PHYSICAL REVIEW B 88, 165203 (2013)

Electronic band structure, phonons, and exciton binding energies of halide perovskites
CsSnCl3 , CsSnBr3 , and CsSnI3

Ling-yi Huang and Walter R. L. Lambrecht

Department of Physics, Case Western Reserve University, Cleveland, OH 44106-7079, USA
(Received 7 August 2013; published 17 October 2013)
The halide perovskites CsSnX3 , with X = Cl, Br, I, are investigated using quasiparticle self-consistent
GW electronic structure calculations. These materials are found to have an “inverted” band structure from
most semiconductors with a nondegenerate s-like valence band maximum (VBM) and triply degenerate p-like
conduction band minimum (CBM). The small hole effective mass results in high hole mobility, in agreement with
recent reports for CsSnI3 . The relatively small band gap changes from Cl to Br to I result from the intra-atomic
Sn s and Sn p characters of the VBM and CBM, respectively. The latter is also responsible for the high
oscillator strength of the optical transition in these direct-gap semiconductors and hence a strong luminescence
and absorption. The band gap change with lattice constant is also anomalous. It increases with increasing lattice
constant, and this results from the decreasing valence band width due to the decreased Sn s with anion p
interaction. It leads to an anomalous temperature dependence of the gap. The changes in band gap in different
lower-symmetry crystallographic phases is studied. The exciton binding energy of the free exciton, estimated
from the Wannier-Mott exciton theory and the calculated dielectric constants and effective masses, is found
to be two orders of magnitude smaller than previously claimed in literature, or of the order of 0.1 meV. The
photoluminescence peak previously assigned to the free exciton is instead ascribed to an acceptor bound exciton.
The phonons at the  point are calculated as well as the related enhancement of the dielectric constants.

DOI: 10.1103/PhysRevB.88.165203 PACS number(s): 71.20.Nr, 71.15.Mb, 74.25.Kc

I. INTRODUCTION gap close to that of Si4,7 and strong absorption at shorter wave-
lengths make CsSnI3 attractive as a solar luminescent concen-
The halide perovskites with the formula CsSnX3 , with
trator material8 or as a solar-cell absorber.9 It was recently
X = Cl, Br, and I, form an interesting family of materials
shown6 that its high hole mobility4 make it an excellent all-
with unusual properties. Like the much more studied oxide
solid-state replacement for the electrolyte in a dye-sensitized
perovskites, they undergo a number of phase transitions to
Grätzel-type solar cell with TiO2 . It was found that CsSnI3
lower-symmetry structures, starting from a cubic perovskite
itself strongly contributes to the optical absorption, although
structure at high temperature.1–4 These phase transitions the overall performance still is aided by the organic dyes.
correspond mostly to rotations and distortions of the SnX6 However, many of its properties are still rather poorly
octahedra, leading to a closer packing. In CsSnI3 these understood. For instance, it is presently not clear whether the
structures correspond to the so-called black phase, but there strong luminescence is intrinsic, i.e., due to free excitons, or
is also a higher band gap yellow phase with a different, due to defect bound excitons. The effects of crystal structure on
even denser structure.4,5 In the perovskite structure, the SnX6 the luminescence strength are poorly understood and are cru-
octahedra corner share, and thus each halide ion counts for cial for the indicated solar-cell-concentrator approach. Until
one half, resulting in the formula CsSnX3 . The cubic structure recently,4 progress was strongly hampered by the difficulty in
can be understood as a CsCl packing of large Cs+ ions and obtaining phase-pure CsSnI3 material and the rapid conversion
SnX3 −1 units. This means that the Sn in the compounds is of the black phase to the yellow phase of CsSnI3 in air. The
divalent. The bonding thus can be viewed as strongly ionic. electrical conductivity is only poorly understood. Although
Nevertheless, the band gaps are relatively small, of the order a high hole mobility of ∼585 cm2 V−1 s−1 was reported,
of 1–3 eV, compared to, for example, Cs halides, and most a relatively low carrier concentration of the order of only
surprisingly, only relatively small changes occur, especially 1017 cm−3 and a metallic-like decreasing conductivity with
from one halide to the next, especially in the Br to I case (1.38 temperature were obtained in Ref. 4.
to 1.01 eV). As we will show in this paper, several unusual Besides these few applications presently under investiga-
properties of this materials family result from the intracluster tion, other possibilities have not yet been explored. In view
SnI6 bonding. Interestingly, Cs in the structure can be replaced of their close similarity to oxide perovskites in structure,
by other large monovalent ions, for example, Rb and Tl, or even these materials may have interesting piezoelectric or even
organic radicals,1 such as methylammonium CH3 NH3 +1 and ferroelectric properties in the lower-symmetry phases. The
formamidinium NH2 CH = NH2 +1 . relatively heavy ions combined with strongly ionic bonding
Recently, a few of the unusual properties of black-phase lead to low-frequency modes,10,11 which may result in large
CsSnI3 have attracted interest for applications. Some of these phonon enhancements of the dielectric constant. Recently,12
result from the low-temperature process ability of the material. third-order enhanced Raman spectra were reported for CsSnI3 .
CsSnI3 is soluble in certain organic solvents and can thus be de- Heavy ions like Cs also make these materials suitable for cap-
posited on various substrates or even inside a porous structure turing γ rays and, combined with their strong luminescence,
by solution processing.6 The strong luminescence at a band may be excellent scintillation materials for detectors. This idea

1098-0121/2013/88(16)/165203(12) 165203-1 ©2013 American Physical Society

LING-YI HUANG AND WALTER R. L. LAMBRECHT PHYSICAL REVIEW B 88, 165203 (2013)

was recently proposed by Sing13 for closely related Tl- instead more accurately. In the GW method of Hedin23 the exchange-
of Cs-containing compounds, TlGeI3 , TlSnI3 , and TlPbI3 . correlation self-energy of the quasiparticle excitations is
Their band structures have been studied theoretically in obtained as a product of the one-electron Green’s function G0
a few previous papers.14–17 These papers focused mostly and the screened Coulomb interaction W 0 , schematically,  =
on the changes in band structure with crystal structure iG0 W 0 . The screening W 0 = ε−1 v = [1 − 0 v]−1 v itself is
and, because they were based on density functional theory calculated in terms of the independent particle polarizability
(DFT) in the local-density approximation (LDA), strongly 0
= −iG0 × G0 given in terms of the Green’s function. The
underestimated the gaps. In the early literature, there is even self-energy provides a shift of the one-electron eigenvalues
significant discussion on whether they are semiconductors or as well as a lifetime through its imaginary part. The differ-
semimetals.14,15 Only the most recent study4 used the screened ence with QSGW is that the independent Hamiltonian H 0
exchange method, which gives larger gaps for CsSnI3 , but it from which G0 is derived is chosen to make the difference
still obtained a smaller gap than experiment for the black phase. between the quasiparticle energies and the independent-
Here we present a study of the electronic band struc- particle energies as small as possible by adding a Hermitian
ture of the CsSnX3 compounds with X = I, Br, Cl. Using but nonlocal exchange correlation potential change to the LDA
the quasiparticle self-consistent GW method (QSGW ).18,19 derived from the GW self-energy, which is then iterated to
We first present the basic band structures in the cubic per- self-consistency. For details, see Refs. 18 and 19. This method
ovskite structure. Analyzing the atomic orbital character of the has been shown to give quite accurate and systematic results
bands provides immediate insights into some of the questions for a large number of materials. In particular, band gaps are
mentioned above. These are all explained in terms of the SnI6 usually slightly overestimated because of the random phase
intracluster nature of the band gap. The changes with crystal approximation treatment of the screening. It was found that this
structure are found to be in fair agreement with previous theory can be further corrected by reducing  by a somewhat arbitrary
work. However, it is only after adding the QSGW quasiparticle ad hoc correction of a factor of 0.8. In the present work,
corrections that we can make a quantitative comparison of however, we report pure QSGW results. Zero-point motion
the gaps with experiment. We next address another anomaly corrections are not included here because we have currently
on the exciton binding energy.20 We will show that a large no reliable way of estimating them in these materials and they
exciton binding energy (of the order of 12 meV) is indeed are expected to be relatively small in these low-gap materials.
obtained if the electron-hole interaction is only screened by the The QSGW approach is used here in conjunction with the
high-frequency dielectric constant. However, we find that the FP-LMTO implementation. An important point here is that
phonons increase the dielectric constant by a factor of 9.9 and it allows us to obtain a real-space representation of  in the
hence lead to two orders of magnitude smaller exciton binding basis of atom-centered muffin-tin orbitals. By a reverse Fourier
energy if their contribution to the screening is included. The transform, this then allows calculating the GW corrections at
analysis of the previous experimental data is revised. arbitrary k points, even if the GW calculation is performed
only on a relatively coarse k-point mesh. This allows us
II. COMPUTATIONAL METHOD to display the full band structure at the GW level and to
extract reliable effective masses from the band dispersion. The
The starting point for our band structure calculations is convergence of the results with the GW k-point mesh used was
the LDA to the DFT. The full-potential linearized muffin-tin carefully studied along with its dependence on the parameters
orbital method (FP-LMTO)21,22 was used for the band structure of the LMTO basis set. Slightly more localized basis functions
calculations. The basis set in this method is characterized than for the total-energy LDA calculations are used, and a mesh
by the smoothed Hankel function decay parameter κ and its of 5 × 5 × 5 was found to be adequate for the cubic perovskite
smoothing radius Rsm . Inside muffin-tin spheres, the wave structure. For the tetragonal β phase, a correspondingly dense
functions are augmented by φ and φ̇ functions, which are mesh is used. Spin-orbit (SO) coupling (calculated at the LDA
solutions to the radial Schrödinger equation and its energy level) can be added as a separate term in the Hamiltonian.
derivative at some chosen linearization energy νRl , specific to The phonon calculations were performed using the den-
each orbital at atomic site R and angular momentum l. In the sity functional perturbation theory (DFPT)24,25 using norm-
present calculations, we use two (κ,Rsm ), the first for spd and conserving pseudo-potentials26 and within the LDA. The
the second for sp orbitals, on Cs, Sn, and the halide ion. For ABINIT code27 was used for these calculations. A plane wave
Cs, we treat the 5p states as bands. A 6p orbital was added as cut-off energy of 100 Hartree was used and a 4 × 4 × 4 k-point
a local orbital in some test calculations but was found to have mesh. Only phonons at the  point are calculated. The phonons
little effect. For Br and I, we include the 4s and 5s semicore were calculated at the equilibrium lattice constants of the LDA,
orbitals as local orbitals and the empty 5s and 6s as band which were found to be about 3% lower than the experimental
orbitals, respectively. Inside the spheres augmentation is done values within the pseudopotential plane-wave method using
up to lmax = 4. Brillouin zone integrations are done with a the ABINIT code.
10 × 10 × 10 k-point mesh. We emphasize that although a
muffin-tin construction is utilized in defining the basis set,
the actual “full potential” has no shape approximations; it is III. RESULTS
neither spherical inside the atoms nor constant in between the
atoms. A. Band structures
Because the LDA underestimates band gaps, we next The LDA usually slightly underestimates the lattice con-
use the QSGW method to recalculate the band structures stants. In the present case, for the cubic phases, we obtained

165203-2
ELECTRONIC BAND STRUCTURE, PHONONS, AND . . . PHYSICAL REVIEW B 88, 165203 (2013)

TABLE I. Lattice constants for CsSnX3 . Note that α, β, and γ mostly I 5p states and a band which is antibonding between
phases are cubic, tetragonal, and orthorhombic, respectively, while Sn 5s and I 5p. The conduction band minimum (CBM) is
the M phase is monoclinic and the Y phase is orthorhombic. dominated by Sn 5p states, but higher up there are also I 5p
antibonding orbital contributions, and above 5 eV, there is a
Material Space group Lattice constants (Å) significant Cs 5d contribution. The energy levels at the R point
α-CsSnCl3 a P m3m a = 5.56 where the minimum direct gap occurs are symmetry labeled
M-CsSnCl3 b P 21 /n a = 16.10, b = 7.425, c = 5.74 for the CsSnI3 case, and the same labels apply for the other
β = 93.2◦ cases.
α-CsSnBr3 a P m3m a = 5.804 We can see that the band gap in all three materials is direct
β-CsSnBr3 c P 4/mbm a = 8.195, b = 5.805 at the point R = (1,1,1)π/a. The VBM is nondegenerate,
α-CsSnI3 d P m3m a = 6.219 while the conduction band minimum is threefold degenerate
β-CsSnI3 d P 4/mbm a = 8.772, b = 6.261 without SO coupling and splits into a doublet and a quadruplet
γ -CsSnI3 d P nam a = 8.688, b = 8.643, c = 12.378 (including spin) when SO is included, as is illustrated for
Y -CsSnCl3 e P nam a = 10.328, b = 17.677, c = 4.765 CsSnI3 in Fig. 2 The spin-orbit splittings are summarized in
a Table III. This is opposite from what is found in most com-
Barrett et al.28
b pound semiconductors with tetrahedral bonding. We therefore
Poulsen and Rasmussen.29
c
Scaife et al.30
call it an “inverted” band structure. We note that this term
d
Yamada et al.3 should not be confused with a so-called band inversion, a term
e
Mauersberger and Huber.5 recently applied to topological insulators for the case where
there is actually a negative gap in the absence of spin-orbit
coupling because the cation s-like conduction band dips below
an underestimate of the lattice constant by about 1.5%–2% the anion p-like valence band.
using the all-electron LMTO method. Therefore we prefer to To obtain further insight into why this happens, we analyze
use the experimental lattice constants for our band structure the atomic orbital character of the bands. This is shown in
calculations. In Tables I and II the lattice constants and internal Fig. 3. In Fig. 3, the color intensity indicates how much a
parameters used in this paper are summarized. For the β phase chosen atomic orbital is present in the eigenstates at each
of CsSnI3 we also performed LDA calculations to minimize k point. One can see that the CBM is composed mostly of Sn
the structure with respect to the internal parameters and found p states. The VBM, on the other hand, is a mixture of I p and
excellent agreement with the experimental values to 0.001 Sn s states. These two form an antibonding combination. The
precision. reason why there is a strong Sn s to anion p orbitals at this
For CsSnCl3 we also performed LDA calculations for particular k point is that at k point R, the p orbitals on opposite
the monoclinic phase, which has a rather different crystal sites of the unit cell differ by a phase factor eik·T = −1, and
structure,29 and we also calculated the band structure of the hence the p orbitals all point with the same lobe of the orbital
yellow phase of CsSnI3 in the LDA. inward toward the Sn s, as shown in Fig. 4. They thus belong
We start analyzing the band structures in the cubic per- to the same irreducible representation a1g of the cubic group
ovskite phase. Figure 1 shows the band structures of CsSnX3 , and have a strong interaction with each other. This means that
with X = Cl, Br, I, obtained in the QSGW method without the gap is to a large extent intra-atomic between Sn s and
spin-orbit coupling and the corresponding partial densities of Sn p.
states. The dominant orbital character of the bands can be In fact, further group theoretical analysis, following the no-
described as follows. At about −15 eV below the valence tation of Bose et al.,15 shows that the CBM has R15 symmetry,
band maximum (VBM), in CsSnI3 , we find the I 5s bands. while none of the I 5p orbital combinations on the faces of the
In the CsSnBr3 and CsSnCl3 cases these bands move farther unit cell belong to this irreducible representation. It also does
down, reflecting the deeper atomic energy levels in the not interact with the Cs 5p below. There is some interaction
earlier elements of the periodic table. The next narrow band with the I 5s states, which do form a R15 linear combination
corresponds to the Cs 5p semicore levels. Then we find a band at point R but which lie very deep so that the upper state has
which is a bonding combination of Sn 5s and I 5p, followed by little I 5s contribution. There is also some interaction with Cs
5d states higher up which also belong to R15 . This interaction
TABLE II. Atomic coordinates of β and γ CsSnI3 . The position will tend to reduce the gap. On the other hand, the I 5p with
column gives the Wyckoff position and site point group. Sn s interaction leading to the two states of symmetry R1 is
clearly very strong because the levels are close to each other
Material Atom Position x y z and thus there is a very strong hybridization. The Cs 6s on the
corners of the unit cell form a R2 irreducible representation,
β-CsSnI3 Cs 2d(mmm) 0 0.5 0
which would only interact with f states at the center.
β-CsSnI3 Sn 2b(4/m) 0 0 0.5
The natures of the CBM and VBM explain a number of
β-CsSnI3 I(1) 2a(4/m) 0 0 0
the intriguing properties of these materials. First, they explain
β-CsSnI3 I(2) 4h(mm) 0.210 0.290 0.5
γ -CsSnI3 Cs 4c(m) 0.006 − 0.040 0.25
the “inverted” nature of the band structure. Second, they
γ -CsSnI3 Sn 4b(T ) 0.5 0 0 show that strong optical transitions are expected between the
γ -CsSnI3 I(1) 4c(m) 0.476 0.000 0.25 VBM and the CBM because they have an allowed l = 1
γ -CsSnI3 I(2) 8d(1) 0.294 0.705 −0.019 dipole character on the same atom. Third, they show that
the gap is essentially controlled by the Sn s to I p covalent

165203-3
LING-YI HUANG AND WALTER R. L. LAMBRECHT PHYSICAL REVIEW B 88, 165203 (2013)

10 15 10
12 Total Total
Cs-p Cs-p
25 Cs-d Cs-d
5 Sn-s 5 Sn-s
2'
Sn-p Sn-p
I-s Br-s
15
I-p Br-p
0 1 I-d
0 Br-d
Energy (eV)

Energy (eV)
15'
12
-5 25' -5

-10 1 -10
25'

-15 15 -15

X M R 5 10 X M R 5 10
PDOS (number of state/eV/cell) PDOS (number of state/eV/cell)
10
Total
Cs-p
Cs-d
5 Sn-s
Sn-p
Cl-s
Cl-p
0 Cl-d
Energy (eV)

-5

-10

-15

X M R 5 10
PDOS (number of state/eV/cell)

FIG. 1. (Color online) QSGW band structures and partial densities of states of α-CsSnI3 , α-CsSnBr3 , and α-CsSnCl3 .

5 antibonding interaction. For instance, with an increasing lattice

constant, one expects this interaction to become weaker, and
4 hence the valence band width will be decreased and the gap
3 will be increased. We will see below that this is confirmed
directly by the calculations, and it also explains the anomalous
2 temperature behavior of the gap. It also explains why to zeroth
Energy (eV)

order the gap is not as strongly dependent on the anion as

1
one might think. In fact, it is in some sense intra-atomic in
0 nature on Sn. In particular, we note that the gaps of CsSnI3
-1
-2 TABLE III. Spin-orbit splitting in α-structure and tetragonal
crystal field splitting in β-structure in meV.
-3
Compound so c
-4
X M R CsSnCl3 448
CsSnBr3 437 215
FIG. 2. (Color online) Energy bands of CsSnI3 in the QSGW CsSnI3 420 254
method near the gap including spin-orbit coupling.

165203-4
ELECTRONIC BAND STRUCTURE, PHONONS, AND . . . PHYSICAL REVIEW B 88, 165203 (2013)

Sn-p
100
3
80
2
Energy (eV)

60
1

0 40

-1 20

-2 0
X M R
FIG. 4. (Color online) Schematic drawing of the anion p orbitals
Sn-s and I-p pointing toward Sn s at the center of the unit cell for the VBM Bloch
state at point R at the corner of the Brillouin zone.
100
3 is observed for the VBM in tetragonally distorted tetrahedrally
80 bonded semiconductors, such as chalcopyrites.31 A further
2 tilting of the tetrahedra away from the z axis results in the
Energy (eV)

60 orthorhombic γ phase. These structural distortions have been

1 discussed in detail in Refs. 17 and 4. The yellow phase, on
40
the other hand, involves a much more involved rearrangement
0 of the atoms. The LDA band structures of all four CsSnI3
phases are shown in Fig. 5. The Brillouin zones (BZ) and
-1 20 the labeling of the high-symmetry points are shown in Fig. 6.
Note that the relation between the α and β phases involves a
-2 0 rotation by 45◦ and a folding. The -M line of α phase folded
X M R about its midpoint, which we labeled B in Fig. 6, becomes
the -X line of the β phase. We can clearly see in the band
FIG. 3. (Color online) Dominant orbital character of (top) CBM structure figures that the bands along M-B of the α phase
and (bottom) VBM for α-CsSnI3 . For the selected orbitals φi are similar in dispersion to those along the -X line in the β
mentioned, the plots show k- and E-resolved “spectral functions,” phase. Point R in α becomes folded onto the Z point in the β
A(k,E) = |φi |ψnk |2 δ(E − Enk ), with the δ function replaced by a BZ. In fact the whole M-R line is folded onto the -Z line.
Gaussian, using the color scale shown on the right. So the light green We included point L in the middle of the M-R line, although
areas and white areas indicate zero contribution from that orbital,
it is not a higher-symmetry point. It becomes equivalent to
while the red areas indicate a strong contribution from that orbital.
point H in the β structure. The transformation from β to γ
The thin red lines show the band superposed on the color graph.
structure involves a doubling of the cell in the c direction or
folding of the bands along -Z about the halfway point H so
and CsSnBr3 are rather close, especially if we consider the that Z of β now becomes folded on to the  point of γ and
fact that the larger spin-orbit coupling in I, which reduces the point H becomes the new Z point where the bands are doubly
gap, partially overcomes the expected increase in gap on going degenerate because of the folding. However, the tetragonal
from I to Br. Still, of course, the degree of antibonding mixing symmetry is also broken, so that the -X and -Y directions
of the anion p orbital in the VBM will decrease from Cl to become distinct.
Br to I because the anion p levels move farther down. This Our results on the band gap changes obtained in LDA are
explains the decreasing trend of the gaps from Cl to I. in good agreement with previous studies. For β-CsSnI3 we
Next, we consider the changes in band gap due to also carried out the full QSGW calculation. This allows us
the structural distortions in the lower-symmetry phases. to check that the GW gap correction is not very sensitive to
The structural phases of interest can be summarized as follows. the structure. Although the structural changes between the α,
Starting from the cubic α phase, a rotation of the octachedra in β, and γ phases are mild, the changes for the monoclinic
the xy plane results in the tetragonal β phase. Chabot et al.16 phase of CsSnBr3 and the yellow phase of CsSnI3 are more
pointed out that this is related to the imaginary frequency of severe. Nonetheless, one may argue that the GW corrections
the lowest acoustic mode at point M in the Brillouin zone. The are mostly dependent on the average electron density and not
symmetry lowering from cubic to tetragonal symmetry leads to too sensitive to structural details. The degree to which this
a crystal-field splitting of the CBM in addition to the spin-orbit holds may be judged from the agreement between our thus
coupling. The crystal-field splittings are included in Table III. estimated gaps for these phases with experiment. Thus, once
When both spin-orbit and crystal-field splittings are included we have obtained the gap correction for the α phase, we can
for the β phase, the VBM splits into three levels, similar to what also apply it to the gaps of the other phases. The QSGW band

165203-5
LING-YI HUANG AND WALTER R. L. LAMBRECHT PHYSICAL REVIEW B 88, 165203 (2013)

4
b3 b3 b3

3
Z
Z U
2 R R T
H A
L X R
1 b2 b2 X Y
Energy (eV)

X M b2
B M b1
0 b1 b1 S

-1

-2
FIG. 6. (Color online) Brillouin zone of cubic (α phase), tetrag-
onal (β phase), and orthorhombic (γ and Y phases) structures, from
-3 left to right.
-4
X B M L R B
β in CsSnI3 are 0.158 and 0.050 eV. These values are in fair
4
agreement with those of Borriello et al.,17 0.133 and 0.080 eV,
3 but are not as well in agreement with those of Chabot et al.,16
0.179 and 0.244 eV.
2
The comparison with experimental gaps requires a little
1 discussion. For Cl and Br materials, these gaps are not as well
Energy (eV)

0 established as for the I material because we rely on rather

old literature data. Voloshinovskii et al.32 presented photo-
-1
luminescence (PL) and photoluminescence excitation (PLE)
-2 as well as optical absorption data for CsSnCl3 in its high-
-3
temperature cubic and low-temperature monoclinic forms. For
the α phase they show a broad PL band centered around
-4 555 nm, or 2.2 eV. This is probably due to defect bound
M X H Z X
excitons rather than being intrinsic. The PLE shows a peak
4 centered at 368 nm, or 3.4 eV; however, it extends clearly
3
to about 400 nm, or 3.0 eV. The absorption extends even
somewhat further to 420 nm, or 2.9 eV. This is in good
2 agreement with our gaps of 2.7 eV (with spin-orbit coupling)
1 and 3.0 eV (without spin-orbit coupling). The monoclinic
Energy (eV)

phase clearly has a higher gap. The peak of the PLE and
0
optical absorption lies near 4.46 eV.32 We only calculated this
-1 material in the LDA (see Fig. 8), but assuming the same gap
-2
shift to GW as for the cubic phase, we would estimate a gap of
4.7 eV. Ohno and Sasaki34 obtained two phase materials and
-3 associated a 2.11-eV PL band with α-CsSnCl3 , with a PLE
-4 peak at 3.5 eV and onset just above 3.1 eV, and a 2.65-eV PL
S Y Z X band with monoclinic CsSnCl3 , with PLE peaking at 4.23 eV.
4

2 4
1
Energy (eV)

0 2
Energy (eV)

-1

-2 0
-3

-4 -2
S Y Z X

FIG. 5. (Color online) LDA band structure of α-CsSnI3 , β-

CsSnI3 , γ -CsSnI3 , and Y -CsSnI3 .
-4

structures of the β phase materials are shown in Fig. 7 for both X M A Z

CsSnI3 and CsSnBr3 .
The results on the band gaps are summarized in Table IV. FIG. 7. (Color online) QSGW band structure of β-CsSnI3
The band gap differences between β and α and between γ and (red solid lines) and β-CsSnBr3 (blue dashed lines).

165203-6
ELECTRONIC BAND STRUCTURE, PHONONS, AND . . . PHYSICAL REVIEW B 88, 165203 (2013)

TABLE IV. Band gaps for CsSnX3 (eV) in various structures. (LCAO) band structure calculation of Parry et al.36 and the
Note that for the α and β phases, actual QSGW calculations were semiempirical pseudopotential calculation of Lefebvre et al.14
performed, while for the other cases, they are estimates assuming a the Sn p-derived states at R are reversed with the Sn s-derived
structure-independent gap correction beyond LDA. states, and hence the material appeared to be semimetallic,
which was thought to be consistent with the high electrical
Material LDA QSGW QSGW + SO Other calc. Expt.
conductivity. Already in Bose et al.’s15 LMTO calculations in
α-CsSnCl3 0.744 2.997 2.693 0.950a 2.9b the atomic sphere approximation (ASA) this was reversed, and
M-CsSnCl3 c 2.743 4.7 ± 0.1 2.845a 4.5b a semiconducting band structure was concluded, but the gap
α-CsSnBr3 0.351 1.690 1.382 0.58d was still strongly underestimated by the LDA with a gap of
β-CsSnBr3 0.574 1.918 1.740 1.8e the order of 0.58 eV.
f a
α-CsSnI3 0.295 1.354 1.008 0.462, 0.348 These authors noted that the main difference between the
β-CsSnI3 0.453 1.494 1.288 0.641,e 0.481a earlier empirical LCAO band structure and theirs was the
γ -CsSnI3 c 0.503 1.3 ± 0.1 0.885,e 0.561a 1.3g absence of any Cs orbitals in the LCAO basis set. However, at
Y -CsSnI3 c 2.026 2.7 ± 0.1 2.53h 2.55g the R point where the minimum gap occurs, the Cs 6s states
a
Borriello et al.17 GGA pseudopotential plane-wave method. belong to irreducible representation R2 and hence cannot
b
Voloshinovskii et al.32 interact with the R1 VBM. The Cs 5d orbitals of symmetry
c
QSGW results are estimates only. t2g can interact with the Sn p at the center because they both
d
Bose et al.15 LDA with LMTO ASA. belong to R15 symmetry. For example, the Cs dxy orbitals at the
e
Clark et al.33 corners of the cell interact with the Sn pz at the center, taking
f
Chabot et al.16 GGA pseudopotential plane-wave method. into account the Bloch function phase factors as discussed
g
Chen et al.20 above. However, since the Cs 5d lie above the Sn p, omitting
h
Chung et al.4 the Cs orbitals or pushing them up will reduce this interaction
and thus move the Sn p CBM up, which would increase the
Clark et al.33 report a band gap of 1.8 eV for α- gap. The Cs 5p states cannot interact with the Sn p or Sn s
CsSnBr3 from optical absorption and low-temperature band either. Thus we conclude that the absence of Cs orbitals did not
gap luminescence at 1.72–1.75 eV depending on temperature. play a major role in the opening of the gap. However, if we shift
Myagkota et al.35 also assigned a peak at 1.72 eV to intrinsic the Sn levels down with respect to Br or I or Cl orbitals, we can
luminescence of CsSnBr3 in a sample that was a mixture with increase the Sn s with X p interaction sufficiently to make the
other phases. It is not entirely clear whether these reports R1 state shift above the R15 state, thereby closing the gap and
correspond truly to cubic CsSnBr3 or to the tetragonal β form, pinning the Fermi level at the R15 state. Note that the X s states
whose structure is reported in Scaife et al.30 We find that at k point R form a state of R15 symmetry, which interacts
our gap for the β phase of CsSnBr3 agrees much better with with the Sn p orbitals at the center of the unit cell, and this
experiment than the for the α phase. We note that our gap interaction tends to increase the gap. However, this interaction
of α-CsSnBr3 is rather close to that of CsSnI3 , and hence this is much weaker because the halogen X s orbitals lie rather
material is also worth considering for photovoltaic applications deep. Thus the origin of the wrong band structures in the early
if it can be stabilized. In fact, as mentioned in Ref. 4, the smaller work is because the position of the Sn s versus the X p levels,
cage size in this material may lead to a stabler bonding of Cs, which depends on charge transfer, was not properly taken into
and hence it avoids the undesirable larger-gap yellow phase of account. In pseudopotential language, this corresponds to a too
CsSnI3 but stays in a perovskite-derived structure. deep Sn pseudopotential, as was speculated by Bose et al.15
We note that in the past literature,14,15,36 it was unclear if this Interestingly, it was recently proposed by Yang et al.37
material was a semiconductor or semimetal. For example, in that some of these materials could be close to a topological
the first semiempirical linear combination of atomic orbitals insulator condition when a strain is applied so as to induce a
band inversion, i.e., negative gap. The gap would then indeed
result purely from spin-orbit coupling. These calculations,
6
however, were done at the generalized gradient approximation
(GGA) level, which underestimates the gaps significantly, as
4 shown here. With our present results, the strains required
to induce a topological insulator behavior would become
Energy (eV)

unrealistic.
2 For CsSnI3 our gap for the γ phase comes rather close to the
experimental value of 1.3 eV recently established. The change
0 in the gap between the β and γ phases is rather small. In any
case, our gaps are clearly all significantly larger than the LDA
gaps and the gaps reported in previous work, including the sx-
-2 LDA gap reported in Ref. 4. With improved materials, it should
become possible to probe the nature of the spin-orbit split gaps
-4
using circularly polarized light. For Y -CsSnI3 , Chung et al.4
Z X C E A Y D found a direct gap of 2.53 eV for the yellow phase, in good
agreement with the observed 2.55 eV onset of absorption, but
FIG. 8. (Color online) LDA band structure of M-CsSnCl3 . they also found a somewhat lower indirect gap of 2.46 eV. They

165203-7
LING-YI HUANG AND WALTER R. L. LAMBRECHT PHYSICAL REVIEW B 88, 165203 (2013)

3.0 TABLE VI. Effective masses at the band gap edges (in electron
mass). These correspond to the bands without spin-orbit coupling.
2.5 -CsSnCl 3 LDA Sample m∗h m∗he [100] m∗he [111] m∗le [100] m∗le [111]
-CsSnCl 3 QSGW -CsSnBr 3 LDA
2.0 α-CsSnCl3 0.140 0.741 0.272 0.088 0.149
Band gap (eV)

-CsSnBr 3 QSGW -CsSnI 3 LDA

-CsSnI 3 QSGW α-CsSnBr3 0.082 0.635 0.201 0.053 0.084
1.5 α-CsSnI3 0.069 0.573 0.154 0.041 0.068

1.0 related to a similar effect. While the overall structure becomes

denser under this transformation, we found that the Sn halogen
0.5 distance actually increases. This reduces the Sn s-halogen–p
interaction, and hence the gap increases. The change from β
0.0 to γ is more difficult to interpret because the tilting of the
0.98 0.99 1.00 octahedra makes the orbital interactions more complex.
% of lattice constant B. Effective masses and exciton binding energies
FIG. 9. (Color online) Band gap vs lattice constant. In Table VI we report the effective masses obtained by
fitting parabolic dispersion curves to our first-principles energy
did not report a value for the gap for the γ phase, but from band structures. For the holes, this is straightforward because
their figure it appears to be about 0.7 eV, thus significantly the band is nondegenerate and the band dispersion is isotropic
smaller than the 1.3 eV measured. Our LDA band structure of and parabolic near the VBM. We can see that rather low hole
Y -CsSnI3 is shown in Fig. 5. Both the direct gap, 2.026 eV, and masses are found decreasing from the Cl to the Br to the
indirect gap, 1.943 eV, are close to the X point. If we add the I compound. These results are obtained for the cubic phase
same amount of energy change shown in the α-CsSnI3 case, we for simplicity, but they do not differ very much in the other
will get around 2.7 eV for the band gap after QSGW and SO phases. The low hole masses clearly play an important role in
correction, which is close to the experimental value of 2.55 eV. the observed high hole mobility, which was reported4 to be of
Finally, we show in Fig. 9 the changes in band gap with the order of ∼585 cm2 V−1 s−1 . In view of the general relation
lattice constant for CsSnI3 , CsSnBr3 , and CsSnCl3 . We confirm μ = eτ/m∗ they indicate a carrier relaxation time τ of about
the already mentioned increase of gap with lattice constant. In 2.29 × 10−14 s.
view of the thermal expansion, this also implies an increasing Because of the threefold degeneracy of the CBM in the
band gap with increasing temperature, which is opposite absence of SO coupling, the masses depend on direction.
most materials. This is because in most semiconductors, Along the [100] direction, there are two bands, and we label
the gap is between anion p- and cation s-dominated bands. these masses as the heavy (he) and light electron (le), in
The increasing covalent interaction between the two when analogy with the usual terminology for holes in tetrahedrally
atoms are brought closer together then increases the gap with bonded semiconductors. Similarly, in the [111] direction we
decreasing lattice constant. However, here the gap is, to first also have a splitting. This situation can best be described by
approximation, intra-atomic. The decreasing lattice constant the Kohn-Luttinger (KL) Hamiltonian based on the theory of
results in an increasing band width of the valence band, invariants. Although we are here dealing with electrons instead
pushing up the VBM by the increased interaction between of holes and with a different k point, namely, R instead of ,
Sn s and anion p. This then results in the anomalous gap the form of the Hamiltonian is still applicable because of the
behavior. Such anomalous gap behavior was reported recently cubic symmetry. The KL Hamiltonian is given by

for CsSnI3 (Ref. 8) but can also be seen in the data for the  · σ + Ak 2 − (A − B) L2x kx2
Ĥ = 13 0 L
1.72-eV CL band in Myagkota et al.35 for CsSnBr3 . We find 
a gap deformation potential ag = dEg /d ln a of 4.65 eV in + L2y ky2 + L2z kz2 − 2C({ Lx ,Ly }kx ky
LDA and 7.35 in QSGW for CsSnI3 . For the other materials, + { Ly ,Lz }ky kz + { Lx ,Lz }kx kz ), (1)
ag are shown in Table V. While it is often assumed that the gap
deformation potentials can be obtained accurately within LDA where {Li ,Lj } = (Li Lj + Lj Li )/2, 0 is the spin-orbit split-
even when the gaps themselves are underestimated because ting, and the parameters A, B, and C are related to the effective
one expects similar errors for each lattice constant, the present masses in different directions as follows:
results indicate that the GW corrections also are required for 1
the gap changes. The increase in gap from α to β structure is A= ∗ ,
mle [100]
1
TABLE V. Gap deformation potential ag in α structure (in eV). B= ∗ , (2)
mhe [100]
Compound ag in LDA ag in QSGW A + 2B − C 1
= ∗ ,
CsSnCl3 5.62 9.05 3 mhe [111]
CsSnBr3 5.34 8.99 A + 2B + 2C 1
CsSnI3 4.65 7.35 = ∗ . (3)
3 mle [111]

165203-8
ELECTRONIC BAND STRUCTURE, PHONONS, AND . . . PHYSICAL REVIEW B 88, 165203 (2013)

TABLE VII. Calculated inverse mass parameters A, B, and C of of about 12 meV for α-CsSnI3 . This is of the same order of
the Kohn-Luttinger Hamiltonian. magnitude as the phonon frequencies but still lower than the
highest phonon frequencies. Therefore, in principle, at least the
Sample A B C ε∞ ε0 Ẽb (meV) Eb (meV) higher-energy phonons should contribute to the screening. It is
α-CsSnCl3 11.4 1.35 3.05 4.80 29.4 37.0 0.696 not entirely clear if we should take into account the full phonon
α-CsSnBr3 18.7 1.57 6.92 5.35 32.4 19.2 0.345 enhancement factor of the phonons as some of the lower
α-CsSnI3 24.2 1.74 8.28 6.05 48.2 12.2 0.124 phonon frequencies are lower than our exciton binding energy.
However, according to the Lyddane-Sachs-Teller relation

In Table VII we list the A, B, and C parameters obtained by   ωLi 2 ε0

= , (5)
fitting to these equations. i
ωTi ε∞
The exciton binding energy is of great importance to un-
derstanding the relation between absorption and luminescence we can think of each mode as contributing a separate factor
in these materials. In a recent paper, Chen et al.20 deduced the to the enhancement of the dielectric screening. The highest
exciton binding energy from the temperature dependence of modes show the highest contribution because they have the
the photoluminescence intensity and found a rather high value highest LO-TO splitting. They already increase the dielectric
of 18 meV. They further remarked that this implies a large constant by a factor of 1.5 and so would reduce the binding
deviation of the empirical Haynes’ rule, which relates exciton energy by a factor of 2.3. But then the next phonon frequency
binding energies to band gaps. In fact on a logarithmic scale is already higher than this binding energy, so we need to
the exciton binding energies appear linear when plotted versus also include it. Ultimately, this means a factor of 9.9 for the
band gap on the linear scale. dielectric constant and a decrease of our exciton binding energy
Using the effective mass approximation, the exciton binding by a factor of 98, or, in other words, two orders of magnitude.
energy within the Wannier exciton theory can be obtained Thus we expect the exciton binding energy to be of the order
from the electron and hole effective masses and the screened of 0.1 meV. While these are all fairly rough estimates because
Coulomb interaction between the electron and hole. In the a full dynamic theory of excitons including electron-phonon
simplest form of the theory, this lead to a hydrogenic model coupling is beyond present-day capabilities of first-principles
with the binding energy being the effective Rydberg. This calculations, the reduction of the exciton binding energy
scales proportionally to the reduced mass and inversely pro- by the phonons by about two orders of magnitude should
portionally to the dielectric constant squared. This model can remain valid. In that case it is not at all in contradiction with
be further elaborated to take into account the degenerate nature Haynes’s rule; in fact, it is even somewhat low. The same
of the CBM described by the Kohn-Luttinger Hamiltonian. For conclusion applies to the other materials which all end up with
the usual situation of a degenerate VBM and a nondegenerate exciton binding energies of about 0.1 meV. This disagrees with
CBM, the theory was worked out in detail using group Chen et al.20
theoretical analysis by Baldereschi and Lipari.38 Within a We now discuss the reasons for the discrepancy. Chen
so-called spherical approximation, they obtained et al.20 claimed that the exciton is a free rather than bound
      exciton based on their excitation power dependence study,
2π 2 μ0 e4 10 μ0 2 5 μ0 2 which shows a linear dependence and no saturation over
Eb = 1+ + , (4)
h2 ε02 7 μ1 28 μ2 several decades, and on the fact that the shape of the exciton
indicates a single peak which is just shifting and broadening
with μ−1 0 =
∗ −1 −1
√ (mh ) + 2(A+2B)/3, μ1 = (A−B)/3, and
with temperature, rather than shifting weight from a bound
−1 exciton at low temperature to a free exciton at higher temper-
μ2 = 2C/ 3.
The only remaining unknown parameter in this theory is ature. However, this is not entirely convincing because their
the dielectric constant. This is a rather subtle point. If the PL is rather broad. Second, their approach for extracting the
exciton binding energy is small, at the time scale of the relative exciton binding energy relies on a number of assumptions.
motion of the electron and hole, the ions have time to relax They fitted the decay of the exciton intensity as a function of
and contribute to the screening. One should then use the static temperature to the expression
dielectric constant including the phonon contributions. On the Rspon (T ) = RG [1 − e−Eb /kB T ], (6)
other hand, if the time scale of the excitons is fast relative to
the vibrational periods, then the phonons cannot contribute, where RG is the generation rate of excitons, kB is Boltzmann’s
and one should use the high-frequency dielectric constant. In constant, and the underlying assumption is that the decay
most semiconductors, where the Wannier theory of excitons is given by the dissociation of excitons as a function of
applies, the static dielectric constant is applicable. temperature only. In other words, no nonradiative processes
However, in CsSnI3 the highest optical phonon energies are are in competition with the radiative one. This equation makes
of the order the of 18 meV, as we discuss in the next section. sense if we assume the exciton is a bound exciton and there
The static dielectric constant is larger than the high-frequency is no kinetic energy related to the exciton moving as a whole:
dielectric constant by a factor of 9.9. The high-frequency the probability at temperature T of the decay of the exciton is
dielectric constant can be obtained from our GW calculations. then given by the Boltzmann equation, and only the excitons
By taking the q → 0 and ω → 0 limit of ε−1 (q,ω) we surviving can give rise to radiation. However, they also tried to
obtain what is usually called ε∞ . Using this high-frequency justify this equation for a free exciton. In order to arrive at the
dielectric constant, we obtain an exciton binding energy Ẽb same expression under the assumption of a free exciton, they

165203-9
LING-YI HUANG AND WALTER R. L. LAMBRECHT PHYSICAL REVIEW B 88, 165203 (2013)

had to assume that the exciton motion is two-dimensional (2D) assuming a single-exciton binding energy but no kinetic energy
so that there is a steplike density of states of exciton kinetic of the excitons, may be considered to be valid as a way to
energies. In that case the number of excitons would be obtain the exciton binding energy, but it is then the energy of
Eb
a bound exciton and is therefore not in contradiction with our
− k ET −k
N(T ) = D(E)e B dE ∝ T 1 − e BT (7) conclusion that the free-exciton binding energy is two orders
of magnitude smaller. Essentially, the energy is lowered by the
if we integrate up to Eb , which presumably one could justify binding energy of the acceptor, which may easily be of the
by assuming that when an exciton gets kinetic energy above order of 20 meV. Their analysis leading to the conclusion of a
the binding energy, it will decay. In the 2D motion, however, is clearly invalid. In fact, there is also no
√ case of a three-
dimensional (3D) distribution D(E) ∝ E, and we would support for 2D motion of the electrons (or excitons) from our
obtain N (T ) ∝ T 3/2 γ (3/2,Eb /kB T )/ (3/2), with γ (a,x) band structure or from the crystal structure.
and (a) being the incomplete gamma function and gamma
function, respectively.39 This N(T ) must then be divided by C. Phonons
the partition function of free electrons, which is proportional The previous section already relied on some of our results
to T in 2D and to T 3/2 in 3D, to obtain the rate of survival of on phonons. Here we present the results of our phonon
excitons. The reason is that we consider electrons to either be calculations. Group theoretical analysis shows that in the cubic
bound to a hole or free, but we only consider minority carriers perovskite structure, there are three infrared active modes
(electrons) since the material is p type. On the other hand, of irreducible representation T1u and one silent mode of
the theory of exciton decay for free excitons in a 2D quantum symmetry T2u . Since each of these are threefold degenerate,
well was considered by Feldmann et al.40 Slightly generalizing they account for all the modes (N − 1)3, with N = 5 being
their approach to either 2D or 3D, the decay rate of excitons is the number of atoms in the unit cell.
given by The calculated frequencies of the modes are shown in
  Table VIII along with some other properties and experiment
1 Eb μγ D2 , kBT results from Donaldson et al.41 We note that Donaldson et al.41
∝   , (8)
τ M D2 reported three infrared bands for CsSnCl3 at 70, 172, and
310 cm−1 and three for CsSnBr3 at 68, 118, and 218 cm−1 .
where D is the dimensionality of the motion, M = me + mh These seem significantly higher than our calculated modes and
is the total mass of the exciton, μ is the reduced mass, and  do not distinguish between TO and LO modes. However, we
is the spread of energies of the exciton due to their nonzero notice that the lowest optical mode in our calculations had
kinetic energy, which arises from interaction with acoustic an oscillator strength that is between 12 and 21 times weaker
phonons. In other words, (T ), which in general depends on than the higher-frequency ones. We thus assume that it may
T , is the linewidth of the exciton peak. This approach appears have been missed in the experiment and ascribe the lowest
similar to that of Chen et al.20 but is arrived at by integrating experimental mode to our second-lowest mode T1u 2
. One can
the density of excitons only up to  rather than to Eb . It turns see a reasonable agreement between the average of our TO
out that for D = 2 this reduces to a form similar to Eq. (6) and LO modes, i.e., the middle of the absorption band as it
because γ (1,x) = [1 − e−x ], but the important point is that would be seen in reflectivity, and the experimental value. For
the relevant parameter , in this equation, is not the exciton CsSnCl3 our values for (ωT O + ωLO )/2 for T1u 2 3
and T1u are 77
binding energy but the linewidth of the exciton. A fit of the −1
and 182 cm , while the two lowest observed IR bands are at
data to this equation with the assumption D = 3 would give 70 and 172 cm−1 . The higher observed band of 310 cm−1 is
 ≈ 30 meV, which agrees roughly with half the linewidth of close to but somewhat lower than 2T1u 3
, although one would
the PL peak. The measured exciton lifetimes as a function of expect two-phonon absorption to be weak. In two-phonon
temperature by Chen et al.20 show a curve which is essentially absorption one rather expects to see a density of states, which
inversely proportional to the exciton-integrated intensity as a may be more influenced by zone-boundary phonons. Based on
function of temperature, as expected. However, if their exciton the phonon-dispersion curves shown by Chabot et al.,16 we
were really free and this theory were applicable, it does not expect these zone boundary phonons to be somewhat lower
tell us anything about the exciton binding energy but merely than our T1u3
. For CsSnBr3 we similarly find 64 and 145 cm−1
2 3
about the exciton linewidth. as average modes for T1u and T1u , which can be compared with
On the other hand, the study of Chung et al.4 showed that the lowest two IR bands at 68 and 118 cm−1 . Our calculation
after high-temperature cycling (300–600 K), the PL increased would overestimate the experiment by about 23%, which is
in intensity and that this increase is also accompanied by an expected in view of the overbinding of LDA. Again, the third
increase in conductivity. This indicates that chemical changes observed IR band at 218 cm−1 is not too far from 2T1u 3
at
are taking place in the material, leading to an enhanced hole −1
290 cm but again is somewhat lower because we expect it to
concentration (since the material was found to have p-type be a two-phonon density of states band. For CsSnI3 we could
conduction). Their calculations, in fact, ascribe this to the not find infrared absorption data in the literature. Donaldson
formation of Sn vacancies VSn . These observations are clear et al.41 also calculated the silent T2u mode by using force
indications that the PL in question is in fact due to an acceptor constants extracted from the observed IR active T2u modes,
bound exciton rather than a free exciton. Even material as but in view of the, in our opinion, erroneous assignment of the
grown before the heat treatments is already p type with modes, these values are not trustworthy.
concentrations of the order of 1017 cm−3 . In that case, Chen We next analyze the Lyddane-Sachs-Teller relation, already
et al.’s20 first analysis in terms of a simple rate equation, mentioned in the previous section in Eq. (5). We can see

165203-10
ELECTRONIC BAND STRUCTURE, PHONONS, AND . . . PHYSICAL REVIEW B 88, 165203 (2013)

TABLE VIII. Phonon frequencies ω (in cm−1 ), oscillator strength S (in a.u., with 1 a.u. = 253 m3 /s2 ), generalized Lyddane-Sachs-Teller
relation, and dielectric constants in α-CsSnX3 . These dielectric constants ε∞ differ from the ones in Table VII because they are obtained from
the DFPT at the LDA level instead of from GW .
2
ωLi ε0
Mode TO LO S Expt.a ωL /ωT i ωT i ε∞
ε∞ ε0

CsSnCl3 1
T1u 25 32 1.4×10−5 1.311
2
T1u 76 78 1.3×10−5 70 1.029
3
T1u 121 243 2.9×10−4 172 2.003
T2u 53 7.29 7.30 4.03 29.4
1 −5
CsSnBr3 T1u 19 26 1.1×10 1.381
2
T1u 59 69 3.7×10−5 68 1.165
3
T1u 108 183 1.6×10−4 118 1.697
T2u 37 7.45 7.45 4.34 32.4
1 −6
CsSnI3 T1u 12 18 9.5×10 1.483
2
T1u 45 63 5.5×10−5 1.412
3
T1u 101 152 1.1×10−4 1.506
T2u 29 9.93 9.93 4.86 48.2
a 41
Donaldson et al.

that the static to high-frequency dielectric constant ratio is IV. CONCLUSIONS

close to the expected value from the Lyddane-Sachs-Teller
In this paper, we have presented quasiparticle self-
relation. We observe indeed a very strong contribution of the
consistent GW band structures of the family of CsSnX3 com-
phonons to the static dielectric constant by a factor of 7–10
pounds, with X = (Cl, Br, I). Spin-orbit coupling effects were
increasing from Cl to I. This is expected in view of the strong
included. The changes in band structure with the structural
ionic bonding. Relatedly, we have rather large Born effective
distortions of the α, β, and γ phases were investigated. The
charges, obtained from DFPT, as shown in Table IX. The Born
band gap of 1.3 ± 0.1 eV was found to be in good agreement
effective charge for the X atom in the direction parallel to the
with experiment for γ -CsSnI3 , for which the best experimental
polarization of the mode is different from that in the direction
data are available. For α-CsSnCl3 , our gap of 2.7 eV is in good
perpendicular to it. It is interesting that even though Sn is
agreement with the PLE data of Voloshinovskii et al.,32 giving
nominally divalent in this material, its Born effective charge
an estimate of 2.9 eV. We also obtained good agreement for
is nonetheless larger than 4. The Cs effective charge, on the
the much higher gap of the monoclinic phase (of the order of
other hand, is close to 1.
4.5 eV) if we assume the same LDA to QSGW gap correction
Returning to the dielectric constants, the values obtained for
applies. For α-CsSnBr3 , our calculations predict a gap of
the high-frequency dielectric constant obtained from the DFPT
1.38 eV, while for the β phase we obtain 1.7 eV, in good
calculations are somewhat larger than those in the previous
agreement with experiment (1.8 eV) by Clark et al.33 While
section. This is not surprising because, first, they are obtained
this gap is a little larger than optimal for solar-cell applications,
in a rather different manner, namely, from the polarization in
the increased stability of the desirable perovskite phase for
response to a static electric field and only including electronic
the Br compound makes this material of further interest for
contributions and calculated via the Berry phase method. In the
photovoltaics. The Br material, having a somewhat smaller
previous section they were obtained from the low-frequency
lattice constant, appears to avoid the competing so-called
limit of the polarizability calculated as function of frequency
yellow phase of CsSnI3 .
in the region above the band gap. Furthermore that calculation
The most important finding is that all these materials have a
was based on the QSGW band structure, while the present
rather unusual “inverted” band structure with a nondegenerate
one uses the LDA band structure. In view of these widely
s-like valence band and a degenerate p-like conduction band.
different approaches, the agreement between the two can be
This was shown to result from the strong Sn s character of
viewed as satisfactory and certainly sufficient for our main
the valence band maximum, which is an antibonding state, or
present purpose of estimating the exciton binding energies in
anion p states with Sn s. The reason why the VBM occurs at the
the previous section.
Brillouin zone corner R is that at this point in k space, the anion
TABLE IX. Born effective charges obtained within DFPT p orbitals on the corners of the octahedron surrounding the Sn
(in units of e). atom point with their lobes of the same sign toward the center
of the octahedron and hence give the strongest interaction. The
Cs Sn X X⊥ strong intra-atomic nature of the gap explains various features:
the relatively weak dependence on the anion, the strong optical
CsSnCl3 1.162 4.462 −3.471 −1.077 luminescence and band gap absorption, and the anomalous
CsSnBr3 1.109 4.701 −3.295 −1.258 temperature dependence, increasing the gap with increasing
CsSnI3 1.060 4.954 −3.110 −1.452 temperature.

165203-11
LING-YI HUANG AND WALTER R. L. LAMBRECHT PHYSICAL REVIEW B 88, 165203 (2013)

The effective masses were determined for the band edges, We presented also results for the phonons at the  point
and the low hole mass in CsSnI3 is found to be responsible of the cubic perovskite and for the static and high-frequency
for the high hole mobility observed in this material. The other dielectric constants. The phonon frequencies for CsSnBr3 and
materials also have relatively low hole masses and may thus CsSnCl3 are in reasonable agreement with experimentally
also be expected to be excellent p-type semiconductors. observed infrared absorption bands, provided we assume that
The free-exciton binding energy was estimated to be of the the lowest vibrational modes have too low an oscillator strength
order of 0.1 meV in CsSnI3 , in contrast to a recent report20 to be observed.41
which claimed an anomalously high value of 18 meV. Our
low value is obtained when using the static dielectric constant
rather than the high-frequency one. This is consistent with the ACKNOWLEDGMENTS
fact that, while the vibrational frequencies are rather low, they
still are expected to contribute to the dielectric screening of This work was supported by the National Science Founda-
the electron-hole interaction. We also find no evidence for 2D tion under Grant No. DMR-1104595 and the Air Force Office
motion of excitons as claimed by Chen et al.20 and showed that of Scientific Research under Grant No. FA-9550-12-1-0441.
their method of determining the binding energy of the excitons Calculations were carried out using the High Performance
is flawed. Further, we propose that their PL should rather be Computing Resource in the Core Facility for Advanced
interpreted as an acceptor bound exciton than as a free exciton. Research Computing at Case Western Reserve University.

1 21
S. Sharma, N. Weiden, and A. Weiss, Z. Phys. Chem. 175, 63 M. Methfessel, M. van Schilfgaarde, and R. A. Casali, in Electronic
(1992). Structure and Physical Properties of Solids: The Use of the LMTO
2
M. Mori and H. Saito, J. Phys. C 19, 2391 (1986). Method, edited by H. Dreyssé, Lecture Notes in Physics Vol. 535
3
K. Yamada, S. Funabiki, H. Horimoto, T. Matsui, T. Okuda, and (Springer, Berlin, 2000), p. 114.
22
S. Ichiba, Chem. Lett. 20, 801 (1991). T. Kotani and M. van Schilfgaarde, Phys. Rev. B 81, 125117 (2010).
4 23
I. Chung, J.-H. Song, J. Im, J. Androulakis, C. D. Malliakas, H. Li, L. Hedin, Phys. Rev. 139, A796 (1965).
24
A. J. Freeman, J. T. Kenney, and M. G. Kanatzidis, J. Am. Chem. X. Gonze, Phys. Rev. B 55, 10337 (1997).
25
Soc. 134, 8579 (2012). X. Gonze and C. Lee, Phys. Rev. B 55, 10355 (1997).
5 26
P. Mauersberger and F. Huber, Acta Crystallogr., Sect. B 36, 683 C. Hartwigsen, S. Goedecker, and J. Hutter, Phys. Rev. B 58, 3641
(1980). (1998).
6 27
I. Chung, B. Lee, J. He, R. P. H. Chang, and M. G. Kanatzidis, X. Gonze, J.-M. Beuken, R. Caracas, F. Detraux, M. Fuchs,
Nature (London) 485, 486 (2012). G.-M. Rignanese, L. Sindic, M. Verstraete, G. Zerah, F. Jollet,
7
K. Shum, Z. Chen, J. Qureshi, C. Yu, J. J. Wang, W. Pfenninger, M. Torrent, A. Roy, M. Mikami, Ph. Ghosez, J.-Y. Raty, and D. C.
N. Vockic, J. Midgley, and J. T. Kenney, Appl. Phys. Lett. 96, Allan, Comput. Mater. Sci. 25, 478 (2002).
28
221903 (2010). J. Barrett, S. R. A. Bird, J. D. Donaldson, and J. Silver, J. Chem.
8
C. Yu, Z. Chen, J. J. Wang, W. Pfenninger, N. Vockic, J. T. Kenney, Soc. A 1971, 3105 (1971).
29
and K. Shum, J. Appl. Phys. 110, 063526 (2011). F. R. Poulsen and S. E. Rasmussen, Acta Chem. Scand. 24, 150
9
Z. Chen, J. J. Wang, Y. Ren, C. Yu, and K. Shum, Appl. Phys. Lett. (1970).
30
101, 093901 (2012). D. E. Scaife, P. F. Weller, and W. G. Fisher, J. Solid State Chem. 9,
10
M. H. Kuok, L. S. Tan, Z. X. Shen, and C. H. Huan, Solid State 308 (1974).
31
Commun. 97, 497 (1996). S. Limpijumnong and W. R. L. Lambrecht, Phys. Rev. B 65, 165204
11
Z. X. Shen, W. L. Loo, M. H. Kuok, and S. H. Tang, J. Mol. Struct. (2002).
32
326, 73 (1994). A. Voloshinovskii, S. V. Myagkota, N. S. Pidzyrailo, and M. V.
12
C. Yu, Z. Chen, and K. Shum, J. Raman Spectrosc. 44, 262 (2013). Tokarivskii, J. Appl. Spectrosc. 60, 226 (1994).
13 33
D. J. Singh, J. Appl. Phys. 112, 083509 (2012). S. J. Clark, C. D. Flint, and J. D. Donaldson, J. Phys. Chem. Solids
14
I. Lefebvre, P. E. Lippens, M. Lannoo, and G. Allan, Phys. Rev. B 42, 133 (1981).
34
42, 9174 (1990). N. Ohno and K. Sasaki, UVSOR Acta Rep. 2001, 122 (2002).
15 35
S. K. Bose, S. Satpathy, and O. Jepsen, Phys. Rev. B 47, 4276 S. V. Myagkota, P. V. Savchin, A. S. Voloshinovskiı̆, T. M. Demkiv,
(1993). Y. V. Boı̆ko, R. S. Vus, and L. S. Demkiv, Phys. Solid State 50,
16
J.-F. Chabot, M. Côté, and J.-F. Brière, in High Performance 1473 (2008).
36
Computing Systems and Applications and OSCAR Symposium D. E. Parry, M. J. Tricker, and J. D. Donaldson, J. Solid State Chem.
(Proceedings), edited by D. Senechal (NRC Research Press, 28, 401 (1978).
37
Ottawa, 2003), p. 57. K. Yang, W. Setyawan, S. Wang, M. B. Nardelli, and S. Curtarolo,
17
I. Borriello, G. Cantele, and D. Ninno, Phys. Rev. B 77, 235214 Nat. Mater. 11, 614 (2012).
38
(2008). A. Baldereschi and N. O. Lipari, Phys. Rev. Lett. 25, 373 (1970).
18 39
M. van Schilfgaarde, T. Kotani, and S. Faleev, Phys. Rev. Lett. 96, M. Abramowitz and I. A. Stegun, Handbook of Mathematical
226402 (2006). Functions (Dover, New York, 1965).
19 40
T. Kotani, M. van Schilfgaarde, and S. V. Faleev, Phys. Rev. B 76, J. Feldmann, G. Peter, E. O. Göbel, P. Dawson, K. Moore, C. Foxon,
165106 (2007). and R. J. Elliott, Phys. Rev. Lett. 59, 2337 (1987).
20 41
Z. Chen, C. Yu, K. Shum, J. J. Wang, W. Pfenninger, N. Vockic, J. D. Donaldson, S. D. Ross, and J. Silver, Spectrochim. Acta, Part
J. Midgley, and J. T. Kenney, J. Lumin. 132, 345 (2012). A 31, 239 (1975).

165203-12