PHYSICAL REVIEW B 93, 035452 (2016)

Low-energy theory for the graphene twist bilayer

D. Weckbecker,1,* S. Shallcross,1,† M. Fleischmann,1 N. Ray,1 S. Sharma,2,3 and O. Pankratov1
1
Lehrstuhl für Theoretische Festkörperphysik, Staudtstrasse 7-B2, 91058 Erlangen, Germany
2
Max-Planck-Institut für Mikrostrukturphysik, Weinberg 2, 06120 Halle, Germany
3
Department of Physics, Indian Institute of Technology, Roorkee, 247667 Uttarkhand, India
(Received 8 September 2015; published 27 January 2016)

The graphene twist bilayer represents the prototypical system for investigating the stacking degree of freedom
in few-layer graphenes. The electronic structure of this system changes qualitatively as a function of angle,
from a large-angle limit in which the two layers are essentially decoupled—with the exception of a 28-atom
commensuration unit cell for which the layers are coupled on an energy scale of ≈8 meV—to a small-angle
strong-coupling limit. Despite sustained investigation, a fully satisfactory theory of the twist bilayer remains
elusive. The outstanding problems are (i) to find a theoretically unified description of the large- and small-angle
limits, and (ii) to demonstrate agreement between the low-energy effective Hamiltonian and, for instance, ab
initio or tight-binding calculations. In this article, we develop a low-energy theory that in the large-angle limit
reproduces the symmetry-derived Hamiltonians of Mele [Phys. Rev. B 81, 161405 (2010)], and in the small-angle
limit shows almost perfect agreement with tight-binding calculations. The small-angle effective Hamiltonian is
that of Bistritzer and MacDonald [Proc. Natl. Acad. Sci. (U.S.A.) 108, 12233 (2011)], but with the momentum
scale K, the difference of the momenta of the unrotated and rotated special points, replaced by a coupling
momentum scale g (c) = √8π3a sin θ2 . Using this small-angle Hamiltonian, we are able to determine the complete
behavior as a function of angle, finding a complex small-angle clustering of van Hove singularities in the density
of states (DOS) that after a “zero-mode” peak regime between 0.90◦ < θ < 0.15◦ limits θ < 0.05◦ to a DOS that
is essentially that of a superposition DOS of all bilayer stacking possibilities. In this regime, the Dirac spectrum is
entirely destroyed by hybridization for −0.25 < E < 0.25 eV with an average band velocity ≈0.3vF(SLG) (where
SLG denotes single-layer graphene). We study the fermiology of the twist bilayer in this limit, finding remarkably
structured constant energy surfaces with multiple Lifshitz transitions between K- and -centered Fermi sheets
and a rich pseudospin texture.

DOI: 10.1103/PhysRevB.93.035452

I. INTRODUCTION intermediate-angle regime in which van Hove singularities

move toward the Dirac point renormalizing the Fermi velocity,
The rich physics associated with the interlayer degree of
and a small-angle regime in which van Hove singularities are
freedom in few-layer graphenes is manifest perhaps most
dense in energy near the Dirac point, and the two layers are
strikingly in the graphene twist bilayer system. This system,
strongly coupled. In real space, this small-angle regime is
consisting of two mutually rotated layers of graphene, exhibits
associated with the emergence of a geometric moiré lattice,
a wide range of electronic effects as a function of the twist
and electron localization on the “AA spots” of this moiré.
angle, and it has attracted sustained attention from both
The large-angle regime has been studied ab initio with the
theory and experiment [1–24]. The electronic structure of
finding that the Dirac point degeneracy generally found at large
the twist bilayer is both diverse at the single-particle level
angles is, for the smallest 28-atom unit cell commensuration,
and challenging to the naive application of band theory. The
lifted on an energy scale of ≈8 meV [18–21]. Within this
failure of a straightforward band theory approach lies in the
window, the band structure shows either a sublattice exchange
fact that the twist angle of the bilayer, which obviously fixes all
(SE) even (C3 ) or a Bernal bilayer (C6 ) form depending on the
physical properties, is insufficient to define the lattice vectors
point-group symmetry of the bilayer, in agreement with a large-
of the system. The classical picture of backfolded bands in a
angle continuum theory of the twist bilayer [15]. The recent
twist bilayer Brillouin zone hybridizing to open a gap leading
work of Kazuyuki et al. has taken the ab initio approach to the
to van Hove singularities in the density of states evidently
beginning of the strongly coupled regime [25], where they find
cannot hold: there is no unique bilayer Brillouin zone. In fact,
the first “magic angle” at which the Fermi velocity vanishes
it has been shown that the system is endowed with an emergent
to be 1.1◦ . However, as the minimum number of atoms to
coupling momentum scale that depends only on the twist angle
realize a supercell is N = 1/2 sin2 θ with θ = cos−1 (3q 2 − 1)/
of the bilayer. Real-space periodicity is thus a physically
(3q 2 + 1), q ∈ N, q odd, the ab initio approach cannot, with
irrelevant property of the twist bilayer for the electronic
the present state of the art, penetrate the small-angle regime.
structure [5]. The electronic structure itself is enormously
Semiempirical tight-binding calculations have been used to
rich and can be characterized by three regimes: a large-angle
explore the small-angle regime [13,16,26] finding a complex
regime in which the two layers are essentially decoupled, an
clustering of moiré bands at the Dirac point. Finite flakes of the
graphene twist bilayer have also been calculated in the tight-
binding scheme, where it has been shown that, remarkably,
*
[email protected] a single moiré unit cell in a finite geometry is sufficient to
† reproduce the density of states and electron density of the
[email protected]

2469-9950/2016/93(3)/035452(15) 035452-1 ©2016 American Physical Society

D. WECKBECKER et al. PHYSICAL REVIEW B 93, 035452 (2016)

extended twist bilayer [2]. For rotation angles smaller than The small- and large-angle limits are thus conceptually unified
about 0.8◦ , however, even the tight-binding approach reaches by the notion of multiple scattering through high-energy
computational limits. To explore the physics below this angle states, with the principal difference between the large- and
a tight-binding scheme with a single-layer graphene (SLG) small-angle limits being simply a vast increase in the number
basis set [5] and effective Hamiltonians obtained from various of scattering paths required in the small-angle limit, which is
expansions of the tight-binding method [8,22,27–29] have encoded in a real-space moiré field S(r).
been developed. Of these effective Hamiltonian approaches, The remainder of this article is structured as follows. In
perhaps the most elegant is that due to Bistritzer et al. [8], Sec. II we rederive the momentum scale g (c) as a selection
which consists of two Dirac-Weyl operators coupled by a real- rule for the coupling of single-layer Bloch states, and in
space moiré field S(r), which√ possesses the peculiar feature of the subsequent two sections we develop fully electronic
having a periodicity exactly 3 larger than the moiré length low-energy continuum approximations based on this selection
D = 2 sin1θ/2 (we present all real-space quantities in units of the rule for both the large- (Sec. III) and small- (Sec. IV) angle
SLG lattice constant a, and reciprocal space quantities in units regimes. We then explore in detail the single-particle properties
of 2π/a). Calculations based on this Hamiltonian revealed a in the small-angle regime: in Sec. VI we calculate the density of
series of angles at which the band velocity at the Dirac point states and electron density in the very small-angle limit, finding
vanished, the so called magic angles. The SLG basis approach remarkable accuracy between the continuum and tight-binding
of Shallcross et al. revealed both the existence of the coupling approaches, and in Sec. VII we consider how the band velocity
momentum scale g (c) = √43 sin θ2 as well as the emergence of behaves as a function of twist angle and energy (of which the
an approximately self-similar peak in the density of states at magic angles represent a special E = 0 case), and we also
the smallest angles that could be reached (0.46◦ ). describe the rich Fermi surface topology as a function of both
While the state of understanding of a twist bilayer band rotation angle as well as doping. Thereafter, we summarize
structure would appear to be impressively complete, a number our results and discuss possible future research in Sec. VIII.
of questions remain. First, the low-energy Dirac-like Hamil-
tonian of Bistritzer et al. [8] has been reported to require II. SELECTION RULE FOR SINGLE-LAYER BLOCH
rescaling [26] in order to reproduce band structures and ve- STATES
locity renormalization found from tight-binding calculations,
In this section, we will recast the two-center tight-binding
throwing into doubt the validity of the low-energy approach.
method in a way that will subsequently provide a transparent
Furthermore, the continuum theories presented in the literature
route to a low-energy description of the twist bilayer, for
do not clearly relate the origin of the low-energy physics of
both large and small twist angles. These two limits will be
twisted bilayer graphene to the crucial role the momentum
conceptually unified by possessing the same fundamental
scale g (c) plays in the interlayer coupling. It also remains
form for the way in which Bloch functions (or single-layer
unclear what connection there is, if any, between the small-
eigenstates) from each layer are coupled by the interaction.
angle [8,22] and large-angle continuum models [29]. These
We consider the Bloch functions
questions are not merely academic: the twist bilayer continues
to be intensely investigated, with more recent work looking at
 (n) 
φ 1  ikn ·(Rn +ν (n)  
α ) R + ν (n) ,
αkn = √ e n α (1)
many-body effects [30–33] or more complex manifestations of N Rn
the twist geometry [34]. It is important, therefore, to clarify the
“fruit fly” model of twist stacking physics: the single-particle where Rn runs over all lattice sites in the √ nth layer, ν (n)
α
physics of the twist bilayer. represents the layer n basis vectors, and 1/ N denotes a
To clarify this situation, in this paper we derive a low-energy normalization factor. We will first consider interlayer matrix
continuum Hamiltonian for twisted bilayer graphene, making elements in this basis. These are given by
direct use of the g (c) coupling momentum scale, and we show  (1)   (2)  1  −ik1 ·(R1 +ν (1)
φαk1 H φβk2 =
(2)
e α ) e ik2 ·(R2 +ν β )
that it agrees perfectly with tight-binding calculations in the
NRR
intermediate and small-angle limit. The form of this low- 1 2

energy Hamiltonian is identical to that derived by Bistritzer    (2) 

× R1 + ν (1)  
et al., but with the momentum scale K = 43 sin θ2 replaced α H R2 + ν β . (2)
 
by g (c) . Interestingly, the resulting moiré field S(r) now has t(|R2 +ν (2) (1)
β −R1 −ν α |)
exactly the periodicity of the real-space moiré lattice. The
numerical efficiency of the low-energy approach allows us Writing the hopping function t in this expression in terms
to completely characterize constant energy surfaces, band of its Fourier transform t(r) = 1/(2π )2 dq e−ir·q t(q) and
velocities, and density of states in the small-angle limit for performing the resulting phase sums with the help of the
0.03◦ < θ < 1.2◦ . Poisson sum formula, we find a reciprocal space representation
Furthermore, we derive from the tight-binding method a of this matrix element given by
 (1)   (2)  
large-angle continuum model in which the interlayer inter- φαk1 H φβk2 = 2
t(k1 + G̃1i )[Mi ]αβ (3)
action is expressed as a multiple scattering series through (2π ) i
high-energy states. This reproduces the large-angle symmetry-
with the area of the single-layer graphene Brillouin zone
derived continuum Hamiltonians [15], but with all parameters
and where the Mi matrices are defined as
fixed by the underlying electronic theory, and again it shows (2)
[Mi ]αβ = ei G̃1i ·ν α e−iRG̃2i ·ν β ,
(1)
near perfect agreement with full tight-binding calculations. (4)

035452-2
LOW-ENERGY THEORY FOR THE GRAPHENE TWIST BILAYER PHYSICAL REVIEW B 93, 035452 (2016)

with G̃1i and RG̃2i representing the ith solution to the TABLE I. The matrices Mi that appear in the formula for the
interlayer momentum-conservation equation interlayer interaction, Eq. (3), in the absence of umklapp scattering
processes. Shown are the resulting matrices for three different types
G1 = RG2 + k2 − k1 , (5) of initial stacking of the bilayer prior to rotation; AA, AB, and AC
stacking.
and where G1 and RG2 represent reciprocal-lattice vectors in
Stacking type G(c) (c)
1 = (0,0) G2 = (−1, +
√1 ) G(c)
3 = (−1, −
√1 )
layers 1 and 2, respectively. It is this equation, and not those 3 3

that describe the lattice commensuration, that determines the 1 1 1 e−i2π/3 1 e+i2π/3
interlayer coupling. To see this, we note that if the momentum AA/SLG
1 1 e+i2π/3 1 e−i2π/3 1
of the Bloch state k1 lies at the high-symmetry K point K1 =
1 1 1 e+i2π/3 1 e−i2π/3
(2/3,0) of the graphene Brillouin zone, then if the second- +i2π/3
AB 1 1 e e−i2π/3 e −i2π/3
e+i2π/3
layer Bloch state has momenta k2 = k1 , we can set G1 =
−i2π/3 +i2π/3
RG2 = 0 to solve Eq. (5) and yield a coupling constant of 1 1 e e e+i2π/3 e−i2π/3
≈ t(|K1 |) = t(K) in the matrix element sum Eq. (3). For k2 − AC 1 1 e+i2π/3 1 e−i2π/3 1
k1 = G(c) (c) (c)
i − RGi , with Gi a reciprocal vector connecting
K1 to one of the other two equivalent K points K2 and K3 of
the Brillouin zone, i.e., K2 = K1 + G(c) 2 and K3 = K1 + G3 ,
(c) matrix elements are thus given by
we will also evidently find a coupling constant of ≈ t(K). All    
(n)   (n) 
other reciprocal vectors G will yield a momentum K1 + G that φαk H φβk2 = δk1 k2 t(k1 + G1i )[Mi ]αβ , (10)
is outside the first Brillouin zone and hence a much reduced
1
(2π )2 i
coupling constant t(K1 + G)  t(K). Thus only layer 1 and
where the sum is now over all first-layer reciprocal-lattice
layer 2 Bloch states that have momenta satisfying
vectors G1 . As before, for Bloch state momenta k1 close to
a high-symmetry K point it will be sufficient to employ the
k2 − k1 = G(c) (c)
i − RGi “first star” approximation and include only those G1 that yield
= g(c) (6) t(k1 + G1 ) ≈ t(K). The Mi matrices are then given by the
i
SLG entry of Table I.
will yield a significant order ≈t(K) contribution to the matrix
element sum and hence a non-negligible matrix element III. LOW-ENERGY THEORY AT LARGE TWIST ANGLES
(1) (2)
φαk |H |φβk . The vectors g(c)
i may easily be calculated from
1
Eq. (6) yielding
2
As a first application of the Bloch state selection rule
described in the previous section, we consider the electronic
spectrum of twist bilayer systems with large rotation angles.
g(c)
1 = 0, (7) As is by now well established, the spectrum of a large-
angle twist bilayer (15◦ < θ < 45◦ ) is essentially that of two
cos(π/3 + θ/2)
g(c)
2 = g
(c)
, (8) degenerate single-layer graphene spectra. This degeneracy is,
sin(π/3 + θ/2) however, broken at the Dirac point such that for the smallest
cos(π/3 − θ/2) commensuration unit cells of 28-carbon atoms (corresponding
g(c)
3 = −g
(c)
, (9)
− sin(π/3 − θ/2) to rotation angles of θ = 21.79◦ and 38.21◦ ), the splitting at
the Dirac point is of the order of 8 meV. The magnitude of
where the length scale√ of the coupling in momentum space this splitting decreases very rapidly with increasing supercell
is given by g (c) = 4/ 3 sin θ2 . In contrast to the momentum size, and it is only for the 28-atom unit cells that it can be
scale provided by the reciprocal-lattice vectors of the twist considered a potentially measurable effect. The band structure
bilayer, which cannot be fixed by the rotation angle, Eqs. near the Dirac point is also interesting as one finds a SE even
(7)–(9) depend only on the twist angle of the bilayer. This is structure for the case in which the twist bilayer possesses C3
exactly the momentum scale derived in Ref. [5] on symmetry symmetry, while an AB bilayer type band structure is found for
grounds, and it represents the fundamental momentum scale on the higher-symmetry C6 case. In this section, we will develop
which Bloch functions (or single-layer basis functions derived a general low-energy theory for the large-angle regime. This
from them) interact through the interlayer interaction. The will yield low-energy effective Hamiltonians that agree with
Mi matrices in Eq. (3) take on a particularly simple form for the form presented by Mele in Ref. [15], where they were
the G(c)
i , and they are given in Table I for the three principal
derived on general symmetry grounds, but with all parameters
stacking types of AA, AB, and AC stacking. It should be of the theory clearly derived from the underlying tight-binding
noted that Eqs. (7)–(9) are valid only near the high-symmetry formalism.
K point given by K = (2/3,0); at other high-symmetry K A complete basis for the solution of the twist bilayer
points, Eqs. (7)–(9) are replaced by g(c) (c)
i = U gi with U the
electronic structure at k0 [a k vector in the twist bilayer first
point-group operation that relates the two high-symmetry K Brillouin zone (BZ)] is given by the union of the set of first-
points. layer Bloch functions with momenta k1 = k0 + i1 g(c) (c)
1 + i2 g2
The intralayer matrix elements have a much simpler that fall within the first BZ of the unrotated reciprocal
structure due to the fact that the equivalent of the interlayer space lattice, and the set of second-layer Bloch functions
momentum conservation, Eq. (5), is simply k1 = k2 . Such with momenta k2 = k0 + i1 g(c) (c)
1 + i2 g2 that fall within the

035452-3
D. WECKBECKER et al. PHYSICAL REVIEW B 93, 035452 (2016)

with | L  = P0 | . This equation is exact, and the price

to be paid for this is that it is no longer linear in E but
instead involves the Green’s function GH (E) = (E − HH )−1 .
Rather than following the usual approach of linearizing Eq.
(12) in E, we instead note that GH (E) may be expressed
as GH (E) = (E − HH(0) − VH )−1 , where HH(0) is the intralayer
part of the high-energy block (and hence is diagonal in
k-vector indices) and VH represents the interlayer coupling
part of the high-energy block. Introducing the bare high-energy
(0) −1
Green’s function G(0)
H (E) = (E − HH ) , which is simply the
high-energy Green’s function with the interlayer interaction
“switched off,” we may write the latter object as a Dyson
series,

GH (E) = G(0) (0) (0)

H (E) + GH (E)VH GH (E) + · · · , (13)
FIG. 1. Backfolding of single-layer graphene states to the bilayer
Brillouin zone, where the k vector in the bilayer Brillouin zone to and insertion of this into Eq. (12) yields
which basis states fold back has been set to KB . Also shown are the   †
high-energy states that couple to K by the coupling momenta g(c) i .
E − HL(0) − Vc G(0) (0) (0)
H (E) + GH (E)VH GH (E) + · · · Vc
For two of these (g(c) (c)
1 and g3 ), these high-energy states lie outside × P0 |  = 0. (14)
the second-layer (rotated) Brillouin zone, and umklapp vectors are
required to backfold them. The three high-energy k vectors involved This is now in a convenient form in which to implement
in the scattering paths used in constructing the large-angle continuum
a low-energy approximation by setting G(0) (0)
H (E) ≈ GH (0) =
Hamiltonians are labeled as “H .” (0) −1
−[HH ] , which then immediately yields an eigenvalue
problem for the low-energy subspace,
second-layer BZ. Each k vector contributes two Bloch func-
 −1 −1 −1  
tions to the basis due to the two-atom primitive cell. An E−HL(0) +Vc HH(0) − HH(0) VH HH(0) + · · · Vc †
illustration of the k vectors contributing to the basis may
be seen in Fig. 1. In this basis, we write the tight-binding × P0 |  = 0, (15)
Hamiltonian as
which is evidently expressed as a multiple scattering expansion
HL Vc (MSE) in terms of interlayer scattering matrices VH and
H = † , (11)
Vc HH low-energy propagators [HH(0) ]−1 , with the multiplication
(n) (m) together of these 2NH × 2NH dimension matrices generating
where a matrix element [H ]αk1 n,βk2 m = φαk 1
|H |φβk 2
. We all possible scattering paths between the NH high-energy k
organize the Hamiltonian such that HL consists of the basis vectors. Premultiplication by the 4 × 2NH dimension Vc , and
elements arising from the low-energy basis functions close to postmultiplication by the 2NH × 4 dimension Vc † , ensures
the K points of each layer (these are K and RK in Fig. 1); there that these paths begin and terminate at one of the low-energy
will be two such basis functions from each layer and thus HL k vectors.
is a 4 × 4 matrix. The HH block consists of matrix elements For an actual calculation, we restrict the high-energy basis
involving the remaining high-energy basis functions, and it is a to the three degenerate lowest-energy states (the three high-
2NH × 2NH matrix with NH the number of high-energy basis energy k vectors are labeled “H ” in Fig. 1) and include only
functions. Finally, Vc describes the coupling of the low- and the first nonzero order in the MSE. The layer off-diagonal
high-energy subspace, and evidently involves matrix elements block of the effective Hamiltonian is thus given by
between low-energy and high-energy basis functions. The key
approximation we now introduce is to retain a dependence  (0) −1  −1 
S= [Vc ]Ki1 HH [V ]
i1 H i1 i2
HH(0) [V † ]
i2 c i2 RK
,
on the k0 vector (in the bilayer Brillouin zone) at which the
i1 i2
electronic structure is evaluated only in the low-energy block
of the Hamiltonian. We fix the high energy and coupling blocks (16)
to the form they take when k0 = KB , the high-symmetry K
where i1 and i2 label the k vectors involved in the scattering
vector of the bilayer Brillouin zone. This approximation is
path, and all objects are now of dimension 2 × 2 (the
justified and, as we will see, it works very effectively, due
pseudospin degree of freedom). Inspection of Fig. 1 reveals
to the large energy separation between the high-energy and
that there are three scattering paths that connect K and RK
low-energy blocks.
and pass through only one of the high-energy k vectors labeled
Let us consider projection operators onto the low-energy
“H .” The [Vc ]Ki1 and [VH ]i1 i2 are scattering operators similar
subspace, P0 = diag(1,1,1,1,0, . . .), and high-energy sub-
to those of Table I, but differing by the inclusion of umklapp
space, P1 = diag(0,0,0,0,1, . . .). A standard down-folding
vectors in the phases, which occur as the high-energy states that
procedure then allows us to recast the eigenvalue equation
couple to K lie outside the first Brillouin zone, and so umklapp
(E − H )|  = 0 in the low-energy subspace as
vectors are needed to backfold these states. The zero-energy
[E − HL − Vc (E − HH )−1 Vc† ]| L =0 (12) propagator is identical for all three high-energy states and has

035452-4
LOW-ENERGY THEORY FOR THE GRAPHENE TWIST BILAYER PHYSICAL REVIEW B 93, 035452 (2016)

the form
dh eh
Hh = , (17)
eh∗ dh
which describes a general high-energy block. For the case of an
initial AA stacking and a rotation of θ = 21.78 corresponding
to the (p,q) = (1,3) commensuration with 28 atoms in the unit
cell, we find for the effective Hamiltonian
H (k)
⎛ ⎞
0 vF keiθk w 0
⎜ ⎟
⎜vF ke−iθk 0 0 0 ⎟
=⎜
⎜
⎟,
⎝ w 0 0 vF ke−i(θk −θ) ⎟
⎠
0 0 vF kei(θk −θ) 0
(18) FIG. 2. Band structure of the two smallest commensurate twist
bilayer structures: θ = 38.21◦ (top panel) and θ = 21.79◦ (bottom
where w is an effective coupling amplitude given by panel). For each of these twist angles, the supercell consists of 28
3tK2 th carbon atoms. The full dark (black) lines display the full two-center
−iπ/3
w= 2 (Re[e eh ] − dh )2 , (19) tight-binding approximation, and the light shaded (red) broken line
|eh |2 − dh2 displays that of the low-energy Hamiltonians Eq. (20) (top panel) and
Eq. (18) (bottom panel).
in which th is the Fourier transform of the interlayer hopping
evaluated at the high-energy k vector, and tK is the interlayer
hopping evaluated at the high-symmetry K point. In a similar those given by the rotation angles θ = 38.21◦ and 21.79◦ ,
way for the case of a 38.21◦ rotation [corresponding to a and as numerous calculations attest it is negligible for other
(p,q) = (3,5) commensuration with again 28 atoms in the angles. In the small-angle limit, therefore, the existence of a
unit cell], we find second Dirac cone that folds back to KB can be neglected as
we have a twofold degeneracy of all bands. For this reason,
H (k) we need to include only two Dirac cones in the theory, one
⎛ ⎞
0 vF keiθk w 0 from each layer. This simplifies the problem considerably, i.e.,
⎜ ⎟ rather than the “four-cone” problem at large angles, we have a
⎜vF ke−iθk 0 0 weiπ/3 ⎟
=⎜
⎜
⎟. simpler “two-cone” problem. On the other hand, the problem
⎝ w 0 0 vF ke−i(θk −θ) ⎟
⎠ is made more complicated (and the electronic structure made
−iπ/3
0 we vF kei(θk −θ) 0 much richer) due to the fact that the momentum scale that
governs the selection rule for Bloch states g (c) is vanishing in
(20) the small-angle limit. This entails the coupling together of a
Equations (18) and (20) are just the two “SE odd” and “SE great many more states, whereas in the large-angle limit the
even” low-energy Hamiltonians obtained by Mele [15] on electronic structure could be understood as a coupling (via
general symmetry grounds for the large-angle twist bilayer, multiple scattering expansions) of a relatively small number
but with now all parameters fixed by the underlying electronic of states. For this reason, we require an approach that includes
structure. The band structure near the K point that results from all scattering paths and, hence, all matrix elements of the
the effective Hamiltonians Eqs. (18) and (20) is presented interlayer interaction.
in Fig. 2 as the light dashed line, along with the full Given that the two Dirac cones that we are interested in are
line, which is the exact tight-binding result using the same now separated by K = |RK − K| = 43 sin θ2 , which becomes
two-center approximation used to derive tK , th , dh , and eh . small in the small-angle limit, and as we are only interested
(Numerical details of our tight-binding method can be found in a low-energy theory, we may consider a basis of Bloch
in Sec. V.) As can be seen, the agreement is rather good, functions that reside in some momentum sphere around K for
with only small deviations being seen between the continuum layer-1 functions and around RK for layer-2 functions. In this
approximations given by Eqs. (18) and (20) and the full case, all Bloch functions will have a momentum not too far
tight-binding calculations. from K, and we may approximate the Fourier transform of
the two-center hopping in Eq. (3) by t(k1 + G̃1i ) ≈ t(K). The
interlayer matrix elements given by Eq. (3) may now be written
IV. LOW-ENERGY THEORY AT SMALL TWIST ANGLES as
We now consider the case for which the rotation angle  (1)   (2)  tK 
of the bilayer is small, θ < 15◦ ; this regime is characterized φαk1 H φβk2 = [Mi ]αβ δg(c) =k2 −k1 , (21)
(2π )2 i i

by the emergence of a geometric moiré pattern and signif-

icant disruption of the Dirac spectrum as θ → 0. As has where we have included in the matrix element explicitly the
been mentioned, this breaking of the fourfold degeneracy is selection rule for single-layer Bloch states via the Kronecker
significant only for two specific commensuration cells, i.e., δ function. With this approximation, we can construct a

035452-5
D. WECKBECKER et al. PHYSICAL REVIEW B 93, 035452 (2016)

continuum theory that reproduces all such matrix elements V. COMPUTATIONAL DETAILS
exactly as follows. We define the plane-wave spinor state from
A. Tight-binding method
layer n as
We use a simple t(r) = A exp (−Br 2 ) for the two-center
 (n)  1 hopping integrals with the constants A and B depending on
φ ikn ·r
σn kn = √ |σn e , (22) whether the hopping is intra- or interlayer. For the former
V −2
case, we use A = 50 eV, B = 0.4373 Å , while for the latter
−2
where the pseudospinor |σn  is either pure pseudospin up, we use A = −8.4542 eV, B = 0.6649 Å . These parameters
|+ = (1,0), or pure pseudospin down, |− = (0,1), in the are obtained by minimizing via a simulated annealing method
layer n portion of the 4-vector (with the other components of the error in eigenvalues (for selected k vectors) between the
the 4-vector zero). Now consider the r-dependent field two-center tight-binding calculation of a number of few-layer
graphene systems, and the corresponding systems calculated
tK  (c) ab initio with the density-functional software package VASP
O(r) = Mi e−igi ·r , (23) [2,35–37]. We found it sufficient to use a database of single and
(2π )2 i
Bernal stacked bilayer graphene along with the two smallest
unit-cell twist bilayer commensurations with rotation angle
where the sum runs over the three coupling vectors of the θ = 21.79◦ and 38.21◦ .
selection rule derived in Sec. II. Matrix elements of the spinor
plane-wave states from each layer with this field are B. Two-center tight-binding calculations
 We use the method described in Ref. [5] in which a
   
dr φσ(1)1 k1 O(r)φσ(2)2 k2 basis of single-layer states is deployed to solve the tight-
binding problem; we refer the reader to that manuscript for

tK  1 (c) methodological details. It should be noted, however, that the
= dr ei(k2 −k1 −gi )·r [Mi ]σ1 σ2 (24) correspondence of the selection rule for Bloch states derived
(2π )2 i V
in Sec. II can be found for single-layer states and leads to a
tK  simple Diophantine problem (Bezóut’s identity) that results
= [Mi ]σ1 σ2 δg(c) =k2 −k1 (25) in an extremely efficient construction of the sparse interlayer
(2π )2 i i

blocks of the twist Hamiltonian. The use of a single-layer basis

further dramatically increases computational efficiency, as to
and thus reproduce exactly the tight-binding matrix elements compute accurately the eigenvalues in some energy window
in Eq. (21). If we define the diagonal 2 × 2 blocks of a 4 × 4 E requires only single-layer states in an energy window of
Hamiltonian to be the Dirac-Weyl operators of each layer, ≈ 1.4E. For a rotation angle of 0.74◦ , (p,q) = (1,89) in the
with the off-diagonal blocks given by O(r) and its Hermitian notation of Shallcross et al. [5,16], we find that to converge all
conjugate, then we may write for the twist Hamiltonian eigenvalues in an energy range from −0.4 to +0.4 eV requires
  a Hamiltonian of dimension 360; the same problem using the
vF σ · p O(r) standard localized orbital basis of the two-center tight-binding
H = . (26)
O(r)† vF σ · R−1 (p − K) leads to a Hamiltonian of dimension 23 764. The resulting
speedup of the former method compared to the latter is of the
order of 105 .
As the corresponding eigenvectors are now 4-vectors given by
the direct product of layer space with pseudospin space, this C. Low-energy calculations
Hamiltonian will reproduce both the layer off-diagonal matrix In a similar way to the method outlined in Ref. [5], we
elements of tight-binding as well as layer diagonal elements (in solve the low-energy Hamiltonian using a basis of Dirac-Weyl
the low-energy Dirac-Weyl approximation). Note that the shift states. That is, the basis consists of all eigenstates of both the
K in the rotated layer Dirac-Weyl operator is required as we unrotated (rotated) Dirac-Weyl operators that have momenta
measure momentum in a global coordinate system centered on k that satisfy |k| < kcut , with kcut the radius of a momentum
the unrotated cone. sphere centered at the Dirac point of the unrotated (rotated)
In contrast to the large-angle case in which the interlayer cone. The vanishing of the momentum scale g (c) as θ → 0
interaction coupled together relatively few Bloch states from obviously implies that for fixed kcut the dimension of the
each layer and led to constant layer off-diagonal matrices, resulting twist bilayer Hamiltonian in this basis will grow
in the small-angle regime the interlayer coupling of many without bound. We find that the dimension is of the order of
Bloch states drives the emergence of the real valued field 102 to converge all eigenvalues in the energy window from
O(r). It should be remarked that Eq. (26) differs from the −0.4 to +0.4 eV at a rotation angle of 0.90◦ , but it increases to
Hamiltonian derived in Ref. [8] only in the length scale g (c) : 105 for 0.03◦ . Given that the matrix elements themselves can be
the analysis in Sec. II yielded for this g (c) = √43 sin θ2 , whereas obtained analytically from Eq. (26), the numerical bottleneck
in derivation of Ref. [8] the coupling scale K = 43 sin θ2 is resides in diagonalizing the resulting Hamiltonian, and toward
found. As we shall demonstrate in subsequent sections, the that end we use the SCALAPACK [38] subroutine PZHEEVR for
use of the correct coupling scale brings the low-energy theory the eigenvalue computations. To access the real-space wave
in the small-angle limit into complete agreement with full functions, i.e., inverse Fourier transforming back to real space,
tight-binding calculations. standard fast Fourier transform (FFT) [39] is deployed.

035452-6
LOW-ENERGY THEORY FOR THE GRAPHENE TWIST BILAYER PHYSICAL REVIEW B 93, 035452 (2016)

FIG. 3. Density of states (DOS) in the small-angle limit. Shown are calculations using the full two-center tight-binding scheme [dark shaded
(red) full lines], along with the corresponding results generated using the low-energy Hamiltonian given by Eq. (26) [light shaded (green) full
lines]. Shown are the DOS for twist angles from 0.90◦ to 0.03◦ ; note that below 0.10◦ computational resources allowed only for calculation
using the low-energy theory. For all cases shown, the DOS of both methods are seen to be in good agreement, particularly near the Dirac
point at E = 0 eV. The light (cyan) shaded regions indicate the energy region of the central peak from which the electron density in the first
column of Fig. 5 is constructed. Similarly, the dark shaded (red) regions illustrate the states summed over in the negative energy neighbor peak,
from which the electron density displayed in the second column of Fig. 5 is constructed. The black dashed line in the θ = 0.03◦ panel is a
superposition DOS constructed from the DOS of all possible interlayer stacking vectors, showing that at this angle the system has effectively
electronically segmented into separate stacking regions.

VI. DENSITY OF STATES AND ELECTRON DENSITY have previously been achieved. Interestingly, we find that this
regime of approximate self-similarity does not persist, and the
To test the veracity of the low-energy theory derived in
Dirac point peak is continuously suppressed as θ decreases
Sec. IV, we first consider the density of states that results
such that by θ = 0.03◦ , the smallest angle that we were capable
from Eq. (26), and, for a number of twist angles, we compare
of calculating, it has entirely vanished.
it to the DOS found using the full two-center tight-binding
This naturally invites the question, “what happens next?”
approach. This is shown in Fig. 3. One finds very good
To answer this question, we compare the θ = 0.03◦ DOS with
quantitative agreement between the low-energy and tight-
a DOS that is the (normalized) sum of the DOS of all possible
binding calculations for all twist angles from 0.90◦ to 0.10◦
stacking arrangements of a graphene bilayer. Numerically, we
shown in Fig. 3. While near the Dirac point the agreement is
consider a 20 × 20 mesh of interlayer shift vectors in the real-
nearly perfect, away from the Dirac point there are deviations
space unit cell of graphene, calculate the bilayer DOS for each
that result from the fact that low-energy theory possesses,
shift vector, and then sum the 400 individual DOS results and
in the density of states, approximate electron-hole symmetry,
normalize. The result of this procedure is shown as the black
and this is not the case for the tight-binding calculation. Note,
broken line in the θ = 0.03◦ panel of Fig. 3. The very close
however, that formal electron-hole symmetry does not exist
agreement between the “summed DOS of all stacking vectors”
for the Hamiltonian Eq. (26). In Ref. [5] it was observed
and the twist bilayer DOS suggests that by this angle the system
that in the small-angle regime there appeared a region of
has effectively electronically segmented into separate stacking
approximate self-similarity in which a peak at the Dirac
regions, and no further significant changes in the DOS will
point persisted at the smallest angles. With the massively
occur for θ < 0.03◦ . On the other hand, it should be stressed
parallel implementation of both the single-layer basis tight-
that this limit is somewhat academic. For a twist angle of 0.03◦ ,
binding and the low-energy approaches, we can now push
the moiré length is already 4698 Å and it may well be that
the DOS calculations to significantly smaller angles than
this is greater than some of the other length scales inevitably

035452-7
D. WECKBECKER et al. PHYSICAL REVIEW B 93, 035452 (2016)

FIG. 4. Band structure of the twist bilayer plotted on a standard M-K- path through the Brillouin zone. Open circles are tight-binding
results, and full lines show the result provided by the low-energy effective Hamiltonian, Eq. (26). The pronounced set of flat “moiré bands”
seen near the Dirac point correspond to the peak in the density of states of the bilayer spectrum shown in Fig. 3.

present: scattering of Bloch states due to the electron-electron Fig. 3, that is, we consider the electron density determined by
interaction, or scattering introduced by disorder. Furthermore, summing | (r)|2 over all the states in the central peak (see the
there may arise structural changes of the moiré in the small- first column of Fig. 5), the satellite peak (the second column
angle limit by the formation of domains separated by screw of Fig. 5), and an energy interval in the linear DOS region not
dislocations. Note, however, that it is not possible to calculate shown in Fig. 3 (the third column of Fig. 5).
the limit for θ → 0 approaching from θ = 0 exactly, as the As may be seen, the electron densities integrated over the
unit-cell size diverges, and regions of AA and AB as well as central peak show a strong localization on the AA spots of the
all other possible shifts exist at each rotation angle. The case of moiré lattice. For a rotation angle of 0.90◦ , the localization
Bernal stacking (θ = 0) may be recovered from Eqs. (26) and effect is most strongly pronounced, while for 0.40◦ and 0.10◦
(23) simply by setting g (c) = √43 sin θ2 in the latter equation. a weaker localization is seen, however even in these cases
The standard zeroth order in momentum approximation to the the density is of the order of 10 times greater in the AA
Bernal bilayer Hamiltonian is then immediately recovered, as regions than in the regions with the lowest electron density.
may be verified from Eq. (23) and Table I: This localization effect is well known and has been previously
⎛ ⎞ reported on a number of occasions; see Refs. [2,5,6]. Turning
1 0 to a comparison of the low-energy approach with that of
⎜ v σ .p t
F ⊥
0 0 ⎟ full tight binding, we see that the electron densities are in
HBernal = ⎜
⎝
⎟
⎠ (27) very good quantitative agreement between the two theories.
1 0
t⊥ vF σ .p It is noteworthy that even the fine structure of the density
0 0
modulations away from the AA spot found in the 0.10◦
(where the out-of-plane hopping constant is denoted by t⊥ ). case are reproduced perfectly by the low-energy approach,
A more detailed comparison between the low-energy theory demonstrating that this approach captures even details of the
and the tight-binding calculation is provided by the band tight-binding calculations.
structure of the twist bilayer, as shown in Fig. 4. The full For electron densities obtained by integrating over the
symbols are the tight-binding results, with the continuous negative energy satellite peak, indicated by the dark shaded
lines the band structure generated by the low-energy theory, region in Fig. 3 and presented in the central column of Fig. 5,
and, as may be seen, for all angles a very good agreement is a similar situation may be observed. For a twist angle of 0.10◦ ,
found. This agreement is particularly good for θ = 0.8◦ and the satellite peak has been suppressed into the prominent side
larger angles, although for the θ = 0.3◦ case the agreement regions around the Dirac point, and even though the DOS
is somewhat worse (although this could not be seen in the is not in perfect agreement at these energies, the intricate
Brillouin zone averaged density of states). One can note the “star-shaped” high-density region on the AA spots is seen
appearance of a detailed set of almost dispersionless “moiré to be in remarkably good agreement between the tight-binding
bands” in the small-angle limit; the band-structure counterpart and low-energy Hamiltonian result. A curious exception to this
of the density of states peak shown in Fig. 3. For all angles, very good agreement at low energies can be seen in the inner-
we find an energy window within which a linear Dirac cone most “hexagon modulation” for the 0.40◦ case, which is rotated
can be found, although already at θ = 0.4◦ this window has by 90◦ between the tight-binding and low-energy calculations.
reduced to 4 meV, below the precision of current experiments. The reason for this curious discrepancy is not known.
To further test the agreement between the low-energy Finally, we show in the third column of Fig. 5 the electron
approach and full tight-binding calculations, we now integrate density in an energy region at which the DOS is linear
the electron densities over the energy intervals highlighted in and indistinguishable from that of single-layer graphene (we

035452-8
LOW-ENERGY THEORY FOR THE GRAPHENE TWIST BILAYER PHYSICAL REVIEW B 93, 035452 (2016)

FIG. 5. Electron density formed by integrating | (r)|2 over all states in the Dirac point peak indicated by the light shaded region in
Fig. 3 (first column), the dark shaded region of the negative energy satellite peak (second column), and over all states in an energy window
−0.5 < E < −0.41 eV for 0.90◦ and 0.40◦ and −0.40 < E < −0.35 eV for 0.10◦ (third column). Shown is the electron density at rotation
angles of 0.90◦ (a), 0.40◦ (b), and 0.10◦ (c) obtained both from tight-binding calculations and the low-energy Hamiltonian of Eq. (26) as
indicated.

035452-9
D. WECKBECKER et al. PHYSICAL REVIEW B 93, 035452 (2016)

FIG. 6. Average band velocity as a function of twist angle of the bilayer and energy [panel (a)]. Valleys of reduced band velocity that arise
from the interlayer coupling of the two Dirac cones may be seen, with such valleys directed toward the small-angle low-energy regime in which
the average band velocity exhibits a complex (E,θ ) dependence shown in panel (b). Panel (c) displays the average band velocity as a function
of energy for two constant angles, and panel (d) shows the band velocity at the Dirac point (not averaged). In this panel, calculations using the
low-energy Hamiltonian of this work [Eq. (26)] are shown as the solid black line, the low-energy Hamiltonian of Bistritzer et al. [8] is shown
as the light dashed line, and full tight-binding calculations are shown both from this work (blue diamonds) and from Ref. [26] (black squares).
The “magic angle” structure seen in panel (d) is to some extent washed out by convolving with a Gaussian, thus it is more difficult to detect in
panel (b).

integrate states in an energy window −0.50 < E < −0.41 eV Eq. (26), allows for easy computation of constant energy
for 0.90◦ and 0.40◦ and −0.40 < E < −0.35 eV for 0.10◦ ). In surfaces and band velocities in the small-angle regime. We
the tight-binding calculations, it is apparent that there is only a motivate the presentation of these results both by their intrinsic
weak modulation of the density induced by the moiré, although usefulness in understanding the twist bilayer, but also by the
this modulation again features the high-density region at the fact that (i) it is likely possible to shift the Fermi energy
AA spots. This density modulation, however, is not well away from the Dirac point by doping without significantly
reproduced by the low-energy Hamiltonian, indicating that changing the electronic structure of the bilayer, and (ii) the
it is now operating outside its region of applicability. We can single-particle band structure at all energies is required for
conclude, therefore, that the fine structure of the tight-binding many-body calculations of the twist bilayer.
calculation is not accessible by the low-energy approach Panel (a) of Fig. 6 shows a density plot of the band velocities
outside an energy window of ≈±0.4 eV about the Dirac point. averaged by simply convolving each eigenvalue by a Gaussian
of width 26 meV and determined for energies −1 < E < 1 eV
and rotation angles 1◦ < θ < 10◦ . The appearance of a series
VII. BAND VELOCITIES AND FERMIOLOGY of “valleys” of reduced average band velocity may be seen.
Having established the excellent agreement between the These result from states from the Dirac cones from each layer
low-energy Hamiltonian, Eq. (26), and full tight-binding that are connected by the coupling vectors g(c) i , in particular
(c)
calculations, we will now explore further the small-angle the vector g1 = 0, which corresponds to simple intersection
physics on the basis of the low-energy approach alone. Of points of the two cones. At such coupling points, bands from
crucial importance to the transport physics of the twist bilayer each layer hybridize, leading to the opening of a local gap
(and more generally the response properties of the bilayer) is in the cone structure and a reduction of the average band
the topology of the constant energy surfaces, and the averaged velocity in this energy region. The dashed lines seen in panel
band velocity on these surfaces. The latter property is also, (a) indicate the lowest-energy intersection between the two
as we shall now demonstrate, very instructive for visualizing cones, and, as may be seen, this corresponds very well with
the complex electronic structure of the twist bilayer. The the most prominent band velocity valley. The band velocity
computational efficiency of the low-energy Hamiltonian, valleys at higher energies correspond to intersections that occur

035452-10
LOW-ENERGY THEORY FOR THE GRAPHENE TWIST BILAYER PHYSICAL REVIEW B 93, 035452 (2016)

FIG. 7. Fermi surfaces at twist angles of θ = 8.0◦ , 3.0◦ , and 0.9◦ for a range of Fermi energies between 10 and 200 meV at both T = 0
(left panel) and T = 300 K (right panel). The color indicates the band velocity of the bands (presented as the ratio v/vF(SLG) ) intersecting the
constant energy surface. For the large twist angle of 8.0◦ , the fermiology is recognizably that of a Dirac cone; at smaller angles, the low-energy
fermiology differs dramatically from the cone topology.

under “backfolding” by the vectors g(c)2,3 . Panel (c) displays the

This can be seen in panel (b), in which an enlarged plot of
average band velocity as a function of energy for two represen- this interesting region is presented. For angles θ < 0.5◦ , the
tative angles indicated by the dashed vertical lines in panel (a). average band velocity for energies −0.25 < E < 0.25 eV is
The reduction in band velocity corresponding to the valleys in ≈0.3vF(SLG) , and evidently the electronic structure has nothing
the density plot can clearly be seen. to do with the Dirac cone of SLG or indeed any conical
The band velocity valleys in panel (a) can all be seen to manifold. In panel (d) we plot the band velocity evaluated at
run toward the origin E = 0, θ = 0. For angles at which a the Dirac point, which is seen to exhibit the “magic angle”
well-defined valley structure exists, it is legitimate to view the structure in which, for certain θ , the band velocity at the
electronic spectrum as that of a Dirac cone manifold disrupted Dirac point approaches zero [8]. This structure is to a large
by interlayer hybridization at a discrete set of energies. extent washed out by convolving with Gaussians, and so it
However, in the region where these valleys converge toward the is not easy to see in panel (b). We should point out that
origin, the very concept of a Dirac cone becomes inapplicable. while our results for the band velocity at the Dirac point

035452-11
D. WECKBECKER et al. PHYSICAL REVIEW B 93, 035452 (2016)

agree very well with the tight-binding calculations of Ref. [26] differs dramatically from that of single-layer graphene. In
(despite a completely different scheme for the tight-binding Fig. 7 we show constant energy surfaces for energies 10 <
parametrization), they are in disagreement with the results E < 200 meV, and for three representative angles, θ = 8◦ , 3◦ ,
of the continuum model of Ref. [8]. This was noticed in and 0.9◦ . The coloring of the constant energy surface indicates
Ref. [26], and it was shown that a simple rescaling would the magnitude of the band velocity |v(k)| (scaled by the Fermi
bring the results into perfect agreement. In the context of the velocity of single-layer graphene).
low-energy model derived in this work, we recognize that this The θ = 8◦ data are presented for comparison with the two
“rescaling problem” simply arises from the wrong choice of smaller angles that we subsequently discuss; at this angle and
momentum scale: K = 43 sin θ2 instead of g (c) = √43 sin θ2 . for energies less than 200 meV, the constant energy surfaces
Despite the fact that we find “magic angles” in the lowest are rather simple, and they are exactly what would be expected
electron band, we note that their physical relevance is very from a Dirac cone, albeit with a somewhat reduced Fermi
limited, as other bands close to the Dirac energy do not velocity as compared to single-layer graphene.
show “magic angles,” or the minima of their band velocities Turning to the case of a twist bilayer with θ = 3◦ one
are at different angles. Consequently, the smoothing over notes that a strong trigonal warping of the cone sets in already
many bands leads to the structure seen in panels (a) and (b) at 100 meV; in single-layer graphene, such trigonal warping
in Fig. 6. occurs only at very high energy. The magnitude of the band
The complex behavior in the strong-coupling small-angle velocity on these constant energy surfaces can be seen to be
low-energy limit implies significant disruption of the Dirac much reduced from that of SLG, and while for the θ = 8◦
cone by the backfolding hybridization mechanism. To gain bilayer |v(k)| showed little variation over constant energy
insight into the qualitative topological changes of the Dirac surfaces, at θ = 3◦ significant variation is seen. This trigonally
manifold, we now present constant energy surfaces in the warped cone persists up to ≈200 meV, at which point the
low-energy regime. For all angles investigated, the E = 0 topology of the constant energy surface changes from a single
constant energy surface is found to be the same zero measure sheet to multiple sheets.
Fermi surface of single-layer graphene. This is a remarkable Finally, for the case of θ = 0.9◦ we see that this transition to
result, as for all E = 0 in the small-angle regime (as may be a multiple sheet topology occurs almost immediately. Already
seen in Fig. 7), the topology of the constant energy surfaces at 10 meV we see the trigonal warping that occurred up to a

FIG. 8. Pseudospin texture of the twisted bilayer in the large- (θ = 30◦ ), intermediate- (θ = 3◦ ), and small- (θ = 0.9◦ ) angle regimes.
The Fermi energies of the Fermi surfaces are 0.9 eV for the large-angle case and 0.13 eV for the intermediate- and small-angle cases. The
large-angle structure resembles the double vortex of a decoupled graphene bilayer, whereas in the intermediate-angle case a trigonally warped
vortex is found. In the small-angle case, the pseudospin texture is seen to “flow” along the sixfold “arms” of the starlike Fermi surface. Note
that the definitions of pseudospin up and pseudospin down differ between the two layers due to the rotation of the basis atoms of the second
layer of the twist bilayer.

035452-12
LOW-ENERGY THEORY FOR THE GRAPHENE TWIST BILAYER PHYSICAL REVIEW B 93, 035452 (2016)

much higher energy of 150 meV for the 3◦ twist bilayer. Above VIII. CONCLUSIONS
10 meV we see a multisheet and rapidly changing topology
We have presented a unified theory of the graphene twist
for all energies. The structure of these constant energy surfaces
bilayer based on the notion of scattering paths that connect the
is remarkably rich, and even more baroque than the famous
K points of the two layers. In the large-angle limit, there are a
low-energy Fermi surface of graphite. Interestingly, while
small number of these scattering paths (three for the 28-atom
capturing the detail of the Fermi surface of graphite required
commensuration cell) that may be conveniently expressed in
very accurate ab initio methods, at these low energies the
real space as a multiple scattering series. In contrast, the
electronic structure is to a very high degree of accuracy
small-angle limit presents an enormously increased number
described by the two-parameter Hamiltonian Eq. (26): the
of scattering paths, which are conveniently encoded in the
complexity of these constant energy surfaces is driven by
moiré field S(r) that couples together the Dirac cones from
reciprocal space geometry of the Dirac cone backfolding. One
each layer. In contrast to the small-angle limit, which involves
should also note again the significant variation of |v(k)| over
only two Dirac cones, in the large-angle limit the scattering
the constant energy surface, ranging from close to the SLG
paths involve all four inequivalent Dirac cones of the twist
value to an order of magnitude less. As one would expect,
bilayer. For large angles, we find this formalism leads to
regions with high |v(k)| tend to be less susceptible to the
electronic versions of the symmetry-derived Hamiltonians
dramatic changes in topology than regions with lower |v(k)|;
of Mele [15], whereas in the small-angle limit we recover
compare, for instance, the constant energy surfaces at 125 and
the Hamiltonian of Bistritzer and MacDonald [8], but with
150 meV. For a more complete visual representation of the
a different momentum coupling scale: instead of K, the
energy dependence of the Fermi surfaces, we provide video
momentum transfer between the unrotated and rotated K
clips as supplemental material [40].
points, a coupling momentum scale g (c) that in real space
Such rapid variations of the constant energy surface
corresponds exactly to the moiré√length D of the problem
topology call into question the very usefulness of the concept
(K leads to a real-space scale of 3D). We demonstrate that
of a Fermi surface in the small-angle regime. As an illustration
the small-angle effective Hamiltonian agrees almost perfectly
of this, we show in the T = 300 K panels of Fig. 7 the plots
with the low-energy electronic structure calculated by a tight-
in which the band velocity is averaged in an energy window
binding method, and thus that the rich electronic structure
E − kB T < E < E + kB T . For the case of an 8◦ twist, at
of the small-angle limit is essentially governed by just two
which a Dirac cone is still well-defined, such a procedure
leads to a “broadened” Dirac cone, but one still occupying a parameters: the single-layer graphene Fermi velocity, vF(SLG) ,
small portion of the Brillouin zone. However, for the case of and the Fourier transform of the interlayer hopping evaluated
a 0.9◦ rotation, the result is dramatic: there is now almost no at the high-symmetry K point, t(K). The baroque complexity
part of the Brillouin zone in which for an interval of 2kB T of the small-angle limit—magic angles, band velocity valleys,
around an energy E there does not exist some single-particle electron localization patterns—is thus essentially geometric
eigenstate. in origin and driven by the reciprocal space juxtaposition
Furthermore, we analyze the dependence of the pseudospin of mutually rotated single-layer Brillouin zones. At large
on the rotation angle, as it has been recently reported that angles, the effective Hamiltonians involve three additional
this quantity may explain measurable physical effects [41]. electronic parameters connected to the high-energy states of
In Fig. 8 we present the pseudospin textures for three twist the scattering paths, however once again we find that the
bilayers: in the large-angle (θ = 30◦ ), intermediate-angle (θ = agreement between the effective Hamiltonian and tight binding
3◦ ), and small-angle (θ = 0.9◦ ) regimes. In the large-angle is nearly perfect.
case, the pseudospin texture is that of single-layer graphene. By calculating the average band velocity systematically as
In layer 1, a vortex of high-intensity pseudospin texture is seen a function of twist angle and energy, we are able to visualize a
on the Fermi circle situated at the K point, while in layer 2 series of “band velocity valleys” that run toward a θ = 0, E =
we find a similar situation but with the high-intensity vortex 0 strong-coupling region. While for twist angles greater than
situated now on the Fermi circle of the K ∗ point. Note that ≈3◦ the electronic spectrum may legitimately be described as
in the second layer, the pseudospin texture does not appear that of Dirac cones hybridized (by the interlayer interaction)
to be tangential to the Fermi surface; this simply results from at a discrete number of energies, in the small-angle region,
the different definitions of pseudospin up and down in the two in particular for θ < 0.5◦ , we find that the Dirac cones are
different layers, and it follows from the rotation of the basis entirely destroyed by hybridization. In this small-angle regime,
vectors in layer 2. by deploying massively parallel calculations we are able to
In the intermediate-angle regime, a pronounced trigo- track the density of states (DOS) of the bilayer all the way
nally warped Fermi surface is found at 0.13 eV, and the to a twist angle of 0.03◦ at which the system electronically
pseudospin texture to some extent follows this warping to segments into distinct stacking regions, and thus no further
produce a “warped vortex” structure with, however, a highly change to the DOS will occur at smaller angles.
pronounced modulation of the pseudospin texture now seen
on the Fermi surface. Finally, in the small-angle regime this
ACKNOWLEDGMENTS
picture again holds with the pseudospin texture appearing
to “flow” on the sheets of the richly structured Fermi This work was supported by the Collaborative Research
surface: one notes that each “arm” of the sixfold starlike Center SFB 953. One of us (S.S.) acknowledges the hospitality
structure situated at the  point consists of a pseudospin of I.I.T. Roorkee, India, during which an early version of the
vortex. manuscript was worked on.

035452-13
D. WECKBECKER et al. PHYSICAL REVIEW B 93, 035452 (2016)

[1] L.-J. Yin, J.-B. Qiao, W.-X. Wang, Z.-D. Chu, K. F. Zhang, Conrad, Why Multilayer Graphene on 4H-SiC(0001) Behaves
R.-F. Dou, C. L. Gao, J.-F. Jia, J.-C. Nie, and L. He, Tuning Like a Single Sheet of Graphene, Phys. Rev. Lett. 100, 125504
structures and electronic spectra of graphene layers with tilt (2008).
grain boundaries, Phys. Rev. B 89, 205410 (2014). [21] J. M. Campanera, G. Savini, I. Suarez-Martinez, and M. I.
[2] W. Landgraf, S. Shallcross, K. Türschmann, D. Weckbecker, and Heggie, Density functional calculations on the intricacies of
O. Pankratov, Electronic structure of twisted graphene flakes, moiré patterns on graphite, Phys. Rev. B 75, 235449 (2007).
Phys. Rev. B 87, 075433 (2013). [22] J. M. B. Lopes dos Santos, N. M. R. Peres, and A. H. Castro Neto,
[3] A. Jorio and L. G. Cançado, Raman spectroscopy of twisted Graphene Bilayer With a Twist: Electronic Structure, Phys. Rev.
bilayer graphene, Solid State Commun. 175–176, 3 (2013). Lett. 99, 256802 (2007).
[4] X. Zou, J. Shang, J. Leaw, Z. Luo, L. Luo, C. La-o vorakiat, L. [23] H. Schmidt, J. C. Rode, D. Smirnov, and R. J. Haug, Superlattice
Cheng, S. A. Cheong, H. Su, J.-X. Zhu, Y. Liu, K. P. Loh, A. H. structures in twisted bilayers of folded graphene, Nat. Commun.
Castro Neto, T. Yu, and E. E. M. Chia, Terahertz Conductivity 5, 5742 (2014).
of Twisted Bilayer Graphene, Phys. Rev. Lett. 110, 067401 [24] B. Roy and K. Yang, Bilayer graphene with parallel magnetic
(2013). field and twisting: Phases and phase transitions in a highly
[5] S. Shallcross, S. Sharma, and O. Pankratov, Emergent mo- tunable dirac system, Phys. Rev. B 88, 241107 (2013).
mentum scale, localization, and van Hove singularities in the [25] K. Uchida, S. Furuya, J.-I. Iwata, and A. Oshiyama, Atomic
graphene twist bilayer, Phys. Rev. B 87, 245403 (2013). corrugation and electron localization due to moiré patterns in
[6] P. San-Jose, J. González, and F. Guinea, Non-Abelian Gauge twisted bilayer graphenes, Phys. Rev. B 90, 155451 (2014).
Potentials in Graphene Bilayers, Phys. Rev. Lett. 108, 216802 [26] G. Trambly de Laissardière, D. Mayou, and L. Magaud,
(2012). Numerical studies of confined states in rotated bilayers of
[7] E. J. Mele, Interlayer coupling in rotationally faulted multilayer graphene, Phys. Rev. B 86, 125413 (2012).
graphenes, J. Phys. D 45, 154004 (2012). [27] C. J. Tabert and E. J. Nicol, Optical conductivity of twisted
[8] R. Bistritzer and A. H. MacDonald, Moiré bands in twisted bilayer graphene, Phys. Rev. B 87, 121402 (2013).
double-layer graphene, Proc. Natl. Acad. Sci. (USA) 108, 12233 [28] J. M. B. Lopes dos Santos, N. M. R. Peres, and A. H. Castro
(2011). Neto, Continuum model of the twisted graphene bilayer, Phys.
[9] A. Luican, G. Li, A. Reina, J. Kong, R. R. Nair, K. S. Novoselov, Rev. B 86, 155449 (2012).
A. K. Geim, and E. Y. Andrei, Single-Layer Behavior and its [29] E. J. Mele, Band symmetries and singularities in twisted
Breakdown in Twisted Graphene Layers, Phys. Rev. Lett. 106, multilayer graphene, Phys. Rev. B 84, 235439 (2011).
126802 (2011). [30] D. Smith and L. von Smekal, Monte Carlo simulation of
[10] M. Kindermann and P. N. First, Phys. Rev. B 83, 045425 (2011). the tight-binding model of graphene with partially screened
[11] G. Li, A. Luican, J. M. B. Lopes dos Santos, A. H. Castro Coulomb interactions, Phys. Rev. B 89, 195429 (2014).
Neto, A. Reina, J. Kong, and E. Y. Andrei, Observation of Van [31] J. F. Dobson, T. Gould, and G. Vignale, How many-body effects
Hove singularities in twisted graphene layers, Nat. Phys. 6, 109 modify the van der waals interaction between graphene sheets,
(2010). Phys. Rev. X 4, 021040 (2014).
[12] G. Trambly de Laissardiére, D. Mayou, and L. Magaud, Nano [32] R. W. Havener, Y. Liang, L. Brown, L. Yang, and J. Park,
Lett. 10, 804 (2010). Van Hove singularities and excitonic effects in the optical
[13] E. Suárez Morell, J. D. Correa, P. Vargas, M. Pacheco, and conductivity of twisted bilayer graphene, Nano Lett. 14, 3353
Z. Barticevic, Flat bands in slightly twisted bilayer graphene: (2014).
Tight-binding calculations, Phys. Rev. B 82, 121407 (2010). [33] J. González, Magnetic and Kohn-Luttinger instabilities near
[14] D. L. Miller, K. D. Kubista, G. M. Rutter, M. Ruan, W. A. a Van Hove singularity: Monolayer versus twisted bilayer
de Heer, P. N. First, and J. A. Stroscio, Structural analysis of graphene, Phys. Rev. B 88, 125434 (2013).
multilayer graphene via atomic moiré interferometry, Phys. Rev. [34] E. Suárez Morell, M. Pacheco, L. Chico, and L. Brey, Electronic
B 81, 125427 (2010). properties of twisted trilayer graphene, Phys. Rev. B 87, 125414
[15] E. J. Mele, Commensuration and interlayer coherence in twisted (2013).
bilayer graphene, Phys. Rev. B 81, 161405 (2010). [35] G. Kresse and J. Hafner, Ab initio molecular dynamics for liquid
[16] S. Shallcross, S. Sharma, E. Kandelaki, and O. A. Pankratov, metals, Phys. Rev. B 47, 558 (1993).
Electronic structure of turbostratic graphene, Phys. Rev. B 81, [36] G. Kresse and J. Furthmüller, Efficiency of ab-initio total energy
165105 (2010). calculations for metals and semiconductors using a plane-wave
[17] F. Varchon, P. Mallet, L. Magaud, and J.-Y. Veuillen, Rotational basis set, Comput. Mater. Sci. 6, 15 (1996).
disorder in few-layer graphene films on 6H-SiC(000-1): A [37] G. Kresse and J. Furthmüller, Efficient iterative schemes for
scanning tunneling microscopy study, Phys. Rev. B 77, 165415 ab initio total-energy calculations using a plane-wave basis set,
(2008). Phys. Rev. B 54, 11169 (1996).
[18] S. Shallcross, S. Sharma, and O. A. Pankratov, Quantum [38] L. S. Blackford, J. Choi, A. Cleary, E. D’Azevedo, J. Dem-
Interference at the Twist Boundary in Graphene, Phys. Rev. mel, I. Dhillon, J. Dongarra, S. Hammarling, G. Henry, A.
Lett. 101, 056803 (2008). Petitet, K. Stanley, D. Walker, and R. C. Whaley, ScaLAPACK
[19] S. Shallcross, S. Sharma, and O. A. Pankratov, Twist boundary in Users’ Guide (Society for Industrial and Applied Mathematics,
graphene: Energetics and electric field effect, J. Phys. Condens. Philadelphia, PA, 1997).
Matter 20, 454224 (2008). [39] M. Frigo and S. G. Johnson, The design and implementation of
[20] J. Hass, F. Varchon, J. E. Millán-Otoya, M. Sprinkle, N. Sharma, FFTW3, Proc. IEEE 93, 216 (2005); Special issue on Program
W. A. de Heer, C. Berger, P. N. First, L. Magaud, and E. H. Generation, Optimization, and Platform Adaptation.

035452-14
LOW-ENERGY THEORY FOR THE GRAPHENE TWIST BILAYER PHYSICAL REVIEW B 93, 035452 (2016)

[40] See Supplemental Material at http://link.aps.org/supplemental/ [41] D. Song, V. Paltoglou, S. Liu, Y. Zhu, D. Gallardo, L. Tang,
10.1103/PhysRevB.93.035452 for visualizations of the energy J. Xu, M. Ablowitz, N. K. Efremidis, and Z. Chen, Unveiling
dependence of the Fermi surfaces at θ = 8.0◦ , 3.0◦ , 0.9◦ , and pseudospin and angular momentum in photonic graphene, Nat.
0.75◦ as video clips. Commun. 6, 6272 (2015).

035452-15