the non diagonal matrix
elements of the velocity operator. In this case the electrons interact with
an electromagnetic field which is applied on the system. They make
transitions between electronic states, for example from the valence
to the conduction band, through the coupling of the electromagnetic
field to the dipole operator, or so to say, to the position operator
𝐫. Indeed, the perturbation hamiltonian is 𝐻 ′ = 𝑒𝐫 ⋅ 𝐄. Writing the
matrix element of the position operator in the Bloch basis |𝐤𝑛⟩, we
can split the representative matrix of the position operator (See Eq.
1.10.6 of [12]) into an intraband term, which gives the gradient in
reciprocal space, and an interband term, which, in modern language,
is the non abelian Berry connexion [13]. When computing the average
value of the electrical current density 𝐣 = (−𝑒∕𝑉 )𝐯 = (−𝑖𝑒)∕(ℏ𝑉 )[𝐻, 𝐫],
those two parts of the position operator will give the intraband (Drude)
conductivity and the interband optical conductivity. Such textbook
derivation is summarized in the Appendix B following [14,15].
In the case of phonon heat transport the logic is formally the
same. If a temperature gradient exists in the system, the perturbation
0 In [17], molecular dynamics calculations considering 107 atoms
hamiltonian can be considered to be 𝐻 ′ = − 𝑇1 ∫−∞ ∫ 𝑑 3 𝑟 ∇𝑇 ⋅ 𝐒[5]. It
means that the temperature gradient couples the harmonic eigenstates
through the heat current operator, and, like the position operator for
electrons, the diagonal and non diagonal parts of the representative
matrix of this operator will give the intraband and interband parts of
the thermal conductivity given by Eqs. (13) and (14).
Being proportional to the lifetime 𝜏𝐪𝑗 , the intraband contribution
may be large when 𝜏𝐪𝑗 is large, for example in systems where the
energy is not too anharmonic. On the other hand, Eq. (14) shows
that the contributions from states to the interband part of the thermal
conductivity may become large for those states which are close in
energy, 𝜔𝐪𝑗 ≈ 𝜔𝐪𝑗 ′ . This may happen for almost degenerated states,
or for avoided crossings. Obviously the bands are on average much
closer of each others in a system with many atoms in the unit cell.
Therefore the interband part of the thermal conductivity is more likely
to be important in systems with a complicated atomic structure.
There is a important physical difference between Eqs. (5) and (8).
In Eq. (5) the intraband contribution (𝑗 ′ = 𝑗) will be divergent for
crystals (𝜂 → 0+ ), and zero for disordered systems where only the 𝐪 = 0
terms are considered and the diagonal elements of the heat current
operators are zero. The interband contribution will contain a Dirac
delta function 𝛿(𝜔𝐪𝑗 ′ − 𝜔𝐪𝑗 ), therefore heat transport will occur though
transitions between degenerate modes. This is possible in disordered
systems, but in crystals, at fixed 𝐪, those states are rare and due to
symmetry, and therefore this contribution will be vanishingly small.
When anharmonicity is taken into account, each mode 𝐪𝑗 acquires the
linewidth 𝛾𝐪𝑗 . It means that for the transitions to happen the modes
need only to be almost degenerated, in an energy window 𝛾𝐪𝑗 + 𝛾𝐪𝑗 ′ .
The anharmonicity introduced through 𝛾𝐪𝑗 therefore makes both the
intraband and interband contribution finite in crystals. It removes the
divergence of the intraband contribution and makes the interband
contribution non zero.
The result in Eq. (5) is not new. It is essentially the result obtained
by Allen and Feldmann. More recently, a similar equation has also
been obtained in [7,8]. In [7] the approach is quite similar to ours
since the time evolution of the creation and annihilation operators is
also assumed to be damped. In [8] the approach is different since the
time evolution equation of the one body density matrix is considered.
3
Journal of Non-Crystalline Solids 621 (2023) 122618
The Wigner–Weyl transform of this equation is taken, and an expan-
sion with respect to small wavevectors is performed to retain only
large scale inhomogeneities. The resulting equation is then solved by
decoupling the intraband and interband components [14], to obtain
the intra and inter band contributions to the thermal conductivity.
The derivation presented in [8] goes beyond the one presented in
this section since it allows considering the full collision kernel in the
transport equation, while in Eqs. (6) and (7) only lifetime effects are
considered. This improvement impacts the intraband contribution to
the thermal conductivity. The interband contribution to the thermal
conductivity has however the same form than the one in Eq. (14). In
this equation, the off diagonal elements of the velocity operator used
in [8] are nevertheless different than the ones derived in [9], which we
use in this work.
3. a-SIO𝟐
3.1. Structural model
In this section we would like to compute the thermal conductivity
of amorphous silica using the model of thermal conductivity obtained
in the last section. For this we will approximate a-SiO2 as a peri-
odic system with a unit cell containing 78 atoms. This unit cell is
relatively large for density functional theory (DFT) calculations, but
small compared to previous numerical studies using classical molecular
dynamics [16,17].
interacting through a Lennard-Jones potential were performed. Such
a large system allows to study wavevectors down to a percent of the
size of the Brillouin zone. It is found that at wavevectors below the
boson peak position (𝑞 ≈ 1.5 nm−1 ), eigenvectors are plane waves like,
the dispersion is linear and the linewidth follows an 𝜔4 behavior. This
is the result obtained in the Debye or continuum approximation, with
a Rayleigh scattering. In glasses this approximation however breaks
down at much smaller wavevectors than in crystals, corresponding to
tens of interatomic spacings. Around the peak in the reduced density
of states 𝜌(𝜔)∕𝜔2 , the so called boson peak [18], at 𝑞 = 1 − 2 nm−1
(𝜔 ≈ 6 THz), the dispersion departs from being linear, leading to an
excess in the density of states. The linewidth also changes its behavior
from an 𝜔4 to an 𝜔2 law, which has been observed experimentally [19–
21]. In this range of wavevectors, the eigenstates are still approximately
plane wave like, but this is not true at larger wavevectors, mirroring the
complicated disordered structure.
The above picture means that the Debye model ceases to be valid
at quite low frequency, or equivalently that the atomic structure start
playing a role at quite low wavevectors, on the order of 𝑞 = 1 − 2 nm−1 .
It means that large simulation cells are needed to study such systems.
For example, in [16], a-SiO2 is studied using classical molecular dy-
namics with the modified Beest–Kramer–van Santen (BKS) potential
in a cell containing 4608 atoms. It is shown that the linear size of
the cell, 4.026 nm, which allows to describe states with wavevectors
larger than 1.5 nm−1 , is not enough to observe the 𝜔4 behavior of the
linewidth, although the 𝜔4 − 𝜔2 transition is observed in this range of
wavevectors [21].
At the DFT level, those unit cell sizes are presently beyond our
computational capabilities. Therefore at low frequency a model needs
to be used to compute the thermal conductivity. In the following
we show that a minimal model can however reproduce the plateau
observed in the thermal conductivity around 10 K[22,23].
In [24] we have shown that a-SiO2 can be modeled using a cubic cell
of linear size 𝑎 = 10.558 Å, containing 78 atoms. Following this initial
study, we have computed the phonon states of this structure using the
phonopy software [25]. The forces on the atoms have been obtained
from the VASP code [26–29] using a 2 × 2 × 2 supercell containing
624 atoms. The PBE exchange correlation potential is used [30,31]. The
M. Ndour et al.
Fig. 1. Phonon density of states computed using the DFT and the BKS potential.
Experiments are from [34] (black squares) and [33] (red squares).
Brillouin zone is sampled using a 2 × 2 × 2 uniform grid and the energy
cutoff is fixed to 500 eV. The forces are converged to 10−6 eV Å .
The density of states is shown in Fig. 1. The non analytical correc-
tion to the dynamical matrix, accounting for dipole–dipole interactions,
is included using Born effective charges. In this figure the density of
∑
states, 𝜌(𝜔) = 𝑁1 𝐪𝑗 𝛿(𝜔 − 𝜔𝐪𝑗 ), is computed using a dense mesh
of 𝑁 = 30 × 30 × 30 points to sample the first Brillouin zone. The
dashed red line shows the result of this calculation when the delta
function is approximated by a gaussian function of width 0.97 THz. The
continuous light blue line shows the result of this calculation when the
Brillouin zone integration of the delta functions is performed using the
tetrahedron method. In this method the Brillouin zone is partitioned
into small tetrahedra, and the frequencies 𝜔𝐪𝑗 are linearly interpolated
within those tetrahedra from their values at the corners. This allows to
perform the integration within each tetrahedron analytically. As it can
be seen, this approach appears to be noisier than the gaussian smearing.
Indeed, while the smearing of the delta functions into gaussian func-
tions broaden the frequencies and therefore lead to smooth density of
states, the interpolation of the tetrahedron method may lead to small
denominators leading to oscillations. The original references and the
details about recent implementations can be found in [32].
It is clear from this figure that it is important to use DFT to perform
the calculations since the peak at 75 THz is not obtained using the
BKS potential. The peak at 𝜔 ≈ 25 THz is also obtained by Buchenau
et al. [33] and is shown to be related to the boson peak around 𝜔 = 6
THz in the reduced density of states 𝜌(𝜔)∕𝜔2 .
At wavevectors smaller than 𝑞 = 2𝜋∕𝑎 ≈ 5.9 nm−1 , corresponding
roughly to frequencies below 𝜔 ≈ 𝑣𝐷 𝑞 ≈ 27 THz, where 𝑣𝐷 is the
Debye velocity, the states we compute should, a priori, not be used
to approximate the vibrational states of a-SiO2 . Nevertheless, in the
following, we will use those states to compute the thermal conductivity.
The argument behind this approximation is that around and below
the boson peak the eigenstates are plane wave like and therefore the
differences with our structural model will mostly be captured by a
lifetime model that we now describe.
In [35], at low frequency, the lifetime of long wavelength phonons
is assumed to have the simple form,
( )3
𝑇
( )
1 1 ℏ𝜔 4 1 ℏ𝜔 ℏ𝜔 1 𝑇0
= + tanh + ( )3
𝜏𝑙𝑜𝑤 (𝜔) 𝜏𝑅0 𝑘𝐵 𝑇0 0 𝑘 𝑇
𝜏𝑟𝑒𝑠 𝐵 0 2𝑘𝐵 𝑇 𝜏 0
𝑟𝑒𝑙 1 + 𝑇
𝑘𝐵 𝑇 0
𝑇0 ℏ𝜔
The first term of the above equation is due to Rayleigh scattering,
the second is the absorption rate by two level systems and the third the
absorption rate of structural relaxations. The parameters of this model
have been obtained experimentally from transport measurements [35],
assuming a Debye model at low frequency. They are given by 𝑇0 = 18.25
K, 1∕𝜏𝑅0 = 0.068 THz, 1∕𝜏𝑟𝑒𝑙
0 = 0.00134 THz, 1∕𝜏 0 = 2∕𝜏 0 .
𝑟𝑒𝑠 𝑟𝑒𝑙
Journal of Non-Crystalline Solids 621 (2023) 122618
To better understand the physics contained in Eq. (15), consider
the two levels model [36,37]. A group of atoms is assumed to be
able to tunnel in between two atomic configurations close in energy.
When a phonon travels through the system, it can be scattered by such
states. We can distinguish two mechanisms to change the number of
phonons [38] . The first is the resonant absorption of phonons by the
group of atoms. This is the origin of the subscript res in the second
term of Eq. (15). The second, called relaxation absorption, results
from the modulation of the level spitting of the two levels system
by a time depend strain. This is the origin of the subscript rel in the
third term of Eq. (15). Indeed, when a sound wave travels through
an ensemble of two level systems, it disturbs their thermal equilibrium
and the two level systems try to reach a new equilibrium by emitting
and absorbing phonons. Basically, the phonons transport energy from
the hot atoms to the cold atoms. The two mechanisms can also be
understood considering the equivalence of the two level system with
the hamiltonian of a spin 1∕2 in a magnetic field [39–41]. Then the
transverse part of the magnetization leads to a resonant absorption
while the longitudinal part leads to a relaxation process. Moreover, in
−1
addition to those scattering mechanisms, in disordered materials the
variation of the environment of the two level systems gives rise to a dis-
tribution of the parameters of the model. The formula given in Eq. (15)
are the scattering times of the phonon averaged over this distribution
of parameters. Since such average cannot be done analytically for the
relaxation absorption we have used the approximation given in [42].
This model only describes the scattering of long wavelength
phonons. At higher frequencies 𝜏𝑙𝑜𝑤 is becoming extremely small. It
is therefore possible to write
𝜏 = 𝜏𝑙𝑜𝑤 + 𝜏ℎ𝑖𝑔ℎ (16)
where 𝜏ℎ𝑖𝑔ℎ is the lifetime at high frequency, which itself is assumed to
be small compare to 𝜏𝑙𝑜𝑤 at low frequency.
𝜏ℎ𝑖𝑔ℎ can in principle be computed using molecular dynamics. There-
fore we performed computations using the supercell defined above. The
system is relaxed during 700 ps is the NVE ensemble, and the data is
collected during 630 ps every 0.7 fs to compute the spectral velocity
function at equilibrium. A normal mode analysis using DynaPhoPy[43]
is performed to obtain the lifetime of the phonon modes at 300 K. The
results are shown in Fig. 2. This figure shows 𝜏 with 𝜏ℎ𝑖𝑔ℎ computed
from the above mentioned molecular dynamics calculations (denoted
MD(a=2.111 nm)), using the results of the molecular dynamics calcu-
lations performed in [16] (denoted MD(a=4.026 nm)) and also using
constant values 𝜏ℎ𝑖𝑔ℎ = 10, 1.44, 0.1, 0.01 THz−1 . 1.44 is in fact the
value which gives the best agreement with the thermal conductivity
computed in the next section, in a least squared sens. To plot the
lifetimes corresponding to the molecular dynamics calculations, we
averaged and interpolated the values obtained in those calculations to
obtain a single valued function of 𝜔.
None of the molecular dynamics calculations can reproduce the
experiments in the THz range, possibly because of the too small cells
used in the calculations, however the model of Eq. (15) performs quite
well. It is worth noting that the constant value 𝜏ℎ𝑖𝑔ℎ = 0.1 THz−1
gives an acceptable fit to the experiments around 10 THz whereas the
molecular dynamics calculations using the BKS potential do not.
3.2. Thermal conductivity model
. (15)
To use the lifetime model of the previous section, it is easier to
rewrite the thermal conductivity as an integral over frequencies. Let
us assume that the lifetime is a function of the frequency, 𝜏𝐪𝑗 = 𝜏(𝜔𝐪𝑗 ),
and that the linewidth is smaller than the band splitting, 𝛾𝐪𝑗 + 𝛾𝐪𝑗 ′ ≪
|𝜔𝐪𝑗 − 𝜔𝐪𝑗 ′ |. Then the thermal conductivity can be rewritten as
1
𝜅= 𝑑𝜔 𝜌(𝜔)𝐶(𝜔)𝐷(𝜔), (17)
3𝛺 ∫
4
M. Ndour et al.
Fig. 2. Lifetime of amorphous silica at 𝑇 = 300 K. Experimental results are from [21],
and molecular dynamics calculations denoted as MD(a=4.026 nm) from [16].
Fig. 3. (Top) Velocity spectral functions (Bottom) Diffusivity spectral function.
with 𝜌(𝜔) the density of states, normalized to 3𝑁𝑎 , where 𝑁𝑎 is the
number of atoms in the cell, and 𝐷(𝜔) the diffusivity, which is the sum
of the intraband, 𝜏(𝜔)𝑣2 (𝜔), and the interband contribution
𝐷(𝜔) = 𝜏(𝜔)𝑣2 (𝜔) + 𝐷𝑖𝑛𝑡𝑒𝑟 (𝜔).
(see the Appendix for details, Eqs. (A.25) and (A.26)). The quantities
appearing in the above equations are shown in Fig. 3. The Brillouin
zone integrations are performed either using a gaussian with a smearing
of 0.97 THz, or with the tetrahedron method to approximate the delta
functions in Eqs. (A.25) and (A.26). In the approximation of small
Journal of Non-Crystalline Solids 621 (2023) 122618
linewidths, 𝐷𝑖𝑛𝑡𝑒𝑟 becomes independent of the lifetime, and then in
Eq. (A.26) 𝜂 is chosen to be equal to the gaussian smearing.
Both integration methods give similar results above 𝜔 = 15 THz. At
smaller frequencies, the convergence is more difficult due to the small
number of points close to the center of the Brillouin zone. At 𝜔 < 5
THz we can nevertheless use the Debye approximation to compute the
contribution to the frequency integral in Eq. (17). Above 5 THz we will
use the results obtained with the tetrahedron method on a 30 × 30 × 30
grid.
To be explicit, in the Debye approximation 𝜔𝐪𝑗 ≈ 𝑞𝑣𝑗 therefore
𝜌(𝜔) = (𝛺∕2𝜋 2 )(3∕𝑣3𝐷 )𝜔2 , 𝑣2 𝜌(𝜔) = (𝛺∕2𝜋 2 )(3∕𝑣𝐾 )𝜔2 , and 𝑣2 (𝜔) =
3 ∑
𝑣3𝐷 ∕𝑣𝐾 with 𝛺 = 1176.91 Å , 𝑣𝐷 ≡ ( 3𝑗=1 1∕𝑣3𝑗 )−1∕3 = 46.46 Å THz and
∑3
𝑣𝐾 ≡ ( 𝑗=1 1∕𝑣𝑗 )−1 = 47.71 Å THz. For the interband contribution we
obtain (see Appendix),
𝛺 ∑ ∑
3 3
1 𝜔2
𝐷𝑖𝑛𝑡𝑒𝑟 𝜌(𝜔) ←←←←←←←←←→
← |𝐯𝑗𝑗 ′ |2 . (19)
𝜔→0 2𝜋 2 𝑗=1 𝑗≠𝑗 ′ =1 𝜂 𝑣3
𝑗
This result is however a consequence of the approximation used for the
lifetimes. In a real system, for the acoustic branches, 𝛾𝐪𝑗 goes to zero
as |𝐪| → 0. If we assume 𝛾𝐪𝑗 ∝ 𝜔𝑛𝐪𝑗 , then 𝐷𝑖𝑛𝑡𝑒𝑟 𝜌(𝜔) goes to zero as 𝜔𝑛 if
𝑛 ≥ 1 and otherwise as 𝜔𝑛+2−2𝑛 . We assume this error to be negligible
and, to be consistent with Eq. (15), we consider 𝐷𝑖𝑛𝑡𝑒𝑟) 𝜌(𝜔) = 𝐷0 𝜔,
2
with 𝐷0 = 1.55 Å THz−1 for 𝜔 < 5 THz. This is shown in the inset
of Fig. 3(Bottom) as a black line.
Following Eq. (17) we have computed the thermal conductivity
using the different values of 𝜏ℎ𝑖𝑔ℎ considered in the previous section.
The Debye approximation for 𝑣2 𝜌 and 𝐷𝑖𝑛𝑡𝑒𝑟 𝜌 is used below 5 THz. We
obtain the results shown in Fig. 4. Fig. 4 shows that above ≈ 30 K
the interband contribution should dominate to have agreement with
experiment. When 𝜏ℎ𝑖𝑔ℎ is too large, as in the 𝜏ℎ𝑖𝑔ℎ = 10 example,
the intraband part of the conductivity gives too large a contribution.
The best agreement is obtained for 𝜏ℎ𝑖𝑔ℎ = 1.44, in the means square
sens. At smaller values of 𝜏ℎ𝑖𝑔ℎ , the visual agreement with experiment
seems still good, but closer inspection reveals that those values give
too small contributions to match with the experimental data at large
temperatures. The molecular dynamics data obtained in a large cell
gives a satisfactory agreement with experiments, in the sens that it
reproduces the plateau around 10 K, while with the data obtained in
the smaller cell the plateau around 10 K is less pronounced. However
both molecular dynamics data has roughly the same high temperature
limit. To better understand this result, it is useful to plot the spectral
functions 𝐷(𝜔)𝜌(𝜔) and 𝐶(𝜔)𝐷(𝜔)𝜌(𝜔). Fig. 5 shows 𝐷𝑖𝑛𝑡𝑒𝑟 (𝜔)𝜌(𝜔) and
𝜏(𝜔)𝑣2 (𝜔)𝜌(𝜔), computed at 300 K (full lines) and 10 K (dashed lines),
for each of the approximations we have considered for 𝜏ℎ𝑖𝑔ℎ .
Above 𝜔 ≈ 1 THz, the Rayleigh scattering dominates in 𝜏𝑙𝑜𝑤 , and
the lifetime is therefore becoming essentially temperature indepen-
dent above this frequency. Therefore the different approximations of
𝜏(𝜔)𝑣2 (𝜔)𝜌(𝜔) using different 𝜏ℎ𝑖𝑔ℎ are themselves temperature indepen-
dent since 𝜏ℎ𝑖𝑔ℎ is temperature independent. Indeed, recall that even
the 𝜏ℎ𝑖𝑔ℎ values obtained from molecular dynamics calculations do not
depend on temperature since we use the results of the calculations at
300 K. Below 𝜔 ≈ 1 THz, 𝜏𝑙𝑜𝑤 dominates in 𝜏 therefore all 𝜏(𝜔)𝑣2 (𝜔)𝜌(𝜔)
values using different approximations of 𝜏ℎ𝑖𝑔ℎ become identical.
Above 10 THz the lattice thermal conductivity is mostly due to
the interband contribution and below 1 THz the different 𝜏ℎ𝑖𝑔ℎ models
contribute to the same amount. It is therefore in the 1 to 10 THz range
that the different 𝜏ℎ𝑖𝑔ℎ models impact the lattice thermal conductivity.
In Eq. (17) the heat capacity function 𝐶(𝜔) acts as a dumb bell window
(18) which gives less weight to the high frequency modes than to the low
frequency ones. The size of this window increases with temperature.
Roughly speaking, at 10 K only modes with frequency lower than 10
THz contribute to the thermal conductivity. 𝐶(𝜔)𝐷(𝜔)𝜌(𝜔) is shown
𝜔
in Fig. 6 as well as its integrated value ∫0 𝑑𝑥 𝐶(𝑥)𝐷(𝑥)𝜌(𝑥) for differ-
ent temperatures, 5, 10 and 50 K. It is clear from those figures that
the lattice thermal conductivity increases much more from 5 to 10
5
M. Ndour et al.
Fig. 4. Thermal conductivity considering different values for the high frequency limit of the lifetime time.
Source: The experimental results are from [44].
Fig. 5. Spectral function 𝐷(𝜔)𝜌(𝜔). Full lines show the values at 𝑇 = 300 K, and dashed
lines at 𝑇 = 10 K. The colored lines show 𝜏(𝜔)𝑣2 (𝜔)𝜌(𝜔) for different values of 𝜏ℎ𝑖𝑔ℎ and
the black line 𝐷𝑖𝑛𝑡𝑒𝑟 (𝜔)𝜌(𝜔).
K in the 𝜏ℎ𝑖𝑔ℎ = 10 model than in the other models. Therefore the
plateau is almost absent in this case, as confirmed in Fig. 4. On the
contrary, the increase of the lattice therm conductivity is moderate
in the 𝜏ℎ𝑖𝑔ℎ = 0.01 and 𝜏ℎ𝑖𝑔ℎ = 0.1 models giving well defined
plateaus in Fig. 4. Therefore, we obtain a well defined plateau around
10 K in the thermal conductivity when the increase of the interband
contribution 𝐷𝑖𝑛𝑡𝑒𝑟 (𝜔)𝜌(𝜔) between 1 and 10 THz is best compensated
by the decrease of the intraband contribution, 𝜏(𝜔)𝑣2 (𝜔)𝜌(𝜔), these
two contributions ultimately crossing in the upper end of this interval.
Notice that in our formulation, as in the one of Allen and Feldman [5],
6
Journal of Non-Crystalline Solids 621 (2023) 122618
the interband contribution to the lattice thermal conductivity does not
depend on the lifetime, and is therefore free of parameters.
4. Conclusions
To summarize, in this work, we have reviewed the assumptions
leading to a formulation of the thermal conductivity which can be used
to study periodic and non-periodic systems. The formulation we obtain
is then approximated, in Eqs. (17) and (A.21) to a form which allows
for an easy coupling with the relaxation time models used to describe
the scattering of low frequency states in amorphous materials.
We applied this formulation to a small cell of SiO2 to compute
the thermal conductivity of a-SiO2 and succeeded to reproduce the
experimental thermal conductivity over a wide temperature range. It
is found that the low temperature part of the thermal conductivity is
due to the usual Peierls–Boltzmann intraband conduction, whereas at
high temperature the thermal conductivity is governed by interband
conduction. Both contributions intersect around 40 K. The plateau
observed in the thermal conductivity around 10 K is explained as a
compensation between the decrease of the intraband conduction and
the increase of the interband conduction in the 1 to 10 THz range. It
is in this range that the boson peak is observed in 𝜌(𝜔)∕𝜔2 , therefore
both phenomena involve the same vibrational states.
An obvious weakness of our approach is the use of a lifetime model,
containing free parameters adjusted from experiments. If it turns out
that the use of such models to describe the scattering of very long
wavelength phonons is unavoidable, the parameters they contain may
however be in the reach of ab initio calculations. We plan to address
these questions in a future publication.
M. Ndour et al. Journal of Non-Crystalline Solids 621 (2023) 122618

Appendix A. Thermal conductivity formulas

Linear response theory gives for the thermal conductivity [45]

𝛽 +∞
𝑉
𝜅𝑥𝑦 = Re 𝑑𝜆 𝑑𝑡𝑒−𝜂𝑡 ⟨𝐒𝑥 (−𝑖ℏ𝜆)𝐒𝑦 (𝑡)⟩ (A.1)
𝑇 ∫0 ∫0
with 𝛽 = 1∕𝑘𝐵 𝑇 . The operators are expressed in the Heisenberg picture,
𝑖 𝑖
𝑂(𝜏) = 𝑒 ℏ 𝐻𝜏 𝑂𝑒− ℏ 𝐻𝜏 (A.2)

and the average is performed over the canonical ensemble,

Tr𝑒−𝛽𝐻 𝑂
⟨𝑂⟩ = . (A.3)
Tr𝑒−𝛽𝐻
The Hamiltonian is in general the sum of an harmonic contribution
∑
𝐻0 = 𝐪𝑗 ℏ𝜔𝐪𝑗 (𝑎†𝐪𝑗 𝑎𝐪𝑗 + 1∕2), and anharmonic ones summarized as 𝐻𝑎 ,
therefore 𝐻 = 𝐻0 + 𝐻𝑎 . In the harmonic approximation, 𝐻 ≈ 𝐻0 ,
Hardy [9] has shown that the heat current operator density is given
by [46]
∑
𝐒= 𝐒𝐪𝑗𝑗 ′ 𝑎†𝐪𝑗 𝑎𝐪𝑗 ′ (A.4)
𝐪𝑗𝑗 ′
1 𝜔𝐪𝑗 + 𝜔𝐪𝑗 ′
𝐒𝐪𝑗𝑗 ′ = ℏ 𝐯𝐪𝑗𝑗 ′ (A.5)
𝑉 2
1 𝜕𝐃(𝐪)
𝐒𝐪𝑗𝑗 ′ = √ ⟨𝐞(𝐪𝑗)| |𝐞(𝐪𝑗 ′ )⟩. (A.6)
2 𝜔𝐪𝑗 𝜔𝐪𝑗 ′ 𝜕𝐪

Moreover, in this approximation 𝐻 ≈ 𝐻0 , the time evolution of the

creation and annihilation operators is given by

𝑎†𝐪𝑗 (𝑡) = 𝑎†𝐪𝑗 𝑒𝑖𝜔𝐪𝑗 𝑡 (A.7)

−𝑖𝜔𝐪𝑗 ′ 𝑡
𝑎𝐪𝑗 ′ (𝑡) = 𝑎𝐪𝑗 ′ 𝑒 (A.8)

Fig. 6. (Top) Spectral function 𝐶(𝜔)𝐷(𝜔)𝜌(𝜔). (Bottom) Integrated spectral function
𝜔
∫ 𝑑𝑥𝐶(𝑥)𝐷(𝑥)𝜌(𝑥). Full lines show the values at 𝑇 = 10 K, dashed lines at 𝑇 = 5 K and
dotted lines at 𝑇 = 50 K. It is clear from this graph that the conductivity will increase
more rapidly in the 5 to 50 K temperature range in the 𝜏ℎ𝑖𝑔ℎ = 10 approximation than
in the other approximations.
and for the expectation values with respect to 𝐻0 we have ⟨𝑎†𝐪𝑗 𝑎𝐪𝑗 ′ ⟩0 =
𝛿𝑗𝑗 ′ 𝑛0𝐪𝑗 and ⟨𝑎𝐪𝑗 𝑎†𝐪𝑗 ′ ⟩0 = 𝛿𝑗𝑗 ′ (1 + 𝑛0𝐪𝑗 ). However, when the Hamiltonian is
not harmonic the time evolution of the operators can in general not be
written in a closed form. Nevertheless, when the anharmonicity is not
too strong, the harmonic excitations are still well defined, and the time
evolution can be described using a damping rate 𝛾𝐪𝑗 ,

CRediT authorship contribution statement 𝑎†𝐪𝑗 (𝑡) = 𝑒𝑖(𝜔𝐪𝑗 +𝑖𝛾𝐪𝑗 )𝑡 𝑎†𝐪𝑗 (A.9)
−𝑖(𝜔𝐪𝑗 −𝑖𝛾𝐪𝑗 )𝑡
𝑎𝐪𝑗 (𝑡) = 𝑒 𝑎𝐪𝑗 . (A.10)
Mbaye Ndour: Performed all numerical simulations, Devised the
𝛾𝐪𝑗 can be obtained from the imaginary part of the phonon self en-
strategy of the study and established the theory. Philippe Jund: De-
ergy, and therefore be computed from many body perturbation theory.
vised the strategy of the study and established the theory. Laurent
Numerical calculations can, for example, be found in [2].
Chaput: Devised the strategy of the study and established the theory.
Substituting Eqs (A.9) and (A.10) in Eq. (A.1), we obtain
𝑉 ∑ ∑
𝜅𝑥𝑦 ≈ Re 𝐒𝑥𝐪 𝑗 𝑗 ′ 𝐒𝑦 ′
Declaration of competing interest 𝑇 ′ ′ 1 1 1 𝐪2 𝑗2 𝑗2
𝐪1 𝑗1 𝑗1 𝐪2 𝑗2 𝑗2
( 𝛽 𝑖[𝜔𝐪 𝑗 −𝜔𝐪 𝑗 ′ ](−𝑖ℏ𝜆) )( +∞ 𝑖[(𝜔𝐪 𝑗 −𝜔𝐪 𝑗 ′ )+𝑖(𝛾𝐪 𝑗 +𝛾𝐪 𝑗 ′ )]𝑡 )
The authors declare that they have no known competing finan- 𝑑𝜆𝑒 1 1 1 1 𝑑𝑡𝑒−𝜂𝑡 𝑒 2 2 2 2 2 2 2 2
∫0 ∫0
cial interests or personal relationships that could have appeared to
× ⟨𝑎†𝐪 𝑎𝐪1 𝑗 ′ 𝑎†𝐪 𝑎𝐪2 𝑗 ′ ⟩0
influence the work reported in this paper. 1 𝑗1 1 2 𝑗2 2

For the imaginary time evolution, we have kept the harmonic ap-
proximation. This means that the spectrum has been modified adding
Data availability a width 𝛾𝐪𝑗 to the phonon states, but thermodynamic averages, like
occupation and heat capacities, are computed in the harmonic approx-
Data will be made available on request. imation. In the above equation, the integrals are easily computed, and
by Wick’s theorem the average of the product of the creation and
annihilation operators is obtained as:
Acknowledgments
⟨𝑎†𝐪 𝑎𝐪1 𝑗 ′ 𝑎†𝐪 𝑎𝐪2 𝑗 ′ ⟩0 = 𝛿𝑗1 𝑗 ′ 𝛿𝑗2 𝑗 ′ 𝑛0𝐪 𝑛0𝐪 + 𝛿𝐪1 𝐪2 𝛿𝑗1 𝑗 ′ 𝛿𝑗 ′ 𝑗2 𝑛0𝐪 (𝑛0𝐪 ′ + 1)
1 𝑗1 1 2 𝑗2 2 1 2 1 𝑗1 2 𝑗2 2 1 1 𝑗1 1 𝑗1

This work has been funded by the Institut Carnot ‘‘Chimie-Balard (A.11)
Cirimat’’, the Institut Carnot ‘‘Energie et Environnement de Lorraine’’
This gives
and the Region ‘‘Occitanie’’. High Performance Computing resources
∑ ( ℏ𝛽(𝜔𝐪𝑗 −𝜔𝐪𝑗 ′ ) )
were partially provided by the EXPLOR centre hosted by the University 𝑉 𝑒 −1 0 0
𝜅𝑥𝑦 = Re 𝐒𝑥𝐪𝑗𝑗 ′ 𝐒𝑦𝐪𝑗 ′ 𝑗 𝑛𝐪𝑗 (𝑛𝐪𝑗 ′ + 1)
de Lorraine. 𝑇 𝐪𝑗𝑗 ′
ℏ(𝜔 𝐪𝑗 − 𝜔 𝐪𝑗 ′ )

7
M. Ndour et al. Journal of Non-Crystalline Solids 621 (2023) 122618
( )
𝑖
(A.12)
(𝜔𝐪𝑗 ′ − 𝜔𝐪𝑗 ) + 𝑖(𝛾𝐪𝑗 + 𝛾𝐪𝑗 ′ )
The first contribution from Wick’s theorem, corresponding to equal
time correlation functions, gives zero, since there is no current at
equilibrium. Finally, the isotropic thermal conductivity is obtained as,
( )( )
1∑ 𝑉 ∑ ∑ 𝑥 2 𝑒ℏ𝛽(𝜔𝐪𝑗 −𝜔𝐪𝑗 ′ ) − 1 0 0
𝜅= 𝜅 = |𝐒 | 𝑛 (𝑛 + 1)
3 𝑥 𝑥𝑥 3𝑇 𝐪𝑗𝑗 ′ 𝑥 𝐪𝑗𝑗 ′ ℏ(𝜔𝐪𝑗 − 𝜔𝐪𝑗 ′ ) 𝐪𝑗 𝐪𝑗 ′
( 𝛾𝐪𝑗 + 𝛾𝐪𝑗 ′ )
. (A.13)
2
(𝜔𝐪𝑗 ′ − 𝜔𝐪𝑗 ) + (𝛾𝐪𝑗 + 𝛾𝐪𝑗 ′ )2

Splitting the intra and inter band contributions, and using the
definition of the current matrix elements,
( ( ℏ𝜔𝐪𝑗 )2 )( )
1 ∑ 1
𝜅= |𝐯𝐪𝑗 |2 𝑘𝐵 𝑛0𝐪𝑗 (𝑛0𝐪𝑗 + 1)
3𝑉 𝐪𝑗 𝑘𝐵 𝑇 2𝛾𝐪𝑗 Fig. A.7. Thermal conductivity considering different values of a constant relaxation
∑ ∑ ( 𝜔𝐪𝑗 + 𝜔𝐪𝑗 ′ )2 time. The total thermal conductivity is shown using full line. The interband contribution
1
+ |𝐯 ′ |2 ℏ is shown using dashed line. The interband contribution is shown using a full black line.
3𝑉 𝑇 𝐪𝑗 𝑗 ′ ≠𝑗 𝐪𝑗𝑗 2 Source: The experimental results are from [44].
( )( )
𝑒ℏ𝛽(𝜔𝐪𝑗 −𝜔𝐪𝑗 ′ ) − 1 0 0 𝛾𝐪𝑗 + 𝛾𝐪𝑗 ′
𝑛 (𝑛 + 1)
ℏ(𝜔𝐪𝑗 − 𝜔𝐪𝑗 ′ ) 𝐪𝑗 𝐪𝑗 ′ (𝜔𝐪𝑗 ′ − 𝜔𝐪𝑗 )2 + (𝛾𝐪𝑗 + 𝛾𝐪𝑗 ′ )2 [ ∑[ ∑ 𝜂 ] ]
1 1
+ 𝑑𝜔 𝐶(𝜔) |𝐯𝐪𝑗𝑗 ′ |2 𝛿(𝜔 − 𝜔𝐪𝑗 )
and if we recognize that 3𝛺 ∫ 𝑁 𝐪𝑗 𝑗 ′ ≠𝑗 2
(𝜔𝐪𝑗 − 𝜔𝐪′ 𝑗 ′ ) + 𝜂 2

1 (A.21)
𝜏𝐪𝑗 = (A.14)
2𝛾𝐪𝑗 1 1
( ℏ𝜔𝐪𝑗 )2 = 𝑑𝜔 𝐶(𝜔)𝜏(𝜔)𝑣2 𝜌(𝜔) + 𝑑𝜔 𝐶(𝜔)𝐷𝑖𝑛𝑡𝑒𝑟 𝜌(𝜔) (A.22)
𝜕 (ℏ𝜔𝐪𝑗 )2 ( 𝜕𝑛0𝐪𝑗 ) 3𝛺 ∫ 3𝛺 ∫
𝐶𝐪𝑗 = 𝐶(𝜔𝐪𝑗 ) = ℏ𝜔𝐪𝑗 𝑛0𝐪𝑗 = 𝑘𝐵 𝑛0𝐪𝑗 (𝑛0𝐪𝑗 + 1) = −
𝜕𝑇 𝑘𝐵 𝑇 𝑇 𝜕ℏ𝜔𝐪𝑗 ≡ 𝜅 𝑖𝑛𝑡𝑟𝑎 + 𝜅 𝑖𝑛𝑡𝑒𝑟 (A.23)
(A.15)
At the second line we have assumed that the lifetime is a function of
(𝑒ℏ𝛽(𝜔𝐪𝑗 −𝜔𝐪𝑗 ′ ) − 1)𝑛0𝐪𝑗 (𝑛0𝐪𝑗 ′ + 1) = 𝑛0𝐪𝑗 ′ − 𝑛0𝐪𝑗 (A.16) the frequency, and defined,

we obtain 1 ∑
𝜌(𝜔) = 𝛿(𝜔 − 𝜔𝐪𝑗 ), (A.24)
𝑁 𝐪𝑗
1 ∑
𝜅= |𝐯 |2 𝐶𝐪𝑗 𝜏𝐪𝑗 1 ∑
3𝑉 𝐪𝑗 𝐪𝑗 𝑣2 𝜌(𝜔) = |𝐯 |2 𝛿(𝜔 − 𝜔𝐪𝑗 ), (A.25)
𝑁 𝐪𝑗 𝐪𝑗
[ ( 𝜔𝐪𝑗 + 𝜔𝐪𝑗 ′ )2 ( 𝑛0𝐪𝑗 ′ − 𝑛0𝐪𝑗 )]
1 ∑∑ 1 1 ∑ ∑
[ 𝜂 ]
+ |𝐯𝐪𝑗𝑗 ′ |2 ℏ − 𝐷𝑖𝑛𝑡𝑒𝑟 𝜌(𝜔) = |𝐯 ′ |2 𝛿(𝜔 − 𝜔𝐪𝑗 ). (A.26)
3𝑉 𝐪𝑗 𝑗 ′ ≠𝑗 𝑇 2 ℏ(𝜔𝐪𝑗 ′ − 𝜔𝐪𝑗 ) 𝑁 𝐪𝑗 𝑗 ′ ≠𝑗 𝐪𝑗𝑗 (𝜔𝐪𝑗 − 𝜔𝐪′ 𝑗 ′ )2 + 𝜂 2
( 𝛾𝐪𝑗 + 𝛾𝐪𝑗 ′ )
(A.17) 𝐷𝑖𝑛𝑡𝑒𝑟 is the non-diagonal/interband contribution to the diffusivity,
(𝜔𝐪𝑗 ′ − 𝜔𝐪𝑗 )2 + (𝛾𝐪𝑗 + 𝛾𝐪𝑗 ′ )2 and 𝑣2 (𝜔) is the velocity spectrum. It is clear that 𝐷𝑖𝑛𝑡𝑟𝑎 (𝜔) = 𝜏(𝜔)𝑣2
The Boltzmann result appears as the intraband contribution. The in- plays the role of diffusivity for the diagonal/intraband contribution,
terband contribution is a purely quantum effect since it involves tun- therefore we define
neling through the non diagonal elements of the velocity operator. 𝐷(𝜔) = 𝐷𝑖𝑛𝑡𝑟𝑎 (𝜔) + 𝐷𝑖𝑛𝑡𝑒𝑟 (𝜔) = 𝜏(𝜔)𝑣(𝜔)2 + 𝐷𝑖𝑛𝑡𝑒𝑟 (𝜔). (A.27)
It can be understood as diffusion. This is best seen if we assume
that |𝜔𝐪𝑗 ′ − 𝜔𝐪𝑗 | ≫ 𝛾𝐪𝑗 + 𝛾𝐪𝑗 ′ . In this limit, considered by Allen and The splitting of the diffusivity into its diagonal and non diagonal
Feldman [5], the Lorentzian becomes a delta function, and the first contributions is sometimes useful to compare to experiments, since
square bracket tends to the mode resolved heat capacity. This gives usually only the diagonal contribution is accessible. Finally we write
for the thermal conductivity
1 ∑ |𝐯𝐪𝑗 |2 1 ∑ ∑ |𝐯𝐪𝑗𝑗 ′ |2
𝜅= 𝐶𝐪𝑗 𝜏𝐪𝑗 + 𝐶𝐪𝑗 𝜋𝛿(𝜔𝐪𝑗 ′ − 𝜔𝐪𝑗 ) (A.18) 1
𝑉 𝐪𝑗 3 𝑉 𝐪𝑗 3 𝜅= 𝑑𝜔 𝐶(𝜔)𝐷(𝜔)𝜌(𝜔). (A.28)
𝑗 ′ ≠𝑗 3𝛺 ∫
1 ∑ |𝐯𝐪𝑗 |2 1 ∑ This is the approximation we use to compute numerically the lattice
= 𝐶𝐪𝑗 𝜏𝐪𝑗 + 𝐶 𝐷 (A.19)
𝑉 𝐪𝑗 3 𝑉 𝐪𝑗 𝐪𝑗 𝐪𝑗 thermal conductivity.
In Fig. A.7, we show the results of the computation for SiO2 as-
where we have defined the mode diffusivity as: suming a constant relaxation time 𝜏. The other parameters used for
∑ |𝐯𝐪𝑗𝑗 ′ |2 this calculation are explained in the main text. It is clear that this
𝐷𝐪𝑗 = 𝜋𝛿(𝜔𝐪𝑗 ′ − 𝜔𝐪𝑗 ). (A.20) constant relaxation time approximation is not able to reproduce the
3
𝑗 ′ ≠𝑗 experimental measurements. Indeed, the calculation for 𝜏 = 10 THz−1
overestimates the thermal conductivity at high temperature, but under-
It is important to realize that in this small linewidth limit, the
estimates this quantity at low temperature. A better approximation is
intraband contribution is diverging as 𝛾 −1 , while the interband contri-
discussed in the main text and shown in Fig. 4.
bution is independent of 𝛾. It is the argument used to claim that the
interband contribution is less sensitive to scattering than the intraband
Appendix B. Electrical conductivity
contribution.
From a computational point of view, it is suitable to rewrite Linear response theory gives the Kubo formula for the electrical
Eq. (A.19) as: conductivity [47],
[ ∑ ]
1 1 +∞ 𝛽
𝜅= 𝑑𝜔 𝐶(𝜔) 𝜏𝐪𝑗 |𝐯𝐪𝑗 |2 𝛿(𝜔 − 𝜔𝐪𝑗 ) 𝜎𝑥𝑦 (𝜔) = 𝑉 𝑑𝑡 𝑑𝜆 𝑒−𝜂𝑡 𝑒𝑖𝜔 ⟨𝐣𝑥 (−𝑖ℏ𝜆)𝐣𝑦 (𝑡)⟩, (B.1)
3𝛺 ∫ 𝑁 𝐪𝑗 ∫0 ∫0

8
M. Ndour et al. Journal of Non-Crystalline Solids 621 (2023) 122618

which, when specialized to independent electron described by Bloch [16] J.M. Larkin, A.J.H. McGaughey, Thermal conductivity accumulation in amor-
states |𝐤𝑛⟩, gives the Kubo–Greenwood formula [15,48,49], phous silica and amorphous silicon, Phys. Rev. B 89 (2014) 144303, http:
//dx.doi.org/10.1103/PhysRevB.89.144303.
∑ ( 𝑓𝐤𝑛 0 − 𝑓0 )
𝐤𝑛′ ⟨𝐤𝑛|𝐣𝑥 |𝐤𝑛′ ⟩⟨𝐤𝑛′ |𝐣𝑦 |𝐤𝑛⟩ [17] G. Monaco, S. Mossa, Anomalous properties of the acoustic excitations in glasses
𝜎𝑥𝑦 (𝜔) = 𝑖ℏ𝑉 − . (B.2) on the mesoscopic length scale, Proc. Natl. Acad. Sci. 106 (2009) 16907–16912.
𝐤𝑛𝑛′
𝜖𝐤𝑛 − 𝜖𝐤𝑛′ ℏ𝜔 + (𝜖𝐤𝑛 − 𝜖𝐤𝑛′ ) + 𝑖(𝛾𝐤𝑛 + 𝛾𝐤𝑛′ )
[18] U. Buchenau, Neutron scattering investigations of the boson peak, Condens.
The computation of the matrix elements of the current density Matter Phys. (2019).
[19] C. Masciovecchio, G. Baldi, S. Caponi, L. Comez, S. Di Fonzo, D. Fioretto, A.
operator 𝐣 = (−𝑒∕𝑉 )𝐯 allow to split the above formula into computable Fontana, A. Gessini, S.C. Santucci, F. Sette, G. Viliani, P. Vilmercati, G. Ruocco,
intra and inter band contributions. We have Evidence for a crossover in the frequency dependence of the acoustic attenuation
𝑖𝑒 in vitreous silica, Phys. Rev. Lett. 97 (2006) 035501, http://dx.doi.org/10.1103/
⟨𝐤𝑛′ |𝐣|𝐤𝑛⟩ = ⟨𝐤𝑛′ |𝐫𝐻 − 𝐻𝐫|𝐤𝑛⟩, (B.3) PhysRevLett.97.035501.
𝑉ℏ
[20] G. Baldi, V.M. Giordano, G. Monaco, B. Ruta, Sound attenuation at terahertz
and using Eq. 1.10.6 in [12] frequencies and the boson peak of vitreous silica, Phys. Rev. Lett. 104 (2010)
195501, http://dx.doi.org/10.1103/PhysRevLett.104.195501.
⟨𝐤′ 𝑛′ |𝐫|𝐤𝑛⟩ = −𝑖𝛿𝑛𝑛′ ∇𝐤 𝛿𝐤𝐤′ + 𝑖𝛿𝐤𝐤′ ⟨𝑢𝐤𝑛′ |∇𝐤 |𝑢𝐤𝑛 ⟩, (B.4) [21] G. Baldi, V.M. Giordano, G. Monaco, Elastic anomalies at terahertz frequencies
and excess density of vibrational states in silica glass, Phys. Rev. B 83 (2011)
with 𝑢𝐤𝑛 the periodic part of the Bloch functions. We obtain
174203, http://dx.doi.org/10.1103/PhysRevB.83.174203.
−𝑒 1 −𝑒 𝜖𝐤𝑛 − 𝜖𝐤𝑛′ [22] P. Jund, R. Jullien, Molecular-dynamics calculation of the thermal conductivity
⟨𝐤𝑛′ |𝐣|𝐤𝑛⟩ = 𝛿′ ∇ 𝜖 + ⟨𝑢𝐤𝑛′ |∇𝐤 |𝑢𝐤𝑛 ⟩ (B.5)
𝑉 (𝑛 𝑛 ℏ 𝐤 𝐤𝑛 𝑉 ℏ ) of vitreous silica, Phys. Rev. B 59 (1999) 13707–13711, http://dx.doi.org/10.
−𝑒 1103/PhysRevB.59.13707.
≡ 𝛿𝑛′ 𝑛 𝐯𝐤𝑛 + (1 − 𝛿𝑛′ 𝑛 )𝐯𝐤𝑛′ 𝑛 , (B.6) [23] R.B. Stephens, Low-temperature specific heat and thermal conductivity of non-
𝑉
crystalline dielectric solids, Phys. Rev. B 8 (1973) 2896–2905, http://dx.doi.org/
and therefore 10.1103/PhysRevB.8.2896.
𝑖𝑛𝑡𝑟𝑎 𝑖𝑛𝑡𝑒𝑟 [24] M. Ndour, L. Chaput, P. Jund, Classical molecular dynamics study of small
𝜎𝑥𝑦 (𝜔) = 𝜎𝑥𝑦 (𝜔) + 𝜎𝑥𝑦 (𝜔) (B.7) samples of amorphous silica : structural and dynamical properties, J. Non Cryst.
( 𝜕𝑓 0 )
𝐯𝑥𝐤𝑛 𝐯𝑦𝐤𝑛
𝑒2 ∑
Solids 569 (2021) 120995, http://dx.doi.org/10.1016/j.jnoncrysol.2021.120995.
𝐤𝑛
=𝑖 − 𝐤𝑛 (B.8) [25] A. Togo, I. Tanaka, First principles phonon calculations in materials science, Scr.
𝑉 𝐤𝑛 𝜔 + 2𝑖𝛾𝐤𝑛 𝜕𝜖 Mater. 108 (2015) 1–5.
( 𝑓0 − 𝑓0 ′ ) [26] G. Kresse, J. Hafner, Ab initio molecular dynamics for liquid metals, Phys. Rev.
𝑒2 ∑ 𝐯𝑥𝐤𝑛𝑛′ 𝐯𝑦𝐤𝑛′ 𝑛 𝐤𝑛 𝐤𝑛
+𝑖 − . (B.9) B 47 (1993) 558–561, http://dx.doi.org/10.1103/PhysRevB.47.558.
𝑉 𝐤𝑛≠𝑛′ ℏ𝜔 + (𝜖𝐤𝑛 − 𝜖𝐤𝑛′ ) + 𝑖(𝛾𝐤𝑛 + 𝛾𝐤𝑛′ ) 𝜖𝐤𝑛 − 𝜖𝐤𝑛′ [27] G. Kresse, J. Hafner, Ab initio molecular-dynamics simulation of the liquid-metal–
amorphous-semiconductor transition in germanium, Phys. Rev. B 49 (1994)
The splitting of the thermal conductivity into intra and inter contribu- 14251–14269, http://dx.doi.org/10.1103/PhysRevB.49.14251.
tions in Eqs. (9) and (14) is analogous to this well known splitting of the [28] G. Kresse, J. Furthmüller, Efficiency of ab-initio total energy calculations for
metals and semiconductors using a plane-wave basis set, Comput. Mater. Sci. 6
(optical) conductivity, which for example allow to separate the Drude
(1) (1996) 15–50.
(intra) peak from the interband contribution in simple metals [15,50]. [29] G. Kresse, J. Furthmüller, Efficient iterative schemes for ab initio total-energy
calculations using a plane-wave basis set, Phys. Rev. B 54 (1996) 11169–11186,
References http://dx.doi.org/10.1103/PhysRevB.54.11169.
[30] J.P. Perdew, K. Burke, M. Ernzerhof, Generalized gradient approximation made
[1] R. Peierls, Quantum Theory of Solids, in: International Series of Monographs on simple, Phys. Rev. Lett. 77 (1996) 3865–3868, http://dx.doi.org/10.1103/
Physics, Clarendon Press, 1955. PhysRevLett.77.3865.
[2] A. Togo, L. Chaput, I. Tanaka, Distributions of phonon lifetimes in Brillouin [31] J.P. Perdew, K. Burke, M. Ernzerhof, Generalized gradient approximation made
zones, Phys. Rev. B 91 (2015) 094306, http://dx.doi.org/10.1103/PhysRevB.91. simple [Phys. Rev. Lett. 77, 3865 (1996)], Phys. Rev. Lett. 78 (1997) 1396,
094306. http://dx.doi.org/10.1103/PhysRevLett.78.1396.
[3] C. Kittel, Interpretation of the thermal conductivity of glasses, Phys. Rev. 75 [32] A. Togo, L. Chaput, T. Tadano, I. Tanaka, Implementation strategies in phonopy
(1949) 972–974, http://dx.doi.org/10.1103/PhysRev.75.972. and phono3py, J. Phys.: Condens. Matter 35 (35) (2023) 353001, http://dx.doi.
[4] A. Auerbach, P.B. Allen, Universal high-temperature saturation in phonon and org/10.1088/1361-648X/acd831.
electron transport, Phys. Rev. B 29 (1984) 2884–2890, http://dx.doi.org/10. [33] U. Buchenau, M. Prager, N. Nücker, A.J. Dianoux, N. Ahmad, W.A. Phillips,
1103/PhysRevB.29.2884. Low-frequency modes in vitreous silica, Phys. Rev. B 34 (1986) 5665–5673,
[5] P.B. Allen, J.L. Feldman, Thermal conductivity of disordered harmonic solids, http://dx.doi.org/10.1103/PhysRevB.34.5665.
Phys. Rev. B 48 (1993) 12581–12588, http://dx.doi.org/10.1103/PhysRevB.48. [34] A. Hiramatsu, M. Arai, H. Shibazaki, M. Tsunekawa, T. Otomo, A.C. Hannon, S.M.
12581. Bennington, N. Kitamura, A. Onodera, Investigation on permanently densified
[6] J.L. Feldman, M.D. Kluge, P.B. Allen, F. Wooten, Thermal conductivity and vitreous silica by means of neutron scattering, Physica B 219 (1996) 287–289.
localization in glasses: Numerical study of a model of amorphous silicon, Phys. [35] J.E. Graebner, B. Golding, L.C. Allen, Phonon localization in glasses, Phys. Rev.
Rev. B 48 (1993) 12589–12602, http://dx.doi.org/10.1103/PhysRevB.48.12589. B 34 (8) (1986) 5696.
[7] L. Isaeva, G. Barbalinardo, D. Donadio, S. Baroni, Modeling heat transport in [36] P.w. Anderson, B.I. Halperin, c.M. Varma, Anomalous low-temperature thermal
crystals and glasses from a unified lattice-dynamical approach, Nature Commun. properties of glasses and spin glasses, Philos. Mag. A J. Theor. Exp. Appl. Phys.
10 (2019). 25 (1) (1972) 1–9, http://dx.doi.org/10.1080/14786437208229210.
[8] M. Simoncelli, N. Marzari, F. Mauri, Unified theory of thermal transport in [37] W.A. Phillips, Tunneling states in amorphous solids, J. Low Temp. Phys. 7 (3–4)
crystals and glasses, Nat. Phys. 15 (8) (2019) 809–813, http://dx.doi.org/10. (1972) 351–360, http://dx.doi.org/10.1007/BF00660072.
1038/s41567-019-0520-x. [38] J. Jäckle, On the ultrasonic attenuation in glasses at low temperatures, Z. Phys.
[9] R.J. Hardy, Energy-flux operator for a lattice, Phys. Rev. 132 (1963) 168–177, 257 (3) (1972) 212–223, http://dx.doi.org/10.1007/BF01401204.
http://dx.doi.org/10.1103/PhysRev.132.168. [39] J. Jäckle, L. Piché, W. Arnold, S. Hunklinger, Elastic effects of structural
[10] L. Chaput, A. Togo, I. Tanaka, G. Hug, Phonon-phonon interactions in transition relaxation in glasses at low temperatures, J. Non-Cryst. Solids 20 (3) (1976)
metals, Phys. Rev. B 84 (2011) 094302, http://dx.doi.org/10.1103/PhysRevB.84. 365–391, http://dx.doi.org/10.1016/0022-3093(76)90119-8.
094302. [40] S. Hunklinger, W. Arnold, in: W.P. Mason, R. Thurston (Eds.), 3 - Ultrasonic
[11] J. Callaway, Quantum Theory of the Solid State, No. ptie. 1, Academic Press, Properties of Glasses at Low Temperatures, in: Physical Acoustics, vol. 12, Aca-
1974. demic Press, 1976, pp. 155–215, http://dx.doi.org/10.1016/B978-0-12-477912-
[12] W. Jones, N. March, Theoretical Solid State Physics, in: Dover books on physics 9.50008-4.
and chemistry, vol. 1, Wiley-Interscience, 1973. [41] R.O. Behunin, P. Kharel, W.H. Renninger, H. Shin, F. Carter, E. Kittlaus, P.T.
[13] D. Vanderbilt, Berry Phases in Electronic Structure Theory: Electric Polarization, Rakich, Long-lived guided phonons in fiber by manipulating two-level systems,
Orbital Magnetization and Topological Insulators, in: Titolo Collana, Cambridge 2016, arXiv:1501.04248.
University Press, 2018. [42] D.P. Jones, W.A. Phillips, Thermal conductivity of vitreous silica, Phys. Rev. B
[14] W. Kohn, J.M. Luttinger, Quantum theory of electrical transport phenomena, 27 (1983) 3891–3894, http://dx.doi.org/10.1103/PhysRevB.27.3891.
Phys. Rev. 108 (1957) 590–611, http://dx.doi.org/10.1103/PhysRev.108.590. [43] A. Carreras, A. Togo, I. Tanaka, DynaPhoPy: A code for extracting phonon
[15] P. Allen, Chapter 6 electron transport, in: S.G. Louie, M.L. Cohen (Eds.), quasiparticles from molecular dynamics simulations, Comput. Phys. Comm. 221
Conceptual Foundations of Materials, in: Contemporary Concepts of Condensed (2017) 221–234, http://dx.doi.org/10.1016/j.cpc.2017.08.017.
Matter Science, vol. 2, Elsevier, 2006, pp. 165–218, http://dx.doi.org/10.1016/ [44] R.C. Zeller, R.O. Pohl, Thermal conductivity and specific heat of noncrystalline
S1572-0934(06)02006-3. solids, Phys. Rev. B 4 (6) (1971) 2029.

9
M. Ndour et al.
[45] G.V. Chester, The theory of irreversible processes, Rep. Progr. Phys. 26 (1) (1963)
411, http://dx.doi.org/10.1088/0034-4885/26/1/311.
[46] L. Chaput, Nanostructured Semiconductors: Amorphization and Thermal Prop-
erties, Jenny Stanford Publishing, 2017 (Chapter Thermal Conductivity of
Disordered Systems),
[47] R. Kubo, Statistical-mechanical theory of irreversible processes. I. General theory
and simple applications to magnetic and conduction problems, J. Phys. Soc.
Japan 12 (6) (1957) 570–586, http://dx.doi.org/10.1143/JPSJ.12.570.
10
Journal of Non-Crystalline Solids 621 (2023) 122618
[48] D. Greenwood, The Boltzmann equation in the theory of electrical conduction
in metals, Proc. Phys. Soc. 71 (4) (1958) 585, http://dx.doi.org/10.1088/0370-
1328/71/4/306.
[49] H. Rathgen, M.I. Katsnelson, Symmetry assumptions, Kramers–Kronig transforma-
tion and analytical continuation in Ab initio calculations of optical conductivities,
Phys. Scr. 2004 (T109) (2004) 170, http://dx.doi.org/10.1238/Physica.Topical.
109a00170.
[50] P.B. Johnson, R.W. Christy, Optical constants of copper and nickel as a function
of temperature, Phys. Rev. B 11 (1975) 1315–1323, http://dx.doi.org/10.1103/
PhysRevB.11.1315.