F. D. M. Haldane PDF
F. D. M. Haldane PDF
F. D. M. Haldane PDF
The quantum Hall effect' (QHE) in two-dimensional mine which type of state is realized.
(2D) electron systems is usually associated with the pres- To model a 2D semimetal, I use the "2D graphite"
ence of a uniform externally generated magnetic field, model investigated previously by Semenoff as a possible
which splits the spectrum of electron energy levels into lattice realization of a (2+I)-D field theory with the
Landau levels. In this Letter I show how, in principle, a anomaly. 2D graphite has the honeycomb net structure,
QHE may also result from breaking of time-reversal consisting of two interpenetrating triangular lattices
symmetry (i.e., magnetic ordering) without any net mag- ("A" and "8" sublattices) with one lattice point of each
netic fiux through the unit cell of a periodic 2D system. type per unit cell (Fig. 1). A 2D inversion (i.e. , a rota-
In this case, the electron states retain their usual Bloch tion in the plane by tr) interchanges the two sublattices.
state character. Since spin-orbit coupling effects will not be included, the
The model presented here is also interesting in that if electron spin will (for the moment) be suppressed.
its parameters are on a critical line at which its ground Semenoff investigated the tight-binding model with
state changes from the normal semiconductor state to one orbital per site and a real hopping matrix element t ~
this new type of QHE state, its low-energy states simu- between nearest neighbors on different sublattices, and
late a "(2+1)-dimensional" relativistic quantum field also considered the effect of an inversion-symmetry-
theory exhibiting the so-called "parity anomaly" and a breaking on-site energy +M on /I sites and M on 8
(2+1)-D analog of "chiral" fermions without the sites. The model has point group Cs(M=O) or C3
opposite-chirality anomaly-canceling partners that usu- (MAO). In this original version of the model, time-
ally accompany them in lattice realizations of field reversal invariance is present, and Semenoff found com-
theories ("fermion doubling" ). plete cancellation of the anomaly in the M =0 model due
In the zero-temperature limit, the transverse conduc- to fermion doubling, and normal semiconductor behavior
tivity o "3' of a periodic 2D electron system with a gap in for MAO.
the single-particle density of states at the Fermi level
takes quantized values ve /h, where v is generally ra-
tional, but can only take i nteger values in the absence of
electron interactions. This property of a pure system is
stable against sufficiently weak disorder effects. Since
a" is odd under time reversal, a nonzero value can only ,bg qb, ~,
occur if time-reversal invariance is broken.
In the usual QHE, the gap at the Fermi level results
from the splitting of the spectrum into Landau levels by
an external magnetic field. The scenario considered here
is different, and involves a 2D semimetal where there is a
degeneracy at isolated points in the Brillouin zone be-
tween the top of the valence band and the bottom of the
FIG. 1. The honeycomb-net model ("2D graphite") showing
nearest-neighbor bonds (solid lines) and second-neighbor bonds
conduction band, that is associated with the presence of
(dashed lines). Open and solid points, respectively, mark the A
both inversion symmetry and time-reversal invariance. and 8 sublattice sites. The Wigner-Seitz unit cell is con-
If inversion symmetry is broken, a gap opens and the sys- veniently centered on the point of sixfold rotation symmetry
tem becomes a normal semiconductor (v=0), but if the (marked "+") and is then bounded by the hexagon of nearest-
gap opens because time-reversal invariance is broken the neighbor bonds. Arrows on second-neighbor bonds mark the
system becomes a v=+ 1 integer QHE state. If both directions of positive phase hopping in the state with broken
perturbations are present, their relative strengths deter- time-reversal invariance.
I now include a second real hopping term tz between tudes are tzexp(+i&) are shown in Fig. 1: It can be
second-neighbor sites (i.e. , between nearest-neighbor seen that the Hamiltonian has acquired a chirality if the
sites on the same sublattice). This does not change the local field is present.
space group, though it does eliminate a particle-hole It is useful to consider a possible model for the origin
symmetry of the energy bands of the original model. To of such an internal magnetic field. Magnetic dipole mo-
break time-reversal invariance, I also add a periodic lo- ments p, ordered ferromagnetically normal to the plane,
cal magnetic-flux density B(r) in the z direction normal are placed at the center of each hexagonal cell of the
to the 2D plane, with the full symmetry of the lattice, honeycomb net, and B(r) is the sum of their dipole
and with zero total flux through the unit cell. fields. Note that ferromagnetic ordering in 2D does not
Since the net flux per unit cell vanishes, the vector po- generate a uniform component of the magnetic-flux den-
tential A(r) can be chosen to be periodic. The effect of sity. The absolute value of p is Ca p/a, where a is the
this local field is to multiply the matrix element for hop- fine-structure constant, p is the dipole moment in Bohr
ping between sites by the unimodular phase factor magneton units, a is the lattice spacing Bohr radii, and C
exp[i(e/l'i) fA dr] where the integral is along the hop- is a dimensionless constant (of order unity) which de-
ping path, which I take to be rectilinear. The phases can pends on the lattice structure. If p and a are of order
be chosen with any consistent convention such that the unity in their natural units, the phase p in this model will
total phase accumulated around a closed path adds up to be a small quantity controlled by the fine-structure con-
the flux enclosed in units of the flux quantum 40 stant.
To diagonalize the Hamiltonian, I use a basis of two-
Since closed paths of first-neighbor hops enclose com- component "spinors" (y~, y~) of Bloch states con-
plete unit cells (and hence no net flux) the t matrix ele-
~
structed on the two sublattices. Let al, a2, a3 be the dis-
ments are unaff'ected. The t2 matrix elements acquire a placements from a B site to its three nearest-neighbor A
phase &=2rr(2@, +Nb)/@o, where @, and @b are the sites, defined so that ia~xa2 is positive. I also define
fluxes through the regions of the unit cell marked a and b~ a2 a3, b2 a3 al, etc. ; the set of displacements to
b in Fig. 1. The hopping directions for which the ampli- the six nearest neighbors on the same sublattice is
[+ b;}. In this representation, the Hamiltonian becomes
r
H(k) =2tzcosy g;cos(k b;) I+i~ g; [cos(k a;)a'+sin(k a;)o ] + M 2t2sinp g;sin(k b;) a
where cr' are Pauli matrices. The Brillouin zone is a
hexagon rotated 90' with respect to the Wigner-Seitz through the thermodynamic relation cr" =err/BBo ,T~
unit cell: At its six corners (k al, k a2, k a3) is a per- evaluated at BO=O, where o is the 2D electric-charge
mutation of (0, 2rr/3, 2rr/3). The two distinct corners density, and Bo is the flux density of a uniform external
k, are defined so that k, b; = (2rr/3) a, a = 1. ~ magnetic field in the z direction. To calculate the in-
The energy bands are easily obtained. There are two duced charge density rr to a weak external magnetic
bands which only touch if all three Pauli matrix terms in field, it is convenient to expand the Hamiltonian in the
(1) have vanishing coefficients. This can only occur at neighborhood of the band extrema at the zone corners k,
zone corners kand then only if M =3 J3at2sinp. I will to linear order in 8k=k k, , and make the Landau-
assume )tz/t~ ~
( ',
, which guarantees that the two Peierls substitution 68k II, where II=(II",IP) is the
dynamical momentum with components satisfying the
bands never overlap, and are separated by a finite gap
unless they touch. commutation relation [II",IIr] =ih, eBO.
If both M and t2sinp vanish, the bands touch at both For weak Bo, coupling between the two distinct zone
zone corners, where the group of the wave vector has corners can be neglected, and two independent effective
the unitary subgroup C3,which contains a reflection Hamiltonians H, are obtained, where
that interchanges the A and 8 sublattices. Apart from
H, =c(II'o 11 o')+m, czcr .
the zone center, these are the only points in the Brillouin
zone where this group has irreducible representations Here c=
', t~ ~a; ~/6 and m, c =M 343at2sinp; II,'
with dimensions greater than unity, and the degenerate and H are Hermitian operators with the commutation
states at these points belong to the two-dimensional rep- relation [II,', II, ] =iaeBoh, defined by
resentation. The touching of the bands at t~o distinct
points in the Brillouin zone is a manifestation of fermion (3)
doubling. ' The degeneracy of the bands at these points After second-quantization, (2) is precisely the Hamil-
is lifted either by nonzero M or nonzero t2sinp, either of tonian of a free-fermion-field theory studied by Jackiw
which reduce the unitary subgroup to C3, which has only as an (2+1)-D analog of the Dirac Hamiltonian.
one-dimensional irreducible representations. The spectrum of (2) is relativistic; for 80 =0,
When the Fermi level lies in a gap between two bands,
a ~ is quantized at T =0, and its value can be obtained e, t(k) = +' [(h ck ) + (m, c ) -] '
2016
VOLUME 61, NUMBER 18 PHYSICAL REVIEW LETTERS 31 OCTOBER 1988
while for Bp&0, relativistic Landau levels are obtained the same sign. This allows 0." in the limit Bp=0 to be
as follows: evaluated as ve /h, where v= 2 [sgn(m ) sgn(m+)l
=+ 1 or 0. The phase diagram of v for the spinless
e,~ = ~ [(rn, c ) +nb eBp c ] ' ~ ~
(n ~ I ), (4a) electron model as a function of M/t2 and p is shown in
e p =am, c sgn(eBp) . (4b) Fig. 2.
I note that when the model has neither an inversion
Every n ~
1 level that evolves out of the upper band as center nor time-reversal invariance (i.e. , when both M
Bp is turned on is balanced by a level that evolves from and t2sinp are nonzero), so ~m~ e~m ~, the spec-
k, and the
~
the lower band. However, the n =0 "zero-mode" energy trum is no longer invariant under k
is not symmetric under Bp Bp. It evolves from the fermion-doubling principle is defeated. In particular,
upper band if am, eBp is positive, and from the lower along the critical lines in the phase diagram where one of
band if it is negative. rrt+ or rrt vanishes, the model has a low-lying massless
In the time-reversal symmetric case t2sin&=0, the two spectrum simulating nondegener ate relativistic chiral
masses m+ and m are equal, and the sum of the fermions.
Landau-level spectra derived from the two distinct zone When m, 0, the fermion field theory derived from
corners is particle-hole symmetric, and invariant under the expansion (2) about the Fermi point with vanishing
Bp Bp. In this case, a" 0 by time-reversal invari- gap has a charge-conjugation symmetry (particle-hole
ance. As the Hainiltonian is changed, tr"i' remains in- symmetry) which is not present in the lattice model with
variant, provided the Fermi level remains in a gap. t2&0 from which it is derived. In the continuum field
When Bp 0, models where the Fermi level is in the gap theory, there is no lower bound to the Dirac sea of filled
and rn ~ and m have the same sign can evolve continu- electron states, and the establishment of absolute as op-
ously from the time-reversal invariant case, and hence posed to relative values of cr"~ is ambiguous. Jackiw in-
have 0'~ 0. vokes the charge-conjugation symmetry of (2) with
To calculate tr"r for models where rn~ and trt have m =0 to assign the value o" =0 in the case of a
opposite signs, I continuously turn on the external field, particle-hole symmetric Fermi level, where the "zero-
then vary m+ and m until they become equal, at the mode" Landau level (4b) is half filled. This would imply
same time varying the Fermi level so at all times it lies in a quantum Hall effect with v= 2 a if the zero mode is
filled, and v = , a if it is empty.
'
a gap. Comparison of the occupation numbers of the This suggests
Landau levels obtained this way with those obtained by "charge fractionalization, "
and violates the principle
continuously applying the field to the time-reversal in- that a noninteracting electron system can only exhibit an
variant system shows that they differ by the complete integral QHE. The model studied here shows how the
filling of one Landau level. Thus at T=O and with a high-energy cutoff structure of a model with undoubled
fixed chemical potential, the application of a weak exter- fermions described by the relativistic Hamiltonian (2) at
nal magnetic field to a system where m~ and m have low energies must break the charge-conjugation symme-
try, and give an extra contribution of +' , to v, restor-
opposite signs induces an extra fteld dependent g-round- '
state charge density Atr ~ e Bp/h relative to the field- ing an integral QHE. Thus even if the low-energy spec-
independent charge density when these parameters have trum consists of undoubled chiral fermions, their
partners must be present at high energies to restore a
properly integral QHE.
When electron spin is included without any other
change, there is an equal contribution from both spin
components, and 0 "r is doubled. However, a periodic lo-
cal magnetic field with the full symmetry of the lattice
will also couple to electrons with a Zeeman term 0'
f2
=y&S', where S' is the azimuthal electron spin. This
term will relatively displace the up-spin and down-spin
bands by an energy ) hp, and if this exceeds the gap at
the Fermi level, the system will become a partially spin-
polarized metal. If , y ii exceeds 3 J3 t2 ~, the QHE
'
~ ~ ~
2017
VOLUME 61, NUMBER 18 PHYSICAL REVIEW LETTERS 31 OCTOBER 1988
geometrical constant of order unity, and g is the Lande g end of their calculation.
factor for the electrons. I thank E. Fradkin and T. A. L. Ziman for very useful
While the particular model presented here is unlikely discussions. The author would like to thank the Alfred
to be directly physically realizable, it indicates that, at P. Sloan Foundation for financial support.
least in principle, the QHE can be placed in the wider
context of phenomena associated with broken time-
reversal invariance, and does not necessarily require
external magnetic fields, but could occur as a conse-
quence of magnetic ordering in a quasi-two-dimensional 'K. Von Klitzing, G. Dorda, and M. Pepper, Phys. Rev. Lett.
system. 45, 494 (1980).
This requirement is not fulfilled by the physical system 2R. Jackiw, Phys. Rev. D 27, 2375 (1984).
(a domain wall in a PbTe-type semiconductor) in which H. B. Nielsen and M. Ninomiya, Nucl. Phys. B185, 20
Fradkin, Dagotto, and Boyanovsky (FDB) have recent- (1981), and B193, 173 (1981).
ly proposed related effects may be realized. In this mod- 4D. J. Thouless, M. Kohmoto, M. P. Nightingale, and
el, spin-orbit coupling is supposed to give rise to the M. den Nijs, Phys. Rev. Lett. 49, 405 (1982).
sG. Semenoff, Phys. Rev. Lett. 53, 2449 (1984).
effect, but this does not break time-reversal symmetry.
The full group of the wave vector at zone corners contains
In fact, in "simplifying" the p bands of the Hamiltonian
additional antiunitary operations, but these do not give rise to
that describes PbTe, FDB introduce an unphysical
extra degeneracies.
effective spin-dependent hopping term that is odd under 7P. Streda, J. Phys. C 15, L717 (1982); A. Widom, Phys.
time reversal, and thus break the time-reversal invari- Lett. 90A, 474 (1982).
ance of the original physically motivated model. This, E. Fradkin, E. Dagotto, and D. Boyanovsky, Phys. Rev.
rather than any topological character of the domain wall, Lett. 57, 2967 (1986), and 58, 961(E) (1987); D. Boyanovsky,
is the reason that FDB find the "parity anomaly" at the E. Dagotto, and E. Fradkin, Nucl. Phys. B285, 340 (1987).
2018