Scaling of entanglement entropy in honeycomb

lattice on a torus
arXiv:1502.01854v1 [cond-mat.str-el] 6 Feb 2015

Wen-Long You
School of Physical Science and Technology, Soochow University, Suzhou, Jiangsu
215006, People’s Republic of China
E-mail: wlyou@suda.edu.cn

Abstract. The entanglement entropy of noninteracting fermionic system confined to

two-dimensional (2D) honeycomb lattice on a torus are calculated. We find that the
entanglement entropy can characterize Lifshitz phase transitions without a local order
parameter. In the noncritical phase and critical phase with nodal Fermi surface, the
entanglement entropy satisfies the area law. The leading subarea term is a constant
in gapped phase rather than a logarithmic additive term in gapless regime. The
tuning of chemical potential allows for a nonzero Fermi surface, whose variation along
particular direction determines a logarithmic violation of the area law. We perform the
scaling of entanglement entropy numerically and find agreement between the analytic
and numerical results. Furthermore, we clearly show that entanglement spectrum is
equivalent to edge spectrum.

PACS numbers: 03.67.Mn,71.10.Fd,05.70.Jk,73.20.At

Scaling of entanglement entropy in honeycomb lattice on a torus 2

1. Introduction

Entanglement entropy is a unifying theme in the interdisciplinary theories during the

past decade. It is a helpful bridge between quantum gravity, quantum information
and condensed matter physics. It is defined as the von Neumann entropy of one of the
reduced states for a pure state. Given a normalized wave function |ΨAB i and a partition
of the system into subsystems A and B, the associated reduced density matrix of A part
is obtained from ρA = TrB |ΨAB ihΨAB | by tracing out the degrees of freedom in B part.
The bipartite entanglement entropy is given by,
SvN = −TrρA log2 ρA . (1)
Related generalized quantities are the Rényi entropies defined as
1
Sn = log2 TrρnA , (2)
1−n
whose limit recovers the von Neumann entropy via relation SvN = limn→1 Sn . Once the
complete values of Sn for all n are known, one can obtain all eigenvalues of ρA , which are
called the entanglement spectrum. The spooky nonlocal nature makes the entanglement
entropy hold great advantage in the context of condensed matter physics. On one hand,
it plays an essential role in determining the efficiency of numerical algorithms based on
matrix product states, as well as an extension to tensor network states [1] and Projected
Entangled Pair States (PEPS) [2]. On the other hand, it is proven that the entanglement
entropy is able to capture the quantum criticality in a many-body Hamiltonian [3, 4].
The constituents become extraordinarily entangled at criticality. Quantum critical
points are successfully pinpointed in both free systems [5, 6] and interacting systems
[7, 8, 9, 10, 11], especially for those critical ground states without prior knowledge of
order parameters, e.g., disordered systems[12, 13, 14] and quantum spin liquids [15].
In particular, the discovery of a topological phase transition, where distinct phases
of matter have the same symmetry, was in urgent need of basis-independent way of
identifying quantum critical points. In this respect, extensive work has utilized the phase
sensitivity of the entanglement entropy to successfully characterize the topological order.
The universal constant term in the bipartite entanglement entropy has been shown to
be a probe of topological contribution [16, 17]. The success of entanglement entropy
as a diagnostic tool is essentially rooted in its peculiar scaling properties in a quantum
state. The growing awareness of the monotonic scaling of von Neumann entropy can
be traced back to the study of black hole physics [18, 19, 20], and then the profound
relations to conformal field theory (CFT) in (1+1)-dimensions were established [21, 22].
Despite the formal simplicity, it is surprisingly difficult to calculate the entanglement
entropy analytically, even for noninteracting models. The entanglement entropy in one-
dimensional system has been extensively studied and the scalings have been explicitly
established by CFT [3, 23]. Intensive research in terms of numerical algorithms unveiled
a few qualitative and quantitative aspects in higher dimensions [24, 25].
In this work, we study the half-space entanglement entropy of free fermion in
the anisotropic honeycomb lattice with fine tuning chemical potential. In order to
Scaling of entanglement entropy in honeycomb lattice on a torus 3

discriminate the relation between various boundaries and entanglement entropy, we

place the fermion system on a torus. The scaling behavior of the entanglement entropy
in a two-dimensional hexagonal lattice demands a comprehensive investigation. The
basic interest in honeycomb lattices comes from the graphene and associated Dirac
fermion physics. More interestingly, a recent influx of ideas and toolboxes can create
the so-called ”artificial honeycomb lattice” [26], such as photonic lattices [27], coupled
molecular systems [28] as well as optical honeycomb lattice [29]. The manipulations of
Dirac points are possible through these artificial methods [30, 31, 32]. The asymmetric
tunneling amplitudes cause a topological Lifshitz transition from a semimetal to an
insulator, which is characterized by the topological change of the Fermi surface in the
Brillouin zone [33]. It was found that the Lifshitz points can be also captured by the
entanglement entropy [34]. Moreover, based on the bulk-edge correspondence [35, 36],
the entanglement entropy and the Berry phase are closely related in a topologically
nontrivial system through the zero-energy edge states [37]. A fascinating aspect of
graphene-like structure is that there are two types of edges, i.e., zigzag and armchair
edges. The role of edges in determining the entanglement entropy in honeycomb lattice
and associated scaling behavior are not fully understood. In addition, the chemical
potential can be well tuned in graphene by a gate voltage, so the dependence of the
entanglement entropy on the chemical potential is a nontrivial issue to be discussed.

2. Tight binding model on honeycomb lattice

The anisotropic fermionic tight-binding model on a two-dimensional honeycomb lattice

reads
X X X X
H= − ti (a†r br+τi + b†r+τi ar ) − µ a†r ar − µ b†r br , (3)
r∈ΛA i=a,b,c r∈ΛA r∈ΛB

where ar (br ) is the annihilation operator on the triangular sublattice ΛA ◦ (respectively

ΛB •) of the rth dimer, τi is the nearest neighbor bond vector, and µ is the chemical
potential. For simplicity, we confine the hexagonal lattice to the brick lattice without
changing lattice topology as shown in Fig.1(a). The nearest-neighbor separations are
given by
τa = (0, 1), τb = (−1, 0), τc = (1, 0), (4)
where the bond length is set to unit. Taking Fourier transformation for all lattice sites,
the Hamiltonian (3) reduces to
!
X † ξ(~k) − µ i∆(~k)
H= − ck ~k) −ξ(~k) − µ ck , (5)
~
−i∆(
k∈BZ
and then the eigenspectrum is
q
ǫ(k) = ± ξ 2 (~k) + ∆2 (~k) − µ,
~ (6)
where ξ(~k)= ta cos kx + tb cos kx + tc cos ky , ∆(~k)=ta sin kx - tb sin kx -tc sin ky , c~†k =(c~†k ,
c† ) and ~k ≡ (kx , ky ) is the Brillouin zone shown in Fig.1(b). The bulk band structure
~k+~
π
Scaling of entanglement entropy in honeycomb lattice on a torus 4

in Eq.(6) is invariant under inversion : ~k → −~k. In the parameter space |ta − tb | < tc <

Figure 1. (Color online) (a) Schematic illustration of brick-type graphene torus. A

virtual bipartition with zigzag edges is made horizontally. The shaded region denotes
one of the subsystems. The dashed (red) box indicates a unit cell. A bipartition with
armchair edges can be implemented by truncating the lattice vertically. (b) BZsq is
the Brillouin zone of square lattice, and the double count for ~k and ~k + ~π indicates the
BZ (region 1-4) is the Brillouin zone of the hexagonal lattice.

ta + tb (containing symmetric case ta = tb = tc ), the valence and conduction bands

intersect linearly at two ~kD -points (known as the Dirac points), given by
t2c −t2a −t2b
!
~kD = 1 2π − arccos( 2ta tb
)
2 2 )t2 −(t2 −t2 )2 , (7)
2 arccos( (ta +tb2t c
t t2
a b
)
a b c

~
which will be suited at K-points, i.e., (± 32 π, 0), of the first Brillouin zone for symmetric
hoppings t1 = t2 = t3 . The spectra and the corresponding Dirac points for a semimetallic
phase is displayed in Fig.2(a). The low-energy electronic states at any Dirac point
are described by the massless relativistic Dirac fermions, which lead to a number of
unconventional and fascinating phenomena, such as anomalous quantum Hall effect
[38]. The asymmetric hoppings make the two Dirac points move away from the K- ~
points and approach each other [39]. At a critical asymmetry the Dirac points merge at
one of the four inequivalent M ~ -points, that is, (±π, 0) and (0, ±π). A band gap opens
upon increasing the asymmetry further; see illustration in Fig.2(b). The topological
invariant that characterizes the topological properties is given by the Zak phase, which
is the integration of Berry connection over in the reduced Brillouin zone [40, 41]. If the
integral loop encloses a gap closing point, the Zak phase gives
Z π
γ = −i hv− (~k)|∂~k v− (~k)i · d~k = π, (8)
−π
!
1
where v− (~k) = √1 is the eigenvector of lower band and θk = Arg[ξ(~k) + i∆(~k)].
2 eiθk
Otherwise it is zero. The merging is a topological transition since the Berry phases ±π
Scaling of entanglement entropy in honeycomb lattice on a torus 5

Figure 2. (Color online) Three-dimensional view of the energy spectrum of the

Hamiltonian (3) for (a) ta = 0.28, tb = 0.32, tc = 0.4; (b) ta = 0.15, tb = 0.25,
tc = 0.6.

associated with the two Dirac points annihilate each other at the critical point.

3. Entanglement entropy of noninteracting fermions

The entanglement entropy in the ground state of free fermions can be determined from
the correlation matrix as a fictitious Hamiltonian defined as [42, 43, 44, 45]
Ĝij = Tr(ρA c†i cj ), (9)
where ci (c†i ) is the annihilation (creation) operator for site i, j ∈ A. There exist an
simple relation between correlation matrix of order O(NA ) and the reduced density
matrix of order O(2NA ) : ρA = det(1 − Ĝ) exp{ ij [ln Ĝ(1 − Ĝ)]ij c†i cj } [46], where NA is
P

the number of sites in A subsystem. Calculating the eigenvalue λn of matrix Ĝ requires

only the diagonalization of NA × NA matrix, and then the entanglement entropy is given
by
SvN = −Tr[Ĝ log2 Ĝ − (1 − G) log2 (1 − Ĝ)]. (10)
The eigenvalue λn (0 ≤ λn ≤ 1) of Ĝ is termed entanglement spectrum, which is an
alternative description of Li and Haldane’s version [47]. Recently the entanglement
spectrum has been proposed as a ground-state property that captures characteristic edge
excitations [48, 49, 50]. In a large subsystem, most of the eigenvalues lie exponentially
close to 0 and to 1 [51]. A special role is played by eigenvalues λn = 1/2 corresponding
to a zero-energy mode [52], and it will contribute a maximal value 1 to the entanglement
entropy by Eq. (10).
In the following, we consider a torus of size Lx × Ly , and a subregion A can be
designated as Ω= [0, z1 Lx ]×[0, z2 Ly ] with zi ≤ 1. Here we measure the entanglement
between a cornerless cylindrical region A and its complement by two virtual cuts along
x or y direction; see illustration in Fig.1(a). Zigzag edges emerge following a horizontal
bipartition, i.e., z1 = 1, z2 = 1/2, and kx labels the momentum running along the zigzag
ribbon. A vertical truncation produces armchair edges, i.e., z1 = 1, z2 = 1/2, and ky
Scaling of entanglement entropy in honeycomb lattice on a torus 6

is well defined instead. One can simultaneously classify the entanglement spectrum
with respect to the momentum parallel to the cut, and an intimate relation between
entanglement spectrum and eigenspectrum of subblock A can be discovered.
In Fig.3 (a), a flat band of zero energy exists for momenta connecting with two
Dirac points along x direction near the zigzag edges. Such zero-energy edge states are
reflected by the maximally entangled modes in entanglement spectrum, as shown in
Fig.3 (b). The fermion doubling in a chiral symmetric system is a universal property
of the bulk, and the bulk-edge correspondence guarantees the edge states appear as
pairs. In other words, such doubly degenerate entanglement modes are protected by the
chiral symmetry. In the semimetallic phase with zigzag edges, the contributions to the
entanglement entropy can be broadly separated into those from the edges and those from
bulk, while only bulk states contribute to the entanglement entropy in the insulating
phase (see Fig.4). However, the situation is quite different for armchair edges. The
interedge interaction of armchair edges in the semimetallic phase makes the localized
states fade away [53], and thus edge states play a trivial role in the entanglement entropy
(see Fig. 5). Nevertheless, as is shown in Fig.6, the anisotropy will induce number of
edge states, which contribute significantly to the entanglement entropy. It is explicit that
the zero-energy states have one-to-one correspondences with the maximally entangled
modes in the entanglement spectrum. In a sense, the entanglement spectrum is a very
promising tool to characterize chiral edge states.

1 1.0
(a) (b)

0.1
(k )
x
(k )
x

0.0
0 0.5
n

-0.1
2.0 2.2 2.4 2.6

-1 0.0
-3 -2 -1 0 1 2 3 -3 -2 -1 0 1 2 3
k k
x x

Figure 3. (a) The energy spectrum for ta =0.28, tb =0.32, tc =0.40 on Lx = 40, Ly = 80
lattice with zigzag edges. (b) The corresponding entanglement spectrum at µ = 0. The
inset amplifies the region in the dashed circle.

We show the half-space entanglement entropy as a function of tc along the

normalized path ta + tb + tc = 1 for ta = 0.15, µ = 0 in Fig.7. The entanglement entropy
changes discontinuously when tc moves across points t1c = |ta − tb | and t2c = ta + tb .
In fact, the abrupt change associated with the topological phase transition can be
grasped more clearly by its first derivative for both types of bipartition and any µ
within the bulk bands. The system is an insulator when tc < |ta − tb | or tc >ta + tb , and
enters the semimetallic phase when |ta − tb |< tc <ta + tb . In the insulating phase, the
Scaling of entanglement entropy in honeycomb lattice on a torus 7

1 1.0
(a) (b)

(k )
(k )

x
x

0 0.5

n
-1 0.0
-3 -2 -1 0 1 2 3 -3 -2 -1 0 1 2 3
k k
x x

Figure 4. (a) The energy spectrum for ta =0.15, tb =0.25, tc =0.6 on Lx = 40, Ly = 80
lattice with zigzag edges. (b) The corresponding entanglement spectrum at µ = 0.

1 1.0
(a)
(b)

0.504

(k )
y
0.500

n
(k )
(k )

y
y

0 0.5 0.496
-0.4 -0.2 0.0 0.2 0.4
n

k
y

-1 0.0
-3 -2 -1 0 1 0.00 2 3 -3 -2 -1 0 1 2 3

k k
y -0.04 y
-0.8 0.0 0.8

Figure 5. (Color online) The energy spectrum for ta =0.28, tb =0.32, tc =0.4 on
Lx = 40, Ly = 80 lattice with armchair edges. (b) The corresponding entanglement
spectrum at µ = 0. The inset amplifies the region in the dashed circle.

1.0
1 (a)
(b)
(k )

(k )
y

0 0.5
n

-1
0.0
-3 -2 -1 0 1 2 3 -3 -2 -1 0 1 2 3
k k
y y

Figure 6. The energy spectrum for ta =0.15, tb =0.25, tc =0.6 on Lx = 40, Ly = 80

lattice with armchair edges. (b) The corresponding entanglement spectrum at µ = 0.
Scaling of entanglement entropy in honeycomb lattice on a torus 8

entanglement entropy is linear with the block length for both edges, and the deviation
seems negligible. However, the finite-size effect is prominent in the gapless phase. There
exists a remarkable correction to the linear dependence on the block length. We note
that the correction is likely a logarithmic form for armchair edges, whereas an oscillating
correction exists for zigzag edges. We attribute this peculiar correction to finite-size
effect of edge modes. The trends in the proliferation of edge modes for both edges are
converse as we approach a quantum critical point, as displayed in the inset of Fig.7.
We find that the existence of edges state is that ta , tb and tc should satisfy the triangle
inequality for zigzag edge [54].

1.00 60
Armchair 3.5
20x20

edge
30x30

n
30
40x40
0.75

zigzag
3.0
0
0.0 0.4 0.8
/L

t
0.50 c
vN
S

2.5

0.25

2.0
0.00
0.0 0.4 0.8
t
c

Figure 7. (Color online) The entanglement entropy through path ta + tb + tc = 1

for ta = 0.15,µ = 0 with zigzag edge (left axis) and armchair edge (right axis). Inset
shows the convolution of the number of edge states on Lx = 30, Ly = 60 lattice and
µ = 0.

Next we investigate the scaling behavior of bipartite entanglement entropy in the

presence of different edges. The scaling of the entanglement entropy has attracted much
attention by a flurry of recent works [55, 56, 57, 58]. A general form of the entanglement
entropy in two-dimensional system is argued to be of the form
SvN = c1 L log2 L + c2 L + c3 log2 L + c4 , (11)
where ci (i=1,· · ·, 4) are size independent coefficients. Eq.(11) includes a universal
logarithmically divergent correction and a nonuniversal area-law term. Which term is
the leading term requires further elucidation of the Fermi surface. We will give both
analytical and numerical results for the logarithmical scaling of entanglement entropy
in one concrete case.
Apparently in the insulating phase, the entanglement entropy grows asymptotically
as L [59], as supported in Fig.8(a). For µ = 0 in the semimetallic phase, the Fermi
surface reduces to zero dimension, and then the entanglement converges to the area law
[55, 60, 61], as is shown in Fig.9(a). We see that the slope of scaling changes for varying
parameters in the semimetallic phase, so the area-law term is nonuniversal. Apart from
Scaling of entanglement entropy in honeycomb lattice on a torus 9

the leading power-law term, the subleading term is of general interest. To this end, we
plot the quantity
sub
SvN = LSvN (L + 2) − (L + 2)SvN (L), (12)
which will cancel the area-law term and hence extract the subleading contribution. It is
sub
remarkable that SvN behave diversely in different phases. If the entanglement entropy
sub
follows a behavior as SvN (L) = c2 L + c4 + O(1/L), SvN will approach a constant term
like
sub
SvN ∼ −c4 , (13)
as is shown in the inset of Fig.8(a). In contrast, SvN displays a divergent oscillation in
the inset of Fig.9 (a). A careful analysis identifies a logarithmic additive term in these
critical phases by following SvN (L) = c2 L + c3 log2 L + O(1), and then the subarea law
of the entanglement entropy yields[61, 62]
sub
SvN ∼ −c3 log2 L. (14)
The subdominant contribution stems from one chiral mode in the zero-dimensional Fermi
surface, which infers c3 = 1/3 by the FisherCHartwig conjecture [63, 64]. The oscillation
can be interpreted as the following: In case of symmetric hoppings, the existence of zero-
energy localized states is restricted for 1/3 of the total momentum parallel to the zigzag
edge, and hence a multiple of 3 in block length favors more edge states. Hence, the
oscillation has a period of 3, as is observed in the inset of Fig.7.
A significant difference of the scaling in the entanglement entropy is found in the
presence of finite density of state at the Fermi surface. In such situation, every low-
energy radial mode is approximately treated as a chiral excitation that contribute a lnL
to the total entanglement entropy [65]. Therefore, a multiplicative logarithmic correction
to the power-law behavior arises for finite Fermi surface (i.e., line nodes), which are
reported in Fig.8(b) and Fig.9(b)[66, 67]. Noticeably, we see that the entanglement
entropy with armchair edge is larger than that with zigzag edge, and the reason will be
accounted soon.
Widom conjecture [68] predicted that the prefactor of the logarithmic term is
determined by chemical potential µ via an accurate formula [69]
1
Z Z
c1 (µ) = dSx dSk |nx · nk |, (15)
24π ∂Ω ∂Γ(µ)

where Ω is the real-space region of A, rescaled by L such that Vol(Ω)=1. Vectors nx

and nk denote the normal vectors on the real-space surface ∂Ω and the Fermi surface
∂Γ(µ). From Eq.(15), we see the topology of the subsystem and the Fermi surface
has intimately related to the entanglement entropy. Thus the existence of edge states
crucially affects the physical properties of graphene system. Besides, it was argued
that Lifshitz quantum phase transitions are accompanied with a nonanalyticity in the
prefactor of the leading-order term [70]. In two dimensions the calculation of c1 (µ) can be
simplified. The normal unit vectors in real space are ±ŷ for zigzag-type bipartition and
Scaling of entanglement entropy in honeycomb lattice on a torus 10

150 300
(a) (b)
armchair
armchair
zigzag
zigzag
100 200
0.05

vN
vN

S
S

sub

vN
0.00

S
50 100
-0.05
20 40 60
L

0 0
0 20 40 60 80 0 100 200 300
L L log L
2

Figure 8. (Color online) The scaling of entanglement entropy at ta =0.15, tb =0.25,

sub
tc =0.60 when (a) µ = 0 and (b) µ = −0.5. Lines are linear fits. SvN for ta =0.15,
tb =0.25, tc =0.60 with zigzag edge is shown in the inset.

150 200
(a) (b)

160 armchair

zigzag

100
120
vN
S

vN
S

70 80
50
sub

vN

0
40
S

-700 40 80

0
L
0
0 20 40 60 80 100 0 100 200 300 400
L L log L
2

Figure 9. (Color online)(a) The scaling of entanglement entropy with zigzag (square)
and armchair edges (circle) at ta =tb =tc =1.0 (solid) and ta =0.28, tb =0.32, tc =0.40
sub
(open) with µ = 0. SvN for ta =tb =tc =1 with zigzag edge is shown in the inset. (b) The
scaling of entanglement entropy with both zigzag and armchair edges at ta =tb =tc =1.0
with µ = −1.0. Lines are linear fits.

±x̂ for armchair-type bipartition. Since the Fermi surface is a curve in two dimensions,
we can parameterize the curve by kx = kx (θ) and ky = ky (θ) like [34]
(ta + tb ) cos kx + tc cos ky ≡ µ cos θ,
(ta − tb ) sin kx − tc sin ky ≡ µ sin θ.
Then we have
 
1  dkx 
Z
c1,zigzag (µ) = dθ  , (16)
12π ∂Ω  dθ 
 
1  dky 
Z
c1,armchair (µ) = dθ  . (17)
12π ∂Ω  dθ 
Eq.(16) and Eq.(17) indicates c1 (µ) are proportional to the variation of Fermi surface
along x and y, respectively. For |µ| is large, the Fermi surface is an oval, while
Scaling of entanglement entropy in honeycomb lattice on a torus 11

the disconnected Fermi components appear with the decrease of |µ|. Without loss
of generality, we give a simple demonstration of the solution for symmetric case
ta = tb = tc = 1. The dispersion has particle-hole symmetry and 90-degree rotational
invariance, so we only consider positive µ in the first quadrant of the Brillouin zone,
and then the Fermi surface can be parameterized by
(
kx (θ) = cos−1 R(µ, θ)
ky (θ) = sin−1 (−µ sin θ)
p
for 0 ≤ µ ≤ 1 and θ ∈ (−π, 0). Here R(µ, θ) = (− 1 − µ2 sin2 θ + µ cos θ)/2.
The longitudinal variation ∆ky is 2 arcsin(µ) and the horizontal variation ∆kx is
√
arccos( −µ−1
2
) − arccos( µ−1
2
). When µ > 5, it is replaced by
(
kx (θ) = cos−1 R(µ, θ)
ky (θ) = sin−1 (−µ sin θ)
2 2
for θ ∈ [− arccos( µ4µ+3 ), 0]. The longitudinal variation ∆ky is arccos µ 4−5 and the
√
horizontal variation ∆kx is arccos( µ−1
2
). Otherwise when 1 < µ < 5,
(
kx (θ) = cos−1 R(µ, θ)
ky (θ) = sin−1 (−µ sin θ)
for θ ∈ [− sin−1 µ1 , 0] and

(
kx (θ) = π − cos−1 R(µ, θ)
ky (θ) = π + sin−1 (µ sin θ)
2 2
for θ ∈ [arccos( µ4µ+3 ) − π, arcsin( µ1 ) − π]. The longitudinal variation ∆ky is arccos µ 4−5
and the horizontal variation ∆kx is arccos( µ−1 2
). The resulting prefactors of leading
logarithmic terms are:
1
c1,zigzag (µ) = ∆kx ,
3π
1
c1,armchair (µ) = ∆ky . (18)
3π
A generalization to anisotropic case and arbitrary edges is straightforward. The
coefficient of the leading term in system of open boundary condition is half of the one of
periodic boundary condition, due to the fact that the boundary between the bipartite
spaces is half of that of stripes [71].
From Fig. 10, we can recognize the entanglement entropy presents a shoulder-like
behavior with respect to the chemical potential µ. The finite-size effect becomes less
dominated for large size. c1 (µ) extracted from fits to our numerical data agrees rather
well with the exact form Eq.(18) in the inset of Fig.10. c1 (µ) goes to zero with µ
approaches Dirac point, where the logarithmic correction is vanishing. When µ deviates
slightly from the critical point, two small Fermi pockets are formed with Fermi vector
kF = |µ|/vF , and Eq.(18) implies that c1 (µ) scales linearly for the low-energy modes
Scaling of entanglement entropy in honeycomb lattice on a torus 12

6
0.8
(a) 0.4
(b)

)
3

c (
20x20 0.2 20x20 0.4

c (
1
40x40 40x40

80x80 0.0 80x80

4 0.0
0 1 2 3
0 1 2 3

2
/L

/L
vN

vN
S

S
2
)

)
2
1
c (

c (
1
2

2
0 0
0 1 2 3 0 1 2 3

0 0
0 1 2 3 0 1 2 3

Figure 10. (Color online) The entanglement entropy between A and B block for
different chemical potential at ta =tb =tc =1.0 with (a) zigzag edges and (b) armchair
edges. The legends mark the half size. Insets show the prefactor c1 (µ), c2 (µ) in Eq.(11)
as a function of the chemical potential µ for the ground state of the two-dimensional
fermionic tight-binding model. c1 (µ) extracted from fits to our numerical data (red
filled square) are compared with the analytical results (solid line) in the upper left
inset. (c) Analytical results c1 (µ) for both edges, and the ratio in the inset.

when system size L ≫ l∗ , where l∗ is a characteristic length l∗ ∼ kF−1 . The ratios

√
of c1 (µ) to c2 (µ) with both edges are above a factor 3, because the Brillouin zone
√
of brick-type lattice along y direction is stretched 3 times longer than that along x
direction comparing with pristine honeycomb lattice. The ratios reach their maxima
passing through a Van Hove singularity, i.e., µ = 1, where two Fermi surfaces merge
to form a single Fermi surface; see illustration in Fig.10(c). Thus at µ = 1 there is a
Lifshitz transition. Besides a logarithmic divergence of the entanglement entropy, the
linear-term coefficient is intricately related to ultraviolet cutoff or correlation length,
and it is generally nonuniversal. Shown in the lower insets of Fig.10 (a-b) are c2 (µ),
which decrease with the increase of |µ|. Extraordinarily, there is a dip at µ = 1 by a
meticulous observation. We attribute such decrease to the enhanced contribution from
a negative subarea term at µ = 1 [61].

4. Spin-orbit interaction

Next, we extend our scope to Kane-Mele model, which is a tight-binding model with
second neighbor spin-orbital interaction defined on a honeycomb lattice,
X X
H = −t (a†r,α br+τi ,α + b†r+τi ,α ar,α )
r∈ΛA ,α i=a,b,c
X X
+ it2 a†r,α [σ · (τ i × τ̃ j )]α,β ar+τi +τ̃ j ,β
r∈ΛA ,α,β i,j=a,b,c
X X
− it2 b†r,α [σ · (τ i × τ̃ j )]α,β br+τi +τ̃ j ,β .
r∈ΛB ,α,β i,j=a,b,c
Scaling of entanglement entropy in honeycomb lattice on a torus 13
X X
−µ a†r,α ar,α − µ b†r,α br,α . (19)
r∈ΛA ,α r∈ΛB ,α

Here τ̃i = −τi is the antiparallel vector of τi in Eq.(4). The Hamiltonian is time
reversal invariant and has a wide range of applicability, such as quantum spin-Hall
(QSH) insulator [72, 73], graphene/graphane interface [74] and silicene [75, 76, 77, 78].
The energy band structure of the brick lattice is given by
q
ǫk = ± Φ2k + Θ2k , (20)
where Φk = −t(eiky + eikx + e−ikx ) and Θk = 4t2 cos ky sin kx . Without the second-
neighbor hopping, the model’s bulk and surface spectra display the non-generic feature
that all energy levels come in pairs at each momentum due to parity symmetry [52].
The presence of the spin-orbit interaction, i.e., t2 6= 0, leads to a gap in the bulk
spectrum, as shown in Fig.11. More interestingly, in the case of zigzag ribbon, the flat
dispersion is replaced by the helical modes intrinsic to the QSH insulator as the spin-
orbital is introduced [79]. The dominant contributions originates from the helical edge
state is displayed in Fig.11(d). The area-law monotonic scaling at µ = 0 is confirmed
in Fig.12(a), and the leading logarithmic divergence is also observed in Fig.12(b).

(a) (b)
2 2
(k )

(k )
x

0 0

-2 -2

-1 0 1 -1 0 1
k / k /
x x

1.0 1.0
(c) (d)
t =0.1
2

t =0.0
(k )

2
(k )

x
x

0.5 0.5

0.0 0.0
-1 0 1 -1 0 1
k / k /
x x

Figure 11. (Color online) (Color online) The energy spectrum of Hamiltonian (19)
for (a) t2 = 0 and (b) t2 = 0.1 on Lx = 40, Ly = 40 lattice with zigzag edges. The
corresponding entanglement spectrum at µ = 0 for (c) t2 = 0 and (d) t2 = 0.1.
Scaling of entanglement entropy in honeycomb lattice on a torus 14

120 300
(a) (b)

t =0.0 t =0.0
2 2

80 t =0.1
2
200 t =0.1
2
vN

vN
S

S
40 100

0 0
0 10 20 30 40 50 0 100 200 300
L L log L
2

Figure 12. (Color online) The scaling of entanglement entropy with zigzag edges at
t2 =0 and 0.1 when (a) µ = 0 and (b) µ = −1. Lines are linear fits.

5. Conclusion

In this paper, we study the von Neumann entropy in free fermionic systems on a two-
dimensional honeycomb lattice. Distinct bipartitions on the honeycomb lattice will
produce different edges, and the edge states will have an effective impact upon the area-
law terms in the entanglement entropy. We exhibit that the zero-energy edge state has
an one-to-one correspondence with the maximal entangled mode in the entanglement
spectrum, and hence we speculate the main contributions to the entanglement entropy
come from the bulk states and edge states. The edge states play an essential role
in semimetallic state with zigzag edges and in insulating phase with armchair edges.
Furthermore, we find the entanglement entropy obeys area law when the Fermi surface is
within the gap or a nodal point. The corrections to the pow-law term behave differently
in critical phase from in noncritical phase. A logarithmic violation to the area law
emerges in the presence of line-nodes Fermi surface. We also show that the logarithmic
scaling highly depends on the topology of the Fermi surface. The prefactor of logarithmic
term is determined by the variation of Fermi surface along the direction parallel the edge.
The scaling laws are independent of the statistics of the microscopical constituents. The
introduction of spin-orbit coupling will cause a gap in the bulk spectrum and a decrease
of entanglement entropy.
Since there is an equivalence between the entanglement entropy and particle number
fluctuations in free-fermion systems [56, 57, 58], it is very promising to measure the
entanglement entropy in the graphene-like systems. Especially experimentally the edges
can be better fabricated in the artificial honeycomb than those of natural electronic
graphene, which tend to be very irregular and contaminated with adsorbates. The
flat edge modes can be observed in the zero-bias conductance measurements [80]. Our
results can be extended to other free fermionic systems, i.e., band topological insulators
and superconductors, as well as weakly interacting systems [81].
Scaling of entanglement entropy in honeycomb lattice on a torus 15

Acknowledgments
W-L.Y. thanks Peter Horsch and Andreas Schnyder for insightful discussions. This
work was supported by the National Natural Science Foundation of China (NSFC)
under Grant No. 11004144.

References

[1] Vidal G 2008 Phys. Rev. Lett. 101, 110501

[2] Verstraete F, Cirac J and Murg V 2008 Adv. Phys. 57, 143
[3] Amico L, Fazio R, Osterloh A and Vedral V 2008 Rev. Mod. Phys. 80, 517
[4] Eisert J, Cramer M and Plenio M B 2010 Rev. Mod. Phys. 82, 277
[5] Osborne T J and Nielsen M A 2002 Phys. Rev. A 66, 032110
[6] Plenio M B, Eisert J, Dreißig J and Cramer M 2005 Phys. Rev. Lett. 94, 060503
[7] Wu L A, Sarandy M S and Lidar D A 2004 Phys. Rev. Lett. 93, 250404
[8] Barthel T, Dusue S and Vidal J 2006 Phys. Rev. Lett. 97, 220402
[9] Haque M, Oleksandr Zozulya O and Schoutens K 2007 Phys. Rev. Lett. 98, 060401
[10] You W L and Lu W L 2011 Int. J. Mod. Phys. B 25, 231-241
[11] You W L, Oleś A M and Horsch P 2012 Phys. Rev. B 86, 094412
[12] Refael G and Moore J E 2004 Phys. Rev. Lett. 93, 260602
[13] Laflorencie N 2005 Phys. Rev. B 72, 140408
[14] Berkovits R 2012 Phys. Rev. Lett. 108, 176803
[15] Zhang Y, Grover T and Vishwanath A 2012 Phys. Rev. Lett. 107, 067202
[16] Kitaev A and Preskill J 2006 Phys. Rev. Lett. 96, 110404
[17] Levin M and Wen X G 2006 Phys. Rev. Lett. 96, 110405
[18] Bombelli L, Koul R K, Lee J and Sorkin R D 1986 Phys. Rev. D 34, 373
[19] Srednicki M 1993 Phys. Rev. Lett. 71, 666
[20] Hawking S, Maldacena J M and Strominger A 2001 DeSitter entropy, quantum entanglement and
AdS/CFT J. High Energy Phys. JHEP05(2001)001
[21] Holzhey C, Larsen F and Wilczek F 1994 Nucl. Phys. B 424, 443-467
[22] Calabrese P and Cardy J 2004 Entanglement entropy and quantum field theory J. Stat. Mech.
0406 P06002 (arXiv:hep-th/0405152)
[23] Vidal G, Latorre J, Rico E and Kitaev A 2003 Phys. Rev. Lett. 90, 227902
[24] Kallin A B, González I, Hastings M B and Melko R G 2009 Phys. Rev. Lett. 103, 117203
[25] Jiang H C, Wang Z and Balents L 2012 Nature Physics 8, 902-905
[26] Polini M, Guinea F,Lewenstein M, Manoharan H C and Pellegrini V 2013 Nature Nanotechnology
8, 625-633
[27] Rechtsman M C, Plotnik Y, Song D, Heinrich M, Zeuner J M, Nolte S, Malkova N, Xu J, Szameit A,
Chen Z and Segev M 2012 Observation of novel edge states in photonic graphene arXiv:1210.5361
[28] Gomes K K, Mar W, Ko W, Guinea F and Manoharan H C 2012 Nature 483, 306
[29] Tarruell L, Greif D, Uehlinger T, Jotzu G and Esslinger T 2012 Nature 483, 302
[30] Lim L K, Fuchs J N and Montambaux G 2012 Phys. Rev. Lett. 108, 175303
[31] Montambaux G, Piéchon F, Fuchs J N and Goerbig M O 2009 Eur. Phys. J. B 72, 509
[32] Montambaux G, Piéchon F, Fuchs J N and Goerbig M O 2009 Phys. Rev. B 80, 153412
[33] Yamaji Y, Misawa T and Imada M 2006 J. Phys. Soc. Jpn. 75 094719
[34] Rodney M, Song H F, Lee S S, Hur K L and Sørensen E S 2013 Phys. Rev. B 87, 115132
[35] Ryu S and Hatsugai Y 2002 Phys. Rev. Lett. 89, 077002
[36] Hatsugai Y 2009 Solid State Comm. 149, 1061
[37] S. Ryu and Y. Hatsugai 2006 Phys. Rev. B 73, 245115
[38] Ostrovsky P M, Gornyi I V and Mirlin A D 2008 Phys. Rev. B 77, 195430
Scaling of entanglement entropy in honeycomb lattice on a torus 16

[39] Wunsch B, Guinea F and Sols F 2008 Dirac-point engineering and topological phase transitions in
honeycomb optical lattices New J. Phys. 10, 103027
[40] Park C H and Marzari N 2001 Phys. Rev. B 84, 205440
[41] Delplace P, Ullmo D an Montambaux G 2011 Phys. Rev. B 84, 195452
[42] Peschel I 2003 J. Phys. A: Math. Gen. 36, L205
[43] Peschel I and Eisler V 2009 J. Phys. A: Math. Theor. 42, 504003
[44] Peschel I and Eisler V 2012 Braz. J. Phys. 42,74
[45] Chung M C and Peschel I 2001 Phys. Rev. B 64, 064412
[46] Cheong S A and Henley C L 2004 Phys. Rev. B 69, 075111
[47] Li L and Haldane F D M 2008 Phys. Rev. Lett. 101, 010504
[48] Fidkowski L 2010 Phys. Rev. Lett. 104, 130502
[49] Läuchli A M, Bergholtz E J, Suorsa J and Haque M 2010 Phys. Rev. Lett. 104, 156404
[50] Thomale R, Sterdyniak A, Regnault N and Bernevig B A 2010 Phys. Rev. Lett. 104, 180502
[51] Chung M C, Jhu Y H, Chen P and Yip S 2011 EPL 95, 27003
[52] Turner A M, Zhang Y, and Vishwanath A 2010 Phys. Rev. B 82, 241102
[53] Arikawa M, Aoki H and Hatsugai Y 2011 AIP Conf. Proc. 1399, 823
[54] Kohmoto M and Hasegawa Y 2007 Phys. Rev. B 76, 205402
[55] Barthel T, Chung M C and Schollwöck U 2006 Phys. Rev. A 74, 022329
[56] I. Klich and L. Levitov 2009 Phys. Rev. Lett. 102, 100502
[57] Song H F, Flindt C, Rachel S, Klich I and Hur K L 2011 Phys. Rev. B 83, 161408
[58] Song H F, Rachel S, Flindt C, Klich I, Laflorencie N and Hur K L 2012 Phys. Rev. B 85, 035409
[59] Wolf M M, Verstraete F, Hastings M B and Cirac J I 2008 Phys. Rev. Lett. 100, 070502
[60] Levine G C and Miller D J 2008 Phys. Rev. B 77, 205119
[61] Ding L, Bray-Ali N, Yu R and Haas S 2008 Phys. Rev. Lett. 100, 215701
[62] Yu R, Saleur H and Stephan Haas S 2008 Phys. Rev. B 77, 140402
[63] Calabrese P, Campostrini M, Essler F and Nienhuis B 2012 Phys. Rev. Lett. 104, 095701
[64] Calabrese P, Mintchev M and Vicari E 2011 Phys. Rev. Lett. 107, 020601
[65] Brian Swingle 2010 Phys. Rev. Lett. 105, 050502
[66] Wolf M M 2006 Phys. Rev. Lett. 96, 010404
[67] Li W, Ding L, Yu R, Roscilde T and Haas S 2006 Phys. Rev. B 74, 073103
[68] Widom H 1982 Operator Theory: Adv. Appl. 4, 477
[69] Gioev D and Klich I 2006 Phys. Rev. Lett. 96, 100503
[70] Cramer M, Eisert J and Plenio M B 2007 Phys. Rev. Lett. 98, 220603
[71] Calabrese P, Mintchev M and Vicari E 2012 EPL 97, 20009
[72] Kane C L and Mele E J 2005 Phys. Rev. Lett. 95, 146802
[73] Kane C L and Mele E J 2005 Phys. Rev. Lett. 95, 226801
[74] Schmidt M J and Loss D 2010 Phys. Rev. B 81, 165439
[75] Vogt P, Padova P D, Quaresima C, Avila J, Frantzeskakis E, Asensio M C, Resta A, Ealet B and
LeLay G 2012 Phys. Rev. Lett. 108, 155501
[76] Fleurence A, Friedlein R, Ozaki T, Kawai H, Wang Y and Yamada-Takamura Y 2012 Phys. Rev.
Lett. 108, 245501
[77] Chen L, Liu C C, Feng B, He X, Cheng P, Ding Z, Meng S, Yao Y and Wu K 2012 Phys. Rev.
Lett. 109, 056804
[78] Feng B, Ding Z, Meng S, Yao Y, He X, Cheng P, Chen L and Wu K 2012 Nano Lett. 12, 3507-3511
[79] Ezawa M and Nagaosa N 2013 Phys. Rev. B 88, 121401
[80] Hu C R 1994 Phys. Rev. Lett. 72, 1526
[81] Ding W, Seidel A and Yang K 2012 Phys. Rev. X 2, 011012