Moiré bands in twisted double-layer graphene Rafi Bistritzer and Allan H. MacDonald 1 Department of Physics, University of Texas at Austin, Austin, TX 78712 Contributed by Allan H. MacDonald, June 7, 2011 (sent for review December 8, 2010) A moiré pattern is formed when two copies of a periodic pattern are overlaid with a relative twist. We address the electronic struc- ture of a twisted two-layer graphene system, showing that in its continuum Dirac model the moiré pattern periodicity leads to moiré Bloch bands. The two layers become more strongly coupled and the Dirac velocity crosses zero several times as the twist angle is reduced. For a discrete set of magic angles the velocity vanishes, the lowest moiré band flattens, and the Dirac-point density-of- states and the counterflow conductivity are strongly enhanced. L ow-energy electronic properties of few layer graphene (FLG) systems are known (1–8) to be strongly dependent on stacking arrangement. In bulk graphite 0° and 60° relative orientations of the individual layer honeycomb lattices yield rhombohedral and Bernal crystals, but other twist angles also appear in many sam- ples (9). Small twist angles are particularly abundant in epitaxial graphene layers grown on SiC (10, 11), but exfoliated bilayers can also appear with a twist, and arbitrary alignments between adja- cent layers can be obtained by folding a single graphene layer (12, 13). Recent advances in FLG preparation methods have attracted theoretical attention (14–20) to the intriguing electronic proper- ties of systems with arbitrary twist angles, usually focusing on the two-layer case. The geometry of the bilayer system is character- ized by a twist angle θ and by a translation vector d. Commensur- ability is determined only by θ. Sliding one layer with respect to the other in a commensurate structure modifies the unit cell but leaves the bilayer crystalline. In this work we find it convenient to regard the AB stacking as the aligned configuration. The posi- tions of the carbon atoms in the two misaligned layers labeled by R and R 0 are then related by R 0 ¼ MðθÞðR − τÞþ d, where M is a 2-D rotation matrix within the graphene plane, and τ is a vector connecting the two atoms in the unit cell. The problem is mathematically interesting because a bilayer forms a two-dimensional crystal only at a discrete set of commen- surate rotation angles; for generic twist angles Bloch’s theorem does not apply microscopically and direct electronic structure calculations are not feasible. For twist angles larger than a few degrees the two layers are electronically isolated to a remarkable degree, except at a small set of angles which yield low-order com- mensurate structures (16, 19). As the twist angles become smal- ler, interlayer coupling strengthens, and the quasiparticle velocity at the Dirac point begins to decrease. Here we focus on the strongly coupled small twist angle regime. We derive a low-energy effective Hamiltonian valid for any value of d and for θ ≲ 10° irrespective of whether or not the bilayer struc- ture is periodic. We show that it is meaningful to describe the elec- tronic structure using Bloch bands even for incommensurate twist angles and study the dependence of these bands on θ. Model We construct a low-energy continuum model Hamiltonian that consists of three terms: two single-layer Dirac–Hamiltonian terms that account for the isolated graphene sheets and a tunneling term that describes hopping between layers. The Dirac–Hamilto- nian (21) for a layer rotated by an angle θ with respect to a fixed coordinate system is h k ðθÞ¼ −vk 0 e iðθ k −θÞ e −iðθ k −θÞ 0 ; where v is the Dirac velocity, k is momentum measured from the layer’s Dirac point, θ k is the momentum orientation relative to the x axis, and the spinor Hamiltonian acts on the individual layer’s A and B sublattice degrees-of-freedom. We choose the co- ordinate system depicted in Fig. 1 for which the decoupled bilayer Hamiltonian is j1ihðθ∕2Þh1jþj2ihð−θ∕2Þh2j, where jiihij projects onto layer i. We derive a continuum model for the tunneling term by assum- ing that the interlayer tunneling amplitude between π-orbitals is a smooth function tðrÞ of spatial separation projected onto the graphene planes. The matrix element T αβ kp 0 ¼hΨ ð1Þ kα jH T jΨ ð2Þ p 0 β i  of the tunneling Hamiltonian H T describes a process in which an electron with momentum p 0 ¼ Mp residing on sublattice β in one layer hops to a momentum state k and sublattice α in the other layer. In a π-band tight-binding model the projection of the wave functions of the two layers to a given sublattice are jψ ð1Þ kα i¼ 1 ﬃﬃﬃﬃ N p ∑ R e ikðRþτ α Þ jR þ τ α i  and jψ ð2Þ pβ i¼ 1 ﬃﬃﬃﬃ N p ∑ R 0 e ipðR 0 þτ 0 β Þ jR 0 þ τ 0 β i:  Here τ A ¼ 0, τ B ¼ τ, and R is summed over the triangular Bravais lattice. Substituting Eqs. 2 and 3 in Eq. 1 and invoking the two- center approximation, hR þ τ α jH T jR 0 þ τ 0 β i¼ tðR þ τ α − R 0 − τ 0 β Þ;  for the interlayer hopping amplitude in which t depends on the dif- ference between the positions of the two carbon atoms we find that T αβ kp 0 ¼ ∑ G 1 G 2 t ¯ kþG 1 Ω e i½G 1 τ α −G 2 ðτ β −τÞ−G 0 2 dδ ¯ kþG 1 ;¯ p 0 þG 0 2 :  Here Ω is the unit cell area, t q is the Fourier transform of the tun- neling amplitude tðrÞ, the vectors G 1 and G 2 are summed over re- ciprocal lattice vectors, and G 0 2 ¼ MG 2 . The bar notation over momenta in Eq. 5 indicates that momentum is measured relative to the center of the Brillouin zone and not relative to the Dirac point. Note that crystal momentum is conserved by the tunneling process because t depends only on the difference between lattice positions.* Author contributions: R.B. and A.H.M. designed research; R.B. performed research; and R.B. and A.H.M. wrote the paper. The authors declare no conflict of interest. *A closely related but slightly different expression appears in ref. 19 in which we chose the origin at a honeycomb lattice point. The present convention is more convenient for the discussion of small rotations relative to the Bernal arrangement. 1 To whom correspondence should be addressed. E-mail: [email protected]. www.pnas.org/cgi/doi/10.1073/pnas.1108174108 PNAS ∣ July 26, 2011 ∣ vol. 108 ∣ no. 30 ∣ 12233–12237 PHYSICS
Moiré bands in twisted double-layer grapheneRafi Bistritzer and Allan H. MacDonald1
Department of Physics, University of Texas at Austin, Austin, TX 78712
Contributed by Allan H. MacDonald, June 7, 2011 (sent for review December 8, 2010)
A moiré pattern is formed when two copies of a periodic patternare overlaid with a relative twist. We address the electronic struc-ture of a twisted two-layer graphene system, showing that in itscontinuum Dirac model the moiré pattern periodicity leads tomoiré Bloch bands. The two layers become more strongly coupledand the Dirac velocity crosses zero several times as the twist angleis reduced. For a discrete set of magic angles the velocity vanishes,the lowest moiré band flattens, and the Dirac-point density-of-states and the counterflow conductivity are strongly enhanced.
Low-energy electronic properties of few layer graphene (FLG)systems are known (1–8) to be strongly dependent on stacking
arrangement. In bulk graphite 0° and 60° relative orientations ofthe individual layer honeycomb lattices yield rhombohedral andBernal crystals, but other twist angles also appear in many sam-ples (9). Small twist angles are particularly abundant in epitaxialgraphene layers grown on SiC (10, 11), but exfoliated bilayers canalso appear with a twist, and arbitrary alignments between adja-cent layers can be obtained by folding a single graphene layer(12, 13).
Recent advances in FLG preparation methods have attractedtheoretical attention (14–20) to the intriguing electronic proper-ties of systems with arbitrary twist angles, usually focusing on thetwo-layer case. The geometry of the bilayer system is character-ized by a twist angle θ and by a translation vector d. Commensur-ability is determined only by θ. Sliding one layer with respect tothe other in a commensurate structure modifies the unit cell butleaves the bilayer crystalline. In this work we find it convenient toregard the AB stacking as the aligned configuration. The posi-tions of the carbon atoms in the two misaligned layers labeledby R and R0 are then related by R0 ¼ MðθÞðR − τÞ þ d, whereM is a 2-D rotation matrix within the graphene plane, and τ isa vector connecting the two atoms in the unit cell.
The problem is mathematically interesting because a bilayerforms a two-dimensional crystal only at a discrete set of commen-surate rotation angles; for generic twist angles Bloch’s theoremdoes not apply microscopically and direct electronic structurecalculations are not feasible. For twist angles larger than a fewdegrees the two layers are electronically isolated to a remarkabledegree, except at a small set of angles which yield low-order com-mensurate structures (16, 19). As the twist angles become smal-ler, interlayer coupling strengthens, and the quasiparticle velocityat the Dirac point begins to decrease.
Here we focus on the strongly coupled small twist angle regime.We derive a low-energy effective Hamiltonian valid for any valueof d and for θ ≲ 10° irrespective of whether or not the bilayer struc-ture is periodic. We show that it is meaningful to describe the elec-tronic structure using Bloch bands even for incommensurate twistangles and study the dependence of these bands on θ.
ModelWe construct a low-energy continuum model Hamiltonian thatconsists of three terms: two single-layer Dirac–Hamiltonian termsthat account for the isolated graphene sheets and a tunnelingterm that describes hopping between layers. The Dirac–Hamilto-nian (21) for a layer rotated by an angle θ with respect to a fixedcoordinate system is
hkðθÞ ¼ −vk 0 eiðθk−θÞ
where v is the Dirac velocity, k is momentum measured from thelayer’s Dirac point, θk is the momentum orientation relative tothe x axis, and the spinor Hamiltonian acts on the individuallayer’s A and B sublattice degrees-of-freedom. We choose the co-ordinate system depicted in Fig. 1 for which the decoupled bilayerHamiltonian is j1ihðθ∕2Þh1j þ j2ihð−θ∕2Þh2j, where jiihij projectsonto layer i.
We derive a continuummodel for the tunneling term by assum-ing that the interlayer tunneling amplitude between π-orbitalsis a smooth function tðrÞ of spatial separation projected onto thegraphene planes. The matrix element
Tαβkp0 ¼ hΨð1Þ
kα jHTjΨð2Þp0βi 
of the tunneling Hamiltonian HT describes a process in which anelectron with momentum p0 ¼ Mp residing on sublattice β in onelayer hops to a momentum state k and sublattice α in the otherlayer. In a π-band tight-binding model the projection of the wavefunctions of the two layers to a given sublattice are
jψ ð1Þkα i ¼
eikðRþταÞjR þ ταi 
jψ ð2Þpβ i ¼
0þτ0βÞjR0 þ τ0βi: 
Here τA ¼ 0, τB ¼ τ, and R is summed over the triangular Bravaislattice. Substituting Eqs. 2 and 3 in Eq. 1 and invoking the two-center approximation,
hR þ ταjHTjR0 þ τ0βi ¼ tðR þ τα − R0 − τ0βÞ; 
for the interlayer hopping amplitude in which t depends on the dif-ference between the positions of the two carbon atoms we find that
Tαβkp0 ¼ ∑
2d�δkþG1 ;p0þG02: 
Here Ω is the unit cell area, tq is the Fourier transform of the tun-neling amplitude tðrÞ, the vectors G1 and G2 are summed over re-ciprocal lattice vectors, and G0
2 ¼ MG2. The bar notation overmomenta in Eq. 5 indicates that momentum is measured relativeto the center of the Brillouin zone and not relative to the Diracpoint. Note that crystal momentum is conserved by the tunnelingprocess because t depends only on the difference between latticepositions.*
Author contributions: R.B. and A.H.M. designed research; R.B. performed research; andR.B. and A.H.M. wrote the paper.
The authors declare no conflict of interest.
*A closely related but slightly different expression appears in ref. 19 in which we chose theorigin at a honeycomb lattice point. The present convention is more convenient for thediscussion of small rotations relative to the Bernal arrangement.
The continuum model for HT is obtained by measuring wavevectors in both layers relative to their Dirac points and assum-ing that the deviations are small compared to Brillouin-zonedimensions. The model’s utility rests centrally on the observa-tion that, although tq is not precisely known, it should never-theless fall to zero very rapidly with q on the reciprocal latticevector scale. This behavior follows from the property that thegraphene layer separation d⊥ exceeds the separation betweencarbon atoms within the layers by more than a factor of 2. Be-cause the two-center integral tðrÞ varies with the three-dimen-sional separation R ¼
processes vary with r on the scale of d⊥. For this reason tq beginsto decline rapidly for qd⊥ > 1. Fig. 2 plots tq values obtainednumerically from the π-band tight-binding models proposed inrefs. 19, 22, and 23. The largest tq values that enter the tunnelingnear the Dirac point have q ¼ kD, the Brillouin-zone corner(Dirac) wave vector magnitude, and correspond to the threereciprocal vectors 0, Gð2Þ, and Gð3Þ where the latter two vectorsconnect a Dirac point with its equivalent first Brillouin-zonecounterparts (See Fig. 1). When only these terms are retained,
TαβðrÞ ¼ w∑3
expð−iqj · rÞTαβj ; 
where w ¼ tkD∕Ω is the hopping energy,
T1 ¼ 1 1
� �; T2 ¼ e−iG
ð2Þ0 ·d e−iϕ 1
T3 ¼ e−iGð3Þ0 ·d eiϕ 1
� �; 
and ϕ ¼ 2π∕3. The three qj’s in Eq. 6 are Dirac model momen-tum transfers that correspond to the three interlayer hoppingprocesses.
For d ¼ 0 and a vanishing twist angle the continuum tunnelingmatrix is 3wδαAδβB, independent of position. By comparing withthe experimentally known (24) electronic structure of an ABstacked bilayer we set w ≈ 110 meV for exfoliated samples,however experiments suggest (25) that w may be smaller in someepitaxial graphene samples. As we show below the spectrum isindependent of d for θ ≠ 0. In the following we thereforeset d ¼ 0.
ResultsMoiré Bloch Bands. In the continuum model hopping is local andperiodic, allowing Bloch’s theorem to be applied at any rotationangle irrespective of whether or not the bilayer is crystalline. Wesolve numerically for the moiré bands using the plane wave ex-pansion illustrated in Fig. 1. Convergence is attained by truncat-ing momentum space at lattice vectors of the order of w∕ℏv. Thedimension of the matrix, which must be diagonalized numerically,is roughly ∼10 θ−2 for small θ (measured in degrees), comparedto the ∼104 θ−2 matrix dimension of microscopic tight-bindingmodels (14, 16).
Up to a scale factor the moiré bands depend on a single para-meter α ¼ w∕vkθ. We have evaluated the moiré bands as a func-tion of their Brillouin-zone momentum k for many different twistangles; results for w ¼ 110 meV, and θ ¼ 5°, 1.05°, and 0.5° aresummarized in Fig. 3. For large twist angles the low-energy spec-trum is virtually identical to that of an isolated graphene sheet,except that the velocity is slightly renormalized. Large interlayercoupling effects appear only near the high energy van Hove sin-gularities discussed by Andrei and coworkers (26). As the twistangle is reduced, the number of bands in a given energy windowincreases and the band at the Dirac point narrows. This narrow-ing has previously been expected to develop monotonically withdecreasing θ. As illustrated in Fig. 3, we instead find that theDirac-point velocity vanishes already at θ ≈ 1.05°, and that thevanishing velocity is accompanied by a very flat moiré band whichcontributes a sharp peak to the Dirac-point density-of-states(DOS). At smaller twists the Dirac-point velocity has a nonmo-
Fig. 1. Momentum-space geometry of a twisted bilayer. (A) Dashed linemarks the first Brillouin zone of an unrotated layer. The three equivalentDirac points are connected by Gð2Þ and Gð3Þ. The circles represent Diracpoints of the rotated graphene layers, separated by kθ ¼ 2kD sinðθ∕2Þ, wherekD is the magnitude of the Brillouin-zone corner wave vector for a singlelayer. Conservation of crystal momentum implies that p0 ¼ kþ qb for atunneling process in the vicinity of the plotted Dirac points. (B) The threeequivalent Dirac points in the first Brillouin zone result in three distincthopping processes. Interference between hopping processes with differentwave vectors captures the spatial variation of interlayer coordination thatdefines the moiré pattern. For all the three processes jqj j ¼ kθ ; however,the hopping directions are ð0; − 1Þ for j ¼ 1, ð ffiffiffi
∕2;1∕2Þ for j ¼ 2, andð− ffiffiffi
∕2;1∕2Þ for j ¼ 3. We interchangeably use 1, 2, 3, b, tr, and tl as sub-scripts for the three momentum transfers qj . Repeated hopping generates ak-space honeycomb lattice. The green solid line marks the moiré bandWigner–Seitz cell. In a repeated zone scheme the red and black circles markthe Dirac points of the two layers.
2 4 6 8 10 12
t q [meV
Fig. 2. Hopping amplitude. The Fourier transform of the hopping amplitudeis plotted vs. momentum (a is the carbon-carbon distance in single-layer gra-phene). The different curves correspond to different models described inrefs. 19 (dotted), 22 (solid), and 23 (dashed). The vertical line crosses the xaxis at kDa.
12234 ∣ www.pnas.org/cgi/doi/10.1073/pnas.1108174108 Bistritzer and MacDonald
notonic dependence on θ, vanishing repeatedly at the series ofmagic angles illustrated in Fig. 4.
Partial insight into the origin of these behaviors can be achievedby examining the simplest limit in which the momentum-spacelattice is truncated at the first honeycomb shell. Including thesublattice degree of freedom, this truncation gives rise to theHamiltonian
Hk ¼hkðθ∕2Þ Tb Ttr Ttl
b hkbð−θ∕2Þ 0 0
T†tr 0 hktrð−θ∕2Þ 0
tl 0 0 hktlð−θ∕2Þ
where k is in the moiré Brillouin-zone and kj ¼ k þ qj. ThisHamiltonian acts on four two-component spinors Ψ ¼ ðψ0;ψ1;ψ2;ψ3Þ. The first (ψ0) is at a momentum near the Dirac point ofone layer and the other three ψ j are at momenta near qj and in the
other layer. The dependence of hðθÞ on angle is parametricallysmall and can be neglected. We have numerically verified that thisapproximation reproduces the velocity with reasonable accuracydown to the first magic angle (Fig. 4, Inset).
The renormalized velocity v⋆ ¼ ∂kϵ⋆k jk¼0 follows from thespectrum ϵ⋆k of the twisted bilayer. The Hamiltonian is expressedas a sum of the k ¼ 0 term Hð0Þ and the k-dependent term Hð1Þ
kand solved to leading order in k.
Consider the k ¼ 0 term in the Hamiltonian. We assume thatHð0Þ has zero energy eigenstates and prove our assumption byexplicitly finding these states. The zero energy eigenstates mustsatisfy
ψ j ¼ −h−1j T†j ψ0: 
Tjh−1j T†j ¼ 0 
the equation for the ψ0 spinor is h0ψ0 ¼ 0, i.e., ψ0 is one ofthe two zero energy states ψ ð1Þ
0 and ψ ð2Þ0 of the isolated layer.
The two zero energy eigenstates of Hð0Þ then follow from Eq. 9.Given that jψ ðjÞ
0 j ¼ 1, the wave functions should be normalizedby jΨj2 ¼ 1þ 6α2. The effective Hamiltonian matrix to leadingorder in k is therefore
hΨðiÞjHð1Þk jΨðjÞi ¼ −v
1þ 6α2ψ ðiÞ†0
�σ · k þ w2
· kh−1j T†j
�ψ ðjÞ0 ¼ −v⋆ψ ðiÞ†
0 σ · kψ ðjÞ0 :
Aside from a renormalized velocity
v¼ 1 − 3α2
1þ 6α2; 
the Hamiltonian is identical to the continuum model Hamilto-nian of single-layer graphene. The denominator in Eq. 11 cap-tures the contribution of the Ψj’s to the normalization of thewave function whereas the numerator captures their contributionto the velocity matrix elements. For small α, Eq. 11 reduces tothe expression v⋆∕v ¼ 1 − 9α2, first obtained by Lopes dos Santoset al. (15). The velocity vanishes at the first magic angle because itis in the process of changing sign. The eigenstates at the Diracpoint are a coherent combination of components in the two layersthat have velocities of opposite sign.
Counterflow Conductivity. The distribution of the quasiparticlevelocity between the two layers implies exotic transport charac-teristics for separately contacted layers. Consider a counterflowgeometry in which currents in the two layers flow antiparallel toone another. We focus on twist angles θ ≳ 2° for which the eight-band model is valid and to the semiclassical regime in whichϵFτ > 1 and find the counterflow conductivity σCF. We assumethat the Fermi momentum is much smaller than kθ and that1∕τ0 < ℏvkθ, where τ0 is single particle lifetime. Using the Kuboformula we find that
σCF ¼ 4e2
vxCF ¼ −v
σx 0 0 0
0 −σx 0 0
0 0 −σx 0
0 0 0 −σx
Fig. 3. Moiré bands. (A) Energy dispersion for the 14 bands closest to theDirac point plotted along the k-space trajectory A → B → C → D → A (seeFig. 1) for w ¼ 110 meV, and θ ¼ 5° (Left,), 1.05° (Middle), and 0.5° (Right).(B) DOS. (C) Energy as a function of twist angle for the k ¼ 0 states. Bandseparation decreases with θ as also evident from A. (D) Full dispersion ofthe flat band at θ ¼ 1.05°.
Fig. 4. Renormalized Dirac-point band velocity. The band velocity of thetwisted bilayer at the Dirac point v⋆ is plotted vs. α2, where α ¼ w∕vkθ
for 0.18° < θ < 1.2°. The velocity vanishes for θ ≈ 1.05°, 0.5°, 0.35°, 0.24°,and 0.2°. (Inset) The renormalized velocity at larger twist angles. The solidline corresponds to numerical results and dashed line corresponds to analyticresults based on the eight-band model.
Bistritzer and MacDonald PNAS ∣ July 26, 2011 ∣ vol. 108 ∣ no. 30 ∣ 12235
is the x component of the counterflow velocity operator (we setthe electric fields along the x axis), Gr
kμðωÞ ¼ ðω − ϵ⋆kμ þ i∕2τ0Þ−1is the retarded Green function with μ labeling the two Diracbands, and ϵ⋆kμ ¼ μv⋆k is the energy dispersion of the twistedbilayer at small momenta. For an electron-doped system thevalence band can be neglected and
σCF ≈ e2gτν⋆ðϵFÞZ
where ν⋆ is the DOS of the twisted bilayer. The vertex function
hψkjvxCFjψki ¼ vCF cos θk; 
where vCF ¼ vð1þ 3α2Þ∕ð1þ 6α2Þ follows directly from theprevious section if we notice the sign differences between thecounterflow velocity operator and the normal velocity operator.The counterflow conductivity is then
σCF ¼ σ0
where σ0 ∼ e2ϵFτ∕π is the conductivity of an isolated single gra-phene layer. As θ is reduced from a large value toward 1°, v⋆ isreduced and the DOS is correspondingly increased. The counter-flow conductivity is enhanced because of an increased densityof carriers, which is not accompanied by a decrease in counter-flow velocity. For a conventional measurement in which thecurrent in the bilayer is unidirectional vCF in Eq. 16 is replacedby v⋆. The increase in the DOS is then exactly compensated bythe reduction of the renormalized velocity and the single-layervalue of the conductivity is regained.
Dependence of the Spectrum on d.We now show that the spectrumof misaligned bilayers is independent of linear translations of onelayer with respect to the other using a unitary transformation thatmakes the Hamiltonian independent of d. Consider HQ where Qis a momentum in the first moiré Brillouin zone. With each mo-mentum on the k-space triangular Bravais lattice (see Fig. 1)
k ¼ Qþ nq1 þmq2;
where q1 ¼ kθð1∕2;ffiffiffi3
p∕2Þ and q2 ¼ kθð−1∕2;
p∕2Þ, we associate
ϕk ¼ nG02 · dþmG0
3 · d:
The phase associated with momentum k − kθ y on the other sub-lattice is ϕk as well. In terms of the new basis states expðiϕkÞjkαithe Hamiltonian HQ is d-independent.
Physically, the lack of dependence on d can be understoodby noticing that varying d just shifts the moiré pattern in space.The bilayer spectrum does depend on d at θ ¼ 0, and at othercommensurate angles. We expect that dependence on d will beobservable only at short period (large θ) commensurate angles.
DiscussionTwisted double-layer graphene is, for most values of θ, a quasi-periodic structure that has no unit cell. Nevertheless, we find thatfor θ ≲ 10° it is meaningful to describe the electronic structure ofthe system in terms of Bloch bands. The hidden periodic structureis shown to be related to the moiré pattern of the overlaidlayers (27).
The leading corrections to the periodic moiré band Hamilto-nian involve hopping amplitudes with the smallest momenta gthat satisfy the crystal momentum conservation condition in Eq. 5and are larger than kD. As we showed in ref. 19, real space com-mensuration between the two rotated hexagonal lattice is conco-
mitant to momentum space commensuration of the Dirac pointsin the extended-zone scheme (see figures 2 and 3 in ref. 19). Thecommensurate vector g can therefore be found using the formulafor the moiré periodicity if the lattice vector of graphene
(where a is the carbon-carbon distance in a single-layer graphene)is replaced by the reciprocal lattice vector G. It follows thatgðθÞ ≈G∕θ. For example gð10°Þ ¼ 24∕a and gð2°Þ ¼ 120∕a. AsFig. 2 demonstrates, the hopping amplitudes for these large wavevectors are indeed negligible compared to the value of tkD . Wetherefore expect the continuum model to be very accurate upto energies of approximately 1 eV and up to angles of approxi-mately 10°.
The Bloch band model has a simple and appealing physicalinterpretation. The hopping Hamiltonian is local in space. Ateach position, its 4 × 4 matrix, describes sublattice-dependentinterlayer hopping, which depends on the local coordinationbetween the atoms in the two layers. In Fig. 5 we have plotted themoiré pattern of atomic positions and the smaller of the two po-sitive eigenvalues of the hopping Hamiltonian as a function ofposition on the same length scale. At each position, the local in-terlayer tunneling Hamiltonian, is that of a system in which thelocal coordination is maintained through all space. At AB and BApoints, for example, the tunneling Hamiltonian is that of AB andBA systems, for which tunneling does not produce a gap so thatthe smallest positive eigenvalue vanishes. On the other hand thegap reaches its maxima (6w) at AA points in the moiré pattern.
In summary we have formulated a continuum model descrip-tion of the electronic structure of rotated graphene bilayers. Theresulting moiré band structure can be evaluated at arbitrary twistangles, not only at commensurate values. We find that the velocityat the Dirac point oscillates with twist angle, vanishing at a seriesof magic angles which give rise to large DOS and to large counter-flow conductivities. Many properties of the moiré bands are stillnot understood. For example, although we are able to explain thelargest magic angle analytically, the pattern of magic angles atsmaller values of θ has so far been revealed only numerically.Additionally the flattening of the entire lowest moiré band atθ ≈ 1.05° remains a puzzle. Interesting new issues arise whenour theory is extended to graphene stacks containing three ormore layers. Finally, we remark that electron-electron interac-tions neglected in this work are certain to be important at magictwist angles in neutral systems and could give rise to counterflowsuperfluidity (28, 29), flat-band magnetism (30), or other types ofordered states.
ACKNOWLEDGMENTS. We acknowledge a helpful conversation with GeneMele. This work was supported by Welch Foundation Grant F1473 and bythe National Science Foundation-Nanoelectronics Research Initiative SouthWest Academy of Nanoelectronics program.
−5 0 5−6
k θ y AB
−5 0 5−6
k θ y
Fig. 5. Moiré period. (Right) Moiré pattern obtained from two graphenelayers overlaid with a relative twist angle θ. Distances are measured in unitsof k−1
θ , where kθ ¼ 2kD sinðθ∕2Þ with kD being the Dirac wave vector. Bluedots denote areas with local AB coordination. (Left) Smallest positive energyof the interlayer Hamiltonian. The energy vanishes for local AB or BA coor-dination and reached a maximum of 3w for local AA coordination.
12236 ∣ www.pnas.org/cgi/doi/10.1073/pnas.1108174108 Bistritzer and MacDonald
1. McCann E, Falko VI (2006) Landau-level degeneracy and quantum hall effect in agraphite bilayer. Phys Rev Lett 96:086805.
2. Koshino M, Ando T (2007) Orbital diamagnetism in multilayer graphenes: Systematicstudy with the effective mass approximation. Phys Rev B 76:085425.
3. Nilsson J, Castro Neto AH, Guinea F, Peres NMR (2008) Electronic properties of bilayerand multilayer graphene. Phys Rev B 78:045405.
4. Min H, MacDonald AH (2008) Chiral decomposition in the electronic structure ofgraphene multilayers. Phys Rev B 77:155416.
5. Koshino M, McCann E (2009) Trigonal warping and Berrys phase Nπ in ABC-stackedmultilayer graphene. Phys Rev B 80:165409.
6. Zhang F, Sahu B, Min H, MacDonald AH (2010) Band structure of ABC-stackedgraphene trilayers. Phys Rev B 82:035409.
7. Wang Y, et al. (2010) Stacking-dependent optical conductivity of bilayer graphene.ACS Nano 4:4074–4080.
8. Ghosh S, et al. (2010) Dimensional crossover of thermal transport in few-layergraphene. Nat Mater 9:555–558.
9. Rong ZY, Kuiper P (1993) Electronic effects in scanning tunneling microscopy: Moirépattern on a graphite surface. Phys Rev B 48:17427–17431.
10. Hass J, et al. (2008) Why multilayer graphene on 4H-SiCð0001Þ behaves like a singlesheet of graphene. Phys Rev Lett 100:125504.
11. First PN, et al. (2010) Epitaxial graphenes on silicon carbide. MRS Bull 35:296–305.12. Schmidt H, Luedtke T, Barthold P, Haug RJ (2010) Mobilities and scattering times in
oriented high resolution lithography of graphene nanoribbons by STM lithography.Phys Status Solidi B 247:896–902.
14. Trambly de Laissardière G, MayouD,Magaud L (2010) Localization of Dirac electrons inrotated graphene bilayers. Nano Lett 10:804–808.
15. Lopes dos Santos JMB, Peres NMR, Castro Neto AH (2007) Graphene bilayer with atwist: Electronic structure. Phys Rev Lett 99:256802.
16. Shallcross S, Sharma S, Kandelaki E, Pankratov OA (2010) Electronic structure ofturbostratic graphene. Phys Rev B 81:165105.
17. Shallcross S, Sharma S, Pankratov OA (2008) Quantum interference at the twistboundary in graphene. Phys Rev Lett 101:056803.
18. Mele EJ (2010) Commensuration and interlayer coherence in twisted bilayer graphene.Phys Rev B 81:161405(R).
19. Bistritzer R, MacDonald AH (2010) Transport between twisted graphene layers. PhysRev B 81:245412.
20. Morell ES, Correa JD, Vargas P, Pacheco M, Barticevic Z (2010) Flat bands in slightlytwisted bilayer graphene: Tight-binding calculations. Phys Rev B 82:121407(R).
21. Castro Neto AH, Guinea F, Peres NMR, Novoselov KS, Geim AK (2009) The electronicproperties of graphene. Rev Mod Phys 81:109–162.
22. Pereira VM, Castro Neto AH, Peres NMR (2009) Tight-binding approach to uniaxialstrain in graphene. Phys Rev B 80:045401.
23. Tang MS, Wang CZ, Chan CT, Ho KM (1996) Environment-dependent tight-bindingpotential model. Phys Rev B 53:979–982.
24. Kuzmenko AB, Crassee I, van der Marel D, Blake P, Novoselov KS (2009) Determinationof the gate-tunable band gap and tight-binding parameters in bilayer grapheneusing infrared spectroscopy. Phys Rev B 80:165406.
25. Sprinkle M, et al. (2009) First direct observation of a nearly ideal graphene bandstructure. Phys Rev Lett 103:226803.
26. Li G, et al. (2010) Observation of van Hove singularities in twisted graphene layers.Nat Phys 6:109–113.
27. Amidror I (2009) The Theory of the Moiré Phenomenon, (Springer, London), Vol I.28. Eisenstein JP, MacDonald AH (2004) Bose–Einstein condensation of excitons in bilayer
electron systems. Nature 432:691–694.29. Min H, Bistritzer R, Su JJ, MacDonald AH (2008) Room-temperature superfluidity in
graphene bilayers. Phys Rev B 78:121401(R).30. Arita R, Suwa Y, Kuroki K, Aoki H (2002) Gate-induced band ferromagnetism in an
organic polymer. Phys Rev Lett 88:127202.
Bistritzer and MacDonald PNAS ∣ July 26, 2011 ∣ vol. 108 ∣ no. 30 ∣ 12237