See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/263006816

Modeling the thermal conductivities of the zinc antimonides ZnSb and

Zn4Sb3

Article in Physical Review B · December 2013

DOI: 10.1103/PhysRevB.89.024304

CITATIONS READS

119 754

3 authors, including:

Georg K. H. Madsen
Interdisciplinary Centre for Advanced Materials Simulation
180 PUBLICATIONS 29,529 CITATIONS

SEE PROFILE

All content following this page was uploaded by Georg K. H. Madsen on 20 February 2015.

The user has requested enhancement of the downloaded file.

 PHYSICAL REVIEW B 89, 024304 (2014)

Modeling the thermal conductivities of the zinc antimonides ZnSb and Zn4 Sb3
Lasse Bjerg and Bo B. Iversen
Center for Materials Crystallography, Department of Chemistry and iNANO, Aarhus University, Denmark

Georg K. H. Madsen*
Department of Atomistic Modelling and Simulation, ICAMS, Ruhr-Univesität Bochum, Germany
(Received 20 July 2013; revised manuscript received 13 December 2013; published 13 January 2014)
ZnSb and Zn4 Sb3 are interesting as thermoelectric materials because of their low cost and low thermal
conductivity. We introduce a model of the lattice thermal conductivity which is independent of fitting parameters
and takes the full phonon dispersions into account. The model is found to give thermal conductivities with
the correct relative magnitudes and in reasonable quantitative agreement with experiment for a number of
semiconductor structures. The thermal conductivities of the zinc antimonides are reviewed and the relatively
large effect of nanostructuring on the zinc antimonides is rationalized in terms of the mean free paths of the heat
carrying phonons. The very low thermal conductivity of Zn4 Sb3 is found to be intrinsic to the structure. However,
the low-lying optical modes are observed in both Zn-Sb structures and involve both Zn and Sb vibrations, thereby
strongly questioning dumbbell rattling. A mechanism for the very low thermal conductivity observed in Zn4 Sb3
is identified. The large Grüneisen parameter of this compound is traced to the Sb atoms which coordinate only
Zn atoms.

DOI: 10.1103/PhysRevB.89.024304 PACS number(s): 63.20.dk, 63.20.Ry, 66.70.Df

I. INTRODUCTION were both conjectures and only limited quantitative models

have been introduced for the thermal conductivity.
Thermoelectric materials are capable of interconverting
Probably, the simplest models with quantitatively predictive
heat and electricity. As for any heat engine, the conversion
power are based on Klemens-Callaway models of the Debye
efficiency is limited by the Carnot efficiency. Moreover, the
solid. While such simple models cannot be expected to
dimensionless figure of merit of a material is limiting,
reach the same precision as a solution of the full Boltzmann
S2σ transport equation [22,23], they have been found to give the
zT = T, (1) correct order of magnitude for the thermal conductivity over
κe + κl
a surprising large range of solids [24]. They are, hence, well
where S is the Seebeck coefficient, σ is the electric conductiv- suited for quantifying a discussion of the structural origins of
ity, κe and κl are the electric and lattice thermal conductivities, a “low” thermal conductivity, as seen, e.g., in a discussion
and T is the temperature. Generally, a zT above 1 is needed of the influence of lone pairs on the thermal conductivity
for sufficient conversion efficiency. [25]. Furthermore, their computational simplicity makes them
Besides having a high zT , a good thermoelectric material interesting in their own right in, e.g., a high-throughput
should be made from abundant and nontoxic components. framework where they can provide a first estimate of thermal
For this reason, Zn and Sb based compounds have attracted conductivity. The aim of the present paper is thus twofold.
interest. Zn4 Sb3 first gained interest because of it having a First of all, we take a step beyond earlier ab initio based
lattice thermal conductivity of just 0.6 W m−1 K−1 at room models, which take only the linear part of the acoustic bands
temperature, whereby zT values above 1 were obtained [1]. into account [26], by integrating over the full set of group
Though progress has been made in increasing its stability velocities, thereby taking the information available from ab
[2–4], the inherent instability of Zn4 Sb3 , which decomposes initio calculations better into account. We validate the model
into ZnSb, Sb, and ZnO upon cycling in air [2,5,6], has by showing how it can reliably reproduce the ordering of the
increased the interest in ZnSb as a thermoelectric material measured thermal conductivities in a number of compounds,
[7–16]. At room temperature, single crystalline ZnSb has been including the structurally complex zinc antimonides.
found to have a thermal conductivity of 3–4 W m−1 K−1 In the second part, we investigate the origins of the
[17], and for polycrystalline and doped samples, values below low thermal conductivity of ZnSb and Zn4 Sb3 . The thermal
2 W m−1 K−1 have been obtained [7,8,11,12,16]. A zT of 0.9 conductivities of the zinc antimonides are reviewed and it is
at 550 K has been reported [7,12]. shown that to explain the observed thermal conductivities,
Based on the observation of abundant Zn vacancies and grain boundary and impurity scattering must be introduced.
interstitials, the low thermal conductivity of Zn4 Sb3 has been However, it is also found that the very low thermal conductivity
attributed to the high crystalline disorder of Zn4 Sb3 [18–20]. of Zn4 Sb3 is inherent of the structure. The low-frequency
A different interpretation related the low thermal conductivity optical peaks in both Zn-Sb compounds can be attributed to
to an observed low-lying optical phonon which was attributed cooperative Zn and Sb motion, questioning dumbbell rattling.
to Sb-Sb dumbbell rattling [21]. However, these mechanisms On the contrary the low thermal conductivity is related to
an exceptionally large squared Grüneisen parameter and it is
shown how it originates from anharmonic vibration of the Sb
*
georg.madsen@rub.de atoms which are not bonded in dumbbells.

1098-0121/2014/89(2)/024304(8) 024304-1 ©2014 American Physical Society

LASSE BJERG, BO B. IVERSEN, AND GEORG K. H. MADSEN PHYSICAL REVIEW B 89, 024304 (2014)

II. BACKGROUND This is double the value used elsewhere in the literature
[27,28]. The factor of 2 was discovered by Julian [30] as a
Within the relaxation-time approximation, the lattice ther-
correction to Leibfried and Schlömann’s expression [29].
mal conductivity is given as
The idea behind the present model is to take the full ab

1 dq 2 initio calculated phonon dispersion into consideration when
κl (T ) = v τiq Ciq , (2) evaluating Eq. (2), but to replace τiq by an ω and T dependent
3 i 8π 3 iq
model τ , given by Eqs. (3) and (7). Based on the discussion by
where the sum is over all phonon bands; the integral is over Slack we use different definitions of the Debye temperature and
all q points in the first Brillouin zone; viq is the group velocity Grüneisen parameters [24]. The Debye temperature is obtained
of a given phonon mode; τiq is the mode relaxation time; and from the second moment of the whole phonon spectrum
Ciq is the mode heat capacity depending only on the mode  
2 ∞ ω2 g(ω)dω
frequency ωiq and the temperature. −1/3 5
θ̃D = n 0
∞ , (10)
Except for the relaxation time, the quantities in Eq. (2) can 3kB2 0 g(ω)dω
be extracted directly from the phonon dispersion relation of
a material. The relaxation-time model used in this paper is where n is atoms per unit cell. The overall Grüneisen parameter
based on the model of three-phonon anharmonic scattering is defined as
developed by Slack and co-workers [24,27,28]. There are, at  dq
i 3 γiq Ciq V ∂ωiq
least, two different models in the literature. One is based on γ = 8πdq , γiq = − , (11)
the empirical finding that a good expression for the relaxation i C
8π 3 iq
ω iq ∂V
time around the Debye temperature θD is [27,28]
and is directly related to the observable thermal expansion
−1 T coefficient. In Slack’s model, the sum over the squared
τM = pω2 e−θD /3T . (3)
θD Grüneisen parameter is used, thereby avoiding a cancellation
between the acoustic modes, which can have a negative
A somewhat different model was given by Slack [24] based on
Grüneisen parameter, and the optical modes
work on noble gas crystals by Leibfried and Schlömann [29],
 dq 2
and Julian [30], γ C
i 8π 3 iq iq
√ γ̃ =
2
 dq . (12)
0.849 × 3 3 4 C
8π 3 iq
κl (θD ) =
Slack i
20π 3 (1 − 0.514γ −1 + 0.228γ −2 )
  Furthermore, since mainly the acoustic bands contribute to the
kB θD 2 kB MV 1/3 thermal conductivity, the sum is be performed only over modes
× , (4) having an energy less than kB θ̃D .
 γ 2
The model gives a fitting-parameter-free expression for
where γ is the Grüneisen parameter, V is the unit cell volume calculating the thermal conductivity of a material. It has the
of and M is the average atomic mass. advantage that it takes the full phonon band structure into
The two models can be compared by considering a Debye account when evaluating the mean velocities. Furthermore,
model where all three branches have the same sound speed s, γ̃ 2 , s, and θ̃D can be evaluated quite straightforwardly from the
relaxation time, and Debye temperature, where Eq. (2) gives phonon band structure, thereby avoiding a certain arbitrariness
   in making a Debye model for a given solid.
Debye kB kB T 3 θD /T x 4 ex dx
κl (T ) = τ x . (5) Using this model, and the computational methods described
2π 2 s  0 (e − 1)2 in the next section, we have calculated the lattice thermal
By inserting Eq. (3) in Eq. (5) and evaluating the thermal conductivity κlM of four reference compounds Si, Mg2 Si, ZnTe,
conductivity at the Debye temperature and CdTe and the zinc antimonides ZnSb and Zn6 Sb5 at 300 K
√ (Table I). We compare the values obtained using our model to
2  1 2 x
3
−1 kB θD e x e dx those of a simpler model given by Slack
κl (θD ) = p . (6)
2π s 0 (ex − 1)2
2
θD
κlSlack (T ) = κlSlack (θD ) . (13)
If we equate Eqs. (6) and (4), we obtain T
1 − 0.514γ −1 + 0.228γ −2 2 γ 2 The factor θD /T is introduced to comply with the observed
p= . (7) T −1 behavior of the thermal conductivity at temperatures
0.0948 kB θD MV 1/3 s
above the Debye temperature [24]. Both models are found
Inserting a Debye temperature of to predict the right order of magnitude of the thermal
 conductivities (Table I). For all compounds, the Slack model
2 s
3 6π predicted the highest κ values and the present model is found
θD = (8)
V kB to give results in better agreement with experiment. Shown in
Fig. 1 is the relative thermal conductivity with respect to mean
into Eq. (7), for ordinary γ values between 0.6 and 1.8, we get
free path for Si and Mg2 Si compared to literature data obtained
the expression
using the full Boltzmann transport equation [31,32]. It is seen
γ 2 that good agreement is obtained, further lending credibility to
p≈2 . (9) the model.
Ms 2

024304-2
MODELING THE THERMAL CONDUCTIVITIES OF THE . . . PHYSICAL REVIEW B 89, 024304 (2014)

TABLE I. Calculated average sound velocity, Eq. (19); Grüneisen parameter, Eq. (12); reduced Debye temperature, Eq. (10); ω2 τM ; and
Expt.
(v/ω)2 . The thermal conductivity calculated using Eqs. (2), (3), and (7), κlM , Eq. (13), κlSlack , and experiment, κl , at 300 K are also reported.

Exp
s θ̃D (ω2 τM ) (v/ω)2 κlM κlSlack κl
Compound (m/s) γ̃ 2
(K) 1015 s−1 10 J m−1 K−1
−15
(W m−1 K−1 ) (W m−1 K−1 ) (W m−1 K−1 )
Si 5260 0.66 521 33.8 5.9 200 234 157 [33]
Mg2 Si 5318 1.39 296 3.5 4.2 15 17 11 [20]
ZnTe 2413 0.62 194 6.1 4.0 25 49 18 [34]
CdTe 1952 1.00 155 2.5 3.6 9 21 8 [34]
ZnSb 2241 0.58 92 2.3 2.9 6.6 10.6 3.5 [8]
Zn6 Sb5 1805 2.59 77 0.3 2.3 0.7 1.4 0.5-1.4 [35,36]

To understand the different results obtained using our and will be used to quantify the discussion of the thermal
method and Slack’s method, we can take ω2 τM out of the conductivity in the following.
integral in Eq. (2), and change the integral to a frequency
integral over a velocity density of states, gv2 ,
III. COMPUTATIONAL METHODS
κlM (T ) = (ω2 τM )(v/ω)2 (14) Since Zn4 Sb3 is highly disordered, a somewhat represen-
with tative unit cell is very large. Therefore, we performed calcu-
lations on the ordered, interstitial-free Zn6 Sb5 unit cell. The
 ∞
C(ω) difference between the structures is given in Table II. A force-
(v/ω)2 = gv2 (ω) dω, (15) constant matrix was obtained with the finite-displacements
0 ω2
 method in 2 × 2 × 2 supercells. The forces were calculated us-
1 dq 2 ing density functional theory as implemented in VASP [37,38].
gv2 (ω) = v δ(ω − ωiq ), (16)
3 i 8π 3 iq From this, the dynamical matrix can be found and diagonalized
to obtain the phonon mode eigenvectors and eigenfrequencies.
where δ is the Dirac delta function. In Fig. 2, the density of We used the Perdew-Burke-Ernzerhof (PBE) functional [39]
states g is shown together with gv2 and gv2 C/ω2 for ZnTe and with projector augmented wave basis sets [40,41] and a cut-off
ZnSb. Clearly, the acoustic modes give the largest contribution energy of 360 eV. For ZnSb, we used a 6 × 5 × 5 k grid,
to κlM , but the lowest optic branches cannot be neglected. The whereas a 4 × 4 × 4 k grid was used for Zn6 Sb5 .
Debye model is depicted by the dashed line in Fig. 2. For a We calculated the phonon dispersion relation using
Debye solid, gv2 C/ω2 = C/(2π 2 s) up to an energy of kB θ̃D . PHONOPY [42]. The phonon modes were calculated on a
As C is almost ω independent at T = 300 K, this corresponds 32 × 26 × 24 q grid for ZnSb and a 32 × 32 × 32 q grid for
to an almost square area. κlSlack corresponds to taking the Zn6 Sb5 . The integrals, Eqs. (2), (10), and (12), were calculated
integral under the dashed curve, and we found that the simpler using a histogram method and are converged at the given grids.
method in all cases gave a too big integral. Furthermore, both We repeated this procedure for unit cell volumes which were
(ω2 τM ) and (v/ω)2 are only weakly temperature dependent ±3% of the relaxed volume. For the changed volumes, the
above the Debye temperature and can thus be viewed as atomic positions were relaxed, and the lattice constants were
materials dependent constants. They are listed in Table I allowed to change while keeping the unit-cell volume constant.

100 Si, our model

Si, Esfarjani 2011
Mg2Si, our model
Mg2Si, Li 2012
80
Cumulative κ (%)

60

40

20

0
10-9 10-8 10-7 10-6 10-5
Λ (m)

FIG. 1. (Color online) Calculated cumulative thermal conductiv-

ity as a function of mode mean free path for Si and Mg2 Si at FIG. 2. (Color online) g, gv2 , and gv2 C/ω2 for (a) ZnTe and (b)
300 K compared to literature data calculated using the full Boltzmann ZnSb. Also shown with a dashed line is gv2 C/ω2 for a Debye model.
equation [31,32]. C is evaluated at T = 300 K and is almost ω independent.

024304-3
LASSE BJERG, BO B. IVERSEN, AND GEORG K. H. MADSEN PHYSICAL REVIEW B 89, 024304 (2014)

TABLE II. Comparison of the experimental structure Zn4 Sb3 10

Böttger 2010
[18], and the theoretical structure Zn6 Sb5 . Zn4 Sb3 , furthermore, Böttger 2011
contains three interstitial sites with 5%–6% occupancy. Eklöf 2013
8 Model
L = 100 nm
Atom Parameter Zn4 Sb3 Zn6 Sb5

κl (Wm K )
−1 −1
6
Space group R3c R3c
a 12.2282 12.3310
c 12.4067 12.3011 4
Zn1 x 0.0792 0.0813
y 0.2439 0.2432 2
z 0.4033 0.4023
Occupancy 0.8999 1
0
Sb1 x 0.3555 0.3564 100 200 300 400 500 600 700
y 0 0 T (K)
z 0.25 0.25
Occupancy 1 1 FIG. 3. (Color online) Calculated and experimental [8,11,16]
Sb2 x 0 0 thermal conductivity of ZnSb. The first sample was unprocessed,
y 0 0 whereas the latter two were ball milled and ground, respectively.
z 0.1364 0.1347 Little anisotropy was found in the calculated values, which are shown
Occupancy 1 1 as averages over the x, y, and z directions.

Here, we have taken boundary (B) and vacancy (V) scattering

The velocities and Grüneisen parameters are calculated using into account. For a crystallite with shortest dimension L, it is
the phonon eigenvectors eiq : normally assumed that the boundary scattering relaxation time
is [43]
1 ∂D
viq = eiq | |eiq , (17) τB−1 = v/L. (21)
2ωiq ∂q
A vacancy (V) will give rise to a mass difference equal to
V ∂D
γiq = − 2
eiq | |eiq . (18) the mass of the missing atom, MV . Furthermore, the breaking
2ωiq ∂V of bonds will contribute in a way corresponding to a mass
difference of 2M. This will cause a relaxation time from
Volume changes of ±1% were also tested and the results vacancies of [44]
changed only little. To determine s, a spatial average of v  2
close to the point was taken for each acoustic branch. Then, −1 MV V ω4
an average was taken over the three acoustic bands τV = c +2 , (22)
M 4π ns 3
−1/3 where c is the vacancy concentration.
1 1
3
s= . (19) Figure 3 shows the calculated thermal conductivity as
3 i=1 si3 a function of the temperature for ZnSb compared to ex-
perimental measurements. Few reports have been made on
IV. RESULTS AND DISCUSSION measurements of the thermal conductivity of ZnSb single
crystals. In 1966 Shaver and Blair reported a value of
Experimentally measured thermal conductivities can be
strongly sample dependent, and it can be necessary to consider
the contributions of grain boundaries and defects to the 3.0
Caillat 1997
scattering of the heat carrying phonons in order to understand Pedersen 2006
Bhattacharya 2006
the low thermal conductivity. At the same time Table I shows 2.5 Wu 2009
Toberer 2010
good overall agreement between the model and experiment Dasgupta 2011
2.0 Model
κl (Wm K )

without including such effects. This is even the case for the
-1 -1

ordered 6:5 model of the Zn4 Sb3 structure, which would

1.5
indicate that the low thermal conductivity is inherent of the
structure. The discussion is thus split into a quantitative
1.0
comparison with experiment and a more qualitative discussion
of the origin of the low thermal conductivity in Zn4 Sb3 .
0.5

A. Thermal conductivity of the zinc antimonides 0.0

100 200 300 400 500 600 700
For the present comparison we write the resistive relaxation T (K)
time as
FIG. 4. (Color online) Calculated and experimental [1,35,36,46–
−1
τ −1 = τM + τB−1 + τV−1 . (20) 48] thermal conductivity of Zn6 Sb5 .

024304-4
MODELING THE THERMAL CONDUCTIVITIES OF THE . . . PHYSICAL REVIEW B 89, 024304 (2014)

100 also clear that the samples not exhibiting a drop at 250 K
have a higher thermal conductivity than the ones that do.
The discrepancy between these results agrees well with the
80
Cummulative κl (%)

study by Dasgupta et al. [47] who performed a systematic

study of synthesis conditions and their effect on the transport
60
properties [47]. It was found that Zn-depleted Zn4 Sb3 has a
thermal conductivity more than double that of more Zn-rich
40 samples. The measurements were only performed above room
temperature, but it is seen (Fig. 4), that the thermal conductivity
20 of the Zn-depleted phases agree well with the results where
ZbSb no drop at 250 K is observed, whereas Zn-rich samples agree
Zn6Sb5
0
Zn6Sb5; 10% vacancies with the low thermal conductivity observed for samples which
10-9 10-8 10-7 10-6 10-5 have a drop at 250 K.
Λ (m) Figure 4 shows that the calculated thermal conductivity is in
good agreement with experiment. Considering the simplicity
FIG. 5. (Color online) Cumulative thermal conductivity as a
of the model and the complexity of the real Zn4 Sb3 structure,
function of mode mean free path for ZnSb and Zn6 Sb5 at 300 K.
the agreement is probably somewhat fortuitous. However, the
general ability of the model to predict the relative magnitudes
of the thermal conductivities is again validated. It would thus
3.5 W m−1 K−1 at room temperature [17]. This value was
appear that the low thermal conductivity of Zn4 Sb3 is indeed
also found by Böttger et al. in 2010 on a polycrystalline
related to the structure itself. To get an estimate of the role
sample [8]. Other experiments show a considerably lower
of defects in lowering the thermal conductivity of Zn4 Sb3 , we
thermal conductivity. Böttger et al. reported on ball-milled
included scattering from the experimentally observed 10% Zn
samples in 2011 [11], and Eklöf et al. on ground samples in
vacancies [18] through Eq. (22). At 300 K this resulted in a
2013 [16]. Both groups reported values below 2 W m−1 K−1 .
47% reduction of κ.
All reported thermal conductivities are the right order of
In order to predict the effect of microstructuring on the ther-
magnitude, but lower than our calculated values. The lower
mal conductivity, we calculated the cumulative contributions
thermal conductivities might arise from smaller grain sizes
to κ as a function of phonon mean free path (MFP)
and less dense samples, and Fig. 3 illustrates how including a
boundary scattering based on L = 100 nm, Eq. (21), leads to = τ v. (23)
a thermal conductivity in good agreement with experiment.
In Fig. 4, a review of the measured thermal conductivities The results are shown in Fig. 5. It is seen that the main
for Zn4 Sb3 is shown. Battacharya et al. [36] did not observe heat carrying phonons have MFPs between 10 and 1000 nm.
a drop in the thermal conductivity at 250 K, where an order- This disagrees with the model of Schweika et al. [21] which
disorder phase transition of the Zn sublattice in Zn4 Sb3 occurs estimated an extremely low MFP of the heat carrying phonons
[45]. Schweika et al. used the absence of this drop to rule out of 0.96 nm. The difference in MFP has important implications
a dominant role from Zn disorder on the thermal conductivity for the effect of nanostructuring. At 300 K, 64% of the heat in
[21]. However, this argument overlooked the measurements by ZnSb is carried by phonons with a mean free path longer than
Pedersen et al. [35], which were also available at the time, and 100 nm, whereas the same is true for 42% of the heat at 800 K,
show a drop in the thermal conductivity at 250 K. This drop which explains the large effect of nanostructuring (Fig. 3).
was later confirmed by measurements by Wu et al. [46]. It is Figure 5 also shows the MFP of Zn6 Sb5 . It is interesting

25 25

20 20

15 15
E (meV)

E (meV)

10 10

5 5
Sb1
Sb Sb2
Zn Zn
0 0
Γ X S Y Γ Z U R T Z0 0.1 0.2 0.3 Γ X L Γ Z B P Z0 0.1 0.2 0.3
-1 -1
Projected DOS per atom (meV ) Projected DOS per atom (meV )

(a) (b )

FIG. 6. (Color online) Phonon dispersion relations, and pure and weighted DOS for (a) ZnSb and (b) Zn6 Sb5 .

024304-5
LASSE BJERG, BO B. IVERSEN, AND GEORG K. H. MADSEN PHYSICAL REVIEW B 89, 024304 (2014)

to note how the heat carrying phonons have a shorter MFP 0.10
ZnSb, Sb
than in ZnSb, which would make the effect of nanostructuring ZnSb, Zn
smaller. It is also seen how the vacancies lower the contribution Zn6Sb5, Sb1
0.08 Zn6Sb5, Sb2
from the short MFP phonons, whereby the relative influence Zn6Sb5, Zn
of nanostructuring is increased.
0.06

|γμ|
B. Origin of the low thermal conductivity
The calculated phonon dispersion relations and projected 0.04
density of states (DOS) for ZnSb and Zn6 Sb5 are shown
in Fig. 6. Both structures have low-lying optical branches 0.02
between 4 and 8 meV. In Table III the frequencies of
the Raman-active modes are listed and compared to the
experimentally observed peak energies [49]. The general 0.00
0 5 10 15 20 25
agreement is similar to what is found for the ZnSb structure
E (meV)
[14] and reasonable when taking into account the overesti-
mated unit-cell volume, and resulting underestimated peak FIG. 7. (Color online) Root mean square of the mode Grüneisen
energies, obtained when using the PBE functional. This is parameters for ZnSb and Zn6 Sb5 .
also seen in the average speed of sound for Zn6 Sb5 , which is
underestimated (Table I), compared to the experimental value
of 2310 m/s [1]. Significant Zn-vibrational contributions are experimentally and in the model. This is mainly due to a shorter
found for the low energy modes, and we found the Sb1 atoms relaxation time in ZnSb, whereas the velocity term (v/ω)2
to have a larger contribution than the Sb2 atoms. We also accounts for about one-third of the reduction. Comparing the
investigated the relative contributions of Zn and Sb vibrations 6:5 zinc antimonide to the 1:1, there is a drop of thermal
to the individual eigenstates in the Brillouin zone, and in conductivity by almost a factor 8, again in good agreement
all cases found Zn contributions above 40%. This would with experiment. It is seen that this is mainly due to a large
disagree with the analysis by Schweika et al., who estimated decrease in (ω2 τM ). Table I shows that this decrease can almost
that in Zn4 Sb3 , these optical branches originated from Sb-Sb entirely be attributed to the increase in the squared Grüneisen
dumbbell rattling [21], but it aligns well with the conclusions parameter of the acoustic bands γ̃ 2 [Eq. (12)]. The very low
drawn by Jund et al. [14]. Because of the high mobility of Zn
in Zn4 Sb3 [50], we would expect the Zn atoms to be loosely
bound in this compound. The same is, apparently, true in ZnSb,
since we also found soft Zn modes in ZnSb, which agrees well
with our earlier observation of a low formation energy of Zn
vacancies [15]. We would argue that the loose binding of the
Zn atoms makes it impossible for the antimony dumbbells to
rattle independently.
The low thermal conductivity of the Zn-Sb compounds
is quantified by (ω2 τM ) and (v/ω)2 [Eq. (14)] in Table
I. Comparing the two compounds ZnTe and ZnSb, the
thermal conductivity of ZnTe is about four times larger both

TABLE III. Energies of the Raman-active modes in meV. The (a)

experimental values are taken from Ref. [49].

PBE
114.1
99.5 2.80
A1g Eg Expt.
94.9
99.8 2.84
4.76 2.79

6.46
7.10 8.03 58.9 80.4 2.68
10.29 11.54 2.74

12.14 12.13 13.52 83.1

117.2

13.19 15.38
14.19
16.79 17.66 18.98 (b) (c)
18.45 19.16 21.22
20.44 21.55 22.49 FIG. 8. (Color online) (a) Structure of the Zn6 Sb5 structure used
22.17 to model the 4:3 phase. (b) and (c) show the coordination of the Sb1
and Sb2 atoms, respectively.

024304-6
MODELING THE THERMAL CONDUCTIVITIES OF THE . . . PHYSICAL REVIEW B 89, 024304 (2014)

thermal thermal conductivity in the 6:5 or 4:3 phase is thus the relative magnitudes of thermal conductivities in a number
due to a strong increase in the anharmonicity. of semiconductor structures. The model is used to rationalize
In Fig. 7, we introduce an atom projected contributions the relatively large effect of nanostructuring on the thermal
Grüneisen DOS conductivity of the zinc antimonides in terms of the mean free
   paths of the heat carrying phonons.
V 2 ∂D 2
dq
γμ2 (ω) = 2
φ μiq | |eiq  δ(ω − ωiq ) 3 , Low-lying optical modes involving both Zn and Sb
2ω i
∂V 8π vibrations are found in both ZnSb and Zn6 Sb5 , strongly
(24) questioning dumbbell rattling. Furthermore, in agreement with
experiment, we found the thermal conductivity of Zn6 Sb5 to be
where φμiq is a version of the eigenvector containing only approximately eight times lower than in ZnSb. A mechanism
nonzero elements on those entries which are related to the for the very low thermal conductivity observed in Zn4 Sb3 has
investigated atom μ. It is seen that the Zn and Sb2 atoms been identified and traced to the anharmonic motion of the Sb1
in Zn6 Sb5 contribute to the Grüneisen parameter in a very atoms which coordinate only Zn atoms.
similar way as those in the ZnSb, which is understandable Since the 6:5 structure is an ordered model of the disordered
by comparing the 6:5 structure (Fig. 8) to that of ZnSb (see 4:3 phase, disorder is not necessary to explain the low
Fig. 2 in Ref. [9]). The coordinations of atoms Zn and Sb2 thermal conductivity of the disordered phase. The low thermal
are very similar to those in ZnSb. Both structures have Sb-Sb conductivity thus seems to be an inherent feature of these
dumbbells with interatomic distances around 2.8 Å, where the structures. However, it is also clear that disorder on the
Sb2 atoms furthermore coordinate to three Zn atoms. The Sb1 zinc lattice further lowers the thermal conductivity in the
atoms coordinate a prism of six Zn nearest neighbors and are experimental phases.
not involved in a Sb-Sb dumbbell. It is interesting to note
how the large γ̃ 2 is directly related to the Sb1 atoms (Fig. 7), ACKNOWLEDGMENTS
which have no equivalent in the 1:1 structure. This provides a
clear indication that the very low thermal conductivity of the This work was supported by the Danish National Re-
Zn4 Sb3 and its model 6:5 phase is directly related to a strongly search Foundation (Center for Materials Crystallography,
anharmonic potential and local environment of the Sb1 atoms DNRF93), the Danish Strategic Research Council (Centre for
[Fig. 8(b)]. Energy Materials), and the Danish Center for Scientific Com-
puting. G.K.H.M. acknowledges financial support through
ThyssenKrupp AG, Bayer MaterialScience AG, Salzgitter
V. CONCLUSION
Mannesmann Forschung GmbH, Robert Bosch GmbH, Ben-
We have introduced a model to calculate the lattice thermal teler Stahl/Rohr GmbH, Bayer Technology Services GmbH
conductivity based on ab initio calculated phonon dispersions. and the state of North-Rhine Westphalia, as well as the
The model has been found to give results in reasonable European Commission in the framework of the European
quantitative agreement with experiment and correctly predict Regional Development Fund (ERDF).

[1] T. Caillat, J. P. Fleurial, and A. Borshchevsky, J. Phys. Chem. [12] K. Valset, P. H. M. Böttger, J. Taftø, and T. G. Finstad, J. Appl.
Solids 58, 1119 (1997). Phys. 111, 023703 (2012).
[2] B. L. Pedersen and B. B. Iversen, Appl. Phys. Lett. 92, 161907 [13] X. Song, P. H. M. Böttger, O. B. Karlsen, T. G. Finstad, and
(2008). J. Taftø, Phys. Scr. T148, 014001 (2012).
[3] B. L. Pedersen, H. Yin, H. Birkedal, M. Nygren, and B. B. [14] P. Jund, R. Viennois, X. Tao, K. Niedziolka, and J.-C. Tedenac,
Iversen, Chem. Mater. 22, 2375 (2010). Phys. Rev. B 85, 224105 (2012).
[4] H. Yin and B. B. Iversen, Sci. Adv. Mater. 3, 592 [15] L. Bjerg, G. K. H. Madsen, and B. B. Iversen, Chem. Mater. 24,
(2011). 2111 (2012).
[5] Y. Mozharivskyj, Y. Janssen, J. L. Harringa, A. Kracher, A. O. [16] D. Eklöf, A. Fischer, Y. Wu, E.-W. Scheidt, W. Scherer, and
Tsokol, and G. J. Miller, Chem. Mater. 18, 822 (2006). U. Häussermann, J. Mater. Chem. A 1, 1407 (2013).
[6] H. Yin, M. Christensen, B. L. Pedersen, E. Nishibori, S. Aoyagi, [17] P. J. Shaver and J. Blair, Phys. Rev. 141, 649 (1966).
and B. B. Iversen, J. Electron. Mater. 39, 1957 (2010). [18] G. J. Snyder, M. Christensen, E. Nishibori, T. Caillat, and B. B.
[7] C. Okamura, T. Ueda, and K. Hasezaki, Mater. Trans. 51, 860 Iversen, Nat. Mater. 3, 458 (2004).
(2010). [19] W. Chen and J. Li, Appl. Phys. Lett. 98, 241901 (2011).
[8] P. H. M. Böttger, K. Valset, S. Deledda, and T. G. Finstad, [20] G. Li, Y. Li, L. Liu, Q. Zhang, and P. Zhai, Mater. Res. Bull. 47,
J. Electron. Mater. 39, 1583 (2010). 3558 (2012).
[9] L. Bjerg, G. K. H. Madsen, and B. B. Iversen, Chem. Mater. 23, [21] W. Schweika, R. P. Hermann, M. Prager, J. Perßon, and
3907 (2011). V. Keppens, Phys. Rev. Lett. 99, 125501 (2007).
[10] D. Benson, O. F. Sankey, and U. Häussermann, Phys. Rev. B [22] A. Ward and D. A. Broido, Phys. Rev. B 81, 085205
84, 125211 (2011). (2010).
[11] P. H. M. Böttger, G. S. Pomrehn, G. J. Snyder, and T. G. Finstad, [23] L. Lindsay, D. A. Broido, and T. L. Reinecke, Phys. Rev. Lett.
Phys. Status Solidi A 208, 2753 (2011). 111, 025901 (2013).

024304-7
LASSE BJERG, BO B. IVERSEN, AND GEORG K. H. MADSEN PHYSICAL REVIEW B 89, 024304 (2014)

[24] G. A. Slack, in Solid State Physics, edited by F. S. Henry [37] G. Kresse and J. Hafner, Phys. Rev. B 47, 558 (1993).
Ehrenreich and D. Turnbull (Academic, New York, 1979), [38] G. Kresse and J. Furthmüller, Phys. Rev. B 54, 11169
Vol. 34, pp. 1–71. (1996).
[25] E. J. Skoug and D. T. Morelli, Phys. Rev. Lett. 107, 235901 [39] J. P. Perdew, K. Burke, and M. Ernzerhof, Phys. Rev. Lett. 77,
(2011). 3865 (1996).
[26] Y. Zhang, E. Skoug, J. Cain, V. Ozoliņš, D. Morelli, and [40] P. E. Blöchl, Phys. Rev. B 50, 17953 (1994).
C. Wolverton, Phys. Rev. B 85, 054306 (2012). [41] G. Kresse and D. Joubert, Phys. Rev. B 59, 1758 (1999).
[27] G. A. Slack and S. Galginaitis, Phys. Rev. 133, A253 (1964). [42] A. Togo, F. Oba, and I. Tanaka, Phys. Rev. B 78, 134106 (2008).
[28] D. T. Morelli, J. P. Heremans, and G. A. Slack, Phys. Rev. B 66, [43] P. G. Klemens, in Kältephysik I/Low Temperature Physics I,
195304 (2002). Handbuch der Physik/Encyclopedia of Physics (Springer-
[29] G. Leibfried and E. Schlömann, Nachr. Akad. Wiss. Göettingen, Verlag, Berlin, 1956), pp. 198–281.
Math.-Phys., Kl. 2A: Math.-Phys.-Chem. Abt. 4, 71 (1954). [44] C. A. Ratsifaritana and P. G. Klemens, Int. J. Thermophys. 8,
[30] C. L. Julian, Phys. Rev. 137, A128 (1965). 737 (1987).
[31] K. Esfarjani, G. Chen, and H. T. Stokes, Phys. Rev. B 84, 085204 [45] J. Nylén, M. Andersson, S. Lidin, and U. Häussermann, J. Am.
(2011). Chem. Soc. 126, 16306 (2004).
[32] W. Li, L. Lindsay, D. A. Broido, D. A. Stewart, and N. Mingo, [46] Y. Wu, J. Nylén, C. Naseyowma, N. Newman, F. J. Garcia-
Phys. Rev. B 86, 174307 (2012). Garcia, and U. Häussermann, Chem. Mater. 21, 151 (2009).
[33] R. Kremer, K. Graf, M. Cardona, G. Devyatykh, A. Gusev, [47] T. Dasgupta, C. Stiewe, L. Boettcher, R. Hassdorf, H. Yin,
A. Gibin, A. Inyushkin, A. Taldenkov, and H.-J. Pohl, Solid B. Iversen, and E. Mueller, J. Mater. Res. 26, 1925 (2011).
State Commun. 131, 499 (2004). [48] E. S. Toberer, P. Rauwel, S. Gariel, J. Taftø, and G. Jeffrey
[34] G. A. Slack, Phys. Rev. B 6, 3791 (1972). Snyder, J. Mater. Chem. 20, 9877 (2010).
[35] B. L. Pedersen, H. Birkedal, B. B. Iversen, M. Nygren, and P. T. [49] R. Viennois, M.-C. Record, V. Izard, and J.-C. Tedenac, J. Alloys
Frederiksen, Appl. Phys. Lett. 89, 242108 (2006). Compd. 440, L22 (2007).
[36] S. Bhattacharya, R. P. Hermann, V. Keppens, T. M. Tritt, and [50] E. Chalfin, H. Lu, and R. Dieckmann, Solid State Ionics 178,
G. J. Snyder, Phys. Rev. B 74, 134108 (2006). 447 (2007).

024304-8

View publication stats