PHYSICAL REVIEW B 76, 045302 共2007兲

Topological insulators with inversion symmetry

Liang Fu and C. L. Kane

Department of Physics and Astronomy, University of Pennsylvania, Philadelphia, Pennsylvania 19104, USA
共Received 14 November 2006; revised manuscript received 17 January 2007; published 2 July 2007兲
Topological insulators are materials with a bulk excitation gap generated by the spin-orbit interaction that are
different from conventional insulators. This distinction is characterized by Z2 topological invariants, which
characterize the ground state. In two dimensions, there is a single Z2 invariant that distinguishes the ordinary
insulator from the quantum spin-Hall phase. In three dimensions, there are four Z2 invariants that distinguish
the ordinary insulator from “weak” and “strong” topological insulators. These phases are characterized by the
presence of gapless surface 共or edge兲 states. In the two-dimensional quantum spin-Hall phase and the three-
dimensional strong topological insulator, these states are robust and are insensitive to weak disorder and
interactions. In this paper, we show that the presence of inversion symmetry greatly simplifies the problem of
evaluating the Z2 invariants. We show that the invariants can be determined from the knowledge of the parity
of the occupied Bloch wave functions at the time-reversal invariant points in the Brillouin zone. Using this
approach, we predict a number of specific materials that are strong topological insulators, including the
semiconducting alloy Bi1−xSbx as well as ␣-Sn and HgTe under uniaxial strain. This paper also includes an
expanded discussion of our formulation of the topological insulators in both two and three dimensions, as well
as implications for experiments.

DOI: 10.1103/PhysRevB.76.045302 PACS number共s兲: 73.43.⫺f, 72.25.Hg, 73.20.⫺r, 85.75.⫺d

I. INTRODUCTION IQHE phases, which is intimately related to their topology, is

the existence of gapless chiral edge states which are robust in
In elementary solid-state physics textbooks, an insulator is the presence of disorder.7,8
described as a material with an energy gap separating filled Recently, new topological insulating phases for systems
and empty energy bands. A more sophisticated definition of with time-reversal symmetry have been discovered.9–15 In
an insulator is that of a material for which all electronic two dimensions, the quantum spin-Hall phase is distin-
phenomena are local.1 Such a material is insensitive to guished from a band insulator by a single Z2 invariant.10 This
boundary conditions, so that in a multiply connected sample, phase exhibits gapless spin-filtered edge states, which allow
such as a ring, there is exponentially small sensitivity to for dissipationless transport of charge and spin at zero tem-
magnetic flux threading the holes. The simplest atomic insu- perature and are protected from weak disorder and interac-
lator, in which all electrons are tightly bound to atoms in tions by time-reversal symmetry. In three dimensions, a time-
closed shells, clearly satisfies both properties. Ionic and co- reversal invariant band structure is characterized by four Z2
valent insulators do too. These band insulators are topologi- invariants.13–15 Three of the invariants rely on the transla-
cally equivalent in the sense that the Hamiltonian can be tional symmetry of the lattice and are not robust in the pres-
adiabatically transformed into an atomic insulator without ence of disorder, leading to “weak topological insulators.”
going through any phase transitions. With regards to their The fourth invariant, however, is robust and distinguishes the
low-energy electronic behavior, conventional insulators are “strong topological insulator” 共STI兲.
equivalent to atomic insulators. Nontrivial Z2 invariants imply the existence of gapless
The existence of a bulk energy gap does not guarantee the surface states. In particular, in the STI phase, the surface
insensitivity to boundary conditions, and there exist phases states form a two-dimensional “topological metal,” in which
with bulk gaps, which are topologically distinct. In addition the Fermi arc encloses an odd number of Dirac points.15 This
to exotic strongly correlated phases,2,3 this fact arises even leads to a quantized Berry’s phase of ␲ acquired by an elec-
for noninteracting electrons described within band theory. tron circling the surface Fermi arc, which does not change
The simplest example is the integer quantum Hall effect under continuous perturbations.16,17 The ␲ Berry’s phase also
共IQHE兲. In two dimensions, a magnetic field introduces a signifies that with disorder, the surface states are in the sym-
cyclotron gap between Landau levels, which may be viewed plectic universality class and exhibit antilocalization.18 Thus,
as energy bands in the magnetic Brillouin zone. This phase the metallic surface states form a unique phase, which cannot
can exist even without Landau levels in the absence of a be realized in a conventional two-dimensional electron sys-
uniform magnetic field,4 though a necessary condition is that tem for which Dirac points must come in pairs.19
time-reversal symmetry be broken. Based on the band struc- The purpose of this paper is twofold. First, we will ex-
ture alone, it is difficult to tell the difference between the plain the formulation of the Z2 invariants in somewhat more
IQHE state and a band insulator. The distinction between the detail than in Ref. 15, emphasizing the physical meaning of
two is a topological property of the occupied bands which is the invariants and their relation to the surface states. Second,
encoded into the Chern integer introduced by Thouless et al.5 we will develop a method for evaluating the Z2 invariants in
Three-dimensional generalizations of the IQHE state, which systems which have inversion symmetry. This allows us to
may be viewed as layered two-dimensional 共2D兲 states, are predict a number of specific materials to be strong topologi-
indexed by a triad of Chern integers.6 A hallmark of the cal insulators.

1098-0121/2007/76共4兲/045302共17兲 045302-1 ©2007 The American Physical Society

LIANG FU AND C. L. KANE PHYSICAL REVIEW B 76, 045302 共2007兲

Most insulators are conventional insulators. In order to N

find topological insulators experimentally, it is necessary to ␦i = 兿 ␰2m共⌫i兲.

m=1
共1.1兲
develop criteria for recognizing them from their bulk band
structure. Clearly, a necessary condition is the existence of a Here, ␰2m共⌫i兲 = ± 1 is the parity eigenvalue of the 2mth occu-
bulk energy gap which owes its existence to the spin-orbit pied energy band at ⌫i, which shares the same eigenvalue
interaction. However, evaluating the Z2 invariants for a given ␰2m = ␰2m−1 with its Kramers degenerate partner. The product
band structure is, in general, a difficult problem. To date, involves the 2N occupied bands. The Z2 invariant ␯ = 0 , 1,
three general approaches have been used. which distinguishes the quantum spin-Hall phase in two di-
mensions and the strong topological insulator in three dimen-
共1兲 One powerful approach is to exploit adiabatic conti- sions, is then given by the product of all the ␦i’s,
nuity to a Hamiltonian which has extra symmetry. We used
this method to identify the quantum spin-Hall phase in 共− 1兲␯ = 兿 ␦i . 共1.2兲
graphene9,10 by arguing that the Hamiltonian for graphene is i
adiabatically connected to a Hamiltonian in which the spin Sz The other three “weak” topological invariants in three di-
is conserved. With this extra conservation law, the system mensions are also determined by ␦i. Since the parity eigen-
can be characterized by a spin Chern number, which de- values ␰n共⌫i兲 are tabulated in the band theory literature, this
scribes the quantized spin-Hall conductivity.20,21 The Z2 in- allows us to identify inversion symmetric topological insu-
variant can then be identified with the parity of the spin lating materials. Moreover, exploiting adiabatic continuity
Chern number. In the presence of Sz nonconserving terms allows us to identify topological insulators which do not
共which are inevitably present兲, the spin Chern number loses have inversion symmetry but are adiabatically connected to
its meaning.22 However, the Z2 invariant retains its value and materials which have inversion symmetry.
characterizes the quantum spin-Hall phase. Applying the above approach, we predict that the follow-
Adiabatic continuity can also be used to establish that a ing narrow gap semiconductors are strong topological insu-
material is a band insulator if an adiabatic path can be found lators: 共1兲 the alloy Bi1−xSbx, which is semiconducting for
which connects the material to an “atomic” limit. Moreover, 0.07⬍ x ⬍ 0.22, 共2兲 ␣-Sn and HgTe under uniaxial strain, and
it can be argued that the Z2 invariant changes at an appropri- 共3兲 the alloy Pb1−xSnxTe under uniaxial strain for x ⬃ xc in
ate quantum phase transition, where the bulk energy gap the vicinity of the band inversion transition. The materials
goes to zero.12,14 In general, this approach requires a continu- 共2兲 and 共3兲 were suggested by Murakami et al.25 as candi-
ous path to be found which connects the Hamiltonian in dates for spin-Hall insulators. Those authors argued that
question to a known phase. those materials share a large spin-Hall conductivity, as cal-
共2兲 It is also possible to evaluate the Z2 invariant directly culated by a Kubo formula. Our analysis of these materials is
with the knowledge of the Bloch wave functions for the oc- rather different, and we will show that PbTe is a conventional
insulator, despite its large spin-Hall conductivity, while
cupied energy bands. In Ref. 22, we established a formula
strained ␣-Sn and HgTe are topological insulators.
for the invariant in terms of a Brillouin-zone integral. This is
In Sec. II, we will present an expanded discussion of our
analogous to the calculation of the Chern number as an inte-
formulation of the Z2 invariants. Then, in Sec. III, we will
gral of the gauge invariant Berry’s curvature.5,23 However, derive Eqs. 共1.1兲 and 共1.2兲 for problems with inversion sym-
unlike the Chern invariant, the integral for the Z2 invariant metry. In Sec. IV, we will apply our method to a class of four
also involves the Berry’s potential and requires a gauge in band tight-binding models, which includes the graphene
which the wave functions are globally continuous. Since model as well as the three-dimensional 共3D兲 model intro-
time-reversal symmetry requires the Chern invariant to van- duced in Ref. 15. In Sec. V, we will apply Eqs. 共1.1兲 and
ish, a globally continuous gauge is guaranteed to exist. How- 共1.2兲 to deduce the Z2 invariants of several real materials
ever, finding a continuous gauge is not always simple. based on their known band structures. Readers uninterested
共3兲 A third approach is to characterize the zeros of Pfaff- in the technical details can skip directly to Sec. V to read
ian function introduced Ref. 10. Though the Pfaffian is not about these applications. Finally, in Sec. VI, we will con-
gauge invariant, its zeros can be determined without speci- clude with a brief discussion of the experimental implica-
fying a continuous gauge. While this approach is tedious tions for the topological insulating phases.
共especially in three dimensions兲, it has been successfully
implemented by Murakami24 to show that two-dimensional II. Z2 INVARIANTS IN TWO AND THREE DIMENSIONS
bismuth bilayers realize a quantum spin-Hall phase.
In this section, we will review our formulation of the
In this paper, we will show that the presence of inversion topological insulating phases. We begin in Sec. II A by de-
symmetry greatly simplifies the problem of identifying the Z2 fining the time-reversal polarization. In Sec. II B, we develop
invariants. We show that the invariants can be determined the Z2 characterization of a band structure as a topological
from the knowledge of the parity of the occupied band eigen- property of the occupied Bloch wave functions. In Sec. II C,
states at the eight 共or four in two dimensions兲 time-reversal we show how the Z2 invariants determine the surface-state
invariant momenta ⌫i in the Brillouin zone. Specifically, we spectrum. In Sec. II C, we consider a more general formula-
will show that the Z2 invariants are determined by the quan- tion of the Z2 invariant as a sensitivity of a bulk crystal to
tities boundary conditions.

045302-2
TOPOLOGICAL INSULATORS WITH INVERSION SYMMETRY PHYSICAL REVIEW B 76, 045302 共2007兲

A. Time-reversal polarization ends correspond to the inner and outer surfaces. This system
In Ref. 22, we introduced the concept of the time-reversal can be threaded by two independent magnetic fluxes, and
polarization, in the same spirit as charge polarization. This they correspond to the two components of the momentum
allows for an interpretation of the Z2 invariants using a perpendicular to G. There are four time-reversal invariant
Laughlin-type gedanken experiment on a cylinder.26 In the surface momenta ⌳a, corresponding to the two fluxes being
ordinary quantum Hall effect, the Chern invariant can be either 0 or h / 2e. The band structure can be characterized by
interpreted in a similar way. To motivate the time-reversal the difference in the time-reversal polarization between any
polarization, we therefore begin with a discussion of the pair.
charge polarization. The Z2 invariants can be deduced from the topological
The charge polarization determines the surface charges structure of the Bloch wave functions of the bulk crystal in
present in a finite system. Since electrons may be added or the Brillouin zone. Consider a time-reversal invariant peri-
removed from a surface, the charge polarization is defined odic Hamiltonian H with 2N occupied bands characterized
only modulo an integer.27–30 However, changes in the charge by Bloch wave functions
polarization induced by adiabatic changes in the Hamiltonian 兩␺n,k典 = eik·r兩un,k典. 共2.1兲
are well defined. In Laughlin’s gedanken experiment for the
integer quantum Hall effect, a quantum of magnetic flux h / e Here, 兩un,k典 are cell periodic eigenstates of the Bloch Hamil-
is adiabatically inserted in a cylindrical quantum Hall sample tonian
at filling ␯ = N. The resulting transfer of N electrons from one
end of the cylinder to the other can be interpreted as a change H共k兲 = e−ik·rHe+ik·r . 共2.2兲
in the charge polarization of the cylinder. The Chern invari-
ant, which characterizes the integer quantum Hall state, pre- We require 兩␺n,k+G典 = 兩␺n,k典 for reciprocal-lattice vectors G,
cisely characterizes this quantized change in charge polariza- so that the Brillouin zone in which k is defined is a torus.
tion. This implies 兩un,k+G典 = e−iG·r兩un,k典. Time-reversal symmetry
The time-reversal polarization is a Z2 quantity, which implies 关H , ⌰兴 = 0, where ⌰ = exp共i␲Sy兲K is the time-
characterizes the presence or absence of a Kramers degen- reversal operator 共Sy is spin and K complex conjugation兲,
eracy associated with a surface. Like the charge polarization, which for spin 1 / 2 particles satisfies ⌰2 = −1. It follows that
its value can be changed by adding an extra electron to the H共−k兲 = ⌰H共k兲⌰−1.
surface. Thus, by itself, the time-reversal polarization is not There are special points k = ⌫i in the Brillouin zone which
meaningful. However, changes in the time-reversal polariza- are time-reversal invariant and satisfy −⌫i = ⌫i + G for a
tion due to adiabatic changes in the bulk Hamiltonian are reciprocal-lattice vector G. There are eight such points in
well defined. Specifically, the change in the time-reversal three dimensions and four in two dimensions. At these
polarization when half a flux quantum h / 2e is threaded points, H共⌫i兲 = ⌰H共⌫i兲⌰−1, so that the eigenstates are Kram-
through a cylinder defines a Z2 invariant, which is analogous ers degenerate. In the following, it will be useful to use two
to the Chern invariant, and distinguishes topological insula- different notations to label the distinct ⌫i. 共1兲 The eight 共or
tors. four兲 ⌫i can be indexed by three 共or two兲 integers nl = 0 , 1
defined modulo 2, which specify half a “mod 2 reciprocal-
lattice vector,”
B. Z2 invariants characterizing a band structure
The topological invariant characterizing a two- 1
⌫i=共n1n2n3兲 = 共n1b1 + n2b2 + n3b3兲, 共2.3兲
dimensional band structure may be constructed by imagining 2
a long cylinder whose axis is parallel to a reciprocal-lattice
vector G and which has a circumference of a single lattice where bl are primitive reciprocal-lattice vectors. Two mod 2
constant. Then, the magnetic flux threading the cylinder reciprocal-lattice vectors are equivalent if they differ by
plays the role of the circumferential 共or “edge”兲 crystal mo- twice a reciprocal-lattice vector. 共2兲 A second notation is use-
mentum kx, with ⌽ = 0 and ⌽ = h / 2e corresponding to two ful when considering a surface perpendicular to reciprocal-
edge time-reversal invariant momenta kx = ⌳1 and kx = ⌳2. lattice vector G. The surface time-reversal invariant mo-
The Z2 invariant characterizes the change in the Kramers menta ⌳a will be projections of pairs of ⌫i which differ by
degeneracy at the ends of this one-dimensional system be- G / 2, as shown in Fig. 1. Given G, we can define ⌫i=共a␮兲,
tween kx = ⌳1 and kx = ⌳2. such that ⌫a1 − ⌫a2 = G / 2.
For a three-dimensional crystal, imagine a “generalized The change in the time-reversal polarization associated
cylinder” which is long in one direction 共parallel to G兲 but, with a cylinder oriented along G due to changing the flux
in the other two directions, has a width of one lattice con- between two values corresponding to ⌳a and ⌳b can be cal-
stant with periodic boundary conditions. While this structure culated by a method analogous to the calculation of the
cannot be pictured as easily as a regular cylinder, a distorted charge polarization as a Berry’s phase.27–30 In Ref. 22, we
共but topologically equivalent兲 version can be visualized as a showed that the result is most simply expressed in terms of
torus with a finite thickness. This “Corbino donut” is analo- the quantities
gous to the generalized cylinder in the same way the Corbino
disk is analogous to the regular cylinder. The “long” direc-
冑det关w共⌫i兲兴
␦i = = ± 1, 共2.4兲
tion corresponds to the thickness of the torus, and the two Pf关w共⌫i兲兴

045302-3
LIANG FU AND C. L. KANE PHYSICAL REVIEW B 76, 045302 共2007兲

FIG. 1. 共a兲 A two-dimensional cylinder threaded by magnetic flux ⌽. When the cylinder has a circumference of a single lattice constant,
⌽ plays the role of the edge crystal momentum kx in band theory. 共b兲 The time-reversal invariant fluxes ⌽ = 0 and h / 2e correspond to edge
time-reversal invariant momenta ⌳1 and ⌳2. ⌳a are projections of pairs of the four bulk time-reversal momenta ⌫i=共a␮兲, which reside in the
two-dimensional Brillouin zone indicated by the shaded region. 共c兲 In three dimensions, the generalized cylinder can be visualized as a
Corbino donut, with two fluxes, which correspond to the two components of the surface crystal momentum. 共d兲 The four time-reversal
invariant fluxes ⌽1, ⌽2 = 0 , h / 2e correspond to the four two-dimensional surface momenta ⌳a. These are projections of pairs of the eight
⌫i=共a␮兲 that reside in the bulk 3D Brillouin zone.

where w is the 2N ⫻ 2N unitary matrix defined by det关w共k兲兴 → det关w共k兲兴e−i共␪k+␪−k兲 , 共2.8兲

wmn共k兲 ⬅ 具um−k兩⌰兩unk典. 共2.5兲
Pf关w共⌫i兲兴 → Pf关w共⌫i兲兴e−i␪⌫i . 共2.9兲
Since 具⌰a 兩 ⌰b典 = 具b 兩 a典 and ⌰2 = −1, the matrix w共k兲 is anti-
symmetric for k = ⌫i, allowing for the definition of the Pfaff- If we choose ␪k = k · R for a lattice vector R, the Brillouin-
ian, which satisfies det关w兴 = Pf关w兴2. Given a surface G, the zone periodicity of unk is preserved. From Eq. 共2.8兲, it is
time-reversal polarization associated with surface momen- clear that det关w共k兲兴 is unchanged. However, if G · R = 2␲, it
tum ⌳a was found to be follows that Pf关w共⌫a1兲兴Pf关w共⌫a2兲兴, and hence ␲a changes
␲a = ␦a1␦a2 . 共2.6兲 sign. Since this gauge transformation changes the sign of ␲a
for all a, however, the product ␲a␲b, which gives the change
This formula is analogous to the expression for the charge in the time-reversal polarization between ⌳a and ⌳b, remains
polarization as a Berry’s phase along a closed cycle in mo- invariant. In general, the product of any four ␦i’s for which
mentum space.28 ⌫i lie in the same plane is gauge invariant and defines a
Because of the square root, the sign of ␦i is ambiguous. topological invariant characterizing the band structure.
However, since we require 兩un,k典 to be continuous, In two dimensions, there are four time-reversal invariant
冑det关w共k兲兴 is defined globally throughout the Brillouin zone. momenta ⌫i and a single Z2 invariant, given by
The sign ambiguity then cancels for ␲a. The existence of the
global square root is guaranteed by the fact that the phase 4
winding of det关w共k兲兴 around any closed loop C is zero, so 共− 1兲 = 兿 ␦i .
␯
共2.10兲
that 冑det关w共k兲兴 has no branch cuts. When C is contractable, i=1
the vanishing phase winding follows from the continuity of
兩un,k典. For noncontractable loops, which can be continuously In three dimensions there are eight time-reversal invariant
deformed to satisfy C = −C, it follows from the fact that points. This leads to four independent Z2 invariants.13–15 One
det关w共−k兲兴 = det关w共k兲兴. of these invariants, ␯0, can be expressed as the product over
The inherent ambiguity of the time-reversal polarization all eight points,
is reflected in the fact that the ␲a are not gauge invariant. For
8
instance, consider a gauge transformation of the form

再 冎
共− 1兲 = 兿 ␦i .
␯0
共2.11兲
ei␪k兩un,k典 for n = 1 i=1
兩un,k典 → 共2.7兲
兩un,k典 for n ⫽ 1.
The other three invariants are given by products of four ␦i’s
Under this transformation, for which ⌫i reside in the same plane.

045302-4
TOPOLOGICAL INSULATORS WITH INVERSION SYMMETRY PHYSICAL REVIEW B 76, 045302 共2007兲

共− 1兲␯k = 兿
nk=1;n j⫽k=0,1
␦i=共n1n2n3兲 . 共2.12兲

␯0 is clearly independent of the choice of primitive

reciprocal-lattice vectors bk in Eq. 共2.3兲. 共␯1␯2␯3兲 are not.
However, they may be viewed as components of a mod 2
reciprocal-lattice vector
G ␯ = ␯ 1b 1 + ␯ 2b 2 + ␯ 3b 3 . 共2.13兲
This vector may be explicitly constructed from the ␦i’s as
follows. A gauge transformation of the form of Eq. 共2.7兲 can FIG. 2. Schematic representations of the surface energy levels of
a crystal in either two or three dimensions as a function of surface
change the signs of any four ␦i for which the ⌫i lie in the
crystal momentum on a path connecting ⌳a and ⌳b. The shaded
same plane. Such transformations do not change the invari-
region shows the bulk continuum states, and the lines show discrete
ants 共␯1␯2␯3兲. By a sequence of these transformations, it is
surface 共or edge兲 bands localized near one of the surfaces. The
always possible to find a gauge in which ␦i = −1 for at most Kramers degenerate surface states at ⌳a and ⌳b can be connected to
one nonzero ⌫i. Define ⌫* = ⌫i if there is one such point. If each other in two possible ways, shown in 共a兲 and 共b兲, which reflect
there is none, then set ⌫* = 0. In this gauge, the mod 2 the change in time-reversal polarization ␲a␲b of the cylinder be-
reciprocal-lattice vector is G␯ = 2⌫*. The remaining invariant tween those points. Case 共a兲 occurs in topological insulators and
␯0 is then determined by ␦i at ⌫i = 0. guarantees that the surface bands cross any Fermi energy inside the
As we will explain below in Sec. II C 2, the latter invari- bulk gap.
ants ␯k are not robust in the presence of disorder. We refer to
them as “weak” topological invariants. On the other hand, ␯0 on the other hand, the ground state remains nondegenerate.
is more fundamental and distinguishes the “strong” topologi- This construction establishes the connection between the sur-
cal insulator. face states and the bulk topological invariants. When ␲a␲b
Formulas 共2.10兲–共2.12兲 are a bit deceptive because they = −1 共+1兲, the surface spectrum is like Fig. 2共a兲 共2b兲.
appear to depend solely on a local property of the wave It follows that when ␲a␲b = −1 共+1兲, a generic Fermi en-
functions. Knowledge of the global structure of 兩unk典, how- ergy inside the bulk gap will intersect an odd 共even兲 number
ever, is necessary to construct the continuous gauge required of surface bands between ⌳a and ⌳b. Thus, when ␲a␲b =
to evaluate Eq. 共2.4兲. The existence of globally continuous −1, the existence of surface states is topologically protected.
wave functions is mathematically guaranteed because the The details of the surface-state spectrum will depend on the
Chern number for the occupied bands vanishes due to time- Hamiltonian in the vicinity of the surface. In Fig. 2, we have
reversal symmetry. However, determining a continuous assumed that surface bound states exist for all momenta. This
gauge is not always simple. need not be the case, since it is possible that by varying the
surface Hamiltonian, the degenerate states at ⌳a and ⌳b can
be pulled out of the gap into the bulk continuum states. This,
C. Surface states
however, does not change our conclusions regarding the
The spectrum of surface 共or edge兲 states as a function of number of Fermi energy crossings. When ␲a␲b = −1, there
momentum parallel to the surface 共or edge兲 is equivalent to still must exist surface band traversing the energy gap.
the spectrum of discrete end states of the cylinder as a func- In the two-dimensional quantum spin-Hall phase, ␲1␲2 =
tion of flux. Figure 2 schematically shows two possible end −1, and there will be an odd number of pairs of Fermi
state spectra as a function of momentum 共or equivalently points.9,10 In the simplest case where there is a single pair,
flux兲 along a path connecting the surface time-reversal in- the states at the Fermi energy will be spin filtered in the
variant momenta ⌳a and ⌳b. Only end states localized at one sense that the expectation value of the spin in the right and
of the ends of the cylinder are shown. The shaded region left moving states will have opposite sign. These states are
gives the bulk continuum states. Time-reversal symmetry re- robust in the presence of weak disorder and interactions be-
quires the end states at ⌳a and ⌳b to be twofold degenerate. cause time-reversal symmetry forbids elastic backscattering.
However, there are two possible ways these degenerate states Strong interactions, however, can lead to an electronic insta-
can connect up with each other. In Fig. 2共a兲, the Kramers bility that opens a gap.31,32 The resulting ground state, how-
pairs “switch partners” between ⌳a and ⌳b, while in Fig. ever, breaks time-reversal symmetry.
2共b兲, they do not. In three dimensions, the Kramers degenerate band cross-
These two situations are distinguished by the Z2 invariant ings that occur at ⌳a in the surface spectrum are two-
characterizing the change in the time-reversal polarization of dimensional Dirac points. While such Dirac points will oc-
the cylinder when the flux is changed between the values cur in any time-reversal invariant system with spin-orbit
corresponding to ⌳a and ⌳b. Suppose that at the flux corre- interactions, the nontrivial structure here arises from the way
sponding to ⌳a the ground state is nondegenerate, and all in which the Dirac points at different ⌳a are connected to
levels up to and including the doublet ␧a1 are occupied. If the each other. This is determined by the relative signs of the
flux is adiabatically changed to ⌳b, then for Fig. 2共a兲, the four ␲a associated with any surface.
doublet ␧b1 is half filled, and the ground state has a twofold In Fig. 3, we depict four different topological classes for
Kramers degeneracy associated with the end. For Fig. 2共b兲, three-dimensional band structures labeled according to ␯0;

045302-5
LIANG FU AND C. L. KANE PHYSICAL REVIEW B 76, 045302 共2007兲

which will necessarily break the time-reversal

symmetry.22,31,32
In the strong topological insulator, it is possible that the
Fermi energy can be tuned to intersect a single Dirac point.
This is a rather unique situation, because lattice Dirac fermi-
ons are generally expected to come in pairs.19 These surface
Dirac fermions are reminiscent of domain-wall fermions
which have been studied in the context of lattice gauge
theories.33 The surface can be viewed as an interface be-
tween the topological insulator and a conventional insulator
共the vacuum兲. These two phases can be characterized in
terms of a three-dimensional Dirac fermion, whose mass has
FIG. 3. Diagrams depicting four different phases indexed by ␯0; opposite signs in the two phases 共see, for example, Sec. III兲.
共␯1␯2␯3兲. The top panel depicts the signs of ␦i at the points ⌫i on the The domain wall between the two is then characterized by a
vertices of a cube. The bottom panel characterizes the band struc- gapless Fermion, or zero mode, which is related to the zero
ture of a 001 surface for each phase. The solid and open circles
energy midgap states that appear in a one-dimensional
depict the time-reversal polarization ␲a at the surface momenta ⌳a,
Peierls insulator at a soliton.34 However, there are some im-
which are projections of pairs of ⌫i that differ only in their z com-
portant differences between our model and the conventional
ponent. The thick lines indicate possible Fermi arcs that enclose
specific ⌳a.
applications of domain-wall fermions. 共1兲 In our problem,
there is no reason to have particle-hole symmetry, so tuning
is required for the Fermi energy to be at the Dirac point. 共2兲
共␯1␯2␯3兲, which are characterized by different values of ␦i for The domain-wall fermion applications have often been used
the eight ⌫i, represented as the vertices of a cube in momen- to model chiral fermions in even-dimensional space-time.33
tum space. The lower panel shows the surface Brillouin zone Our 共2 + 1兲-dimensional surface Dirac fermions are not chi-
for a 001 surface with the four ⌳a labeled by either filled or ral. Nonetheless, they realize the 共2 + 1兲-dimensional “parity
solid circles, depending on the value of ␲a = ␦i=共a1兲␦i=共a2兲. The anomaly.”35
surface band structure will resemble Fig. 2共a兲 on paths con- The parity anomaly arises for a single 共i.e., undoubled兲
necting two filled circles or two empty circles, and will re- species of massless Dirac fermion in 2 + 1 dimensions. When
semble Fig. 2共b兲 on paths connecting a filled circle to an the response to the electromagnetic field is naively computed
empty circle. This allows us to draw a minimal Fermi arc in this model, one finds35
共denoted by the thick lines兲, which separates regions contain-
ing the filled circles from regions containing the empty
circles. e2
J␭ = ± ⑀ ␮␯␭F ␮␯ , 共2.14兲
4h
1. Strong topological insulator
For the strong topological insulator, ␲1␲2␲3␲4 = −1, so
that a single ␲a=a* differs in sign from the other three. The where J␭ is the three current and F␮␯ is the electromagnetic
simplest Fermi arc, shown in Fig. 3共d兲, thus encloses ⌳a* field tensor in 2 + 1 dimensions. This appears “anomalous” in
once. As in the two-dimensional case, this Fermi arc cannot the sense that the electromagnetic field gives rise to currents
be eliminated. In general, time-reversal symmetry requires which appear to violate the symmetries of the Dirac Hamil-
that any time-reversal invariant Fermi arc C satisfies C = tonian. The sign ambiguity in Eq. 共2.14兲 is due to the regu-
−C. It follows that the Berry’s phase acquired by an electron larization procedure, in which a finite mass is included to
circling the Fermi arc can only be either 0 or ␲. Since the control divergences and taken to zero at the end. The origin
Fermi arc of the strong topological insulator encloses a of the singular behavior is the subtlety of this limiting pro-
single Dirac point, an electron circling the Fermi arc acquires cedure.
a Berry’s phase of ␲. Since this cannot be changed by con- In a magnetic field, the Dirac equation leads to a Landau
tinuous changes to the Hamiltonian, we conclude that the ␲ level at exactly zero energy. At exactly half filling, the sys-
Berry’s phase is a generic feature of the surface Fermi arc in tem is thus at a critical point separating two quantum Hall
the strong topological insulator phase. Such a Fermi arc de- states with ␴xy = ± 共1 / 2兲e2 / h. This explains the singular be-
fines a “topological metal,”22 which is topologically pro- havior described above. Indeed, the regulator mass term dis-
tected and, unlike an ordinary metal, cannot be depleted. cussed above which opens a gap necessarily violates time-
In the presence of weak disorder, the ␲ Berry’s phase reversal symmetry because it lifts a Kramers degeneracy.
changes the sign of the weak localization correction to the This leads to quantum Hall states even in zero applied mag-
conductivity and gives rise to antilocalization, as in the sym- netic field.
plectic universality class.16,18 We suspect that in the absence For our problem, in the absence of time-reversal
of electron-electron interactions, these surface states cannot symmetry-breaking perturbations, we do not expect anoma-
be localized even for strong disorder 共provided the bulk to- lous currents to occur. However, in a magnetic field, the par-
pological phase is not destroyed兲. As in the 2D case, how- ity anomaly shows up in the quantum Hall effect because the
ever, electron interactions can lead to a localized phase, surface Hall conductivity will be quantized in half integers,

045302-6
TOPOLOGICAL INSULATORS WITH INVERSION SYMMETRY PHYSICAL REVIEW B 76, 045302 共2007兲

冉 冊
␴xy = n +
1 e2
2 h
. 共2.15兲
associated with the plane kz = 0. The topological invariants
will then be given by ␯0 = 0 and G␯ = 共2␲ / a兲ẑ. This structure
will remain when weak coupling between the layers is intro-
It is interesting to ask whether such a “fractional” integer duced. More generally, quantum spin-Hall states stacked in
quantum Hall effect could be measured. Unfortunately, in a the G direction will have G␯ = G mod 2. This implies that
standard transport experiment in which currents and voltages quantum spin-Hall states stacked along different directions
are measured by attaching leads to a sample, the 1 / 2 cannot G1 and G2 are equivalent if G1 = G2 mod 2.
be directly measured. The problem is that in a slab geometry, The surface states can also be understood in this manner.
there is no simple way to isolate the “top” surface from the When the coupling between the layers is zero, it is clear that
“bottom” surface. The two will always be measured in par- the gap in the 2D system implies that there will be no surface
allel, and the two half integers will always add up to an states on the top and bottom surfaces. On the sides, however,
integer. For such a transport experiment, there is no getting the Fermi points guaranteed that the edges of the two-
around the “fermion doubling problem.” What is required is dimensional system will become straight Fermi lines in the
a method for measuring the current on the top surface with- kz direction. The two Fermi lines will enclose two time-
out measuring the bottom surface. If it can be established reversal invariant momenta, which occur at kz = 0 and kz
that the currents are flowing on both surfaces, then dividing = ␲ / a, as in Fig. 3.
the measured Hall conductance by 2 could, in principle, Since the surface Fermi arc encloses an even number of
demonstrate the half quantization. surface time-reversal invariant momenta 共and hence an even
A lattice realization of the parity anomaly was proposed number of 2D Dirac points兲, it follows that the Berry’s phase
by Fradkin and co-workers,36,37 who studied a tight-binding associated with the Fermi arc is zero. Thus, the surface states
model for PbTe in the presence of a domain wall where the of the weak topological insulators do not enjoy the same
Pb and Te atoms are interchanged. They showed that in their level of topological protection as those of the strong topo-
model, the domain wall exhibits massless Dirac fermions and logical insulator. Below we will argue that in the presence of
has a finite Hall conductivity even at zero magnetic field. disorder, the weak topological invariants lose their meaning.
Their model, however, is rather different from ours. In the
presence of the domain wall, their Hamiltonian explicitly
violates time-reversal symmetry,4 and it leads to an even D. Z2 invariant and boundary-condition sensitivity
number of species of Dirac fermions on the domain wall. Niu et al. generalized the topological characterization of
Haldane introduced a model of the quantum Hall effect on the integer quantum Hall effect to express the Chern invari-
honeycomb lattice in a periodic magnetic field.4,38 This ant in terms of the sensitivity of the ground state of a bulk
model, which also breaks time-reversal symmetry, realizes crystal to phase twisted boundary conditions.39 This is more
the parity anomaly 共with a single Dirac fermion兲 when the fundamental than the characterization in terms of Bloch
Hamiltonian is tuned to the transition between the quantum wave functions because it does not rely on the translational
Hall phase and the insulator. In this model, however, the Hall symmetry of the crystal. It explains the topological stability
conductivity is an integer. of the Hall conductance in the presence of weak disorder. In
The surface of the strong topological insulator is thus this section, we consider a corresponding generalization of
unique in that it can generate a single Dirac fermion without the Z2 invariant.
violating time-reversal symmetry and, in principle, exhibits To do so, we consider large 共but still finite兲 crystal with
the half quantized quantum Hall effect. periodic boundary conditions in all but one direction. A
phase twist ei␪i is associated with each periodic boundary
2. Weak topological insulator
condition. This has the same structure as the cylinder 共and
When ␯0 = 0, states are classified according to G␯. We re- generalized cylinder兲 considered in Sec. II B, but now the
fer to the states with G␯ ⫽ 0 as weak topological insulators.15 circumferences are much larger. The fluxes now correspond
␯0 = 0 implies that for any surface, the associated time- to the phase twists ␪i = ⌽ie / ប. Since the cylinder is still finite,
reversal polarizations will satisfy ␲1␲2␲3␲4 = + 1. This im- the discrete states associated with the ends can be character-
plies that either 共1兲 all of the ␲a⬘s are the same or 共2兲 two will ized by their degeneracy. This allows us to characterize the
be positive and two will be negative. The former case occurs change in time-reversal polarization when the phase twists
for surfaces G = G␯ mod 2, where G␯ is given in Eq. 共2.13兲. are changed by ␲. For noninteracting electrons, the invari-
For these surfaces, there are no topologically protected sur- ants characterizing a large cylinder can be deduced from the
face states. For G ⫽ G␯ mod 2, two of the ⌳a’s are positive band-structure invariants by treating the entire sample to be a
and two negative. The Fermi arc encloses the two ⌳a’s which unit cell of an even larger crystal. It is therefore necessary to
have the same sign for ␲a. consider the effect of enlarging the unit cell on the topologi-
These states can be interpreted as layered two- cal invariants.
dimensional quantum spin-Hall states. To see this, consider The 2D invariant ␯ is preserved when the unit cell is
two-dimensional planes in the quantum spin-Hall state enlarged. This is easiest to see by considering the effect of
stacked in the z direction. When the coupling between the doubling the unit cell on the surface spectra of Fig. 2. The
layers is zero, the electronic states will be independent of kz. effect of doubling the unit cell will be to fold the momenta
It follows that the four ␦i’s associated with the plane kz ⌳a and ⌳b back on top of each other. If after enlarging the
= ␲ / a will have product −1 and will be the same as the four unit cell we add a small random potential, which lowers the

045302-7
LIANG FU AND C. L. KANE PHYSICAL REVIEW B 76, 045302 共2007兲

translational symmetry, then all “accidental” degeneracies in- version and time-reversal symmetries is that the Berry cur-
troduced by the zone folding will be lifted, while the Kram- vature F共k兲 must vanish because it must simultaneously be
ers degeneracies at the time-reversal invariant momenta will odd under time-reversal 关F共−k兲 = −F共k兲兴 and even under in-
remain. It is then clear that the manner in which the degen- version 关F共−k兲 = + F共k兲兴.17 The Berry curvature is given by
erate states at ⌳a are connected to each other is preserved ⵜk ⫻ A共k兲, where the Berry’s potential is
when the bands are folded back. Since the invariant ␯ is
2N
unchanged when the unit cell is enlarged, it may be used to
characterized the bulk system. A共k兲 = − i 兺 具un,k兩ⵜk兩un,k典, 共3.2兲
In three dimensions, the strong topological invariant ␯0 is n=1

also invariant under enlarging the unit cell. This can be seen where the sum is over the 2N occupied bands. The fact that
by noting that if the surface Fermi arc is folded back, the F共k兲 = 0 suggests that it is possible to choose a globally con-
number of time-reversal invariant momenta enclosed is pre- tinuous “transverse” gauge in which A共k兲 = 0. We will show
served modulo 2. The weak topological invariants ␯k, how- that in this special gauge, the ␦i defined in Eq. 共2.4兲 are given
ever, are not preserved by this procedure. It is possible that by Eq. 共1.1兲, so the Z2 invariants can be easily evaluated.
for a Fermi arc which encloses two time-reversal invariant The transverse gauge may be explicitly constructed as fol-
momenta, the two points can be folded back on top of each lows. In an arbitrary gauge, consider the 2N ⫻ 2N matrix
other. When the two bands are coupled to each other, a gap
can then open at the Fermi energy, so that the Fermi arc can vmn共k兲 = 具um,k兩P⌰兩un,k典. 共3.3兲
disappear.
We thus conclude that the weak topological invariants are Since 具a兩b典 = 具⌰b兩⌰a典 and ⌰2 = −1, it follows that v共k兲 is
only defined for the band structure of a perfect crystal and antisymmetric. Since 关P⌰ , H共k兲兴 = 0, v共k兲 is unitary. Thus,
rely on the lattice translational symmetry. For a clean system, the Pfaffian of v共k兲 is defined and has unit magnitude. The
they have important implications for the surface-state spec- phase of Pf关v共k兲兴 depends on the gauge, and its gradient is
trum, but the topological distinction is eliminated by disor- related to A共k兲:
der. The strong topological invariant ␯0, however, is robust
and characterizes a bulk three-dimensional phase. i
A共k兲 = − Tr关v共k兲†ⵜkv共k兲兴 = − iⵜk log Pf关v共k兲兴.
The fragility of the weak topological invariants can also 2
be understood by considering stacked two-dimensional quan- 共3.4兲
tum spin-Hall states. If two identical quantum spin-Hall
states are coupled together, the resulting two-dimensional The first equality is derived by differentiating Eq. 共3.3兲 and
system will be a simple insulator and will not have topologi- using the unitarity of v共k兲. The second equality follows from
cally protected edge states. Thus, a stack of an even number the facts that det关v兴 = Pf关v兴2 and ⵜk log det关v兴
of layers will be equivalent to a stack of insulators, while a = Tr关ⵜk log v共k兲兴 = Tr关v†共k兲ⵜkv共k兲兴.
stack of an odd number of layers will define a 共thicker兲 2D To set A共k兲 = 0, we thus adjust the phase of 兩unk典 to make
quantum spin-Hall state. This sensitivity to the parity of the
number of layers indicates that the weak topological invari- Pf关v共k兲兴 = 1. 共3.5兲
ants do not characterize a robust three-dimensional phase
but, rather, are only properties of the band structure. This can be done, for instance, by a transformation of the
This formulation of the Z2 invariant in terms of the form of Eq. 共2.7兲, under which Pf关v共k兲兴 → Pf关v共k兲兴e−i␪k. In
change in the time-reversal polarization of a large system as this gauge, the problem of continuing 冑det关w共k兲兴 between
a function of twisted boundary conditions can be generalized the ⌫i in Eq. 共2.4兲 is eliminated because det关w共k兲兴 = 1 for all
to interacting systems. This suggests that the strong topologi- k. This can be seen by noting that v共k兲 has the property
cal insulator phase remains robust in the presence of weak v共−k兲 = w共k兲v共k兲*w共k兲T and using the identity Pf关XAXT兴
interactions.22 = Pf关A兴det关X兴.
All that remains for Eq. 共2.4兲 is to evaluate Pf关w共⌫i兲兴. To
III. Z2 INVARIANTS WITH INVERSION SYMMETRY this end, we note that

In this section, we show how the presence of inversion wmn共⌫i兲 = 具␺m,⌫i兩P共P⌰兲兩␺n⌫i典. 共3.6兲
symmetry greatly simplifies the evaluation of the Z2 invari-
ants. We will prove Eq. 共1.1兲, which allows topological in- Here, we have used P2 = 1, along with the antilinearity of ⌰,
sulators to be identified in a straightforward manner. which allows us to replace 兩un⌫i典 by 兩␺n⌫i典 = 兩␺n−⌫i典 in Eq.
Suppose that the Hamiltonian H has an inversion center at 共2.5兲. Since 关H , P兴 = 0, 兩␺n⌫i典 is an eigenstate of P with ei-
r = 0. It follows that 关H , P兴 = 0, or equivalently H共−k兲 genvalue ␰n共⌫i兲 = ± 1. After changing 兩␺n⌫i典 back to 兩un⌫i典, it
= PH共k兲P−1, where the parity operator is defined by follows that

P兩r,sz典 = 兩− r,sz典. 共3.1兲 wmn共⌫i兲 = ␰m共⌫i兲vmn共⌫i兲. 共3.7兲

Here, r is the 共three-dimensional兲 coordinate and sz is the The Pfaffian can be deduced from the following argument,
spin which is unchanged by P because spin is a pseudovec- which uses the fact that the Pf关w兴 will be a polynomial in ␰n.
tor. The key simplification for problems with combined in- First, note that

045302-8
TOPOLOGICAL INSULATORS WITH INVERSION SYMMETRY PHYSICAL REVIEW B 76, 045302 共2007兲

2N HgTe/ CdTe quantum wells introduced recently by Bernevig

Pf关w兴 = det关w兴 = det关v兴 兿 ␰n .
2
共3.8兲 et al.40
n=1
A. General model
Due the Kramers degeneracy, the distinct states 兩u2m,⌫i典 and
We assume that each unit cell associated with Bravais
兩u2m+1,⌫i典 ⬅ ⌰ 兩 u2m,⌫i典 share the same parity eigenvalue. Thus,
lattice vector R has four states 兩R , n典. If the unit cell is parity
each eigenvalue appears twice in the product in Eq. 共3.8兲.
invariant, then the parity operator P may be represented as a
Taking the square root, we find
4 ⫻ 4 matrix as
N
Pf关w兴 = Pf关v兴 兿 ␰2m . 共3.9兲 P兩R,n典 = 兺 P̂nm兩− R,m典. 共4.1兲
m=1 m

The sign of the square root is fixed by the special case in In Secs. IV B and IV C, we will consider examples in which
which all ␰n = 1, so that w = v. Since Pf关v兴 = 1, we conclude each unit cell consists of two sublattices 共denoted by Pauli
that in the transverse gauge, matrix ␴z兲 which are interchanged by inversion and two spin
degrees of freedom 共denoted by sz兲. Therefore,
N
␦i = 兿 ␰2m共⌫i兲.
m=1
共3.10兲 P̂ = ␴x 丢 I, 共4.2兲

where I is the identity for the spin indices. In Sec. IV D, P̂

Equation 共3.10兲 is a central result of this paper. It means will have a slightly different form. The time-reversal opera-
that with inversion symmetry, the Z2 topological invariants tor acting on the four component basis states is represented
can be deduced from the knowledge of the parity of each pair by
of Kramers degenerate occupied energy bands at the four 共or
eight in three dimensions兲 time-reversal and parity invariant ˆ = i共I 丢 sy兲K,
⌰ 共4.3兲
points in the Brillouin zone. This provides a simple method
for determining the topological phase of any inversion sym- where K is complex conjugation and I acts on the orbital
metric insulator, without having to know about the global indices.
properties of the energy bands. Given a lattice Hamiltonian H in the 兩R , n典 basis, we now
In Eq. 共3.10兲, it appears as though each of the four 共or consider the Bloch Hamiltonian
eight兲 ␦i have gauge independent meaning and thus provide H共k兲 = eik·RHe−ik·R , 共4.4兲
extra topological invariants in addition to the one 共or four兲
time-reversal symmetry based invariant discussed in Sec. which for lattice periodic Bloch functions now becomes a
II B. These extra invariants, however, rely on the presence of 4 ⫻ 4 matrix. Note that this transformation is slightly differ-
inversion symmetry and lose their meaning in the presence ent than the standard transformation of a tight-binding model
of surfaces, disorder, or other perturbations which violate for which R in Eq. 共4.4兲 is replaced by r = R + dn, where dn is
inversion symmetry. In contrast, the invariants obtained from a basis vector. The difference is a choice of gauge. With this
the product of four ␦i’s do not rely on inversion symmetry choice, H共k兲 has the properties H共k + G兲 = H共k兲 and H共−k兲
for their existence. They depend only on time-reversal sym- = P̂H共k兲P̂−1. Thus, at the time-reversal invariant momenta,
metry, so they retain their value in the presence of inversion 关H共k = ⌫i兲 , P̂兴 = 0.
symmetry-breaking perturbations.
It is convenient to express the 4 ⫻ 4 matrix H共k兲 in terms
of the identity I, five Dirac matrices ⌫a, and their ten com-
IV. TIGHT-BINDING MODELS mutators ⌫ab = 关⌫a , ⌫b兴 / 共2i兲.41 The Dirac matrices satisfy the
In this section, we construct a class of inversion symmet- Clifford algebra, ⌫a⌫b + ⌫b⌫a = 2␦abI. In this section, in order
ric tight-binding models that exhibit topological insulating to avoid confusion of notation, the Dirac matrices ⌫a will
states and apply the method presented in Sec. III to deter- always appear with a superscript, and the time-reversal in-
mine their topological classes. We will consider minimal variant momenta will always be written as k = ⌫i.
models with four bands which result from four degrees of The choice of Dirac matrices is not unique. For example,
freedom per unit cell. We will focus on lattices in which the in Ref. 10, the Dirac matrices were chosen to be even under
unit cell can be chosen to be inversion symmetric. We will time reversal, ⌰ ˆ −1 = ⌫a. In the presence of both inversion
ˆ ⌫ a⌰
see that this latter assumption makes the analysis of the to- and time-reversal symmetries, it is more convenient to
pological phases particularly simple. While this assumption choose the Dirac matrices to be even under P̂⌰ ˆ . Given the
can always be satisfied for continuum models, it rules out ˆ
form of P̂ and ⌰, the five matrices are
certain inversion symmetric lattice models, such as the rock-
salt lattice. It is satisfied, however, for the specific examples ⌫共1,2,3,4,5兲 = 共␴x 丢 I, ␴y 丢 I, ␴z 丢 sx, ␴z 丢 sy, ␴z 丢 sz兲.
we will consider. 共4.5兲
In Sec. IV A, we study the general structure of this class
of models, and then in Secs. IV B and IV C consider the With this choice of Dirac matrices, the commutators are odd
specific examples of the honeycomb lattice of graphene and under P̂⌰ˆ , 共P̂⌰ ˆ 兲−1 = −⌫ab. Note that ⌫1 = P̂. It fol-
ˆ 兲⌫ab共P̂⌰
the diamond lattice. In Sec. IV D, we analyze a model for lows that

045302-9
LIANG FU AND C. L. KANE PHYSICAL REVIEW B 76, 045302 共2007兲

ˆ ⌫ a⌰
⌰ 再
ˆ −1 = P̂⌫a P̂−1 = + ⌫ for a = 1
a

− ⌫a for a ⫽ 1.
冎 共4.6兲

Time-reversal and inversion symmetries imply that

ˆ 兴 = 0. The most general Hamiltonian matrix is then
关H共k兲 , P̂⌰
5
H共k兲 = d0共k兲I + 兺 da共k兲⌫a . 共4.7兲 FIG. 4. 共a兲 Honeycomb lattice of graphene, with a unit cell
a=1
indicated by the dashed lines. 共b兲 Brillouin zone with the values of
Written in this form, the energy eigenvalues naturally come ␦i associated with the time-reversal invariant momenta labeled. ␶1/2
in Kramers degenerate pairs with energy describes the loop enclosing half the zone used in Eq. 共4.13兲.

E共k兲 = d0共k兲 ± 冑兺 d 共k兲 .

a
a
2
共4.8兲
equivalent under threefold rotations, are called the M points.
Using Eqs. 共1.1兲, 共1.2兲, and 共4.10兲, it is then straightforward
At the time-reversal invariant points k = ⌫i, only ⌫1 = P̂ is
to see that ␦i=共00兲 = ␦i=共10兲 = ␦i=共01兲 = −1 while ␦i=共11兲 = + 1. The
even under P̂ and ⌰ˆ . Therefore,
product is negative, so ␯ = 1, and provided that the energy
gap is finite throughout the Brillouin zone, the system is a
H共k = ⌫i兲 = d0共k = ⌫i兲I + d1共k = ⌫i兲P̂. 共4.9兲
topological insulator in the quantum spin-Hall phase. The
The parity eigenvalues ␰n for the states at k = ⌫i are given by finite gap follows from the fact that d0共k兲 = 0 and there are no
the eigenvalues of P̂. It then follows from Eq. 共1.1兲 that, values of k for which all da共k兲 = 0.
provided there is an energy gap throughout the Brillouin The reason the three M points are not all the same is that
zone, the Z2 invariants characterizing the valence band are the center of inversion defined by our unit cell is at the center
determined by of a bond, which does not have the threefold rotational sym-
metry. By choosing a different unit cell, with inversion cen-
␦i = − sgn„d1共k = ⌫i兲…. 共4.10兲
ter at the center of a hexagon, the M points would be equiva-
We will use the above equation to determine the topological lent. Our conclusion about the topological class, however,
class of specific tight-binding models in the following. remains the same.
It is interesting to note that the value of ␦i does not appear
to have anything to do with the spin-orbit interaction. The
B. Graphene role that the spin-orbit interaction plays is simply to ensure
Graphene consists of a honeycomb lattice of carbon atoms that the energy gap is finite everywhere in the Brillouin zone.
with two sublattices. A tight-binding model which incorpo- We will now argue for a parity and time-reversal invariant
rates the symmetry allowed spin-orbit interactions was intro- system that if the spin-orbit interaction is absent, then the
duced in Refs. 9 and 10, negative product of ␦i implies that the energy gap must van-
ish somewhere in the Brillouin zone. This gives insight into
H = t 兺 c†i c j + i␭SO 兺 c†i s · êijc j . 共4.11兲 the topological stability of the Dirac points in graphene in
具ij典 具具ij典典 the absence of spin-orbit interactions.
The first term is a nearest-neighbor hopping term, and the We prove this by contradiction. In the absence of the spin-
second term is spin-orbit interaction which involves spin de- orbit interaction, we can consider spinless fermions. Suppose
pendent second-neighbor hopping. This term is written in a there is a finite gap everywhere, and the valence band is well
way which can easily be generalized to three dimensions. s is defined throughout the Brillouin zone. Then, on the one
the spin, and we have defined the unit vector hand, the Berry curvature F = ⵜ ⫻ A is identically zero due to
inversion and time-reversal symmetries. On the other hand,
d1ij ⫻ d2ij we will show that the Berry’s phase for the path ␶1/2 shown
êij = , 共4.12兲 in Fig. 4 which encloses half the Brillioun zone satisfies
兩d1ij ⫻ d2ij兩
where d1ij and d2ij are bond vectors along the two bonds the
electron traverses when going from site j to i. Thus, êij · s TABLE I. Parameters for tight-binding model of graphene with
= ± s z. xl = k · al in a notation slightly different from Ref. 10.
Choosing the unit cell shown in Fig. 4, the Hamiltonian
matrix H共k兲 can be determined using Eq. 共4.4兲 and expressed d0 0
in terms of Dirac matrices as in Eq. 共4.7兲. The coefficients d1 t关1 + cos x1 + cos x2兴
da共k兲 are displayed in Table I. The time-reversal invariant d2 t关sin x1 + sin x2兴
momenta, in the notation of Eq. 共2.3兲, occur at k = ⌫i=共n1n2兲 d3 0
for nl = 0 , 1. The Hamiltonian at these points can be deduced d4 0
by noting that at k = ⌫i=共n1n2兲, xl ⬅ k · al = nl␲. ⌫i=共00兲 is com- d5 2␭SO关sin x1 − sin x2 − sin共x1 − x2兲兴
monly referred to as the ⌫ point. The other three, which are

045302-10
TOPOLOGICAL INSULATORS WITH INVERSION SYMMETRY PHYSICAL REVIEW B 76, 045302 共2007兲

TABLE II. Parameters for diamond lattice tight-binding model, have 共n1n2n3兲 = 共100兲, 共010兲, 共001兲, and 共111兲. The 111 dis-
with xk = k · ak. tortion makes the first three L points distinct from the fourth,
which will be referred to as T.
d0 0 From Table II, we can deduce the sign of d1共k兲 and,
d1 t + ␦t1 + t关cos x1 + cos x2 + cos x3兴 hence, ␦i at these points. We find ␦i = −1 at ⌫ and the three L
d2 t关sin x1 + sin x2 + sin x3兴 points, while ␦i = + 1 at T. At the X points, ␦i = −sgn共␦t1兲.
␭SO关sin x2 − sin x3 − sin共x2 − x1兲 + sin共x3 − x1兲兴 Combining these, we then find that

再 冎
d3
d4 ␭SO关sin x3 − sin x1 − sin共x3 − x2兲 + sin共x1 − x2兲兴
共1;111兲 for ␦t1 ⬎ 0
d5 ␭SO关sin x1 − sin x2 − sin共x1 − x3兲 + sin共x2 − x3兲兴 共 ␯ 0 ; ␯ 1␯ 2␯ 3兲 = 共4.15兲
共0;111兲 for ␦t1 ⬍ 0.

冋冖 册
When the 111 distorted bond is stronger than the other three
bonds, so that the system is dimerized, the system is a strong
exp i A共k兲 · dk = ␦1␦2␦3␦4 . 共4.13兲 topological insulator. When the 111 bond is weaker than the
␶1/2
other three, so that the system is layered, it is a weak topo-
Thus, if ␦1␦2␦3␦4 = −1, it would violate Stoke’s theorem and logical insulator with G␯ = 共2␲ / a兲共1 , 1 , 1兲, which can be
leads to a contradiction. The ␲ Berry’s phase thus requires viewed as two-dimensional quantum spin-Hall states stacked
that there either be a Dirac point in each half of the Brillouin in the 111 direction.
zone or a Fermi arc enclosing a Dirac point. In Ref. 15, we computed the two-dimensional band struc-
To obtain Eq. 共4.13兲 for spinless electrons, we consider ture for the diamond lattice model in a slab geometry. The
the unitary matrix results displayed the expected surface states, which behave
according to the general principles discussed in Sec. II C.
mij共k兲 = 具ui,−k兩P兩u j,k典, 共4.14兲
which is related to the Berry’s potential via D. Bernevig-Hughes-Zhang model
ⵜk log det关m共k兲兴 = −i关A共k兲 + A共−k兲兴. Equation 共4.13兲 is then After this paper was originally submitted, an interesting
obtained by breaking the line integral into segments connect- proposal appeared for the 2D quantum spin-Hall effect in
ing the time-reversal invariant momenta and using the fact quantum well structures in which a layer of HgTe is sand-
that det关m共k = ⌫i兲兴 = ␦i. wiched between crystals of CdTe.40 Bernevig-Hughes-Zhang
共BHZ兲 showed that for an appropriate range of well thick-
C. Diamond lattice ness, the HgTe layer exhibits an inverted band structure,
where the s and p levels at the conduction- and valence-band
We now consider the tight-binding model on a diamond
edges are interchanged. In this inverted regime, the structure
lattice introduced in Ref. 15. This model exhibits both weak
exhibits a 2D quantum spin-Hall effect. BHZ introduced a
and strong topological insulator phases.
simple four band tight-binding model which captures this
The diamond structure consists of two interpenetrating
effect. Though real HgTe does not have inversion symmetry,
face-centered-cubic lattices displaced from each other by a
their toy model does. In this section, we analyze this model
basis vector d = a共1 , 1 , 1兲 / 4. The primitive translation vectors
and directly evaluate the Z2 topological invariant using Eq.
a1, a2, and a3 are a共0 , 1 , 1兲 / 2, a共1 , 0 , 1兲 / 2, and a共1 , 1 , 0兲 / 2, 共1.1兲.
respectively. Our model has the same form as Eq. 共4.11兲 and BHZ considered a four band model on a square lattice in
includes a nearest-neighbor hopping term as well as a which each site has two s1/2 states 兩s , ↑ 典 and 兩s , ↓ 典 and two of
second-neighbor spin-orbit interaction. the crystal field split p3/2 states 共with m j = ± 3 / 2兲, 兩px
It turns out that with this spin-orbit interaction term, the + ipy , ↑ 典 and 兩px − ipy , ↓ 典. The Hamiltonian is
valence bands and conduction bands meet at 3D Dirac points
at the three inequivalent X points on the 100, 010, and 001
faces of the Brillouin zone. In order to lift the degeneracy
H= 兺 ␧␣ci†␣␴ci␣␴ − ia兺
i,␴,␣ ␴␣␤
†
ta␴,␣␤ci+a ␣␴c i␤␴ , 共4.16兲

and obtain a gapped phase, we introduced a distortion, which

changes the nearest-neighbor hopping amplitudes. For sim- where a labels the four nearest neighbors, ␴ = ± 1 describes
plicity, we will focus here on a distortion in the 111 direc- the spin, and ␣ , ␤ = s , p is the orbital index. The hopping term
tion, which changes the nearest-neighbor bond in the 111 involves the matrix
direction but leaves the other three bonds alone. The result-
ing model can be expressed in the form of Eq. 共4.7兲, and the
resulting da共k兲 are listed in Table II. For ␭SO , ␦t ⫽ 0, the gap
t a␴ = 冉 tss
tspe−i␴␪a
冊
tspei␴␪a
− t pp
, 共4.17兲

is finite throughout the Brillouin zone. where ␪a gives the angle of nearest-neighbor bond a with the
As in the previous section, the time-reversal invariant mo- x axis.
menta occur at k = ⌫i=共n1n2n3兲 as in Eq. 共2.3兲. At these points, As in Sec. IV A, it is convenient to express this Hamil-
xl ⬅ k · al = nl␲. At the ⌫ point, k = 0, 共n1n2n3兲 = 共000兲. The tonian in the form of Eq. 共4.7兲 using Dirac matrices which
three inequivalent X points 关at k = 共2␲ / a兲共1 , 0 , 0兲 and related are even under P̂⌰ ˆ . The form of the parity operator, how-
points兴 have 共n1n2n3兲 = 共011兲, 共101兲, and 共110兲. The four in- ever, is slightly different in this model, and Eq. 共4.2兲 is re-
equivalent L points 关at k = 共␲ / a兲共1 , 1 , 1兲 and related points兴 placed by

045302-11
LIANG FU AND C. L. KANE PHYSICAL REVIEW B 76, 045302 共2007兲

TABLE III. Parameters for the BHZ model with xk = k · ak. TABLE IV. Symmetry labels for the five valence bands of bis-
muth and antimony at eight time reversal invariant momenta ac-
d0 共␧s + ␧ p兲 / 2 − 共tss − t pp兲共cos x1 + cos x2兲 cording to Ref. 45. The parity eigenvalues can be read from ⫹/⫺
d1 共␧s − ␧ p兲 / 2 − 共tss + t pp兲共cos x1 + cos x2兲 and s / a. Using Eqs 共1.1兲 and 共1.2兲, they determine the topological
class. The indices 共111兲 define a mod 2 vector 关Eq. 共2.13兲 in the
d2 2tsp sin x1
direction of the T point.
d3 0
d4 0 Bismuth
d5 2tsp sin x2 1⌫ ⌫+6 ⌫−6 ⌫+6 ⌫+6 ⌫+45 ⫺
3L Ls La Ls La La ⫺
3X Xa Xs Xs Xa Xa ⫺
P̂ = ␴z 丢 I, 共4.18兲 1T T−6 T+6 T−6 T+6 T−45 ⫺
where ␴z = + 1共−1兲 describes s 共p兲 states. The Dirac matrices Z2 class 共0;000兲
are then the same as Eq. 共4.5兲, except that ␴x and ␴z are
Antimony
interchanged. The coefficients of these new Dirac matrices
1⌫ ⌫+6 ⌫−6 ⌫+6 ⌫+6 ⌫+45 ⫺
for this model are displayed in Table III.
The analysis between Eqs. 共4.6兲 and 共4.10兲 proceeds ex- 3L Ls La Ls La Ls ⫹
actly the same as before, and ␦i = −sgn关d1共k = ⌫i兲兴. We con- 3X Xa Xs Xs Xa Xa ⫺
clude that for k = 共␲ / a兲共n1 , n2兲, 1T T−6 T+6 T−6 T+6 T−45 ⫺

再
␦i=共n1n2兲 = − sgn
␧s − ␧ p
2
− 共tss + t pp兲关共− 1兲n1 + 共− 1兲n2兴 . 冎 Z2 class 共1;111兲

The conduction band crosses the Fermi energy near the three
共4.19兲 equivalent L points, which reside at the other three-body cen-
For ␧s − ␧ p ⬎ 4共tss + t pp兲, all of the ␦i=共n1n2兲 are negative, so that ter zone faces, giving rise to pockets of electrons. At the L
the product ␯ = + 1. The system is a simple insulator. In this points, the bottom of the conduction band, which has Ls sym-
regime, the bands have a conventional ordering throughout metry, is only slightly higher in energy than the next lower
the Brillouin zone, with the s states in the conduction band band, which has La symmetry. In antimony,45 the electrons
and the p states in the valence band. For ␧s − ␧ p ⱗ 4共tss + t pp兲, are again near the L point. However, unlike bismuth, the
bottom of the conduction band has La symmetry. The holes
the bands near k = 0 become inverted, and ␦i=共00兲 becomes
are not at the T point but, rather, at the lower-symmetry H
positive, signaling a transition into the quantum spin-Hall point.
phase in which ␯ = −1. Despite the fact that bismuth and antimony have negative
indirect gaps, the finite direct gap throughout the Brillouin
V. TOPOLOGICAL PHASES IN SPECIFIC MATERIALS zone allows for the topological characterization of the va-
lence energy bands. Since both bismuth and antimony have
In this section, we apply our method for evaluating the
inversion symmetry, we can apply Eqs. 共1.1兲 and 共1.2兲 by
topological invariants to identify specific three-dimensional
reading off the parity eigenvalues ␰n共⌫i兲 from published band
materials that should exhibit a strong topological insulating
phase. structures.44,45 In Table IV, we display the symmetry labels
for the five occupied valence bands at the eight time-reversal
invariant momenta 共⌫i = ⌫, T, and the three equivalent L and
A. Bismuth-antimony alloy
X points兲. The parity eigenvalue ␰n共⌫i兲 can be read from the
Bi and Sb are group-V semimetals in which there is a superscripts ⫾ or the subscripts s / a = + / −. 共For an explana-
finite direct energy gap throughout the Brillouin zone but a tion of this notation, see Ref. 46兲. The right column displays
negative indirect gap due to band overlap. They have very the product ␦i from Eq. 共1.1兲. Based on these data, we con-
close lattice parameters and form the solid alloy clude that the valence band of bismuth is equivalent to that
Bi1−xSbx.42,43 For 0.07⬍ x ⬍ 0.22, the indirect gap becomes of a conventional insulator, while the valence band of anti-
positive, leading to semiconducting behavior, with a maxi- mony is that of a strong topological insulator. The difference
mum energy gap of order 30 meV for x = 0.18. In this sec- between the two is due to the inversion between the Ls and
tion, we will argue, based on the known band structure of La bands.
these materials, that this alloy is a strong topological insula- The evolution of the band structure of the alloy Bi1−xSbx
tor, which will have topological metal surface states. as a function of x has been well studied42,43,47 and is summa-
Bulk bismuth and antimony have the rhombohedral A7 rized in Fig. 5. As the Sb concentration is increased, two
structure, which consists of two interpenetrating, face- things happen. First, the gap between the Ls and La bands
centered-cubic lattices which are displaced in the 111 direc- decreases. At x = 0.04, the bands cross and the gap reopens
tion and slightly distorted in the 111 direction. In with the inverted ordering. Second, the top of the valence
bismuth,44,45 the valence band crosses the Fermi energy in band at T comes down in energy and crosses the bottom of
the vicinity of the T point, which is located on the 111 face the conduction band at x = 0.07. At this point, the indirect gap
of the Brillouin zone, giving rise to a small pocket of holes. becomes positive, and the alloy is a semiconductor. At x

045302-12
TOPOLOGICAL INSULATORS WITH INVERSION SYMMETRY PHYSICAL REVIEW B 76, 045302 共2007兲

FIG. 6. 共a兲 Band structure of ␣-Sn near the ⌫ point, which

describes zero gap semiconductor due to the inverted ⌫+8 and ⌫−7
bands. 共b兲 In the presence of uniaxial strain, the degeneracy at ⌫ is
FIG. 5. Schematic representation of band energy evolution of
lifted, opening a gap in the spectrum. The parity eigenvalues remain
Bi1−xSbx as a function of x. Adapted from Ref. 43.
unchanged.

= 0.09 the T valence band clears the Ls valence band, and the degenerate set of states with ⌫+8 symmetry, which can be
alloy is a direct-gap semiconductor at the L points. As x is derived from p states with total angular momentum j = 3 / 2.
increased further, the gap increases until its maximum value The fourfold degeneracy at the ⌫+8 point is a consequence of
of order 30 meV at x = 0.18. At that point, the valence band at the cubic symmetry of the diamond lattice. Applying uniaxial
H crosses the Ls valence band. For x ⬎ 0.22, the H band strain lifts this degeneracy into a pair of Kramers doublets
crosses the La conduction band, and the alloy is again a and introduces an energy gap into the spectrum.53 For pres-
semimetal. sures of order 3 ⫻ 109 dyn/ cm2, the induced energy gap is of
Since the inversion transition between the Ls and La bands order 40 meV. We now argue that this insulating phase is, in
occurs in the semimetal phase adjacent to pure bismuth, it is fact, a strong topological insulator.
clear that the semiconducting Bi1−xSbx alloy inherits its to- Table V shows the symmetry labels for unstrained ␣-Sn
pological class from pure antimony and is thus a strong to- associated with the four occupied valence bands at the eight
pological insulator. Of course, this conclusion is predicated time-reversal invariant momenta.54 Uniaxial strain lowers the
on a “virtual-crystal approximation⬙ in which the disorder symmetry, so the cubic symmetry labels no longer apply.
due to the random mixture is ignored, so that inversion sym- However, since the strain does not violate inversion symme-
metry is preserved in the alloy. However, since this inherent try, the parity eigenvalues are unchanged. The only effect is
disorder does not destroy the bulk energy gap, it is unlikely to split the degeneracy of the ⌫+8 level into two sets of even-
to change the topological class, which does not require in- parity Kramers doublets 关see Fig. 6共b兲兴. In Table V, ⌫+* 8 re-
version 共or translation兲 symmetry. We thus conclude that in- fers to the occupied doublet. Based on the parity eigenvalues,
trinsic Bi1−xSbx, despite its bulk energy gap, will have con- we conclude that strained gray tin is a strong topological
ducting surface states, which form a topological metal. insulator.
Semiconducting Bi1−xSbx alloys have been studied experi- HgTe is a II-VI material with the zinc-blende
mentally because of their thermoelectric properties, which structure.52,55 It is a zero gap semiconductor with an elec-
make them desirable for applications as thermo- tronic structure closely related to gray tin. The Fermi energy
couples.43,48–50 Transport studies have been carried out both is in the middle of the fourfold degenerate ⌫8 states, whose
on bulk samples43 and epitaxial thin films.49 For T ⬎ 50 K, degeneracy follows from the cubic symmetry of the zinc-
semiconducting behavior is observed, while at lower tem- blende lattice. As in gray tin, uniaxial strain lifts this degen-
peratures, the resistivity saturates at a value in the range eracy and opens a gap at the Fermi energy.
5 – 50 ␮⍀m. This observed residual resistivity is probably Though HgTe lacks inversion symmetry, we now argue
too small to be explained by surface states. It has been at- based on adiabatic continuity that the gap induced by
tributed to residual charged impurities,43 which act as shal- uniaxial strain leads to a strong topological insulator. The
low donors, making the alloy slightly n type. In order to electronic structure of II-VI materials can be understood by
separate the surface properties from the bulk transport, it will adding an inversion symmetry-breaking perturbation to a in-
be necessary either to improve the purity of the samples or version symmetric group-IV crystal.52,56 Provided that this
perhaps use gating in a heterostructure to push the Fermi perturbation does not lead to any level crossings at the Fermi
energy into the gap. energy, we can conclude that the II-IV material is in the same

TABLE V. Symmetry labels for the four valence bands of gray

B. Gray tin and mercury telluride tin at eight time-reversal invariant momenta, according to Ref. 54.
Tin is a group-IV element, which in its ␣ 共or gray兲 phase
has the inversion symmetric diamond structure. Unlike car- 1⌫ ⌫+6 ⌫+7 ⌫−7 ⌫+*
8 ⫺
bon, silicon, and germanium, though, it is a zero gap semi- 3X 2X5 2X5v ⫹
conductor, in which the ordering of the states at the 4L L−6 L+6 L−6v L−45 ⫺
conduction- and valence-band edge is inverted 关see Fig. Z2 Class 共1;000兲
6共a兲兴.51,52 The Fermi energy lies in the middle of a fourfold

045302-13
LIANG FU AND C. L. KANE PHYSICAL REVIEW B 76, 045302 共2007兲

TABLE VI. Symmetry labels for the five valence bands of tin
telluride and lead telluride at eight time-reversal invariant momenta,
according to Ref. 59.

Tin Telluride
1⌫ ⌫+6 ⌫+6 ⌫−6 2⌫+8 ⫺
3X X+6 X+6 X−6 X−6 X−7 ⫺
4L L−6 L+6 L+6 L+45 L−6 ⫹
Z2 class 共0;000兲

Lead telluride
1⌫ ⌫+6 ⌫+6 ⌫−6 2⌫+8 ⫺ FIG. 7. Schematic uniaxial strain-composition phase diagram
3X X+6 X+6 X−6 X−6 X−7 ⫺ for Pb1−xSnxTe. Away from the inversion transition at x ⬃ 0.4, the
material is a conventional insulator 共I兲. Near the transition, it is a
4L L−6 L+6 L+6 L+45 L+6 ⫺
strong topological insulator 共STI兲.
Z2 class 共0;000兲

which is inevitably present in the alloy, blurs the transition.

topological class as the group-IV crystal. The band structures However, uniaxial strain oriented along the 111 direction will
of gray tin and HgTe are very similar, and the cubic symme- distinguish one of the L points 共call it T now兲 from the other
try labels of the energy bands show how the bands evolve three L points. It follows that the inversion transitions at the
between the two. This allows us to conclude that strained L and T points will occur at different values of x. Thus, there
␣-Sn and HgTe will be in the same topological class, which will be an intermediate phase in which L is inverted, but T is
is that of the strong topological insulator. not 共or vice versa兲. From Eqs. 共1.1兲 and 共1.2兲, this interme-
In Ref. 25, Murakami et al. introduced a four band tight- diate phase will be a strong topological insulator with surface
binding model based on p3/2 atomic levels on a fcc lattice to states forming a topological metal. Note that this direction
describe strained ␣-Sn and HgTe. As argued in Ref. 10, this depends on the orientation of the uniaxial strain. For in-
model predicts that these materials are simple insulators in stance, strain along the 100 direction will distinguish two L
the 0; 共000兲 class. This can be understood by noting that points from the other two and will not lead to an intermediate
since the model includes only p3/2 atomic levels, the parity topological phase. In Fig. 7, we show a schematic phase
eigenvalues in Eq. 共1.1兲 are all ␰i = −1. This contradicts the diagram as a function of x and 111 strain.
known band structure of these materials, as displayed in The band inversion between SnTe and PbTe has been dis-
Table V. This model correctly describes the electronic states cussed by a number of authors previously. Volkov and
near the ⌫ point, but it gets the global topology of the bands Pankratov60 argued that PbTe and SnTe can be described by
wrong. To capture the global topology, a tight-binding model a low-energy field theory consisting of three-dimensional
of these materials must include both s and p levels. The more Dirac fermions with masses of opposite signs. They con-
recent theory40 of the 2D quantum spin-Hall effect in cluded that a planar interface between PbTe and SnTe will
HgTe/ CdTe quantum wells discussed in Sec. IV D correctly exhibit interface states described by a two-dimensional mass-
incorporates s and p levels. less Dirac equation. The appearance of such domain-wall
fermions is similar to the appearance of midgap states in a
one-dimensional Peierls insulator at a soliton.34 A related
C. Lead-tin telluride
proposal was made by Fradkin and co-workers,36,37 who con-
PbTe and SnTe are narrow gap IV-VI semiconductors sidered a domain wall in PbTe in which on one side the Pb
with the rocksalt structure.57 The band gap in these materials and Te atoms are interchanged. This was also modeled as 3D
is direct and occurs at the four equivalent L points in the fcc Dirac fermions with a mass which changes sign at the inter-
Brillouin zone. PbTe has an inverted band structure relative face.
to SnTe, in which the relative ordering of the L+6 and L−6 The domain-wall fermions which appear in these theories
bands at the conduction- and valence-band edges are are similar to the states we predict at the surface of a strong
switched. Nonetheless, both of these materials are conven- topological insulator. Indeed, if one views the vacuum as a
tional insulators. In Table VI, we display the symmetry labels band insulator with a large gap, then the surface can be
at the eight time-reversal invariant points 共⌫, three equivalent viewed as an interface between a band insulator and a topo-
X points, and four equivalent L points兲.59 Since the inversion logical insulator, which can be described as an inversion
occurs at an even number of points in the Brillouin zone, transition, where there is a change in the sign of the mass of
both materials belong to the conventional insulator topologi- a 3D Dirac fermion. However, there is an important differ-
cal class. ence between the proposals discussed above and the surface
PbTe and SnTe form an alloy Pb1−xSnxTe. At x ⬃ 0.4, states of the topological insulator: the strong topological-
there is an inversion transition where the band gap at the four insulator–band-insulator interface involves a sign change in
L points vanishes, giving rise to three-dimensional Dirac an odd number of Dirac points, while the interface models
points.57,58 The phases on either side of this transition are above involve four Dirac points. Having an odd number is
only distinct if inversion symmetry is present. Thus, disorder, crucial for the topological stability of the surface states.

045302-14
TOPOLOGICAL INSULATORS WITH INVERSION SYMMETRY PHYSICAL REVIEW B 76, 045302 共2007兲

D. Other materials must be low: T Ⰶ Eg / log共W / a兲, where a is the lattice con-
The materials we have proposed above should not be con- stant. Moreover, since the materials we have suggested have
sidered to be an exhaustive list. In general, it is necessary to rather small energy gaps, samples with high purity will be
consider insulators composed of heavy elements. Another required to reach the intrinsic limit. As discussed in Sec.
candidate for a topological insulating phase is Bi2Te3, which, IV A, the low-temperature behavior of Bi1−xSbx is dominated
like Bi1−xSbx, is known for its thermoelectric properties.61 by a low concentration of charged impurities, which form an
This material is also a narrow gap semiconductor, with an impurity band.43 This is a ubiquitous problem for narrow gap
energy gap of order 0.13 eV. Though the crystal structure of semiconductors due to their low effective mass and high di-
this material is inversion symmetric, we have been unable to electric constant. Clearly, it would be desirable to have a
locate band theory calculations which display the parity ei- transport geometry which probes the surface states while
genvalues. controlling the position of the bulk Fermi energy. Perhaps,
Another possible candidate is the zinc-blende semicon- this may be possible in a clever heterostructure geometry,
ductor ␤ − HgS. The electronic structure of this material has where the bulk Fermi energy can be adjusted with a gate.
been a subject of some controversy. According to Delin,62 it An alternative probe of the surface states would be to map
is a semiconductor which has an unusual band ordering, with the surface Fermi arc using angle-resolved photoemission.
the ⌫6 and ⌫8 levels in the valence band and the ⌫7 level in Such measurements could establish that the surface Fermi
the conduction band. If this is the case, we expect the mate- arc encloses an odd number of time-reversal invariant mo-
rial to be a strong topological insulator. However, this con- menta in the strong topological insulator phase. Detailed
clusion has been challenged by Moon and Wei,63 who find a angle-resolved photoemission spectroscopy studies have
more conventional band ordering with the ⌫6 level in the been carried out on the surfaces of bismuth67–69 and
conduction band and the ⌫7 and ⌫8 levels in the valence antimony.70 However, the presence of the bulk Fermi surface
band, leading to a conventional insulator. complicates the analysis of these materials. It would be in-
teresting to see how the results are modified in the semicon-
ducting Bi1−xSbx alloy.
VI. EXPERIMENTAL IMPLICATIONS Finally, since the surface states are spin filtered, electrical
currents flowing on the surface will be associated with spin
We now briefly consider possible experimental probes of accumulation, leading to a spin-Hall effect. In GaAs, spin
topological insulators. We will focus here on the three- accumulation on a surface has been measured.71,72 The nar-
dimensional strong topological insulator phase, for which we row energy gaps in our proposed materials make detection of
suggested several materials in the previous section. the spin accumulation more difficult. Perhaps, a heterostruc-
The most direct probe of the topological insulators is ture geometry could make this possible.
transport. Since there is a bulk excitation gap, transport in
intrinsic samples at very low temperature will be dominated
VII. CONCLUSION
by the surfaces, which can be probed by the geometry de-
pendence of the conductance. For example, for a wire geom- In this paper, we discussed topological insulator phases in
etry, the conductance will be proportional to the circumfer- two and three dimensions. We discussed in detail how the Z2
ence of the wire, rather than the area. topological invariants characterizing these phases influence
In addition, since the topological metal phase is in the the surface-state spectrum. In particular, the quantum spin-
symplectic universality class the conductivity is expected to Hall phase in two dimensions and the strong topological in-
increase logarithmically at low temperature: ␴共T兲 sulator in three dimensions have robust conducting surface
⬀ log关Lin共T兲 / ᐉ兴,64 where Lin is the inelastic scattering length states, which we have characterized as a topological metal.
and ᐉ is the mean free path. We showed that the Z2 invariants can be determined easily in
An interesting prediction for the surface states is that due parity invariant crystals from the knowledge of the parity
the ␲ Berry’s phase associated with the surface Fermi arc, eigenvalues for states at the time-reversal invariant points in
the surface quantum Hall effect in a perpendicular magnetic the Brillouin zone. Using this method, we deduced that the
field should be quantized in half odd integers, ␴xy = 共n semiconducting alloy Bi1−xSbx is a strong topological insula-
+ 1 / 2兲e2 / h. As discussed in Sec. II C 1, this is difficult to tor, as are ␣-Sn and HgTe in the presence of uniaxial strain.
measure directly without separately measuring the currents There remain a number of further issues which need to be
flowing on the top and bottom surfaces of the sample. How- understood better. High among them are the effects of disor-
ever, if the parallel combination of the two surfaces could be der and interactions. These are important both for the topo-
measured, the resulting Hall effect would be quantized in logical metal surface states as well as for the bulk topological
odd multiples of e2 / h. This is similar to the quantum Hall phases. Numerical work by Onoda et al.73 has suggested that
effect in graphene,65,66 which is quantized in odd multiples the transition between the conventional insulator and the
of 2e2 / h. The difference is due to the fact that graphene has quantum spin-Hall phase in two dimensions belongs to a new
four Dirac points, including spin. universality class. It will be of interest to understand this
A practical difficulty with transport measurements is that transition better, along with the related transition between the
it is necessary to distinguish surface currents from bulk cur- topological insulator and the Anderson insulator, which pre-
rents. Since bulk currents scale with the sample width W, sumably occurs when disorder is increased beyond a critical
even though there is a bulk energy gap Eg, the temperature value.

045302-15
LIANG FU AND C. L. KANE PHYSICAL REVIEW B 76, 045302 共2007兲

Finally, it would be desirable to develop a field theory for ACKNOWLEDGMENTS

the topological insulating phases analogous to the Chern- It is a pleasure to thank Eduardo Fradkin and Gene Mele
Simons theory of the quantum Hall effect. Perhaps, this may for helpful discussions. This work was supported by the NSF
lead to analogs of the fractional quantum Hall effect for the Grant No. DMR-0605066 and by ACS PRF Grant No.
topological insulators. 44776-AC10.

1
W. Kohn, Phys. Rev. 133, A171 共1964兲. 240 共1987兲.
2
X. G. Wen, Phys. Rev. B 44, 2664 共1991兲. 38
G. W. Semenoff, Phys. Rev. Lett. 53, 2449 共1984兲.
3 E. Demler, C. Nayak, H. Y. Kee, Y. B. Kim, and T. Senthil, Phys. 39 Q. Niu, D. J. Thouless, and Y. S. Wu, Phys. Rev. B 31, 3372

Rev. B 65, 155103 共2002兲, and references therein. 共1985兲.

4 F. D. M. Haldane, Phys. Rev. Lett. 61, 2015 共1988兲. 40 A. Bernevig, T. Hughes, and S. C. Zhang, Science 314, 1757
5 D. J. Thouless, M. Kohmoto, M. P. Nightingale, and M. den Nijs, 共2006兲.
Phys. Rev. Lett. 49, 405 共1982兲. 41
S. Murakami, N. Nagaosa, and S. C. Zhang, Science 301, 1348
6
M. Kohmoto, B. I. Halperin, and Y.-S. Wu, Phys. Rev. B 45, 共2003兲; Phys. Rev. B 69, 235206 共2004兲.
13488 共1992兲. 42 L. S. Lerner, K. F. Cuff, and L. R. Williams, Rev. Mod. Phys. 40,
7 B. I. Halperin, Phys. Rev. B 25, 2185 共1982兲. 770 共1968兲.
8
Y. Hatsugai, Phys. Rev. Lett. 71, 3697 共1993兲. 43
B. Lenoir, M. Cassart, J.-P. Michenaud, H. Scherrer, and S.
9 C. L. Kane and E. J. Mele, Phys. Rev. Lett. 95, 226801 共2005兲. Scherrer, J. Phys. Chem. Solids 57, 89 共1996兲; B. Lenoir, A.
10
C. L. Kane and E. J. Mele, Phys. Rev. Lett. 95, 146802 共2005兲. Dauscher, X. Devaux, R. Martin-Lopez, Yu. I. Ravich, H. Scher-
11 B. A. Bernevig and S. C. Zhang, Phys. Rev. Lett. 96, 106802 rer, and S. Scherrer, Proceedings of 15th International Confer-
共2006兲. ence on Thermoelectrics, 1996 共unpublished兲, pp. 1–13.
12 R. Roy, arXiv:cond-mat/0604211 共unpublished兲. 44 S. Golin, Phys. Rev. 166, 643 共1968兲.
13 J. E. Moore and L. Balents, Phys. Rev. B 75, 121306共R兲 共2007兲. 45 Y. Liu and R. E. Allen, Phys. Rev. B 52, 1566 共1995兲.
14 R. Roy, arXiv:cond-mat/0607531共unpublished兲. 46 L. M. Falicov and S. Golin, Phys. Rev. 137, A871 共1965兲.
15 L. Fu, C. L. Kane, and E. J. Mele, Phys. Rev. Lett. 98, 106803 47 S. Golin, Phys. Rev. 176, 830 共1968兲.

共2007兲. 48 D. T. Morelli, D. L. Partin, and J. Heremans, Semicond. Sci.

16
H. Suzuura and T. Ando, Phys. Rev. Lett. 89, 266603 共2002兲. Technol. 5, S257 共1990兲.
17
F. D. M. Haldane, Phys. Rev. Lett. 93, 206602 共2004兲. 49
S. Cho, A. DiVenere, G. K. Wong, J. B. Ketterson, and J. R.
18 S. Hikami, A. I. Larkin, and Y. Nagaoka, Prog. Theor. Phys. 63,
Meyer, Phys. Rev. B 59, 10691 共1999兲.
707 共1980兲. 50
Y. Lin et al., Appl. Phys. Lett. 79, 677 共2001兲.
19 H. Nielssen and N. Ninomiya, Phys. Lett. 130B, 389 共1983兲. 51 S. Groves and W. Paul, Phys. Rev. Lett. 11, 194 共1963兲.
20 D. N. Sheng, Z. Y. Weng, L. Sheng, and F. D. M. Haldane, Phys. 52 I. A. Tsidikovski, Electron Spectrum of Gapless Semiconductors,

Rev. Lett. 97, 036808 共2006兲. Springer Series in Solid State Sciences Vol. 116 共Springer-
21
T. Fukui and Y. Hatsugai, arXiv:cond-mat/0607484 共unpub- Verlag, Berlin, 1997兲.
53 B. J. Roman and A. W. Ewald, Phys. Rev. B 5, 3914 共1972兲.
lished兲.
22 L. Fu and C. L. Kane, Phys. Rev. B 74, 195312 共2006兲. 54 F. H. Pollak, M. Cardona, C. W. Higginbotham, F. Herman, and J.
23 M. Kohmoto, Ann. Phys. 共N.Y.兲 160, 343 共1985兲.
P. Van Dyke, Phys. Rev. B 2, 352 共1970兲.
24 S. Murakami, Phys. Rev. Lett. 97, 236805 共2006兲. 55 R. Dornhaus and G. Nimtz, Narrow Gap Semiconductors,
25
S. Murakami, N. Nagaosa, and S. C. Zhang, Phys. Rev. Lett. 93, Springer Tracts in Modern Physics Vol. 98 共Springer-Verlag,
156804 共2004兲. Berlin, 1983兲.
26
R. B. Laughlin, Phys. Rev. B 23, R5632 共1981兲. 56
F. Herman, Electronics 1, 103 共1955兲.
27 E. I. Blount, Solid State Phys. 13, 305 共1962兲. 57 G. Nimtz and B. Schicht, Narrow Gap Semiconductors, Springer
28
J. Zak, Phys. Rev. Lett. 62, 2747 共1989兲. Tracts in Modern Physics Vol. 98 共Springer-Verlag, Berlin,
29 R. D. King-Smith and D. Vanderbilt, Phys. Rev. B 47, 1651
1983兲.
共1993兲. 58
J. Dimmock, I. Melngailis, and A. J. Strauss, Phys. Rev. Lett. 16,
30 R. Resta, Rev. Mod. Phys. 66, 899 共1994兲.
1193 共1966兲.
31
C. Wu, B. A. Bernevig, and S. C. Zhang, Phys. Rev. Lett. 96, 59
Y. W. Tung and M. L. Cohen, Phys. Rev. 180, 823 共1969兲.
106401 共2006兲. 60
B. A. Volkov and O. A. Pankratov, Pis’ma Zh. Eksp. Teor. Fiz.
32 C. Xu and J. E. Moore, Phys. Rev. B 73, 045322 共2006兲.
42, 145 共1985兲 关JETP Lett. 42, 178 共1985兲兴.
33
D. B. Kaplan, Phys. Lett. B 288, 342 共1992兲. 61
S. K. Mishra, S. Satpathy, and O. Jepsen, J. Phys.: Condens.
34 W. P. Su, J. R. Schrieffer, and A. J. Heeger, Phys. Rev. Lett. 42,
Matter 9, 461 共1997兲.
1698 共1979兲. 62
A. Delin, Phys. Rev. B 65, 153205 共2002兲.
35 R. Jackiw, Phys. Rev. D 29, 2375 共1984兲. 63 C. Y. Moon and S. H. Wei, Phys. Rev. B 74, 045205 共2006兲.
36
E. Fradkin, E. Dagotto, and D. Boyanovsky, Phys. Rev. Lett. 57, 64
T. Senthil and M. P. A. Fisher, Phys. Rev. B 61, 9690 共2000兲.
2967 共1986兲. 65 K. S. Noveselov, A. K. Geim, S. V. Morozov, D. Jiang, M. I.
37
D. Boyanovsky, E. Dagotto, and E. Fradkin, Nucl. Phys. B 285, Katsnelson, I. V. Grigorieva, S. V. Dubonos, and A. A. Firsov,

045302-16
TOPOLOGICAL INSULATORS WITH INVERSION SYMMETRY PHYSICAL REVIEW B 76, 045302 共2007兲

Nature 共London兲 438, 197 共2005兲. 70 K. Sugawara, T. Sato, S. Souma, T. Takahashi, M. Arai, and T.
66
Y. Zhang, Y.-W. Tan, H. L. Stormer, and P. Kim, Nature 共London兲 Sasaki, Phys. Rev. Lett. 96, 046411 共2006兲.
438, 201 共2005兲. 71
Y. K. Kato, R. C. Myers, A. C. Gossard, and D. D. Awschalom,
67
C. R. Ast and H. Höchst, Phys. Rev. Lett. 87, 177602 共2001兲. Science 306, 1910 共2004兲.
68 J. E. Gayone, S. V. Hoffmann, Z. Li, and P. Hofmann, Phys. Rev. 72 J. Wunderlich, B. Kaestner, J. Sinova, and T. Jungwirth, Phys.

Lett. 91, 127601 共2003兲. Rev. Lett. 94, 047204 共2005兲.

69 P. Hofmann, J. E. Gayone, G. Bihlmayer, Y. M. Koroteev, and E. 73 M. Onoda, Y. Avishai, and N. Nagaosa, Phys. Rev. Lett. 98,

V. Chulkov, Phys. Rev. B 71, 195413 共2005兲. 076802 共2007兲.

045302-17