arXiv:1012.3484v2 [cond-mat.str-el] 26 Feb 2012 Contents Stat Current Graphene: in Interactions Electron-Electron I.Qaiatce nGraphene in Quasiparticles III. V h olm rbe n hre Impurities Charged and Problem Coulomb The IV. I hreplrzto n ierscreening linear and polarization Charge II. .Introduction I. .Pyia Observables Physical D. screening and modes Collective D. .Lweeg eairna h ia point Dirac the near behavior Low-energy A. spectrum Tight-binding A. .Fnt est em-iudregime Fermi-liquid density Finite C. function Polarization C. .Sotnosms generation mass Spontaneous B. Hamiltonian fermion Dirac B. .Ifiiesako layers of stack Infinite E. .Oeve fmi results main of Overview E. .fsmrule f-sum F. ngahn,w hwta ltoao neetn susre st is issues research interesting graphene of of plethora field the a that that and show experimentally, as br under we main fully also the the graphene, We reviewing of in to some physics. addition review In mesoscopic als and bilayers. layers as is graphene single invariance well in Lorentz as fields spe effects emerging magnetic whole edge an the and of span a that size effect effects is The We to graphene metals. leads infrared. in and ordinary role in adatoms fundamental effects and/or a many-body impurities of ana models of pro the classical The effect analyze problem. also Kondo impurity We Coulomb and the regime. and coupling problem reg strong many-body coupling the e weak in the the discuss states in we quasi-particles systems, invariant these in Lorentz in interactions interactions range electron-electron long of of problem the review We 2018) 22, October (Dated: Nation Physics, of Department 117542 and Singapore Centre Commonwealt Research 590 University, Graphene Boston Physics, of Department Neto In´e Castro Juana H. Sor Madrid, A. de Materiales de Ciencia de Instituto Guinea Nation F. Physics, of Department 117542 and Singapore Centre Research Graphene Pereira M. Urbana-Ch Vitor at Illinois of University Physics, of Department Uchoa Universit Bruno 82 Vermont, of University Physics, of Department Kotov N. Valeri .Conductivity response spin 2. and Charge 1. generation mass Excitonic mass 2. explicit Finite 1. lifetime Quasiparticle 3. analysis Strong-coupling/RPA 2. analysis Weak-coupling 1. 22 22 21 19 19 18 16 16 16 15 13 10 10 10 9 9 7 6 5 4 4 2 to set ftemn-oyproblem many-body the of aspects stood m,adsrnl orltdelectronic correlated strongly and ime, itneo neegn ia iudof liquid Dirac emerging an of xistence rpee trigfo h screening the from Starting graphene. l xiigadvibrant. and exciting ill mag,11 etGenSre,Ubn,Ilni 61801 Illinois Urbana, Street, Green West 1110 ampaign, icse ntecneto finite of context the in discussed o lc,Brigo,Vrot05405 Vermont Burlington, Place, y et ftemn-oypolmin problem many-body the of pects hwta oet naineplays invariance Lorentz that show anoe,bt hoeial and theoretically both open, main el rz3 -84 ard Spain Madrid, E-28049 3, Cruz la de s I.Itrcineet nmssoi systems mesoscopic in effects Interaction VII. oyadcnetosbtenthe between connections and logy tu,fo h lrvoe othe to ultraviolet the from ctrum, I neatosa onaisadltiedefects lattice and boundaries at Interactions VI. eydsusteeet fstrong of effects the discuss iefly vne otn ascuet 21 and 02215 Massachusetts Boston, Avenue, h .Srn orltosi graphene in correlations Strong V. lmo h antcinstability magnetic the of blem s icse naaoywith analogy in discussed lso lUiest fSnaoe cec rv 3, Drive Science 2 Singapore, of University al 3, Drive Science 2 Singapore, of University al .Knoeffect Kondo D. Experiments in Physics Supercritical D. .Mgeimi unu dots quantum in Magnetism A. states Surface A. lattice honeycomb the in gaps Mass A. Problem Coulomb the of Solution Exact A. .Mda ttsadRno ag Fields Gauge Random and States Midgap C. moments magnetic Local C. Interactions Particle Many to Single From C. .Sae tvcnisadcracks and vacancies at States B. instabilities magnetic and Charge B. Screening and Charge Induced B. .RK interaction RKKY E. .Superconductivity F. .Ectn n pnaeu asGeneration Mass Spontaneous and Excitons problem Body 2. Two Interacting 1. Mass Finite ( 3. Coupling Strong ( 2. Coupling Weak Sections 1. Cross Transport and Scattering DOS, 3. Instabilities Supercritical Spectrum and 2. Equations Wave 1. sadPerspectives and us g < g g > g c ) c ) 44 44 43 43 42 42 38 38 36 34 32 32 32 31 31 31 31 30 29 26 26 25 23 23 23 2

B. Charging effects. Coulomb blockade 45 ergy given by

2 VIII. Interactions in strong magnetic fields 46 p K0 = , (1.1) 2m∗ IX. Interactions in bilayers 48 A. Charge polarization 50 where p is the electron momentum and m∗ is a free pa- B. Quasiparticles 51 rameter of the theory that, for lack of a better name, C. Many-body instabilities 52 is called effective mass. Fermi-Dirac statistics implies that electrons carry spin 1/2 and that, in the ground X. Conclusions 53 state, all states with energy below the so-called Fermi energy, EF , are occupied, and all the states above it are Acknowledgments 55 empty. With these two basic assumptions and simple considerations about electron scattering by defects, the References 55 Drude-Sommerfeld model was capable of describing ex- perimental data of several generations of scientists. The understanding of why these two assumptions are I. INTRODUCTION valid for a strongly interacting problem, such as electrons in a metal, had to wait for the development of two major One of the most important problems in theoretical concepts: (i) the band structure theory that explains that physics is the understanding of the properties of quantum the interaction of the electrons with a periodic lattice of systems with an infinitely large number of interacting de- ions produces states that, as the plane waves described by grees of freedom, the so-called many-body problem. In- (1.1), are extended over the entire lattice (Bloch, 1928); teractions are present in almost all areas of physics: soft and (ii) the theory of screening, that is, that metals and hard condensed matter, field theory, atomic physics, are dynamically polarizable materials and that electrons quantum chemistry, nuclear physics, astrophysics, and so act collectively to screen electric fields in their interior on. Interactions between particles are responsible for a (Lindhard, 1954). Hence, long range Coulomb interac- plethora of effects and many-body states, from the band tions become effectively short ranged and weak enough structure of crystals to superconductivity in metals, from to give substance to Drude’s assumptions. In this case the quark-gluon plasma in heavy ion collisions to asymp- the effective mass m∗ reflects the change in the inertia of totic freedom in quantum chromodynamics (QCD). It is the electron as it moves around in an effective medium. the competition between the kinetic energy of the parti- Nevertheless, there are situations when these assump- cles, that is, their inertia, and interactions among them tions fail even in crystalline systems, and that is when that leads to the richness and complexity of these differ- interesting things happen, namely, the free electron pic- ent phases. For these reasons, many-body interactions ture breaks down. are very specific, and the hardest to describe theoreti- In fact, there are many instances where the Fermi liq- cally. uid ground state becomes unstable. Electrons not only One of the greatest theoretical achievements of the interact with static ions but also with their vibrations, last century, the Landau theory of the Fermi liquid the phonons. Electron-phonon interactions, in the pres- (Baym and Pethick, 1991), asserts something very sim- ence of strong screening, can lead to an effective at- ple but, at the same time, very deep: that the excita- tractive interaction between electrons producing a catas- tions of a large (indeed, infinite) collection of strongly trophic Fermi surface instability towards a superconduct- interacting particles can be described as an equally large ing ground state (Tinkham, 1996). Fermi surface insta- collection of weakly-interacting quasi-particles that carry bilities also happen in special situations in the presence the same quantum numbers as the original particles. This of Fermi surface nesting which can lead to charge and statement is far from trivial. Consider, for instance, the spin density wave ground states (Gruner, 1994). Crys- behavior of electrons in a metal. The electrons inter- tals with inner shell electrons, such as transition metals, act among themselves and with the ions in the crystal can also have many-body instabilities due to the strong via strong long-range Coulomb interactions. It is not local interactions between the electrons, leading to insu- at all clear what is the outcome of this complex inter- lating states with magnetic properties as in the case of acting problem. Without having any deep theoretical re- Mott insulators (Mott, 1949). Another important case of sources to treat this problem, except an extraordinary in- Fermi liquid breakdown is when the electron density is tuition, visionaries like Paul Drude (Drude, 1900a,b) and very low and the screening disappears. Arnold Sommerfeld (Hoddeson et al., 1987) settled the Notice that in quantum mechanics the momentum of foundations for the understanding of this complex prob- the particle relates to its wavelength, λ, by p = ~/λ lem by postulating, shamelessly, that (1) electrons prop- and hence the kinetic energy (1.1) behaves as K = ~2 2 agate freely in a non-relativistic (Galilean invariant) way /(2m∗λ ). If the average distance between electrons is (Drude’s contribution), and (2) electrons obey Fermi- ℓ we see that the average kinetic energy per electron has ~2 2/d d Dirac statistics (Sommerfeld’s contribution). Galilean to be of the order EK nd (2m∗) where nd =1/ℓ is invariance dictates that the electrons have a kinetic en- the average electron density≈ in d spatial dimensions. On 3 the other hand, the Coulomb interaction is given by: graphene’s electron density as n. As the electronic prop- erties of graphene are sensitive to environmental condi- e2 tions, they will be modified by the presence of other lay- V (r)= , (1.2) ǫ0r ers. In fact, as we are going to show, bilayer graphene has properties which are rather different than its mono- where e is the electron charge, and ǫ0 the dielectric con- layer counterpart. Furthermore, due to the same peculiar stant of the medium. Notice that the Coulomb energy per dispersion relation, the electronic density of states, ρ(E), 2 1/d 2 electron is of the order EC e n /ǫ0. Thus, the ratio vanishes at the Dirac point, ρ(E) E /v , and hence ≈ d ∝ | | F of Coulomb to kinetic energy is given by rs = EC /EK graphene is a hybrid between an insulator and a metal: 1/d 2 2 d ∝ (n0/nd) , where n0 = [m∗e /(~ ǫ0)] depends only on neutral graphene is not a metal because it has vanishing material properties. Therefore, at high electron densities, density of states at the Fermi energy, and it is not an n n , the kinetic energy dominates over the Coulomb insulator because it does not have a gap in the spectrum. d ≫ 0 energy, which can be disregarded, and the Fermi liq- This means that pristine (or lightly doped) graphene can- uid description is safe. At low densities, n n the not screen the long range Coulomb interaction in the d ≪ 0 Coulomb energy is dominant and new electronic phases, usual (metallic) way, although it is possible to produce such as ferromagnetism and Wigner crystallization, can electronic excitations at vanishingly small energy. This become stable (Ceperley, 1978). Therefore, the rela- state of affairs makes of graphene a unique system from tive strength of the kinetic to Coulomb interactions in the point of view of electron-electron interactions. The Galilean invariant systems is completely controlled by long-range interactions lead to non-trivial renormaliza- the electron density. Notice that in all the cases dis- tion of the Dirac quasiparticle characteristics near the cussed above the Galilean invariance was kept intact and charge neutrality point, and the resulting electronic state the driving force for the many-body instabilities was the can be called Dirac liquid, to be distinguished from the enhancement of the Coulomb relative to the kinetic en- Fermi liquid behavior at finite chemical potential (away ergy. from the Dirac point, where conventional screening takes With the advent of graphene (Novoselov et al., 2004a), place.) a two dimensional crystal of pure carbon, this picture has The unusual relation between kinetic and Coulomb en- changed and a new example of Fermi liquid breakdown ergies not only affects the electron-electron interactions, has emerged in a big way. In graphene, due to its peculiar but also the interactions of the electrons with charged lattice structure, the electrons at the Fermi energy are impurities, the so-called Coulomb impurity problem. In described in terms of an effective Lorentz invariant theory a metal described by a Galilean invariant theory of the where the kinetic energy is given by the Dirac dispersion form (1.1), screening also makes the interaction with the (Castro Neto et al., 2009a) impurity short ranged, and hence the scattering problem effectively reduces to the one of a short range impurity. K = v p , (1.3) G ± F | | In graphene, because of the lack of screening the situ- ation is rather different, and one has to face the prob- where v is the Fermi-Dirac velocity, and the signs F lem of the effect of the long range part of the poten- refer to two linearly dispersing bands. If we take± (1.3) at tial. Scattering by long range interactions has a long face value and reconsider the argument given above on history in physics and it leads to the issue of logarith- the relevance of the Coulomb interactions we reach very mic phase shifts (Baym, 1969). In graphene, because different conclusions. For one, the form of the Coulomb of its emergent Lorentz invariance, this issue is magni- interaction remains the same as in (1.2), since v is a ma- F fied. Since Coulomb interactions between electrons and terial’s property and hence much smaller than the speed electron scattering by Coulomb impurities are closely re- of light, c. This means that the photons which mediate lated issues, one expects that many of the anomalies of the Coulomb interaction are still much faster than the one problem are also reflected in the other. electrons and, thus, the electron-electron interaction can be considered as instantaneous. Therefore, the Coulomb Another interesting consequence of the scaling of the interaction (1.2) actually breaks the Lorentz invariance kinetic energy with momentum is related to the issue of of (1.3). Secondly, because of the linear scaling of the ki- electron confinement. If electrons are confined to a re- netic energy with momentum, we see that the average ki- gion of size L the energy of the states is quantized, no 1/2 netic energy per electron has to scale like EG ~vF n matter whether the electrons obey Galilean or Lorentz and consequently the ratio of Coulomb to kinetic≈ energy invariance. However, the quantization of energy is rather is given by different in these two cases. In a Galilean invariant sys- tem, like the one described by (1.1) the energy levels 2 2 EC e are spaced as ∆E0 1/L while in graphene Lorentz α = = , (1.4) invariance, (1.3), implies∝ ∆E 1/L. Hence, the size EG ǫ0~vF G dependence of the energy levels in∝ sufficiently small sam- and is independent of the electronic density n, depend- ples of graphene is rather different than one would find ing only on material properties and environmental con- in normal metals. Moreover, since the Coulomb energy ditions, such as ǫ0. Here, and from now on, we refer to scales like 1/L we expect Coulomb effects to be stronger 4 in nanoscopic and mesoscopic graphene samples. a) b)

4 −

Furthermore, the fact that graphene is a two dimen- 0 sional (2D) system has strong consequences for elec- −4 − tronic motion in the presence of perpendicular magnetic Energy (eV) fields. Since a perpendicular magnetic field B leads to Γ M K Γ a quantization of the energy in terms of Landau lev- els, and the electrons cannot propagate along the di- rection of the field, its effect is singular, in the sense c) d) E that the problem has a massive degeneracy. So, strong M magnetic fields can completely quench the kinetic energy K K of the electrons that become dispersionless. The elec- tronic orbits are localized in a region of the size of the Γ magnetic length: ℓB = ~c/(eB). For a Galilean in- variant system, such as the one described by (1.1), for K ’ p p ~/ℓB the kinetic energy per electron is of order K ≈ ~ω B where ω = ~/(m ℓ2 ) is the cyclotron C C ∗ B FIG. 1 (Color online) a) Honeycomb lattice with the two frequency.≈ ∝ On the other hand, for graphene, using (1.3), sublattices in graphene. The red arrows are nearest neighbor one has E ~ω √B where ω = √2v /ℓ , which ∗ G ≈ G ∝ G F B vectors. b) Tight-binding spectrum for the π π bands. The is a consequence of the Lorentz invariance. Notice that horizontal line intersecting the K point corresponds− to the in both cases the Coulomb energy per electron scales like Fermi level at half-filling. c) Brillouin zone centered around 2 EC e /(ǫ0ℓB) √B. Hence, in a Galilean invariant the Γ point. d) Dirac cone resulting from the linearization of system∝ the Coulomb∝ energy is smaller than the kinetic the tight-binding spectrum around the K points (blue circles). energy at high fields while for Lorentz invariant systems they are always comparable. Thus, one expects Coulomb interactions to be hugely enhanced in the presence of II. CHARGE POLARIZATION AND LINEAR SCREENING these magnetic fields. In the 2D electron gas (2DEG) this unusual state of affairs is what leads to the fractional A. Tight-binding spectrum quantum Hall effect (FQHE) (Laughlin, 1983). In isolated form, carbon has six electrons in the orbital configuration 1s22s22p2. When arranged in the honey- comb crystal shown in Fig.1(a), two electrons remain in Given all these unusual circumstances, many questions the core 1s orbital, while the other orbitals hybridize, come to mind: How does screening of the long range 2 2 forming three sp bonds and one pz orbital. The sp Coulomb interaction work in graphene? Can graphene orbitals form the σ band, which contains three localized be described in terms a Lorentz invariant theory of quasi- electrons. The bonding configuration among the pz or- particles? Is the Coulomb impurity problem in graphene bitals of different lattice sites generates a valence band, or the same as in a normal metal? In what circumstances π-band, containing one electron, whereas the antibonding is graphene unstable towards many-body ground states? configuration generates the conduction band (π∗), which Are there quantum phase transitions (Sachdev, 1999) in is empty. the phase diagram of graphene? Do magnetic moments From a kinetic energy point of view, the electronic sin- form in graphene in the same way as they do in normal gle particle dispersion in graphene is essentially defined metals? What is the ground state of graphene in high by the hopping of the electrons between nearest neighbor magnetic fields? carbon sites in the honeycomb lattice. Unlike square or triangular lattices, the honeycomb lattice is spanned by two different sets of Bravais lattice generators, forming The objective of this review is not to cover the basic a two component basis with one set for each triangular aspects of graphene physics, since this was already cov- sublattice. Defining a label for electrons sitting in each ered in a recent review (Castro Neto et al., 2009a), but of the two sublattices, say A and B, the free hopping to try to address some of these questions while keeping Hamiltonian of graphene is others open. The field of many-body physics will always be an open field because a seemingly simple question al- 0 = t a† (Ri)bσ(Rj ) + h.c. µ nˆσ(Ri), H − σ − ways leads to another question even more profound and σ, ij σ,i Xh i X harder to answer in a definitive way. In many ways, what   (2.1) we have done here is to only scratch the surface of this where aσ(Ri), bσ(Ri) are fermionic operators for sublat- rich and important field, and leave open a large number tices A and B respectively,n ˆσ(Ri) is the number op- of interesting and unexplored problems. erator, σ = , labels the spin and ij means summa- ↑ ↓ h i 5 tion over nearest neighbors. The two energy scales in around the valleys centered at K gives rise to an the Hamiltonian are t 2.8 eV, which is the hopping effective low energy description± of the electrons that energy between nearest≈ carbons, and µ, the chemical mimics the spectrum of massless Dirac particles. In potential away from half-filling [see Fig.1(b)]. In a ho- this effective theory, the elementary excitations around mogeneous system, deviations from half-filling (µ = 0) the Fermi surface are described by a Dirac Hamiltonian are routinely induced either by charge transfer from a (Semenoff, 1984), substrate (Giovannetti et al., 2008), by application of a back gate voltage (Novoselov et al., 2005, 2004a,b), or else by chemical doping (Calandra and Mauri, = Ψ† [vk γ µτ σ ]Ψk , (2.6) H0 kσ · − 0 ⊗ 0 σ 2007; Gr¨uneis et al., 2009; McChesney et al., 2010; σk Uchoa et al., 2008b). X In momentum space the free Hamiltonian of graphene where is

Ψkσ = (aK+k,σ,bK+k,σ,b K+k,σ,a K+k,σ) (2.7) µ tφp − − 0 = Ψp† ,σ − − Ψp,σ , (2.2) H tφp∗ µ p,σ − − X   is a four component spinor for sublattice and valley de- grees of freedom. In this representation, γi = τ3 σi where Ψp,σ = (ap,σ,bp,σ) is a two component spinor and , where τ and σ are the usual Pauli matrices, which⊗ 3 operate in the valley and sublattice spaces respectively ip ai (i =1, 2, 3 correspond to x, y and z directions, and τ =1 φp = e · (2.3) 0 i=1 and σ0 = 1 are identity matrices). The form of the spec- X trum mimics the relativistic cone for massless fermions is a tight-binding function summed over the nearest (Wallace, 1947), neighbor vectors E (k)= v k µ (2.8) a √3 a √3 ± ± | |− a1 = ax,ˆ a2 = xˆ + a y,ˆ a3 = xˆ a yˆ , − 2 2 − 2 − 2 where the Fermi velocity v = (3/2)ta 6eVA˚ is nearly (2.4) 300 times smaller than the speed of light,≈ i.e. v 1 where a 1.42A˚ is the carbon-carbon spacing. The di- 106m/s. From now on we set ~ = k = 1 everywhere,≈ × agonalization≈ of Hamiltonian (2.2) yields the spectrum B except where it is needed. For simplicity of notation, we of the two π-bands of graphene in tight-binding approx- call the Fermi velocity v (i.e. v v) throughout this imation (Wallace, 1947), F review. ≡

E (p)= t φp µ . (2.5) The Hamiltonian (2.6) is invariant under a pseudo- ± ± | |− time reversal symmetry operation, = i(τ0 σ2) , 1 S ⊗ C H − = H, ( is the complex conjugation operator), The +( ) sign in the spectrum corresponds to the con- whichS S is equivalentC to a time reversal operation for each duction− (valence) band. valley separately. It is also invariant under a true time The hexagonal Brillouin zone (BZ) of graphene shown reversal symmetry (TRS) operation, which involves an in Fig.1(c) has three high symmetry points: the Γ point, additional exchange between the valleys, = (τ σ ) . T 1 ⊗ 1 C located at the center of the BZ, the M point, which In the absence of back scattering connecting the two indicates the position of the Van Hove singularities of valleys, the Hamiltonian can be decomposed in two in- the π-π∗ bands, where the density of states (DOS) is dependent valley species of Dirac fermions with opposite logarithmically divergent, and the K points, where the chiralities: π-bands touch, and the DOS vanishes linearly. An ex- tensive description of the band structure of graphene = Ψ† [vk σ µ]Ψ k , (2.9) H0,+ +,kσ · − +, σ and its electronic properties is reviewed in detail by σ,k Castro Neto et al., 2009a. X 0, = Ψ† ,kσ [ vk σ∗ µ]Ψ ,kσ, (2.10) H − − − · − − k Xσ, B. Dirac fermion Hamiltonian where Ψ ,kσ = (a K+k,σ,b K+k,σ) are two component The topology of the Fermi surface in undoped spinors.± In this review,± unless± otherwise specified, we graphene is defined by the six K points where the con- will arbitrarily choose one of the two cones and assume duction and valence bands touch, E (K) = φK = 0. an additional valley degeneracy in the Hamiltonian. So These special points form two sets± of nonequivalent±| | valley indexes will be generically omitted unless explicitly points, K and K′, with K = K′ and K =4π/(3√3a), mentioned. A more detailed description of the symmetry which cannot be connected− by the generators| | of the properties of the graphene Hamiltonian can be found in reciprocal lattice. The linearization of the spectrum (Gusynin et al., 2007). 6

FIG. 2 Diagram for¡ the polarization bubble corresponding to eq. (2.12).

C. Polarization function

The Green’s function of graphene is a 2 2 matrix represented in the sublattice basis by ×

G G Gˆ(k, τ)= aa ab , Gba Gbb   where Gaa = T [ak(τ)ak† (0)] and so on, with τ as the imaginary time.−h In the low energyi sector of the spectrum, close to the Dirac points, the non-interacting Green’s 1 function is Gˆ(0)(k,iω) = [iω + µ vk σ]− , or equiv- alently, in a chiral representation, − ·

1 1+ sσˆ Gˆ(0)(k,iω)= k , (2.11) 2 iω + µ sv k s= X± − | | whereσ ˆk = σ k/ k is twice the quantum mechanical helicity operator· for| a| Dirac fermion with momentum k, and s = labels the two branches with positive and negative energy± in one cone. It is clear that the positive and negative branches within the same cone have also opposite helicities. The polarization function in one loop is calculated di- FIG. 3 (Color online) Polarization bubble Π(1)(q,ω) for rectly from the bubble diagram shown in Fig. 2, graphene, within the Dirac approximation. Panels (a) and (b) show, respectively, a density plot of the real and imagi- (1) Π (q,iω)= N s,s′ (p, q) nary parts of the polarization bubble, Π(1)(q,ω), defined in p ′ F × X Xs,s eq. (2.12), and normalized to the DOS at the Fermi level, f[Es′ (p + q)] f[Es(p)] ρ(µ). Panels (c) and (d) present constant frequency cuts at − , (2.12) ω/µ = 0.5, 1.0, 1.5, 2.0, 2.5, 3.0. In panel (e) we show the Es′ (p + q) Es(p) iω − − static limit, Π(1)(q, 0), whose closed form expression is writ- 1 ten in eq. (2.16). Notice the transition from a constant value where f(E)= eE/T +1 − is the Dirac-Fermi distribu- (q < 2kF ) to the linear in q dependence at large momenta. tion, with T as temperature, N = 4 is the degeneracy for The derivative of the polarization is shown in the same panel,
two spins and two valleys, and and can be seen to vary continuously. In (f) we plot the real (black/solid) and imaginary (red/dashed) parts of the uni- 1 form limit (2.17). ′ (p, q)= tr(1 + ss′σˆpσˆp q) (2.13) Fs,s 4 + are the matrix elements due to the overlap of wavefunc- the screening of charge is completely suppressed, and the tions for intraband (s = s′) and interband (s = s′) tran- − polarization function describes the susceptibility of the sitions. ’tr’ means trace over the sublattice indexes. In vacuum to particle-hole pair production, exactly as in a more explicit form, s,s′ (p, q) = [1 + ss′ cos θp,p+q] /2, the diagonal time component of the polarization tensor where θ is the angle betweenF p and p + q. The full mo- in massless Quantum Electrodynamics (QED), QED2+1 mentum, frequency, and chemical potential dependence (Appelquist et al., 1988; Gonz´alez et al., 1994; Pisarski, of (2.12) is shown in panels (a-d) of Fig. 3. 1984), In metals, screening is a many-body property directly related to the polarizability of the electrons around the 1 q2 Π(1)(q,ω)= . (2.14) Fermi surface. In graphene, because the density of −4 v2q2 ω2 states (DOS) vanishes linearly around the Dirac points, − ρ(E) E µ /v2, exactly at the neutrality point (µ = 0) Here we have performed a Wickp rotation to real frequen- ∝ | − | 7 cies, iω ω +0+. Since the Fermi surface in this case for ω < vq, which is due only to intraband transitions. is just a→ point, there is no phase space for intraband ex- The polarization function in graphene is a regular func- citations at zero temperature due to the Pauli principle. tion everywhere except at ω = vq, where it has an on- The process of creation of particle-hole pairs involves in- shell singularity delimiting the| | border of the particle-hole coherent excitations of electrons from the lower to the continuum. upper band. The continuum of particle-hole excitations The polarization was derived originally by Shung, is well defined for all virtual transitions with ω >vq. 1986a and later rederived by a number of authors (Ando, For finite µ there is a crossover in the behavior of the 2006; Barlas et al., 2007; Hwang and Das Sarma, 2007; polarization function. The DOS around the Fermi level is Wunsch et al., 2007). These results rely on the cone ap- finite and the intraband excitations dominate the infrared proximation, which ignores contributions coming from behavior of the polarization. For vq µ and ω µ , ≪ | | ≪ | | the non linear part of the spectrum. In addition, the band the leading term in the polarization function is (Shung, width is assumed to be infinite. Although the charge po- 1986a) larization for Dirac fermions in 2D is well behaved and does not require cut-off regularization in the ultraviolet, (1) 2 µ ω the physical cut-off of the band, D, generates small cor- Π (q,ω) | 2| 1 . (2.15) ≈− πv − ω2 v2q2 ! rections that vanish only in the D limit. In this − sense, the ‘exact’ expression for the→ static ∞ polarization As in a Fermi liquid, there isp a particle-hole continuum function (ω = 0) for arbitrary momentum is

2 (1) 2kF q 2kF 2kF 1 2kF π Π (q, 0) = + θ(q 2k ) 1 + sin− , (2.16) − πv − F 2πv  q − q q − 2  s      

where kF = µ /v is the Fermi momentum, and θ(x) interband transitions is ω > 2 µ vq for q < 2kF , as is a step function.| | The static polarization is plotted in shown schematically in Fig. 5. | |− Fig. 3(e). At q 2k the static polarization exhibits a crossover ≈ F from a two dimensional electron gas (2DEG) to Dirac D. Collective modes and screening fermion behavior. For details of the polarization func- tion in the 2DEG please refer to Fig. 4. As in the 2DEG, The Coulomb interaction among the electrons in the polarization of graphene is constant for q < 2kF . graphene gives rise to collective modes and metallic For q > 2kF , it eventually becomes linear in q for large screening when the Fermi level is shifted away from the momenta. At the crossover, the static polarization and Dirac points. In a 2D system, the bare Coulomb interac- its first derivative are continuous at q = 2kF . The dis- tion is given by continuity only appears in the second derivative. This is distinct from the 2DEG case, where the first derivative is 2πe2 V (q)= , (2.18) discontinuous. The difference will affect the spacial de- ǫ q pendence of the Friedel oscillations in the two systems. 0 In the opposite limit, for arbitrary ω and q 0, the where e is the charge of the electron and ǫ is the effec- → 0 polarization function becomes tive dielectric constant of the medium. For graphene in contact with air and a substrate with dielectric constant q2 2 µ 1 2 µ ω κ, ǫ = (1+ κ)/2. In most of the experiments, graphene Π(1)(q 0,ω)= | | + ln | |− , 0 → 2πω ω 2 2 µ + ω lies on top of some substrate like SiO2 or SiC, where di-    | | (2.17) electric effects are moderate (for instance, the dielectric which is shown in Fig. 3(f). The presence of a pocket constant of SiO2 is κ 4). The background dielectric ≈ of electrons (holes) around the Dirac points opens a gap constant can be significantly enhanced in the presence of in the particle hole continuum for interband excitations substrates in contact with strong dielectric liquids such (ω > vq). From Eq. (2.17), it is clear that the imagi- as ethanol (κ 25) or water (κ 80) (Jang et al., 2008; ≈ ≈ nary part of the polarization function at small momen- Ponomarenko et al., 2009). tum is zero unless ω > 2 µ [Fig. 3(b)]. This is so be- As usual, the collective modes follow from the zeros of cause the phase space for| vertical| interband excitations the dielectric function is Pauli blocked for ω < 2 µ , generating a gap for optical absorption in the infrared.| | At finite q, the threshold for ǫ(q,ω)= ǫ [1 V (q)Π(1)(q,ω)] , (2.19) 0 − 8

ω ω

interband 2µ

intraband

q q 2kF

FIG. 5 (Color online) Colored regions represent the particle- hole continuum of graphene due to interband (gray area) and intraband (green) transitions. On the left: half-filled case; right: finite µ case, away from half filling. Dashed line: acous- tic plasmon for the single layer (ωp √µq). ∝

tric constant (Shung, 1986a),

qe2 2 µ ω (q) ǫ (q) ǫ ln | |− p . (2.21) 0 ≈ 0 − 2ω (q) 2 µ + ω (q) p  | | p  As in the 2DEG, the screened Coulomb interaction for q< 2kF is V (q) 1 2πe2 = (2.22) ǫ(q, 0) ǫ0 q + qT F

2 where qT F =4πe kF /(vǫ0) is the Thomas-Fermi momen- tum (kF = µ /v), which sets the size of the screening cloud. In the| | presence of an external charged impurity Ze, the induced charge, δZ, has a non-oscillatory com- FIG. 4 (Color online) Polarization bubble Π(1)(q,ω) for the ponent coming from the q 0 limit of the polarization conventional 2DEG. Panels (a) and (b) show, respectively, a 3 1 → that decays as (kF r )− (as in a 2DEG), and an oscilla- density plot of the real and imaginary parts of the polarization tory part which corresponds to the Friedel oscillations at bubble, Π(1)(q,ω), normalized to the DOS at the Fermi level. q = 2kF . The Friedel oscillations in graphene decay as Panels (c) and (d) present constant frequency cuts at ω/µ = 3 cos(2kF r)/(kF r ), differently from the 2DEG case, where 0.5, 1.0, 1.5, 2.0, 2.5, 3.0. In panel (e) we plot the static limit, 2 Π(1)(q, 0), and in (f) the uniform limit, Π(1)(0,ω). the decay is of the form cos(2kF r)/r . The difference is caused by the fact that the static polarization function in the 2DEG has a cusp at q = 2kF , whereas in graphene, the first derivative is continuous [cfr. Figs. 3(e) and 4(e)]. (1) 2 calculated here in the Random Phase Approximation For undoped graphene, V (q)Π = (π/2)[e /(vǫ0)] − (RPA). Since graphene is a 2D system, the collective [see Eq. (2.14)], and the static dielectric function is a plasmon mode is gapless. The leading term in the po- constant. The effective Coulomb interaction in this case larization for small frequency and momenta (compared is to k ) is shown in Eq. (2.15). From it one can easily F V (q) 1 2πe2 extract the infrared dependence of the plasmon, = , (2.23) ǫ(q, 0) ǫRPA q

2 2 where ǫRPA = ǫ0 + (π/2)(e /v) is the effective back- ωp(q)= (2µe /ǫ0)q , (2.20) ground dielectric constant, renormalized by the inter- p band transitions. Additional many body effects resulting which follows the same dispersion as the plasmon en- from self-energy insertions in the bubbles logarithmically countered in the 2DEG. The √q dependence of the plas- renormalize this correction to zero in the q 0 limit, mon was recently confirmed by a high resolution en- as will be clear in Sec. III of this review. On→ the dy- ergy loss spectroscopy (EELS) measurement in graphene namical side, inserting Eq. (2.14) into Eq. (2.19), one (Liu et al., 2008). Additional corrections due to the in- can easily see that no collective modes are allowed in un- terband excitations (which are absent in the 2DEG) can doped graphene, at zero temperature, within the RPA be absorbed into the definition of the background dielec- framework. At half-filling, RPA is justified in the limit 9 of large number of fermionic species, N, which favors di- with k = (q, kz), q is an in-plane momentum, and agrams with maximal number of bubbles at each order of (Fetter, 1974) perturbation theory. In graphene, the physical number of species is N = 4, and additional corrections beyond sinh(qd) S(q, kz)= (2.28) RPA coming from the exciton channel near the on-shell cosh(qd) cos(kzd) singularity of the bubble, ω vq, were shown to gener- − ate a new acoustic plasmon| mode|∼ (Gangadharaiah et al., is the structure factor for a stack with an infinite num- 2008). In the static limit (ω 0), vertex corrections in ber of layers. In the limit when the distance between the the bubble are perturbatively→ small and RPA can be jus- layers d is small, Eq. (2.27) recovers the isotropic case 2 2 2 tified in the calculation of the dielectric function even at V (k)=4π(e /ǫ0)/(q +kz ), whereas in the opposite limit 2 et al. (d ) one gets the 2D case, V (k)=2π d (e /ǫ0)/q. half-filling (Kotov , 2008b). The structure of per- → ∞ turbation theory in graphene will be discussed in detail In any case, the polarization function must be integrated in Sec. III. over a cylindrical Fermi surface of height 2π/d, and so (1) In addition to the low energy acoustic mode due to Π (q,ω) acquires an additional factor of 1/d compared intraband transitions, graphene has also two high en- to the single layer case. The extension of this problem to ergy optical plasmons generated by interband excitations include the interlayer hopping dispersion in the polariza- around the Van-Hove singularities of the π π∗ bands, tion was considered by Guinea, 2007. − Away from the neutrality point (µ = 0), instead of and also by optical transitions between σ π∗ and π σ∗ 6 bands (Eberlein et al., 2008; Kramberger− et al., 2008− ). a single acoustic mode as in the monolayer, the ze- The measured optical gaps of the π and π σ band roes in the dielectric function of the multilayer gener- plasmons in graphene are 4.5 eV and 15 eV, respectively.− ate a plasmon band, where the modes are labeled by k [ π/d,π/d]. For q 1/d, the plasmon dispersion Similar modes were also observed in graphite, where they z ∈ − ≪ appear blue shifted to 7eV and 24 eV respectively, ac- is (Shung, 1986a) cording to optical data (Taft and Philipp, 1965), X-ray 2 measurements (Shulke et al., 1988), and ab-initio calcu- 2 2µe ωp(q, kz)= qS(q, kz). (2.29) lations (Marinopoulus et al., 2004). ǫ0

In the kz = 0 mode, the charge fluctuations be- E. Infinite stack of layers tween different layers are in-phase, and the result- 2 2 ing plasmon mode is optical, ωp(q, 0) (4µe /ǫ0d) + 3 2 ≈ In the case of an infinite stack of graphene layers, the 4 (vq) . For ωp(q) > 2µ, this mode is damped by Hamiltonian term for the Coulomb interaction among all the particle-hole continuum due to interband transitions the electrons can be written in real space as (see Fig. 6), in agreement with energy loss spectroscopy data (Laitenberger and Palmer, 1996). The out-of-phase 2 e 3 3 1 modes (for k = 0) are acoustic. At the edge of the plas- C = d rd r′ nˆ(r) nˆ(r′) , (2.24) z 6 H ǫ0 r r′ mon band, the mode k = π/d disperses linearly with Z | − | z ± the in-plane momentum, ω (q, π/d) = µe2d/ǫ q, in wheren ˆ(r) is the 3D particle density operator. In the p ± 0 contrast with the 2DEG dispersion (ωp √q) present absence of interlayer hopping, as in the case for exam- ∝p ple of several graphite intercalated compounds, the elec- in the single layer. Except for the lack of an inter- trons remain confined in each layer, but the unscreened band particle-hole continuum and the associated damp- Coulomb lines fill the entire space in between the layers, ing, similar plasmon band features are also expected in coupling all the electrons in the system. In that case we the 2D layered electron gas, for fermions with quadratic may constrain the local density operatorn ˆ to be in the dispersion (Hawrylak, 1987). form (Visscher and Falikov, 1970)

∞ F. f-sum rule nˆ(r) d nˆ(r)δ(z ld) (2.25) → − l= X−∞ The f-sum rule is a generic statement about conser- where l is an integer labeling the layers, and d is the vation of the number of particles and results from the distance between layers. In momentum space, making analytical properties of the retarded charge susceptibil- a discrete sum over the layers, the Coulomb interaction ity. It can be generically defined as (Nozi`eres, 1964) between all the electrons is ∞ R 2 dωωImχ (k,ω)= π [[ , nˆ( k)] , nˆ(k)] , (2.30) e 3 h H − i C = d k nˆ( k)V (k)ˆn(k) , (2.26) Z−∞ H ǫ0 − Z where is the Hamiltonian,n ˆ is the particle density op- where erator,HχR is a retarded charge susceptibility, χ(k, τ) = e2 T [ˆn(k, τ)ˆn( k, 0)] , and ... is an expectation value cal- V (k)=2π d S(q, kz) (2.27) h − i h i ǫ0q culated in some basis. 10

ω







































for a Dirac Hamiltonian, where D is the ultraviolet cut-



























































off. A similar dependence with the cut-off also occurs







































ω







































in the true 3D relativistic problem, where the sum rule p















































































reflects the number of particles contained in the vacuum















































































of the theory, which is formally divergent (Ceni, 2001;















































































Goldman and Drake, 1982). In graphene, as in any two



























































band semi-metal or semiconductor, the validity of the f-



















Intraband



























































sum rule is physically recovered when the periodicity of



























































q the electronic spectrum is restored back into the Hamil- 2kF tonian. FIG. 6 (Color online) Plasmon band (hatched region) for an infinite stack of graphene layers. Red line: optical mode III. QUASIPARTICLES IN GRAPHENE kz = 0. Dashed line: acoustic mode kz = π/d, ωp √µq, with linear dispersion, at the edge of the band. All the∝ other modes in between are acoustic. Adapted from Shung, 1986a. The quasiparticle properties of graphene are modi- fied by the presence of long-range Coulomb interactions. Their effects are especially pronounced when the Fermi As in any solid, the exact electronic Hamiltonian of energy is close to the Dirac point (µ 0), and can re- graphene can be decomposed into a Hamiltonian of free sult in strong renormalization of the Dirac≈ band struc- electrons, plus a periodic potential due to the lattice, and ture (the Fermi velocity v), and the quasiparticle residue interactions. If the interactions depend only on densities, (Z). Consequently, many physical characteristics, such the commutators in Eq. (2.30) can be calculated exactly, as the compressibility, spin susceptibility and the spe- and the only term that survives is the Kinetic energy due cific heat can be strongly affected by interactions. Even to the free electrons, when the Fermi surface is large and the system is a Fermi k2 liquid, there are strong modifications of the physics near [[ , nˆ( k)] , nˆ(k)] = Ne , (2.31) the Dirac point due to the presence of additional peaks in h H − i m the quasiparticle decay rate, related to plasmon-mediated where m is the bare electron mass and Ne is the number decay channels. Even reconstruction of the Dirac cone of fermions in the band. Choosing, for example, a basis structure near the charge neutrality point appears possi- of non-interacting fermions, the sum rule in graphene is ble, as indicated by recent Angle-Resolved Photoemission 2 Spectroscopy (ARPES) measurements. All these effects ∞ Nek dωωImΠ(1)(k,ω)= π , (2.32) are sensitive to the value of the Coulomb interaction con- m Z−∞ stant in graphene, α. as in metals, where Π(1)(k,ω) is the bare polarization bubble, calculated using the full non-interacting spec- trum (dictated by the lattice symmetry). The validity A. Low-energy behavior near the Dirac point of the f sum-rule does not require Galilean invariance of the quasiparticles, but of the free electrons, which are not 1. Weak-coupling analysis relativistic and hence obey the Schrodinger equation. For low energy effective Hamiltonians, such as the The interaction parameter which characterizes the Dirac Hamiltonian in graphene (which do not include strength of the Coulomb interaction in graphene is the periodicity of the spectrum in the Brillouin zone), (Eq. (1.4)) the f-sum rule above is still formally satisfied when ap- e2 plied for the electrons (holes) in the conduction (valence) α = . (3.1) band only, as can be explicitly checked by direct substi- ǫ0v tution of the polarization due to intraband transitions, At k = 0 screening is absent, and the interaction po- Eq. (2.15), into Eq. (2.32). The number of electrons F 2 2 tential in momentum space: (holes) in this band, Ne = kF A/π, where A = 3√3a /2 is the unit cell area, is set by the size of the Fermi surface, 2πe2 and the verification of the sum rule follows as in a Fermi V (p)= . (3.2) ǫ p liquid. 0 The Dirac Hamiltonian, however, violates the f-sum The value of α = 2.2/ǫ0 depends on the dielectric envi- rule (2.32) when interband transitions are taken into ac- ronment since, as previously discussed, ǫ0 = (1+ κ)/2 count. In that case, the left hand side of Eq. (2.32) be- for graphene in contact with air and a substrate with comes independent of the chemical potential, consistent dielectric constant κ. In vacuum, α =2.2. with the fact that (Sabio et al., 2008) In the case of small coupling, α 1, we can employ ≪ D standard perturbation theory, involving the perturbative [[ , nˆ( k)] , nˆ(k)] = k2 (2.33) h H − i 4 computation of the self-energy Σ(k,ω), which enters in a 11

a) b) A) B) ¡

c) ¡ d) FIG. 8 (a) Self-energy and (b) Vertex corrections to the po-

larization bubble ¡

¡ While the linearity of the spectrum in graphene was FIG. 7 Self-energy diagrams: (a) First order Hartree-Fock, realized a long time ago (Wallace, 1947), in the con- (b) Second order loop diagram (first diagram in the RPA se- text of studying graphite formed by layers of graphene, ries), (c) Second order exchange (vertex correction) diagram, the self-energy correction Eq. (3.8) due to interac- (d) Rainbow diagram. tions was first investigated perturbatively much later by Gonz´alez et al., 1994. The non-trivial velocity standard way the Dirac fermion Green’s function (GF), renormalization is due to the unscreened, long-range for a given valley: Coulomb interactions. Similar logarithmic divergencies were also found in gapless 3D semiconductors, where 1 the Dirac spectrum originated from special symmetries k G( ,ω)= + . ωσ0 vσ k Σ(k,ω)+ iσ00 sign(ω) (Abrikosov and Beneslavskii, 1971). − · − (3.3) The above calculation forms the basis of the Renor- malization Group (RG) analysis. In the RG spirit one It is convenient to decompose the self-energy into two integrates out the high momentum degrees of freedom, pieces with different pseudo-spin structure i.e. regions of momenta Λ > p > Λ1, and the results vary with the quantity ln(Λ/Λ |) | l. Here we denote by l 1 ≡ Σ(k,ω)=Σ0(k,ω)+Σv(k,ω), Σ0 σ0, Σv σ k, the RG parameter, so that the infrared limit corresponds ∝ ∝ · (3.4) to l (i.e. one integrates down to the infrared scale → ∞ where σ0 = 1 is the unit matrix, which from now on will k 0, l = ln(Λ/k)). From Eq. (3.9) we obtain not be written explicitly. Then we have → dv α e2 Z = v = . (3.10) G(k,ω)= , (3.5) dl 4 4ǫ0 ω Z(vσ k +Σv) − · This equation has to be supplemented with an additional where Z is the quasiparticle residue equation reflecting the absence of charge (e2) renormal-

1 ization: Z− =1 ∂Σ0/∂ω , (3.6) − de2 =0 . (3.11) and Σv is responsible solely for the velocity renormaliza- dl tion. The first order diagram shown in Fig. 7(a) is the There are several ways to understand this. It was ar- Hartree-Fock exchange contribution, and can be readily gued early on that the vertex function does not acquire evaluated (we denote by G(0) the non-interacting GF): any divergent contributions, which is related to the ex- pected regular behavior of the polarization operator to d2pdε all orders in graphene (Gonz´alez et al., 1994). More re- Σ(1)(k,ω)= i G(0)(k + p,ω + ε)V (p) , (3.7) (2π)3 cently, explicit calculations up to two loop order were Z performed (de Juan et al., 2010; Kotov et al., 2008b); it which at low external momenta exhibits a logarithmic was confirmed that the vertex function is finite in the low- singularity energy limit. In addition, direct examination of the po- larization function at two loop level (Kotov et al., 2008b) α Σ(1)(k,ω)=Σ(1)(k)= vσ k ln(Λ/k), Λ/k 1. found that the self-energy correction, Fig. 8(a), acquires v 4 · ≫ (3.8) a logarithmic divergence which can be absorbed into the renormalized velocity v(k) (Eq. (3.9)), while the vertex At this order we have Σ0 = 0, i.e. Z = 1 due to the frequency independence of the interaction potential, and correction of Fig. 8(b) is finite: the quasiparticle velocity increases: 2πe2 Π(2b)(q, 0) = finite = 0.53α2 . (3.12) α ǫ q − v(k)= v 1+ ln(Λ/k) , Λ/k 1. (3.9) 0 4 ≫   Incidentally, this contribution leads to enhancement of The ultraviolet cutoff Λ 1/a represents the momentum the dielectric static screening (i.e. the dielectric constant ∼ π 2 scale up to which the spectrum is Dirac-like. beyond linear (RPA) order becomes ǫ = 1+ 2 α+0.53 α .) 12

Alternatively, one can argue that in two-dimensional possible at αc 0.8. This fixed point is infrared unstable ≈dv field theories with Coulomb interactions the charge since near αc, dl = C(α αc)v, C> 0, i.e. for α > αc, 2 − − e does not flow because it appears as a coefficient v flows towards zero (α flows to ) while for α < αc, in a nonanalytic term in the action (Herbut, 2006; v flows towards (α flows to zero.)∞ Of course it is not Ye and Sachdev, 1998). The conclusion then is that only clear that this estimate∞ is reliable since the fixed point the quasiparticle velocity and residue (see below) are value αc is not small, and we used perturbation theory renormalized. In particular, at first order we can combine (α 1) to derive this result. On the other hand, a flow ≪ Eqs. (3.10),(3.11) into a single one reflecting the renor- towards strong coupling for α > αc is consistent with malization (running) of the coupling α: the formation of an excitonic insulator (mass generation), for which strong evidence has accumulated by now, as dα α2 = . (3.13) we discuss in Section III.B. Recent numerical simulations dl − 4 give the value αc 1 (see Section III.B). Finally, we also∼ find that Z is renormalized at second Therefore we have an infrared stable fixed point at α = 0, order, since the self-energy is frequency dependent. From and the flow towards it is logarithmic: Eq. (3.6) we can expand to second order of bare pertur- 2 4 Nα bation theory Z 1 24 ln(Λ/k), which would lead us α(k) , k 0 . (3.14) ≈ − 2 ∼ ln(Λ/k) → dZ Nα to an RG equation for Z: dl = 24 Z, to be solved together with Eq. (3.13), or Eq. (3.17− ), depending on the Thus the Coulomb interactions are marginally irrelevant. desired level of approximation. Alternatively, Eq. (3.6) This is equivalent to a logarithmically divergent velocity: is already written in a “nonperturbative” way. Ignoring v(k) (e2/4) ln(Λ/k), k 0. ∼ → for the moment the running of α, we have at low energies 1 24 a. Two-loop results. It is instructive to examine Z = 2 , k/Λ 0. 1+ Nα ln(Λ/k) → Nα2 ln(Λ/k) → corrections beyond first order (Mishchenko, 2007; 24 (3.18) Vafek and Case, 2008), since additional effects appear, (2b) such as renormalization of Z. For example the first This result, along with the previous one for Σ0 , brings diagram in the RPA series shown in Fig. 7(b) is us to the infrared behavior (we use ω and k interchange- ably in the infrared limit): d2pdε Σ(2b)(k,ω)=i G(0)(k+p,ω+ε)(V (p))2Π(1)(p,ε) . 1 (2π)3 Z , Σ α2ω ln(ω) , ω 0. (3.19) Z ∼ α2 ln(ω) 0 ∼ | | → (3.15) | | An explicit evaluation at low energies and momenta gives This is characteristic of a marginal Fermi liquid a single logarithmic divergence (Das Sarma et al., 2007; Gonz´alez et al., 1994). How- 2 ever, this regime is never achieved if the running of α is (2b) Nα Σ (k,ω)= (ω + vσ k) ln(Λ/k), k/Λ 0, taken into account, as is intuitively clear from the above − 24 · → (3.16) equations. As we will see later from the solution of the (2b) 2 (2b) RG equations for Z and α, in fact Z tends to level off in i.e. Σ0 = (Nα /24)(ω) ln(Λ/k), and Σv = (Nα2/24)vσ k−ln(Λ/k). Because the polarization bub- the infrared, and the system has well-defined quasiparti- −ble is proportional· to the number of fermion flavors cles. N = 4 (valley+spin), we have explicitly written the It is interesting to note that trigonal distortions, N dependence. By comparing with Eq. (3.5), we find which change the band structure away from the Dirac that the velocity is changed by an amount ( N/24 equation, are modified by the electron-electron interac- N/24)α2v ln (Λ/k). − − tion, and their irrelevance at low energies is enhanced In addition, other diagrams at second order have to be (Foster and Aleiner, 2008). As a result, the linear dis- added, such as the vertex correction of Fig. 7(c). Most persion becomes an even more robust feature of graphene importantly, this diagram is also proportional to ln Λ. (Rold´an et al., 2008). Collecting all contributions one finds the RG equation for the velocity flow (Vafek and Case, 2008) b. Influence of disorder. Before we proceed, let us briefly dv α N address the effect of disorder. Two major sources of = v δ α2v , (3.17) dl 4 − 12 − disorder are scalar potential random fluctuations (e.g.   formation of electron-hole puddles), and vector gauge with δ 0.03. One observes that the contribution of the field randomness, related to formation of ripples. Start- “RPA”≈ diagram is numerically dominant at second order ing with the latter, i.e. a gauge field coupled to the (it is larger than the rest by a factor of 10 for N = 4.) Dirac fermion pseudospin σ A, and characterized by · In addition, the second order tendency is a decrease of variance ∆, Aµ(r1)Aν (r2) = ∆δµν δ(r1 r2), one can the velocity. Consequently a finite coupling fixed point is readily deriveh the correspondingi RG equations− in the 13

α a)

b)

= ¡+

¡ ¡

FIG. 10 RPA self-energy,¡ which includes an infinite resum- mation of polarization bubbles.

The RPA potential is given by





Disorder ∆ 2πe2 RPA p V ( ,ε)= 2 (1) . (3.22) FIG. 9 An attractive line of fixed pints for interactions and ǫ0p 2πe Π (p,ε) gauge field disorder. − Quite remarkably, at low momenta one can evaluate the singular contribution to the self-energy analytically weak disorder and interactions limit (Herbut et al., 2008; Stauber et al., 2005) 8 Σ(RPA)(k,ω)= [ F (λ)ω + F (λ)vσ k] ln(Λ/k) , Nπ2 − 0 1 · d∆ dα α2 ∆ =0, = + α . (3.20) (3.23) dl dl − 4 π where we have defined Gauge field disorder itself is not renormalized, while the π λ = Nα. (3.24) interplay of disorder and interactions leads to a line of 8 4 attractive fixed points located at: α∗ = π ∆, as shown in Fig. 9. Physically the variance is related to the char- This parameter is measuring the importance of polariza- acteristic height h, and length L of the corrugations of tion loop contributions relative to the bare Coulomb term the surface, ∆ h4/(L2a2). Thus weak disorder generi- (i.e. the ratio of the second term to the first in the de- cally shifts the∼ fixed point away from α = 0, while strong nominator of Eq. (3.22)). The RPA is generally expected disorder can have an even more profound effect (Section to be valid when the loops dominate over other diagrams, i.e. N 1. Provided this condition is satisfied, we can VI.C). ≫ In addition, for weak interactions, the inclusion of also analyze the strong-coupling regime λ 1, and the crossover toward the weak-coupling one (λ≫ 1), i.e. we scalar (density fluctuations) disorder turns out to be ≪ a relevant perturbation which grows under renormal- can hope to cover a wide range of α values. ization, and thus away from the perturbative regime The calculated functions F0 and F1 in Eq. (3.23) are (Aleiner and Efetov, 2006). Moreover, gauge field disor- √1 λ2 π der, when combined with strong-enough interactions, can − arccos λ 1+ , λ< 1, F (λ)= − λ − 2λ cause the interactions to grow (Vafek and Case, 2008). It 1  √λ2 1 π has been argued that the strong-coupling regime for dis-  − ln λ + λ2 1 1+ ,λ> 1, order and interactions generically occurs when all types λ − − 2λ  p  (3.25) of disorder consistent with graphene’s symmetries are in-  cluded (Foster and Aleiner, 2008). 2 λ2 π − arccos λ 2+ , λ< 1, A detailed analysis of this complex situation is beyond − √ 2 − λ F (λ)=  λ 1 λ the scope of this work, and from now on we continue our 0 λ2 −2 π discussion of clean graphene.  − ln λ + λ2 1 2+ ,λ> 1.  λ√λ2 1 − − λ −  p  (3.26)  This leads to the system of RG equations for v and Z, to 2. Strong-coupling/RPA analysis leading order in 1/N The full RPA treatment was performed by many au- dv 8 et al. = (F1(λ) F0(λ)) v , (3.27) thors (Das Sarma , 2007; Foster and Aleiner, 2008; dl Nπ2 − Gonz´alez et al., 1999; Kotov et al., 2009; Polini et al., 2007; Son, 2007). Here we mostly follow Son, 2007. The dZ 8 RPA self-energy is shown diagrammatically in Fig. 10, = F0(λ)Z. (3.28) and corresponds to the equation dl −Nπ2 d2pdε At strong-coupling, λ 1, one finds Σ(RPA)(k,ω)= i G(0)(k + p,ω + ε)V RPA(p,ε) . ≫ (2π)3 dv 8 Z = v , (3.29) (3.21) dl Nπ2 14

dZ 8 5 = ln (2λ)Z. (3.30) dl −Nπ2 4 The first equation, after integration, leads to the low- energy result (k 0) → 3 Λ η 8 λ v(k)/v = , η = , (3.31) 2 k Nπ2   1 which implies that the quasiparticle dispersion is of the form 0 0 2 4 6 8 8 ω(k) kz, z =1 . (3.32) l ∼ − Nπ2 The existence of the anomalous velocity dimension, η, 1 and consequently z = 1, is characteristic of the strong- coupling regime Nα6 (Son, 2007). However this 0.8 strongly-coupled fixed→ point ∞ is infrared unstable, since, due to the velocity increase, the RG for α flows towards 0.6 weak coupling. (One also expects that for certain N < Z 0.4 Nc and α 1 an excitonic gap can appear, which will be discussed≫ in Section III.B.) In this regime Z can be approximated perturbatively (in 1/N) as 0.2

8 1 0 Z 1 ln(Nαπ/4) ln(Λ/k), Nα 1, (3.33) 0 5 10 15 20 25 30 ≈ − π2 N ≫ l which can be obtained from Eq. (3.30) by ignoring the FIG. 11 RG flow of the coupling λ and the quasiparticle scale dependence of λ. residue Z as a function of the RG scale l; the infrared limit is at l . From Gonz´alez et al., 1999. In the weak-coupling limit λ 1, it is easy to verify →∞ that we recover the previous result≪ (3.10) for the velocity v (leading to a flow for α towards zero), and the previ- ously encountered perturbative equation for Z At higher energies however (away from the fixed point but still much lower than the bandwidth vΛ), the system dZ 8 λ2 N exhibits marginal Fermi liquid behavior. = Z, i.e. Z 1 α2 ln(Λ/k). (3.34) dl −Nπ2 3 ≈ − 24 At finite (but still small) density away from the Dirac point, i.e. k = 0, the logarithmic behavior in The last formula is written to first order in Nα. the infrared is cut-off6 by the Fermi momentum, i.e. Eqs. (3.33),(3.34) allow us to have a qualitative under- ln(Λ/kF ), kF /Λ 0, and the RG stops away from standing of the behavior of Z as a function of the RG the fixed point. For→ comparison with experiments, the scale l. If the initial value of α is large, at the initial flow toward this stable fixed point should be stopped at RG steps Z decreases logarithmically fairly fast (due to a scale set by the (small) density, temperature, or fre- the weak ln(α) dependence in Eq. (3.33), even though α quency, whichever is higher. itself decreases). Eventually, when α has decreased sub- One can also perform a numerical evaluation of the 1 stantially (α (ln(Λ/k))− ), Z is governed by Eq. (3.34), main RPA equation Eq. (3.21) (Polini et al., 2007). For meaning that∼Z will stop decreasing, and will level off for small density, and with logarithmic accuracy (ln(Λ/kF )), l = ln(Λ/k) . → ∞ this is equivalent to evaluating, by using the notation of A numerical evaluation of the system of equations Eq. (3.23), and taking into account Eqs. (3.4),(3.5),(3.6) (3.27),(3.28) confirms the anticipated behavior and is shown in Fig. 11, (Gonz´alez et al., 1999). (The equa- 1 Z = (1 ∂Σ(RPA)/∂ω) 1 = , tion for the coupling λ = π Ne2/(ǫ v) is obtained by − 8 8 0 − 1+ 2 F0(λ) ln(Λ/kF ) 2 π 2 Nπ observing that (dλ/dl) = ( 1/v ) 8 N(e /ǫ0)(dv/dl), due (3.35) to the lack of charge renormalization.)− We conclude that the flow of λ is towards weak coupling, no matter how 8 v∗/v = Z 1+ F (λ) ln(Λ/k ) . (3.36) large its initial value is. Z does not renormalize to zero at Nπ2 1 F low energy due to the RG decrease of λ. Thus, near the   weak-coupling infrared fixed point, the marginal Fermi Here v∗ is the renormalized velocity. At any finite den- liquid (Eq. (3.19)) is ultimately not reached, and the sity the numerical evaluation of Σ(RPA) also picks up system behaves as a Fermi liquid (although the quasi- finite (subleading) contributions, while it can be shown particle decay rate is non-Fermi liquid like, see below.) (Polini et al., 2007) that the leading perturbative results 15

1 (a) (b) Λ = 101 Λ = 102 Λ = 103 4 0.8 Λ = 10 Λ = 105 2DES Z

0.6

(a) 0.4 0 1 2 3 4 5 f FIG. 13 (Color online) (a) Density dependence of the velocity 1.8 for suspended graphene, from (Elias et al., 2011). The solid (b) line is the result of RG treatment within RPA (Eq. (3.27)). 1.6 (b) Reshaping of the Dirac cone due to the interaction-driven renormalization (increase) of the Fermi velocity at low mo- 1.4 menta. The outer cone represents the linear Dirac spectrum without many-body effects. /v ⋆

v 1.2 logarithmic renormalization of the velocity predicted by 1 theory fits the data fairly well, and thus offers a direct proof that the Dirac cones can be reshaped by long-range 0.8 electron-electron interactions near the Dirac point, as 0 1 2 3 4 5 schematically shown in Fig. 13(b). Finally, ARPES mea- f surements of quasi-freestanding graphene grown on the FIG. 12 (Color online) Exact evaluation of the RPA equations carbon face of SiC have also detected logarithmic velocity for (a) the quasiparticle residue, and (b) the Fermi velocity. renormalization (Siegel et al., 2011). On the horizontal axis f is defined as f Nα. Λ is in units 2 1 ≡ of kF . Values of Λ from 10 to 10 correspond to density 11 −2 ∼ 13 −2 3 n from n 10 cm to n 10 cm (while Λ 10 is 3. Quasiparticle lifetime ultra low density∼ n 109cm∼−2). The values of Λ (in∼ units ∼ of kF ) can be converted into density n via: Λ/kF 220/√n˜, 10 −2 ≈ The inverse quasiparticle lifetime (decay rate) due n˜ = n/(10 cm ). The curves labeled 2DES refer to the to electron-electron interactions, 1/τ , is an important √ ee case of 2DEG with parabolic bands, where f = 2rs, and quantity which is relevant to many properties of graphene rs 1/√n. From (Polini et al., 2007). ∼ (and Fermi systems in general). In particular the depen- dence of 1/τee on energy (or temperature) determines the importance of the electron-electron interaction con- such as Eqs. (3.33),(3.34) are readily reproduced. The tribution, relative to other processes, to transport, and RPA results are shown in Fig. 12, and exhibit the natu- interpretation of spectroscopic features, such as ARPES. ral density dependence tendency, i.e. the strongest renor- The decay rate is determined by the imaginary part of malization occurs at the lowest densities. Similar RPA the self-energy, ImΣ(k,ω). The first diagram which has results have been obtained by Das Sarma et al., 2007. energy dependence, and thus a non-zero imaginary part, A significant velocity enhancement was observed in the is the one bubble diagram of Fig. 7(b), whose real part infrared conductivity (Li et al., 2008), which reported is given by Eq. (3.16), i.e. behaves as in Eq. (3.19) at around 15% increase of the Fermi velocity, having value low energies. We can therefore deduce, for energies and 6 as high as v∗ 1.25 10 m/s at the lowest densities momenta close to the mass shell (Gonz´alez et al., 1996), (compared to v≈ 1.1 ×106m/s at higher density). The ≈ × system is at a finite Fermi energy µ 0.2eV. However ImΣ(2b)(k,ω) α2θ(ω vk) ω, ω vk , (3.37) the velocity renormalization is not logarithmic,≈ and it is ∼ − ≈ not clear what is the origin of this effect. i.e. the decay rate is linear in energy. In addition, there A recent study of suspended graphene which measures is an on-shell (“light cone”, ω = vk) discontinuity, where the cyclotron mass (Elias et al., 2011) has detected sig- the rate experiences a jump. This on-shell behavior is nificant logarithmic renormalization of the Fermi veloc- due to the fact that, for ω 4 10 cm− ), Fig. 13(a). The ω>vk. × 16

The above behavior is valid at the Dirac point and T = by confining the electrons into finite-size configurations, µ = 0, while for small T,µ, it is valid for energies of order such as quantum dots (Ponomarenko et al., 2008). In max(T,µ). Notice also that the linear energy behavior of these cases the gap generation mechanism is not intrinsic Eq. (3.37) is very different from the conventional Fermi to graphene, and the value of the gap depends strongly on liquid result ImΣ ω2 (Das Sarma et al., 2007), which the external conditions. However even in such situations would occur for a finite∼ Fermi surface (µ = 0) and is due interactions can play an important role by increasing the to intra-band particle-hole excitations. 6 gap. The on-shell discontinuity present at the one-loop Consider a gap arising from an external potential that level Eq. (3.37) disappears when the full RPA self- alternates between the two sublattices energy is evaluated (Fig. 10). In this case one obtains (Khveshchenko, 2006) mass = ∆0 nσ(Ri) ∆0 nσ(Ri) . (3.40) H − σ,i A σ,i B X∈ X∈ ImΣ(RPA)(k,ω) ln (πα)θ(ω vk)(ω vk), ω vk . ∼ − − ≈(3.38) Consequently an additional pseudospinor structure re- Away from the mass shell, the energy dependence is nat- lated to σ3 is generated, and the new Green’s function urally linear: has the form 1 ImΣ(RPA)(k,ω) ln (πα) ω, ω vk . (3.39) G(k,ω)= . (3.41) ∼ ≫ ω vσ k ∆0σ3 Σ(k,ω) − · − − (RPA) k The full dependence ImΣ ( ,ω) has to be evalu- Here ∆ is the explicit “mass” of the graphene elec- ated numerically (Das Sarma et al., 2007), and the re- 0 (RPA) trons (while Σ(k,ω) contains the information about sults confirm the smooth rise of ImΣ from the point interactions, assumed to perturbatively renormalize ω = vk. all the other terms.) The new spectrum is then In the limit of zero doping µ 0, when the system ap- E(k) = v2k2 + ∆2, with a gap of 2∆ . Com- proaches the fixed point α = 0,→ we argued previously that 0 0 puting the± Hartree-Fock interaction correction to ∆ the residue Z does not approach zero (i.e. the marginal 0 leads to a renormalizedp mass ∆˜ (Kane and Mele, 2005; Fermi liquid behavior ultimately does not manifest itself.) 0 Kotov et al., 2008a) On the other hand the marginal Fermi-liquid behavior is expected to be much more robust as far as the inverse α ∆˜ /∆ 1+ ln (D/∆ ). (3.42) lifetime, ImΣ ω, is concerned, because the running of 0 0 ≈ 2 0 the coupling α∼(ω) only introduces logarithmic variation on top of a much stronger linear energy dependence. The above enhancement can be substantial. For ex- The linear decay rate discussed above is consis- ample for a bare gap due to spin-orbit coupling ∆0 3 ∼ tent with ARPES experiments (Bostwick et al., 2007; 10− meV (Min et al., 2006; Yao et al., 2007), and taking Zhou et al., 2008), and STM measurements of graphene into account the bandwidth D = vΛ 7eV, the logarith- ≈ on graphite (Li et al., 2009a) (see also the discussion in mic factor is around 15. In fact one should integrate the ˜ Grushin et al., 2009). RG equation for the renormalized mass ∆0 as a function of ln(Λ) simultaneously with the equation for the running coupling α(ln(Λ)), Eq. (3.13), down to the lowest infrared B. Spontaneous mass generation scale ∆0 (bare gap). This leads to the stronger de- ∼ ˜ α 2 pendence ∆0/∆0 = (1+ 4 ln (D/∆0)) (Kane and Mele, It is an intriguing possibility that graphene can un- 2005), and the perturbative expansion of this result is dergo a metal-insulator transition for strong enough Eq. (3.42). It is interesting to note that the logarithmic Coulomb interaction α, due to an excitonic pairing mech- mass renormalization formula in graphene Eq. (3.42) is anism. We restrict ourselves to the charge neutrality similar to the well-known expression for the electromag- point µ = 0 since the excitonic pairing tendency de- netic mass of the electron (accounting for radiative cor- creases quickly beyond that. rections) in 3D relativistic QED (Weisskopf, 1939).

1. Finite explicit mass 2. Excitonic mass generation

Before we outline the main results, let us mention that We now turn to the possibility of spontaneous gap gen- an explicit gap can also open in graphene under certain eration due to long-range Coulomb interactions (we set conditions that depend on graphene’s environment. For the explicit gap ∆0 = 0 in Eq. (3.41)). In relativistic example there are suggestions of a detectable gap in situ- QED in two space (plus one time) dimensions, QED2+1, ations when graphene is on a substrate with specific sym- the study of this phenomenon, called chiral symmetry metry, creating sublattice asymmetry in the graphene breaking, started quite a while ago (Appelquist et al., plane, and thus making the graphene electrons massive 1986; Pisarski, 1984), and is still going strong today. (gapped) (Zhou et al., 2007). Gaps can also be produced Graphene is actually different from QED2+1 because only 17 α = 8 (Khveshchenko, 2009). At strong coupling α the gap is non-zero only below a critical number of→ fermion ∞ Excitonic flavors (since the effective interaction scales as 1/N in this limit); for example N 7.2 (Khveshchenko, 2009), α Insulator c ≈ Nc 7 (Liu et al., 2009). Near≈ the critical coupling the low-momentum gap α scales as c C ∆(0) D exp , (3.44) ∝ −√α α Gapless semimetal  eff − eff,c  where C is a constant, the critical αeff,c = 1/2, and the form of the effective coupling αeff depends on the level of approximation used — for example an improve- ment over the static RPA potential leads to: αeff = 0 4 N N α/(1 + Nπα/8√2) (which gives N 7.2, α 1, and c c ≈ ≫ αc = 1.13, N = 4 (Khveshchenko, 2009)). The form of FIG. 14 Schematic phase diagram in the α N plane. Eq. (3.44) suggests that the transition is of infinite or- − der (Berezinskii-Kosterlitz-Thouless type). Even though Eq. (3.44) is only valid near the critical coupling, numer- the fermions are confined to a 2D plane, while the field ical results find that the gap in units of the bandwidth, lines extend through the whole 3D space. In addition, the ∆(0)/D, is exponentially small in a wide range of cou- Coulomb interaction in graphene can be considered in- plings (Khveshchenko and Leal, 2004). Since D 7 eV, stantaneous since the speed of light c is much larger than this implies ∆(0) meV, i.e. a rather small gap≈ value. ∼ the Fermi velocity (v c/300). Hence, Lorenz invari- Finally, recent work that takes into account the renor- ance is not respected, which≈ reflects the non-relativistic, malization of the coupling constant and the quasiparticle purely band origin of the Dirac quasiparticles. The anal- residue suggests that αc could be much larger than pre- ysis in relativistic QED reveals that dynamical mass can viously found (Gonz´alez, 2010; Sabio et al., 2010a). be generated below a critical number of fermion flavors The above results are based on various approxima- Nc, with the mass scale set by the coupling itself, which tion schemes and it is therefore important to com- has dimension of energy in pure QED2+1. A transition is pare them with direct numerical simulations of the lat- also found in non-relativistic graphene, where the gener- tice field theory model. Recent Monte Carlo calcula- ated mass scale is related to the ultraviolet energy cutoff tions (Drut and L¨ahde, 2009a,b,c; Hands and Strouthos, (bandwidth D = vΛ) since the coupling α is dimension- 2008) provide strong evidence that spontaneous mass less in this case. generation does occur, and give comparable values The gap equation can be obtained as a self-consistent for the critical couplings: N 9.6, α 1 c ≈ ≫ solution for the self-energy within RPA (i.e. vertex (Hands and Strouthos, 2008), αc = 1.1, N = 4 corrections are neglected), and is referred to as the (Drut and L¨ahde, 2009b). Unfortunately the Monte Schwinger-Dyson equation. It has the form Carlo simulations do not allow for an exact determina- tion of the gap size, and for that we can only rely on the d2kdω V RPA(p k,ε ω)∆(k,ω) previously described Schwinger-Dyson equation (leading ∆(p,ε)= i − − . (2π)3 ω2 v2k2 ∆2(k)+ i0+ to small gaps). For graphene deposited on SiO the value Z − − (3.43) of αSiO2 0.79 and is therefore not enough to generate The structure of the solution has been analyzed ex- a gap; only≈ experiments on ultrahigh mobility suspended tensively (Gamayun et al., 2010; Gorbar et al., 2002; samples can potentially reveal the insulating state. Khveshchenko, 2009; Khveshchenko and Leal, 2004; The overall phase diagram of graphene in the α N Liu et al., 2009) at different levels of approximation. plane is expected to look as shown in Fig. 14,− with The equation is simplified significantly if the static RPA αc 1 and Nc 7 9. At finite temperature one RPA ≈ ≈ − potential is used V (p, 0) (Khveshchenko and Leal, expects the existence of a critical temperature Tc 2004), while the dynamical equation has also been stud- ∆(0), while finite doping µ very quickly destroys the∼ ied on-shell (∆(p,ε = vp)) (Khveshchenko, 2009), as well gap (Liu et al., 2009). Application of magnetic field as numerically (Liu et al., 2009). perpendicular to the graphene layer leads to enhance- The mass gap ∆(p) has strong momentum dependence, ment of the excitonic instability due to the formation of due to the long-range nature of the Coulomb interaction. Landau levels (Gorbar et al., 2002; Gusynin et al., 2006; ∆(p) decreases at large momenta and reaches maximum Khveshchenko, 2001a). In addition, it has been suggested value at small momenta where it levels off. For fixed that an in-plane magnetic field favors a gapped excitonic physical value of N = 4, a transition to a gapped state is state (Aleiner et al., 2007), due to the instability of a found above a critical coupling αc. Some of the calculated system of electrons and holes polarized in opposite direc- values are: αc =0.92 (Gamayun et al., 2010), αc =1.13 tions. 18

power of energy (or temperature), as in a Fermi liquid, while the quasiparticle residue is finite at the Fermi sur- face.

The existence of a plasmon-related peak in the quasi- particle decay rate, which originates from intraband transitions in which an electron can decay into a plas- mon, was pointed out in the context of intercalated graphite, where the physics is dominated by graphene layers (Lin and Shung, 1996; Shung, 1986b). For n-doped graphene (µ > 0), which is relevant to ARPES ex- periments, a double-feature is found in the decay rate ImΣ: a peak at positive energies, signaling an on- set of plasmon emission, and a sharp spectral feature at negative energies, below the Dirac point, and sepa- rated from it by an amount proportional to the plasmon frequency (Hwang and Das Sarma, 2008b; Polini et al., 2008a). This is the so-called “plasmaron” — a resonance which consists of a quasiparticle strongly coupled to plas- mons (Lundqvist, 1967). Plasmaron features have been previously detected for example in optical measurements FIG. 15 (Color online) ARPES data from (Bostwick et al., of Bismuth (Tediosi et al., 2007). 2010), showing strong features at the Dirac point, which is below the Fermi energy (at 0). The splitting shown in (H) is The above calculations were done within RPA theory. attributed to the presence of “plasmarons” — quasiparticles Line widths have also been analyzed via ab-initio many- strongly bound to plasmons — and depends on the value of body methods (Park et al., 2009; Trevisanutto et al., α (α 0.5 fits the data.) ≈ 2008). Experiments generally show a well-pronounced linear quasiparticle spectrum (Bostwick et al., 2007; Sprinkle et al., 2009; Zhou et al., 2007, 2008), with ad- The physical structure of the gapped state depends on ditional features near the Dirac point which seem to de- the nature of pairing between the valleys — for example pend on the way graphene is prepared, and its purity. one can have charge density wave states (Khveshchenko, For example, gap-like features have been observed near 2001b) with modulation of the electronic density around the Dirac point (Zhou et al., 2007), and attributed to the two sublattices (which corresponds to intravalley par- external, substrate-related factors. Bending of the Dirac ing), or Kekule dimerization (Hou et al., 2007) which spectrum (kink-like feature) was attributed to plasmons corresponds to tripling of the unit cell (intervalley pair- (Bostwick et al., 2007). Most recently manifestations of ing). One generally expects that interactions beyond the sharp plasmaron spectral intensities have been ob- the long-range Coulomb potential, such as short-range served in quasi-freestanding graphene (Bostwick et al., repulsion, would favor particular states, including time- 2010), where a reconstruction of the Dirac point crossing reversal symmetry broken (spin) states. Further discus- seems to take place, as shown in Fig. 15. A diamond- sion appears in Section V.A. like shape appears due to crossing of charge and plas- maron bands. Comparison of the RPA calculation for the energy splitting with experiment leads to the value of C. Finite density Fermi-liquid regime α 0.5 (Fig. 15.) Bostwick et al., 2010 also suggest that the≈ plasmaron features were obscured in earlier measure- As the density increases above half-filling, i.e. ments on non free-standing graphene (Bostwick et al., graphene is at a finite, not necessarily small, chemical po- 2007), due to the several times stronger screening (and tential µ, with a finite Fermi surface, a crossover towards consequently smaller α.) Perhaps most importantly, all a Fermi liquid regime takes place. In this case the lower the current activity in ARPES on different graphene sam- (hole) band becomes irrelevant and the physics near the ples reveals that the electron-electron interactions can af- Fermi surface is dominated by intra-band transitions in fect strongly the physics around the Dirac point, even for the conduction (upper) band (assuming µ> 0). However relatively large density (Fermi energy). the physics near the Dirac point can still be very strongly affected due to the presence of plasmon and “plasmaron” Tunneling spectroscopy measurements, combined with features in the quasiparticle spectral function. ab-initio calculations, have also found evidence for The quasiparticle width near kF is quite simi- density-dependent interactions effects in the tunneling lar to the case of an ordinary 2D electron gas current (Brar et al., 2010) which arise from the sharp (Das Sarma et al., 2007; Hwang and Das Sarma, 2008b; spectral features in the quasiparticle decay rate below Polini et al., 2008a), and is proportional to the second the Dirac point, as discussed above. 19

this gives (1/κ0)= v π/(4n) — behavior which can be clearly seen in experiment Fig. 16. The interaction effectsp in the ground state energy ac- quire divergent contributions in the limit of small den- sity kF /Λ 0, similarly to the previously discussed self- energy (velocity)≈ renormalization. Ignoring any finite (non-diverging) terms, one finds (Barlas et al., 2007) α E /n = (vk ) ln(Λ/k ), (k /Λ) 0, (3.45) ex 6 F F F →

Nα2 E /n = G(α)(vk ) ln(Λ/k ), (3.46) RPA − 6 F F where the function G(α) is defined as G(α) = ∞ 2 2 √ 2 1 (1/2) 0 dx(1+x )− ( x +1+Nπα/8)− , and, in par- FIG. 16 (Color online) Inverse compressibility, measured by ticular, at zero coupling G(0) = 1/3. The above results Martin et al., 2008. The red line is the compressibility of non- exactlyR follow the velocity renormalization, i.e. are equiv- interacting Dirac fermions. alent to the substitution v v(k ) in the free compress- → F ibility (1/κ0) = v π/(4n), where v(kF ) is the running velocity calculated within RPA at the infrared scale kF . D. Physical Observables The result is particularlyp simple at the Hartree-Fock (ex- change) level (when the velocity follows Eq. (3.9)): The interaction-driven singular logarithmic structure near the Dirac point (for µ 0) encountered in the 1 π α ≈ = v 1+ ln(Λ/k )+ O(α2) , (3.47) fermion self-energy, and in particular the renormaliza- κ 4n 4 F tion of the Fermi velocity, can manifest itself in numer- r   ous physical observables, such as the charge compressibil- and was obtained by a number of authors (Barlas et al., ity and the spin susceptibility, which exhibit non Fermi- 2007; Hwang et al., 2007; Sheehy and Schmalian, 2007). liquid behavior. Interactions can also affect the conduc- The above results are valid at zero temperature. We tivity near the Dirac point, leading to deviations from also point out that exactly at zero density kF = 0, but 2 1 2 the celebrated quantized value σ0 = e /4~ expected for T = 0, the compressibility behaves as: κ− (v /T )(1+ 6 2 ∼ free Dirac fermions (Castro Neto et al., 2009a). (α/4) ln(T0/T )) , where T0 is the temperature related to the ultraviolet cutoff; since Λv 7 eV, then T0 8 104K. This is easily understood≈ since in the infrared≈ 1. Charge and spin response limit× near the “critical point” n = T = 0 it’s the larger scale, either vkF , or the temperature T , which enters the a. Compressibility. First we discuss the compressibility physical observables (Sheehy and Schmalian, 2007). κ, which was recently measured (Martin et al., 2008), Of course Eqs. (3.45),(3.46) are valid only asymptoti- Fig. 16, and it was concluded that no interaction effects cally (kF 0), and at any finite density the compressibil- were clearly visible in those samples. Theory predicts ity should→ be calculated numerically. This was achieved significant (α dependent) deviations from the free elec- by expressing the ground state energy via the charge re- tron behavior (Barlas et al., 2007; Hwang et al., 2007; sponse function (Barlas et al., 2007). Polini et al., 2008b; Sheehy and Schmalian, 2007). Fig. 17, upper panel, illustrates the variation of 1/κ The computation of the compressibility requires knowl- with density for fixed interaction. Most notably, 1/κ edge of the ground state energy, which contains the first is larger than the free value 1/κ0. Also, the full order Hartree-Fock exchange contribution Eex, and the RPA implementation weakens the first order Hartree- correlation energy ECorr, describing all the higher order Fock (exchange) result, due to the different signs in effects. Keeping in mind applications of the theory for Eqs. (3.45),(3.46). For example, at α = 0.8 the RPA fairly strong coupling (α 1), the contribution of ECorr term is approximately 1/2 of the exchange, and thus has can be substantial. The correlation∼ energy can be readily to be taken into account (although the RPA effects be- 1 1 calculated within the RPA approximation, i.e. we take come weaker for α 0). Asymptotically, (κ− /κ0− ) E = E . The total ground state energy E, per ln(Λ/k ), as k /Λ → 0. The lower panel gives the vari-∼ Corr RPA F F → unit area, is the sum E = Ekin +Eex+ERPA. The kinetic ation κ/κ0 as a function of the interaction for different 2 energy Ekin = (2/3)vkF n, and n = (kF ) /π is the parti- densities; naturally the deviation from the free limit in- cle density. The inverse compressibility is then calculated creases with increasing interaction and decreasing den- as 1/κ = ∂2E/∂n2, which is equivalent to the usual def- sity. inition involving the variation of the chemical potential The increase of the inverse compressibility, κ0/κ, as with density, (1/κ) = ∂µ/∂n. For free Dirac particles a function of the interaction α (at fixed density), and 20

1.5 5 of the full RPA analysis led us to conclude that α< 0.1. 4 α = 0.8 It has also been argued that exchange and correlation ef- ] 2 RPA 3 fects vanish and do not manifest themselves at all in the 1 (leading logs) 2 compressibility (Abergel et al., 2009). These discrepan- cies indicate that the issue is still unsettled, while it’s eV cm 1

-12 0 also possible (indeed, quite probable) that interaction ef- 0.5 0.001 0.01 0.1 1 [10 fects are obscured by charge inhomogeneities (electron-

κ) Hartree-Fock hole puddles) in these samples. Nevertheless theory pre-

(1/ dicts strong systemic (albeit logarithmic) deviations from 0 Kinetic energy Fermi-liquid theory, and it would be important to test these predictions in cleaner, more uniform, high-mobility, 0.1 0.2 0.3 0.4 0.5 0.6 12 -2 low-density samples. density n [10 cm ]

1 b. Spin susceptibility. The paramagnetic spin suscepti- bility, χs, shows behavior very similar to the charge com- 0.9 1 10 pressibility, i.e. (χs/χs,0) decreases as the interaction in-

0.8 2 creases (Barlas et al., 2007). This is again related to the 10 fact that χ 1 is calculated via the ground state energy, 0 s− 0.7 and is proportional to the Fermi velocity v. It was also 3 κ/κ 10 pointed out that the same effect, i.e. the logarithmic 0.6 growth of the exchange energy, Eq. (3.45), can lead to

4 suppression of ferromagnetism in graphene at low densi- 0.5 10 ties (Peres et al., 2005). The full calculation of χs within 0.4 RPA was carried out by Barlas et al., 2007. 0 1 2 3 4 5 Nα On the other hand the orbital diamagnetic susceptibil- 2 ity, χdia, is proportional to v , because the quasiparti- FIG. 17 Upper panel: Inverse compressibility calculated at cle current that couples to the vector potential contains different levels of approximation as a function of density. The v (the magnetic field is perpendicular to the graphene inset enlarges the low-density region. Lower panel (adapted plane). Therefore interaction corrections lead to an in- from Barlas et al., 2007): Compressibility calculated within RPA, relative to the free level for different couplings and den- crease of χdia (Sheehy and Schmalian, 2007) and, conse- sities. Here N = 4 is the Dirac fermion degeneracy. The quently, orbital effects are expected to dominate in the numbers refer to the values of Λ/kF , which can be converted susceptibility. At the Dirac point, kF = 0, one finds at 10 −2 into density n via: Λ/kF 220/√n˜,n ˜ = n/(10 cm ). This finite temperature 2 ≈ 11 −2 implies (Λ/kF ) 10 for n 10 cm , and (Λ/kF ) 10 for − n 1013cm 2. ∼ ∼ ∼ α 2 ∼ χdia/χdia,0 = 1+ ln(T0/T ) , (3.48) 4   where the non-interacting χ = e2v2/(6πc2T ) with decreasing density (for fixed interaction), represents dia,0 − non-Fermi liquid behavior, and reflects the lack of screen- (Ghosal et al., 2007). Here c is the speed of light. At T = 0, n = 0, we have χ e2v/(c2√n), and interaction ing. By contrast, in a 3D (and 2D) Fermi liquid with 6 dia,0 ∼− a screened potential κ0/κ decreases; for example within corrections readily follow from the v dependence. This re- sult is, strictly speaking, valid for T B µ = v√πn, Hartree-Fock, κ0/κ 1 rs/6 < 1, and eventually goes ≪ ≪ through zero, signaling≈ an− instability (Mahan, 2000) (al- whereas for B = 0 the orbital susceptibility is zero for µ =0as T 0, and is finite only when the Fermi energy though the critical value of rs depends strongly on the 6 → level of approximation.) Such an instability does not oc- is at the Dirac point. It has been suggested that an in- cur in graphene, which is related to the impossibility of teraction driven positive (paramagnetic) contribution to Wigner crystallization (Dahal et al., 2006). It should be the orbital susceptibility can therefore become dominant in doped graphene, χ [e2v2/(µc2)]α ln α , α 1 noted that for larger densities (larger than the density orb ∼ | | ≪ range shown in Fig. 17) the logarithmic corrections be- (Principi et al., 2010). come unimportant and the system recovers the Fermi liq- uid behavior, i.e. eventually κ/κ0 becomes larger than 1. c. Specific heat. The specific heat is logarithmically sup- 2 Fits of the experimental data for κ with adjusted pressed due to the suppression of the DOS v− . Con- 6 2 ∼ (slightly larger) velocity v = 1.1 10 m/s show that sequently CV CV,0/(ln(T0/T )) ,T/T0 1, where × 6 2 2 ∼ ≪ α 0 (Fig. 16), while the use of v = 10 m/s by CV,0 T /v is the free Dirac fermion specific heat. Sheehy≈ and Schmalian, 2007 at the Hartree-Fock level The full∼ RPA calculation, valid also for large coupling, produced α 0.4. On the other hand, the application was carried out by Vafek, 2007. ≈ 21 d. Graphene as a quantum critical system. A uni- ductivity should have the form fied view of the above behavior is presented in Sheehy and Schmalian (2007), where it was stressed that Cα˜ σ(ω)/σ0 =1+ α , (3.49) the logarithmic corrections are manifestations of scal- 1+ 4 ln (Λv/ω) ing behavior around the quantum critical point at n = 0,T = 0. As discussed previously, at finite chemical where the constant C˜ 0.01, as argued by Mishchenko, potential, T = 0,n = 0, graphene behaves as a Fermi 2008; Sheehy and Schmalian≈ , 2009. The smallness of liquid, whereas at T6 = 0, a quantum critical region C˜ reflects the near cancellation of self-energy and ver- fans out of the point n6 = 0,T = 0. In the critical re- tex corrections, and thus the effect of interactions is gion it is natural to call graphene a Dirac liquid, where small. This value is also consistent with optical measure- the proximity to the Dirac point is important for phys- ments on suspended samples (Nair et al., 2008), as well ical phenomena at finite T . This puts graphene’s be- as graphene on a substrate (Li et al., 2008), which find havior into the general framework of quantum critical σ(ω) to be very close to σ0, and frequency independent phenomena (Sachdev, 1999). In practical terms, it im- in a wide range of energies. plies that the logarithmically divergent velocity contri- In the strict DC limit ω = 0, the presence of disorder, butions are cut-off by the largest scale: temperature T , in combination with interactions, can alter the conduc- kF √n, or magnetic field. Computing physical quanti- tivity. For example, for weak gauge field disorder (∆) ties∼ in perturbation theory (Hartree-Fock or RPA) nat- where an attractive line of fixed points exists (Fig. 9) 4 urally involves these infrared scales. The separation be- with α∗ = π ∆, calculations show that the conductivity tween the Dirac liquid and the Fermi liquid regimes in (on the fixed line) increases relatively to the free limit the n T plane is defined by the crossover tempera- (Herbut et al., 2008): σ = [π/2 +(4 π)∆]e2/h. For − − ture T ∗(n) = vkF (1 + (α/4) ln(Λ/kF )), kF = √πn, and stronger scalar and vector disorder/interactions where thus the temperature dependencies quoted previously, the couplings run away to infinity the problem is non- are valid for T0 >T >T ∗(n). The ultraviolet tem- perturbative, and a complex variety of behavior is ex- 4 perature scale T0 8 10 K, while for typical graphene pected (Foster and Aleiner, 2008). 12≈ ×2 2 densities n . 10 cm− , T ∗(n) 10 K. For clean graphene at µ = vkF = 0 it was pointed out ∼ (Fritz et al., 2008; Kashuba, 2008; M¨uller et al., 2008) that at high temperature (compared to the frequency), the conductivity is expected to have the form:

2. Conductivity 2 0.76 e 2 σ = 2 , T α ω, (3.50) The behavior of the electrical conductivity in graphene α h ≫ has been extensively reviewed (Das Sarma et al., 2011; where α(T )=4/ ln(Λv/T ) is the running Coulomb cou- Peres, 2010). It is believed that charged impurities pling. This form reflects electron-electron inelastic colli- 2 and resonant scatterers are the main sources of scat- sions with scattering rate 1/τee α T . The linear tem- tering away from the Dirac point, and to extent the perature dependence is characteristic∼ for Dirac particles. long- or short-range part of the Coulomb potential con- The above formula is valid as long as 1/τee is the dom- tributes to scattering is a matter of ongoing debate inant scattering mechanism (collision-dominated trans- (Chen et al., 2008; Monteverde et al., 2010; Ni et al., port), and implies that clean graphene at the neutral- 2010; Ponomarenko et al., 2009; Reed et al., 2010). ity point should exhibit a universal, interaction-limited Here we will only mention effects related to long-range conductivity, reflecting essentially the quantum critical electron-electron interactions near the Dirac point. Inter- behavior of graphene in this regime (T µ). With ≫ action corrections to the minimum metallic conductivity increased doping (µ/T ), a crossover takes place to a 2 π 2 of free Dirac fermions, σ0 = e /(4~) = e /h (Fradkin, Fermi liquid regime with screened interactions, where 2 1 2 2 1986; Lee, 1993), are more involved, because this expres- τ − α T /µ, (M¨uller et al., 2008) and the conductiv- ee ∼ sion does not contain the quasiparticle velocity, while the ity is dominated by charged impurity scattering. electric charge is not renormalized. The debate was fu- It has also been pointed out that for µ = 0 graphene eled in part by electrical measurements of the minimum behaves as an almost “perfect” fluid, in a sense that conductivity (at the Dirac point) which turned out to its shear viscosity, η, relative to the entropy density 2 ~ be somewhat larger than σ0 (Geim and Novoselov, 2007; s is anomalously small: η/s = (0.13/α (T ))( /kB) Tan et al., 2007). Theoretically, at T = 0 (or T ω (M¨uller et al., 2009). This ratio measures how strongly where ω is the external frequency), it is expected≪ that the excitations in a fluid interact. At room temperature any interaction effect should have sub-leading character, η/s of graphene is smaller than η/s of any known cor- and the frequency can enter only through the running related quantum fluid, and is close to the lower bound of 1 ~ proposed to exist for a large class of strongly of the coupling α(ω). Even though some debate still ex- 4π kB ists (Herbut et al., 2008; Juricic et al., 2010; Mishchenko, interacting quantum field theories (Kovtun et al., 2005). 2008; Sheehy and Schmalian, 2009) as to the implemen- Therefore, due to its quantum critical nature near the tation of the cut-off regularization procedure, the con- Dirac point, graphene is suggested to behave as a strongly 22 correlated quantum liquid and should exhibit signatures found to be amplified by excitonic effects, improv- of electronic turbulence (M¨uller et al., 2009). ing screening of interactions between quasiparti- cles. This analysis leads to values of α ranging from α 1/7 in the static limit to α 2 at high E. Overview of main results frequencies.≈ Very recent measurements≈ of the cy- clotron mass in suspended graphene (Elias et al., Before we proceed with further topics related to inter- 2011) have found logarithmic velocity renormaliza- actions in graphene, let us broadly summarize the main tion and extract, within the RPA scheme, an ef- findings and questions raised so far: fective value of graphene’s dielectric constant ǫG 3.5. One can also expect that near the Dirac point,≈ 1. For clean graphene at the neutrality point µ = 0, where interactions lead to singular effects, addi- interactions are not screened and are marginally tional factors can be important such as disorder, irrelevant; the fixed point α = 0 is approached ∗ inhomogeneities, rippling, etc., and thus obscure logarithmically (or, equivalently, the quasiparticle the clean behavior. velocity increases logarithmically). From a theory standpoint, the approach towards this fixed point 5. In the Fermi-liquid regime, where interactions are is well understood both from weak and strong- screened, the physics near the Dirac point can still coupling (RPA) perspectives. Since in graphene be strongly affected — this is due to resonant fea- one can have α 1 under rather conventional ex- ∼ tures in the quasiparticle self-energy, reflecting in- perimental conditions, our understanding of RPA teractions of quasiparticles with plasmons. calculations is important. RPA is justified only in the limit of large number of fermion species (N 1), while for N = 4 it should work for IV. THE COULOMB PROBLEM AND CHARGED weak≫ to moderate coupling; however there are indi- IMPURITIES cations, coming mostly from two-loop calculations, that vertex corrections are numerically small, and The consideration of non-interacting Dirac electrons thus RPA should work well. Disorder generally in 2D under a Coulomb field is of paramount relevance drives the system away from the clean fixed point, for graphene, and for several reasons. First of all, the towards finite or even strong coupling, depending Coulomb problem for relativistic fermions has many fea- on disorder type. tures that are unfamiliar in condensed matter systems, and which resemble long standing predictions made in the 2. The resulting behavior near the Dirac point is that context of QED in strong fields. As such, and given that of a non-Fermi-liquid with a quasiparticle decay having α 1 makes graphene intrinsically strongly cou- rate which is linear in energy, and decreasing quasi- pled, it can∼ provide the first experimental ground for test- particle residue. All physical characteristics related ing many elusive predictions from strong-coupling QED. to the quasiparticle velocity (which increases loga- On the other hand, the single particle Coulomb prob- rithmically) are affected, and predicted to exhibit lem constitutes the first step in addressing nontrivial fea- systemic, interaction dependent, deviations from tures of the full, many-body interacting problem. Char- their non-interacting values as the Dirac point is acteristics like non-linear screening, or the supercritical approached, either as a function of density or tem- instabilities, provide valuable insight in grasping some perature. proposed many-body effects, like exciton condensation, 3. Can graphene be driven into an excitonic insulating or spontaneous mass generation in graphene. state? At the Dirac point the long-range Coulomb Historically, however, the motivation for studying the interactions can lead to bound electron-hole pairs, Coulomb problem comes from the seminal experimental creating a gap. There has been intense debate observations (Novoselov et al., 2004a) that the field effect whether this can happen under realistic conditions in graphene prepared on SiO2 is characterized by carrier — since the critical interaction strength appears to mobilities that do not depend on the Fermi energy or carrier density (the DC conductivity, σ = me n , with be αc 1, it seems possible to occur in suspended ∼ m const.), and that carriers are chiral Dirac| fermions| samples (α = 2.2). So far no experimental indica- ≃ tions have been observed. in 2D (Novoselov et al., 2005; Zhang et al., 2005). Early semiclassical investigations (Adam et al., 2007; Ando, 4. What is the value of the interaction α? Clearly, 2006; Nomura and MacDonald, 2007, 2006) showed that since α = 2.2/ǫ0 is dielectric constant depen- such linear-in-density conductivity could be explained by dent, working with different substrates could scattering of unscreened Coulomb impurities, which are 10 2 lead to changes in interaction-dependent effects typically seen in silica in concentrations of 10 cm− (Jang et al., 2008). There are also suggestions that (Ando et al., 1982). As a result, transport∼ in the pres- graphene has an “intrinsic” value of α (Reed et al., ence of charged impurities rapidly became one of the 2010), arising from dynamical dielectric screen- most studied topics in the quest for the ultimate mobil- ing. The polarizability of the Dirac fermions was ity in graphene. Since, as we saw before, Coulomb’s law 23 is exactly preserved in undoped graphene, and approx- (Khalilov and Ho, 1998; Novikov, 2007a) imately preserved for small and moderate doping, the scattering processes are essentially governed by the bare m ε g/r Fj (r) + [∂r + j/r]Gj (r) = 0 (4.2a) − − Coulomb problem, unlike conventional metals, where  ∂r j/rFj (r)+ m + ε + g/r Gj (r)=0. (4.2b) screening is perfect. A thorough understanding of this − problem is therefore important not only for its theoreti- This coupled pair of first order equations can be straight- cal relevance and its import on electron-electron interac- forwardly reduced to two decoupled second order equa- tions, but also for its experimental implications, and our tions. Free solutions (g =0)of (4.1) exist when ε > m , understanding of transport in graphene. and are simple spherical waves whose k-normalized| | | ver-| Finally, it is highly significant that this is an exactly sion reads solvable problem. This means that most quantities can be obtained exactly, allowing us to unveil many interact- k ε + m Jj 1/2(kr)Φj 1/2 Ψj = | | − − (4.3) ing and non-interacting effects that are not within reach s2 ε isε ε m Jj+1/2(kr)Φj+1/2 of the perturbative approaches already discussed. We | |  p | − |  p proceed to show several such features. On account of (sx sgn(x)). For nonzero g, one readily sees from (4.2) the long range nature of the Coulomb field, inter-valley that≡ the solutions at r 0 behave as processes are not relevant, and hence we will solve the ∼ γ 2 2 problem within each (independent) valley in the Dirac F (r), G(r) r± , γ = j g . (4.4) description of fermions in graphene. ∼ − The general exact solution is given inp terms of confluent hypergeometric, or Whittaker’s functions, both in the massive (Gamayun et al., 2009; Gupta and Sen, 2008; A. Exact Solution of the Coulomb Problem Gupta et al., 2010; Khalilov and Ho, 1998; Novikov, 2007a; Pereira et al., 2008a), and massless cases 1. Wave Equations and Spectrum (Gupta and Sen, 2009; Pereira et al., 2007; Shytov et al., 2007b). In the massless case, one can map (4.2) into A Coulomb center of charge Z e generates the poten- the familiar Coulomb radial Schrodinger equation in 3D 2 | | tial U(r) = Ze /(ǫ0r) for the electrons. Without any (Pereira et al., 2007): loss of generality let us consider Z > 0. The electronic 2 2 2 dynamics is governed by the wave equation ∂r f + ε +2gε/r γ(γ 1)/r f (r)=0, (4.5) ± − ∓ ± g where the f are linear combinations of F and G, ε2 v iσ ∇ + σ Mv Ψ(r)= EΨ(r). (4.1) ± − · − r 3 takes the place of the Schrodinger energy, and γ plays   the role of angular momentum. Since the solution is for- 2 2 mally the same, the appearance of ε instead of ε means Here we use g = Zα = Ze /(ǫ0v), with ǫ0 reflect- ing the effective dielectric constant of the embedding that the massless case admits no bound solutions, as we medium, and the mass M accounts for the more gen- expect on account of the absence of a spectral (mass) eral possibility of a symmetry breaking gap. Through- gap. The massive case, however, has a well defined infi- nite spectrum of bound solutions when ε < m , given out this chapter we shall use the scaled energy and | | | | mass ε = E/v, m = Mv, and k = √ε2 m2. by (Khalilov and Ho, 1998) Even though m = 0 for ideal graphene without− in- n + j2 g2 teractions, nonzero m can be induced in many ways. εn,j = sgm − , (4.6) 2 One of them is through interaction with suitable sub- g2 + np+ j2 g2 strates, of which some experimental hints have been − q reported (Gr¨uneis and Vyalikh, 2008; Li et al., 2009a;  p  lowest level is given by ε ε = s m 1 (2g)2. Martinazzo et al., 2010; Zhou et al., 2007). In terms of G ≡ 0,1/2 g − the original tight-binding Hamiltonian, the mass M aris- p ing from a sublattice symmetry is related to the parame- 2 2. Supercritical Instabilities ter ∆0 introduced in eq. (3.40) via Mv = ∆0. The axial symmetry of the potential allows us to use the eigen- Consideration of eq. (4.4) immediately reveals a com- states of the total pseudo angular momentum, Jz = Lz + plication if g>gc = 1/2, because γ becomes imaginary σz/2, which is conserved (DiVincenzo and Mele, 1984). for the lowest angular momentum channels (j = 1/2). 1/2 ± We write Ψj† = r− [Fj (r)Φj 1/2(φ), iGj (r)Φj+1/2(φ)], The solution (4.4) is neither regular nor divergent, but − where j = 1/2, 3/2,... are the eigenvalues of Jz, rather oscillates endlessly towards r = 0. This is patho- ± ± ipφ and the cylindrical harmonics read Φp(φ) = e /√2π. logical because the space of solutions is of dimension 2, A detailed derivation of the 2D Dirac equation for gen- and we can no longer discard an irregular contribution eral radial potentials is given by Novikov, 2007a. In our since both linearly independent solutions are square in- case, Eq. (4.1) reduces to the following radial equations tegrable. In other words, there is no boundary condition 24

the pure Coulomb case, εG(g) decreases towards zero in a singular way at g = gc. In a regularized poten- tial, εG depends also on the cutoff radius R, and is al- lowed to monotonically penetrate the negative energy re- gion, until eventually touching the lower continuum at ε = m. If g is further increased, εG dives into the hole (positron)− continuum and becomes a resonance. Other levels will sequentially follow at higher g. The diving point for εG(g) defines a renormalized critical coupling, g˜c > gc that is characterized by a log singularity at 2 2 mR 0:g ˜c gc + π / log (mR) (Gamayun et al., 2009; Khalilov∼ and≃ Ho, 1998; Pereira et al., 2008a; Zhu et al., FIG. 18 (Color online) Schematic drawing of the level diving 2009), strongly depending on the regularization. process in the supercritical regime, and of the resulting quasi- This diving of bound levels entails a complete re- spectrum of levels for massive and massless fermions. structuring of the vacuum. If the level was empty, an electron-hole pair will be immediately created: the elec- tron remains tightly bound and shielding the center, at the origin to univocally select the solution. Secondly, while the hole is ejected to infinity (Greiner et al., 1985; in the massive case the level ε becomes imaginary, sig- G Zeldovich and Popov, 1972). The supercritical regime naling a loss of self-adjointness of the Dirac Hamiltonian is thus characterized by spontaneous pair creation, or for g > 1/2. a spontaneous Schwinger mechanism (Schwinger, 1951). Physically, both effects are a symptom that the po- One expected consequence is a strong signature of these tential has such a strong divergence that particles are resonances in the hole sector of the scattering and trans- inexorably attracted and “fall” into the origin, leading port cross sections. to a collapse of the system (for example, the endless os- An essential detail is that these resonances are not cillations can be read as an infinite phase shift). This usual bound levels diluted inside a continuum, where “fall to the center” is a general characteristic of diverg- their lifetime essentially disappears. One consequence ing potentials in any dimension of space. For power law of the chiral nature of Dirac fermions, combined with potentials, one particular power signals the threshold of the long range tail of the Coulomb potential, is that the criticality. The Coulomb potential is the marginal case supercritical levels in the relativistic Coulomb remain for the Dirac equation (both in 2D and 3D), just like the sharply defined, with diverging lifetime. For example, potential 1/r2 is the marginal case of the 3D Schrodinger for S states (j = 1/2), one shows that these resonances equation (Landau and Lifshitz, 1981). This, of course, follow (Gamayun et al., 2009) begs the question of regularization. Regularizing the po- tential introduces an additional boundary condition at 3π π/√2ξ 3π(β βc) some short distance R, which allows a formal solution, εn m 1+ξ+i e− , ξ = − , (4.7) ≈− 8 8ββc and cures the total collapse of the system (Case, 1960;   2 Perelomov and Popov, 1970). In graphene the lattice is when g & g˜c, and where β = iγ, βc = g˜c 1/4. In the natural regulator and there are no ultraviolet issues. real space the localization of the supercritical− levels is p But the physics in the supercritical regime depends ex- controlled by the reduced Compton wavelength: λC = plicitly on the short range details. 1/(mv). The modulus squared of their wavefunction de- This supercritical collapse has a long history in the cays as Ψ†Ψ exp( 8gr/λC ) and, consequently, even context of QED, where the Dirac equation stands as the inside the continuum,∝ − such levels retain a highly localized basis for understanding the stability of matter. In QED nature, which is whyp they are so relevant, in particular the collapse would occur for ZαQED > 1, which lead to in their potential for screening (Pereira et al., 2008a). extensive investigations regarding the stability of heavy nuclei having Z>Zc = 137 (Case, 1960; Greiner et al., 1985; Popov, 1971a,b; Zeldovich and Popov, 1972). Af- b. Massless Electrons. The spectrum in this case is con- ter regularization Zc 170, which makes the problem tinuous everywhere, and thus there is no sequential div- highly academic, and→ QED’s predictions untestable. In ing and restructuring of the hole continuum as described graphene, on the contrary, Zc 1, which opens the real above. But the pathology associated with Eq. (4.4) still possibility of testing the supercritical∼ instability in a con- exists. Physically, the massless situation is rather more densed matter setting. catastrophic since the solution in a regularized potential reveals an infinite number of quasi-localized resonances in the hole sector (Gamayun et al., 2009; Pereira et al., a. Massive Electrons. To understand the physics in the 2007; Shytov et al., 2007b). This is a highly non-trivial supercritical regime we can follow the level εG as effect for several reasons: (i) in the massless case there the coupling increases (Fig. 18) (Greiner et al., 1985; is no natural length scale in the problem to characterize Pereira et al., 2008a; Zeldovich and Popov, 1972). For such localized states; (ii) the system abruptly develops 25 an infinite quasi-bound spectrum at g >gc, when its spectral fingerprint is rather featureless for g

a + ib πn/√g2 g2 ε e− − c , (a,b) (g), (4.8) n ≈− R ∼ O which has an essential singularity at gc, an energy scale/lower bound set explicitly by the regularization dis- tance, R, and diverging lifetimes close to the critical point. Since the width of these states vanishes linearly, they are practically bound states (hence the designation quasi-bound states). In real space, the localization scale is determined by the regularization distance R itself. Since meso and nanoscopic devices are of high interest, it is pertinent pointing out that massless Dirac fermions FIG. 19 (Color online) (a) LDOS, N(ε,r) at r = a for several in a finite-sized system mimic in all aspects the physics couplings g 3. DOS, Scattering and Transport Cross Sections gc. The inset shows the oscillating LDOS correction for ε > 0. (d) Energy dependence of the phase shifts (top) and the Here and in the coming sections we shall be concerned supercritical contributionn ¯j (ε,r) to the LDOS (bottom) for mostly with massless Dirac fermions, except when explic- g = 1.0. itly stated otherwise. The local density of states (LDOS) and cross sections are useful quantities insofar as they are directly accessible in local probe and transport experi- now clear that strong resonances, decaying rapidly with ments. The LDOS per unit area and spin is isotropic, and distance, appear in the vicinity of the Dirac point, signal- can be written in closed form in terms of partial waves ing the presence of the quasi-bound levels (Pereira et al., as N(ε, r)= j nj (ε, r), (Pereira et al., 2007) with 2007; Shytov et al., 2007b). Their exponential accumu-

2 lation at ε = 0 is confirmed in Fig. 19(d) where we show Pj 2gsε 2 2 the supercritical contributionn ¯j (ε, r) as a function of nj(ε, r)= Fγ 1 + Fγ + Fγ Fγ 1 (4.9) 2π2γ2r − j − log( ε ). At positive energies the LDOS exhibits peri-  | |  odically| | decaying oscillations in εr [inset of Fig. 19(c)], for g < g , and F represents the Coulomb func- c l with extrema separated by nπ, within logarithmic ac- tion F ( gs , ε r) (Abramowitz and Stegun, 1964). The l ε curacy (Shytov et al., 2007a≈). When directly measured function−N(ε,| r)| is plotted in Fig. 19(a) for different cou- in STM such oscillations can be used to extract the elec- plings and distances. Apart from the evident particle- tronic dispersion, as done by Ouyang et al., 2002. hole asymmetry, the LDOS remains rather featureless, We point out that, since the solution of the supercrit- even at the shortest distances. If g >gc the corre- sponding analytical expression obtained in the regular- ical problem involves a nontrivial ad-hoc regularization, ized potential is more complex, but still has a closed form these results have been checked numerically against ex- (Pereira et al., 2007). In this case, supercritical channels act solution of the full tight-binding problem in the hon- ( j < 1/2) need to be isolated from undercritical ones eycomb lattice, being found that the analytical Dirac re- (|j| > 1/2), yielding two contributions to the LDOS: sults reproduce the full lattice problem down to distances | | as small as the lattice scale (Pereira et al., 2007).

N(ε, r)= n¯j (ε, r)+ nj (ε, r). (4.10) The striking differences between the two regimes and j < g j > g the violent modification of the ground state at strong cou- | X| | | | X| | | pling are likewise evident in the behavior of the scattering The total LDOS for this case is shown in Fig. 19(c) for phase-shits, δj (ε). They admit closed formed expressions g = 1.0, and at different distances to the impurity. It is at both g

Shytov et al., 2007a) and g > gc (Castro Neto et al., recognized that conventional procedures of the theory of 2009b; Shytov et al., 2007b). For example, the under- metals, like self-consistent screening, linear response or critical S-matrix reads (Novikov, 2007a) Friedel sum rules, are not straightforward in this sys- tem. For example, within the Dirac (effective mass) ap- iπ(j γ) 2iδj (ε) je − Γ(1 + γ igsε) proximation, the ultraviolet cutoff scale enters explic- Sj (ε)= e = − , (4.11) γ igsε Γ(1 + γ + igsε) itly in Friedel’s sum rule, and Levinson’s theorem is − modified (Lin, 2006) (Levinson’s theorem is one of the which is energy independent, but considerably asymmet- fundamental results in quantum scattering theory, as- ric with respect to the sign of g. The corresponding δj serting that in the Schr¨odinger’s equation with a non- are shown in Fig. 19(b)(inset) as a function of coupling singular spherically symmetric potential the zero en- strength. Note how δ1/2 (the most important partial ergy scattering phase-shift exactly counts the number of wave) behaves rather differently from the others: only bound states: δl(0) = Nlπ). One consequence is that δ1/2 shows the expected sign for the attractive/repulsive a na¨ıve application of Friedel’s sum rule can yield di- situations. On the other hand, in the supercritical regime vergent displaced charges (DiVincenzo and Mele, 1984). there is a strong ε-dependence of δj . In the top row of Even though these divergences are artificial in the target Fig. 19(d) we present (δj mod π) as a function of log(ε). lattice problem, they point, already at a single particle In the attractive sector (ε< 0 if g > 0) the abrupt steps level, to the anomalous screening properties of graphene. centered around π/2 mark the position of the infinite quasi-bound spectrum (which, as per (4.8), accumulates exponentially at ε = 0), whereas in the attractive sector 1. Weak Coupling (g 0 and g < 0 can be used to vided−| | in (4.9). One difficulty with this approach is that extract the density of positively and negatively charged the resulting density per partial wave behaves asymptot- impurities (ni±) from a single measurement of the elec- ically as trical conductivity, σ, as a function of carrier density

(Novikov, 2007b). This technique has been used in some 1 g 0 2 δn (r ) D D + (r− ) , (4.13) experiments (Chen et al., 2009a,b, 2008), but the asym- j → ∞ ∼ r − r − O metry effect can be easily masked by other spurious in- h i fluences (Barraza-Lopez et al., 2010; Huard et al., 2008; which diverges upon summation over j (a reminiscence Nouchi and Tanigaki, 2010). Moreover, on account of the of the problems associated with the ultraviolet scale al- 0 ε-independence of δj in (4.11), the corresponding Drude luded to above). In the above expression D and D 2 2 conductivity, σ =4πe µ/(vniΛtrh ), is immediately seen represent the cutoff in the presence and in the absence to scale linearly with density: σ µ2 n. Therefore, of the coulomb center, respectively. Since the sublead- the linear-in-density conductivity,∝ which∝ appears already ing terms in (4.13) are convergent in j, we regularize it g in the first Born approximation, remains when the cross by taking a position dependent cutoff: D D0 + . → r section is calculated exactly. As a result, the total induced density acquires the form For supercritical potentials, and similarly to the δn(r) H(D0r)/r3, where H(x) is a constant-amplitude ∼ LDOS, there will be undercritical and supercritical par- oscillating function (Pereira et al., 2007). Since it is de- tial waves contributing to Λtr(ε) [cfr. eq. (4.10)]. The sirable to have control over the validity of the regular- latter give rise to strong peaks in the transport cross- ization procedure outlined above, we have calculated the section at densities for which the Fermi energy matches total induced density δn(r) in the full tight-binding prob- the levels εn (Shytov et al., 2007b), tallying with the be- lem, via exact diagonalization. The result is plotted in havior of the DOS. Fig. 20(a), and unequivocally shows the predicted 1/r3 decay, with oscillations on the scale of the lattice. Such fast decay implies that the induced charge concentrates B. Induced Charge and Screening within a small vicinity of the impurity. Moreover, the numerical results in the lattice further suggest that such First attempts at understanding screening in graphene distance is of the order of the lattice parameter a: the date back to DiVincenzo and Mele, 1984, where it was inset in Fig. 20(a) reveals that the total charge pulled in- 27

(a) (b) 0.5 RPA 1 Dirac Eq (GF) 0.4 Exact (lattice) 0.3 α eff 0.5 Z 0.2

Z = 1 0.1 (total induced charge) Z = 2

Q Z = 3 0 0 0 0.1 0.2 0.3 0.4 0.5 1 2 3 4 g = Zα g = Zα

FIG. 21 (Color online) (a) Total integrated charge in the vicinity of the impurity, Q, obtained: from exact diagonal- ization in the lattice (dots), from RPA (4.16) (blue), and from the exact Green’s function in the Coulomb field (4.19) (red). (b) The self-consistent Zeff, obtained from Eq. (4.21) (Terekhov et al., 2008). Numerical data (dots) is plotted after accounting for finite-size renormalization of gc (Pereira et al., 2008a).

in the usual sense, because it merely renormalizes the strength of the impurity: Z Zeff = Z Q. This leaves Coulomb’s law unaltered,→ except for the− substitu- tion Z Z . → eff

b. Linear (RPA) Screening. Single particle results, like the one above, are not generally sufficient to draw con- clusions about screening. Consider now the same prob- lem in linear response, at the RPA level, which is jus- tified for small, undercritical couplings. Within the FIG. 20 (Color online) (a) Induced electron density, δn(r), RPA, the Fourier transform of the statically screened (1) plotted as a function of distance to the Coulomb center, for potential is given by Us(q) = U0(q)/[1 Π (q)V (q)] − 2 different impurity strengths, gatoms. Black lines are 1/r , and guides for the external impurity potential. From (2.14) we know that eye. Inset shows the saturation of∝ the integrated charge ac- Π(1)(q 0) q/(4v), and hence cumulated inside r < Rmax, as a function of Rmax. (b) Same → ≈− as (a), but for the supercritical case, g>gc, and the dashed 1 line is now 1/r2. π − U0(q) ∝ Us(q) U0(q) 1+ α = . (4.15) ≈ 2 ǫRPA   Therefore linear response confirms the absence of screen- side a region r < Rmax saturates within very few lattice spacings. In fact, since D0 1/a, in the limit a 0 ing, except for the trivial renormalization of the static ∝ → dielectric constant: ǫ ǫ = ǫ (1 + πα/2) (Ando, (where the effective mass description is meaningful) the 0 → RPA 0 analytical expression δn(r) H(D0r)/r3 can be seen as 2006). Likewise, the induced density can be computed ∼ in linear response from δn(q) = ZV (q)Π(q) or, in the a representation of the 2D Dirac-delta function. In other − words, we expect the induced charge density to behave RPA: as (1) Π (q)V (q) iq.r a 0 δn(r)= Z dq e , (4.16) δρ(r)= e δn(r) → Q e δ(r). (4.14) − 1 Π(1)(q)V (q) −| | −→ − | | Z − The same conclusion follows from a modified Friedel ar- yielding δρ(r a) δ(r)Z e πα/2 to linear order gument (Shytov et al., 2007a), and from the exact cal- in α (Kolezhuk≫et al.,∼2006 − ). This| | is exactly what was culation of the non-interacting Green’s function in the obtained in (4.14) from a single particle, wavefunction, Coulomb field (see below) (Terekhov et al., 2008). The perspective. In addition, the argument that the Fourier induced charge has a screening sign, as expected, but transform of δn(r) is dimensionless can be used to show the strongly localized distribution of the induced charge that it should be a pure constant in undoped graphene, (4.14) implies that undoped graphene cannot screen for which there is no natural length scale. As a result, 28 that δρ(r) δ(r) remains true in all orders of pertur- QED (Mil’shtein and Strakhovenko, 1982). The main re- bation theory∝ (Biswas et al., 2007). For consistency, the sult is that total induced charge, Q, introduced in (4.14) is then given δρ(r)= Qδ(r)+ δρ (4.19) by − dist π where δρdist(r) represents a positive charge distributed Q = Zα + (higher orders in Zα). (4.17) 2 at r = (needed to satisfy the constraint of total zero induced∞ charge). It is significant that this approach af- To verify this correspondence we can compare (4.17) with fords an exact expression for the dependence of Q upon the value of Q extracted from the non-interacting exact g = Zα, which is shown in panel (a) of Fig. 21. A series diagonalization in the honeycomb lattice. As shown in expansion of this dependence yields the following: Fig. 21(a), the numerical Q for different values of Z fol- π 3 5 lows the relation (4.17) for most of the range 0 gc, self-consistent screening the approximation becomes increasingly unwarranted as delays the supercritical threshold because the condition g nears the critical threshold, gc = 1/2, which is non- Zeffα = 0.5 requires a higher bare Z. This phenomenon perturbative. Rather than analyze this limit on the ba- is most striking for Z = 1, in which case the supercritical sis of exact wavefunctions in the Coulomb field, as was point disappears altogether (˜gc < 1/2 even as Z ), Z=2 Z=3 → ∞ done in Sec. IV.B.1.a, we now describe the solution ob- whereas gc = 1.136 and gc = 0.798. The predic- tained by Terekhov et al., 2008. These authors bypass tion of this self-consistent Hartree renormalization of Zeff the solution of the Dirac equation, obtaining instead an would then be that impurities with Z = 1 can never be- exact integral expression for the Green’s function in a come supercritical. In addition, Hartree screening is suf- Coulomb field, using a proper-time approach common in ficient to suppress the tendency for over-shielding of the 29

Coulomb center: as seen in the inset of Fig. 21(b), Zeff equation for g>gc. Now we address the correspond- remains always positive. ing induced charge obtained using the same procedure An alternative approach to the Hartree screening con- as in Sec. IV.B.1.a. Consideration of the exact wave- sists in treating the induced charge in linear response, functions (Pereira et al., 2007) or the exact phase-shifts δρ(q) = ZV (q)Π(q), but taking into account electron- (Shytov et al., 2007a) leads to the conclusion that the electron interactions perturbatively, via the renormal- supercritical partial waves contribute with an induced ization of the coupling constant (Biswas et al., 2007). charge 1/r2. This could be expected on dimensional This is valid for small α (weak interaction), and leads grounds:∝ δ(r) and 1/r2 are the only dimensionally con- to a result formally equivalent to (4.19), but where sistent possibilities in the absence of any intrinsic length δρdist now arises from the electronic correlations. The scale in massless graphene. The exact induced density distributed charge in the interacting case also has an per partial wave reads (Shytov et al., 2007a) anti-screening sign, but decays as 1/r2, while the non- 2sg 2 2 interacting δρdist(r) is zero everywhere, except at infin- δn¯j(r)= g j , (4.22) ity. π2r2 − p Even though the above considerations pertain to un- and, like the undercritical contributions, has a screening doped graphene, since all screening charge accumulates sign. The full induced charge is obtained from δρ(r) = completely within a narrow distance, finite densities are e δn(r), n(r)= δn¯j + δnj , and has the −| | j gc not expected to alter the picture for as long as qs =4αkF general form | | | | remains large compared to the lattice scale a. P P 1 δn(r)= s A + Bs δ(r). (4.23) g r2 g 2. Strong Coupling (g>g ) c If 1/2 < g < 3/2 eq. (4.23) reduces to δn(r) = 2 2 2 2 In Sec. IV.B.1.c Hartree screening was shown to renor- (πg/2)δ(r)+2sg g gc /(π r ). The general behavior (4.23) is also confirmed− numerically by exact diagonaliza- malize gc and delay the critical threshold. Two important p questions naturally arise: (i) since the self-consistent so- tion of the tight-binding Hamiltonian in the honeycomb lution of (4.21) is uncontrolled, how certain can one be lattice, whose results are plotted in Fig. 20(b). that the critical regime is reachable at all? (ii) So far we looked only at screening from the undercritical side (i.e. b. Supercritical Protection. Unlike the undercritical as long as Zeffα < 1/2). How can one address screening from the supercritical side, given that this regime cannot regime, the additional power law decay in (4.23) causes be reached perturbatively? a modification of Coulomb’s law at large distances. But since we have a quasi-atom with all levels (4.8) filled, The answer to these questions is far from trivial. In the non-interacting result in Eq. (4.23) cannot be the fi- QED it is related to the ground state and stability of nal answer. Each level is quasi-localized on the lattice super-heavy nuclei (Z & 170), when the bound spectrum scale, and should contribute significantly to shield the dives into the positron continuum (Fig. 18). Despite hav- Coulomb center. For g not too much above g we can ing received considerable attention throughout the 1970- c follow an argument advanced by (Shytov et al., 2007a) 80’s (Greiner et al., 1985), the fact that these systems that assumes electrons at some distance r feel the effect require such high Z’s, has turned it largely into an aca- of a point charge consisting of the impurity subtracted demic problem. The exciting prospect about graphene is from all the accumulated screening charge up to r. In that impurities with Z = 1, 2 might already display su- other words, we introduce a distance dependent impu- percritical physics, in which case it would afford a bench- r rity strength, Z (r) = Z δn(r)dr, and substitute top test of some yet untested QED predictions. eff R (4.23) for δn(r): − The essence of the difficulties in treating the super- R critical regime clearly lies in its non-perturbative nature. 2 2 π 4 g gc r Graphene, being gapless, is even more pathological be- Zeff(r)= Z g − log (4.24) cause of the infinite quasi-spectrum that appears in the − 2 − p π R hole channel [Fig. 18]. This quasi-spectrum is akin to an Since the log term represents the renormalization com- atom filled with infinitely many electrons and, as known ing from screening at distances away from the center, we from studies of heavy atoms (Landau and Lifshitz, 1981), should replace (g Zα) (Zeffα geff). This leads to it requires full consideration of correlations and interac- a self-consistent renormalization≡ → of≡ the coupling that can tions, and self-consistent techniques like Thomas-Fermi be written in an appealing RG fashion as dgeff/d log(r)= (Fermi, 1927; Thomas, 1927). 4α g2 g2. In this way, it can be immediately seen − eff − c that the coupling geff will “flow” to the constant value gc withinp a finite distance [see also (Gupta and Sen, 2009) a. Non-interacting Induced Charge. In Sec. IV.A.3 we for a related renormalization procedure]. As such, irre- saw some unusual consequences for the DOS and cross- spective of the bare Z, the system self-consistently re- sections extracted from the exact solution of the Dirac arranges itself so that electrons at large distances never 30 feel a supercritical effective coupling. The undercritical precisely, it applies for 1/Z α 1/√Z, and pro- (stable) situation is therefore protected. This reasoning vided that log(r/R) < 1/α.≪ Otherwise,≪ for intermedi- agrees with expectations for the corresponding problem ate electron-electron coupling (α 1), the asymptotic in QED, where it was shown that, within Thomas-Fermi, screened potential should follow V∼ Z e2/(ǫ r), with eff ≈ c 0 the vacuum polarization charge in super-heavy nuclei be- Zc = gc/α = 1/(2α). This result embodies the under- haves in such a way as to reduce Z to the threshold value critical protection discussed above in Sec. IV.B.2.b, in- (M¨uller and Rafelski, 1975). sofar as the supercritical core is always self-consistently This is quite different from a metal, to the extent that screened so that Zeff Zc. Moreover, within the su- graphene always leaves an universal amount of charge percritical core region,→r < 2Zα2R, the effective poten- 3/2 (Zc = gc/α) unscreened at large distances. Such behav- tial decays as 1/r . This obtains treating graphene ior derives from the sharp transition between the under as an ideal classical∝ metal, under the assumption of and supercritical regimes. On the one hand, the system quasi-complete screening in the core region (Fogler et al., wishes to screen as much charge as it possibly can. But, 2007). on the other, it cannot screen if g

We now briefly address the differences expected in c. Nonlinear Thomas-Fermi and Beyond. While the above the screening properties of charged impurities in massive approach is valid in principle only for g & gc, the fact that graphene. We shall consider only the undoped situation, qualitatively supercritical graphene resembles a super- and assume µ = m, such that none of the bound levels heavy atom suggests the use of TF theory, which is (4.6) are occupied.− exact for atoms with Z (Lieb, 1981), and af- → ∞ fords an approximation from the opposite limit g gc. If we wish to calculate how Coulomb’s law is modi-≫ fied in this regime we can calculate the total potential a. Weak Coupling (g < gc). It is clear that at weak ′ e2 δn(r ) coupling one can directly rely on perturbative results V (r)= V (r)+ δV (r), where δV (r)= ′ dr is eff ǫ0 r r ′ | − | (Sec. IV.B.1), and obtain the induced density from the potential induced by the screening charge. Within (1) R δn(q) = ZV0(q)Π(q). Π (q) has been calculated TF we replace δn(r′)= n[µ V (r)] n(µ), and the ho- in (2.14),− and simple substitution yields the following − − 2 2 mogeneous density depends on µ via n = sEµ /(πv ). asymptotics: Solution of the resulting integral equation leads to the correction to Coulomb’s law, which asymptotically reads π δ(r) r a 0 (Katsnelson, 2006) 2 ≃ → r 2 λC δn( ) Zα  λC− log r a r λC , (4.26) 2 ∼ − 3 ≪ ≪ e Z  λC r− r λC r Veff( ) 2 , (4.25a) − ≫ ≈ ǫ0r 1+2Zα log(r/R)    e2 Z where λC = 1/(mv) is the Compton wavelength, and V (r) , (4.25b) a the lattice parameter of graphene. The short distance eff ≈ ǫ r(q r)2 1 2Zα2 log(q R) 0 s  − s  term is the same as found in the massless case (4.14,4.17), which makes sense given that when r λC the sys- valid for µ = 0, r R and µ = 0, rqs 1 respectively, ≫ 6 ≫ tem does not “feel” the mass yet. It has≪ a screening where qs = 4αµ/v is the screening length (4.18). One notes that the overall space dependence is formally the sign. However as the distance increases screening is in- same as the one obtained within RPA, both at zero and creasingly suppressed, first weakly up to λC , and then finite density. Hence the bracketed coefficients in (4.25) strongly, beyond λC . In fact, since here δn(q = 0) = 0, can be interpreted as a renormalization of the valence. we have exactly δn(r)dr = 0. The meaning of this is simple: the total induced charge is zero. The system The important difference is that, in the limit Z of R interest in the context of TF, the nominal valence→Z ∞dis- cannot screen beyond r & λC because it is essentially an insulator (or a semiconductor with µ in the middle of appears from Veff(r), which thus becomes universal (and undercritical). Hence, even for strong impurities one can the gap). Notwithstanding, unlike a conventional insula- formally use perturbative expressions for the screened po- tor, gapped graphene shows a novel screening behavior at tential, corrected for this renormalization of Z. short distances, reflected in the live dependence of δn(r) It is important to emphasize that, since at this stage we on the distance up to λC . are concerned with screening and corrections to the in- duced charge coming from electron-electron interactions, g = Zα is no longer the relevant parameter alone, but b. Strong Coupling (g>gc). In gapped graphene, screen- both Z and α (that controls the interaction) indepen- ing in the supercritical regime is qualitatively easier to dently. For this reason, Fogler et al., 2007 have argued understand, at least when g & g˜c. If the first level that the result (4.25) is valid only for small α. More has just merged inside to hole continuum, its effective 31 probability density, Ψ (r) 2, remains exponentially lo- general effect present in the two body problem. In this | c | calized, as described in IV.A.2.a. Invoking complete- case the critical coupling occurs at αc = 1 and αc =2.24 ness of the set of single-particle states, one can eas- for s and p channels, respectively. The interacting two- ily show that the non-interacting induced charge follows body problem usually encodes much of the physics that (Pereira et al., 2008a) the many-body system displays. One example is the study of pairing, pair condensation, and other processes 2 δn(r) Ψc(r) + δnpol(r), (4.27) which are dominated by two particle channel events. This ≈ | | has a clear relation with the issue of spontaneous gap 2 0 where δnpol(r) E< m χE(r) χE(r) represents generation, discussed in Sec. III.B. The prospect of exact ≈ − | | − | | the vacuum polarization (i.e.: the induced charge coming solution of the two particle problem would afford more P from the full set of plane wave states), and is the same controllable means to explore this instability in graphene. quantity that obtains in RPA (4.26). Clearly, the contri- bution from the supercritical state alone makes δn(r) in (4.27) highly localized within the Compton wavelength, 2. Excitons and Spontaneous Mass Generation λC . For all purposes, this state screens like a bound state would, and consequently one expects the impurity It is noteworthy that the value α = 1 quoted valence to be reduced by one unity times the degeneracy, c above is tantalizingly close to recent calculations of N, of the level. But since N = 4, this would imply, for the critical coupling which precipitates a spontaneous the experimentally significant cases of Z 1, a tendency mass generation and metal-insulator transition in un- to over-screen the Coulomb center. This∼ bring us again doped graphene. Those values range from α = 0.8 to the role of interactions. The above would be true in c (Vafek and Case, 2008), to α = 1.1 obtained within the limit of weak interaction α 1. But, in that case, c Monte Carlo (Drut and L¨ahde, 2009b) or by using the the supercritical regime would require≪ Z 1, which is Schwinger-Dyson equation (Khveshchenko, 2009). As de- not feasible. In the end, if supercritical systems≫ are to be scribed at length in Sec. III.B, this metal-insulator tran- produced, electron-electron interactions should be strong sition in graphene has been ascribed to the emergence of which, besides requiring the computation of the vacuum an excitonic instability beyond α . polarization in strong-coupling, brings the question of the c Recently the excitonic problem has been considered renormalization of the bound levels themselves (Lamb vis-a-vis the supercritical instability of the Coulomb shift). This situation, however, is completely analogous center. Instabilities in the particle-hole channel ap- to the problem of super-heavy nuclei in QED, and an ex- pear at critical couplings consistent with the above tensive account of its particular features and difficulties (Gamayun et al., 2009; Wang et al., 2010). For exam- can be found in Greiner et al., 1985. ple, Gamayun et al., 2009 show that solving the Bethe- Salpeter equation in graphene leads to instability-prone 2 C. From Single to Many Particle Interactions tachyonic states (E < 0) at αc = 1.6. Such states are the analogue in the two channel many-body language of Coupling to an external Coulomb field can be seen as the quasi-bound resonances for supercritical impurities, the zero-th order approach to the full many body elec- and a glimmer of supercritical effects in the fully inter- tron interactions in graphene. The decisive difference acting problem. that leaves graphene apart from standard electronic sys- tems is the existence of the supercritical region, which, for the Coulomb field, has the peculiarities discussed so D. Supercritical Physics in Experiments far. Since the coupling constant in vacuum is α 2, one can justifiably ask whether supercritical effects≈ carry to The non-perturbative nature of supercritical Coulomb electrons interacting among themselves. After all, even if impurities, and the associated analytical difficulties, pre- a simplification, from a reference frame moving with an clude unequivocal predictions regarding the possibility of electron the problem becomes an impurity one again. crossing the supercritical threshold. Experimental inves- tigation of this problem requires the ability to vary the strength of the Coulomb impurity and/or the electron- 1. Interacting Two Body problem electron interactions. Control over the dielectric environ- ment provides a handle to tune interactions and impurity The two particle problem has traditionally provided strength at the same time, via selection of ǫ0. Experi- valuable insights into the full many-body phenomena ments in this vein have been performed by Jang et al., in condensed matter [e.g. the Cooper pairing (Cooper, 2008 and Ponomarenko et al., 2009, showing that it is 1956)]. The chiral nature of the electronic states, how- possible to controllably tune the value of ǫ by explor- ever, precludes the usual decoupling between center-of- ing substrates with different dielectric properties. Vari- mass and relative coordinates, except for s-states in a ation of Z is a more delicate issue. Chen et al., 2008 quiescent center-of-mass (Sabio et al., 2010b). Even so, have devised a way to add monovalent ions to graphene these authors show that the supercritical collapse is a via K-irradiation, in quantities that can be controlled 32 with some precision. But exploration of the supercritical 2007), it can be drastically enhanced either by curva- regime might require higher valences. For real impurities, ture effects (Huertas-Herno et al., 2006), or by impu- the valence is determined by the nature of the impurity rities (Castro Neto and Guinea, 2009). The spin-orbit atom and the host system, and cannot be changed. One coupling is also logarithmically enhanced by Coulomb can, in principle, use ions of different valence, but here interactions (Kane and Mele, 2005), as discussed in Sec. the difficulty lies in the fact that valences higher than III.B. When the mirror symmetry is broken either by a Z = 2 are very unlikely. One possible alternative to this substrate or external electric field, an additional Rashba constraint imposed by nature, would be to resort to sharp term is allowed STM tips, whose strong local field could mimic a strong local charge. As mentioned in the beginning of this chap- R = λR Ψ† τ3 (σ1 s2 σ2 s1)Ψk,σ , (5.2) H k,σ ⊗ ⊗ − ⊗ k ter, the experimental exploration/confirmation of the su- X,σ percritical state would be rather important a milestone. Not only in understanding the physics of graphene, but where λR > 0 is the Rashba coupling. The induced gap is 2(∆ λ ) for λ < ∆ , closing to zero when because it would afford a glimpse to what might happen SO − R R SO in the more fundamental QED situation. λR > λSO (Kane and Mele, 2005). Kekule lattice distortions (Hou et al., 2007), which break the translational symmetry of the lattice, also lead V. STRONG CORRELATIONS IN GRAPHENE to the opening of gaps in graphene, whereas lowering the rotational symmetry of the C3v group, by stretching A. Mass gaps in the honeycomb lattice the honeycomb lattice in one direction, does not. In the presence of topological defects in the order parameter, Graphene is a semi-metal (SM) with gapless quasipar- such as vortices, the midgap states which are bounded ticles. The Dirac points in graphene are protected by to them allow the emergence of excitations with frac- the combination of sublattice and translational symme- tional statistics under vortex exchange (Chamon et al., tries of the honeycomb lattice. The point group symme- 2008a,b; Hou et al., 2007; Seradjeh and Franz, 2008). In try of the honeycomb lattice, C6v, can be decomposed the superconducting case, the vortex core may sustain into the point group of the triangular sublattice and the a quantum Hall state in the presence of a strong Zee- Z2 sublattice symmetry group, C3v Z2. Violation of man coupling of the electrons with the magnetic field, sublattice symmetry leads to the opening⊗ of a mass gap which lifts the spin degeneracy (Herbut, 2010). In the in the Dirac Hamiltonian. This broken symmetry can most general case, where any spin, valley and pairing be physically implemented either by the Semenoff gap symmetries are allowed, 36 different types of instabilities (Semenoff, 1984), which is induced by a staggered scalar that generate mass gaps in graphene have been classified potential that breaks the sublattice inversion symmetry, (Ryu et al., 2009). as previously discussed in Eq. (3.40), or by the Haldane gap (Haldane, 1988), where there is an additional broken time reversal symmetry (TRS) induced by the inclusion B. Charge and magnetic instabilities of circulating current loops with zero magnetic flux per unit cell, corresponding to a staggered magnetic field. In Although no evidence of mass gaps has been found in particular, a system that breaks inversion and TRS is graphene, numerical results have predicted a semi-metal- susceptible to a “parity” anomaly, where the application insulator (SM-I) transition in the presence of strong cor- of an electric field generates a net axial current flowing relations. Quantum Monte Carlo (QMC) calculations on between the two valleys in graphene (Jackiw, 1984). the Hubbard model for the honeycomb lattice at half In the presence of mirror symmetry along the z- filling predicted the opening of a Mott gap above the axis, the spin-orbit interaction in graphene has the form critical ratio U/t & 5 (Martelo et al., 1997; Paiva et al., (Kane and Mele, 2005) 2005; Sorella and Tosatti, 1992), where t 2.8 eV is the hopping energy and U is the on-site electronic≈ re-

= ∆ Ψ† τ σ s Ψk , (5.1) pulsion. A more recent QMC calculation has found HSO SO k,σ 0 ⊗ 3 ⊗ 3 ,σ k,σ a gapped AF state at half filling for U/t > 4.3, pre- X ceded by an intermediate coupling insulating phase for where ∆SO is the spin orbit coupling gap, and s3 is 3.5 < U/t < 4.3, which has been attributed to a gapped the diagonal Pauli matrix in spin space. The other spin liquid state formed by short-range resonating va- matrices follow the convention in the Dirac Hamil- lence bonds (Meng et al., 2010). An insulating antifer- tonian (2.6). The spin-orbit interaction in graphene romagnetic (AF) ground state has been also predicted breaks the spin degeneracy in the valleys, giving rise above U/t & 4 (Furukawa, 2001; Martelo et al., 1997). to spin polarized currents that flow along the edge Variational (Hanish et al., 1995) and mean field calcula- states of the system — a quantum spin Hall state tions (Peres et al., 2004) predicted the possibility of Na- (Kane and Mele, 2005). Although the spin orbit cou- gaoka ferromagnetism (where the polarization is maxi- 3 pling gap in graphene is rather small, ∆SO 10− meV, mal) above a critical coupling both in the half filled and (Huertas-Herno et al., 2006; Min et al., 2006≈; Yao et al., in the doped regimes. Although the validity of the Hub- 33 bard model in graphene may be questioned since it does U not include long range Coulomb interactions, it could be in principle justified if one accounts for strong screening AF effect from a substrate which can deplete the long range Uc part of the interactions (or also, perhaps, by account- ing for dynamical screening effects from graphene itself SM CDW (Reed et al., 2010)), leaving only the short-range part of the electron-electron interactions. The extent of validity Uc 2U c V of the Hubbard model in graphene is a subject of ongoing debate. FIG. 22 Semi-metal (SM) insulator transition predicted by The bare spin polarization in graphene is a 2 2 tensor the renormalization group analysis of the extended Hubbard (Peres et al., 2004), × model, in large N expansion. U is the on-site Hubbard cou- pling and V is the nearest neighbor site repulsion. Uc stands + + for the critical coupling. AF: antiferromagnetic phase; CDW: Πx,y−(q, τ)= Sx (q, τ)Sy−( q, 0) , (5.3) h − i charge density wave state (Herbut, 2006). + where Sx and Sx− are the spin raising and lowering oper- ators in the two sublattices, x = a,b. Written in terms of the Green’s function (2.11) with additional spin labels, of quantum Hall ferromagnetic states at integer values of the filling factor (Nomura and MacDonald, 2006). The ′ (1)+ 1 s s magnetic field has been also proposed as a source of a Π −(q,iω) = (k) (k + p) x,y −4 Ax,y Ay,x × charge density wave (CDW) Peierls distortion in the zero k,s,s′= X ± Landau level in graphene, breaking the parity symmetry f [Es, (k)] f [Es′, (k + q)] between the valleys (Fuchs and Lederer, 2007). For a ↑ − ↓ , (5.4) iω + Es, (k) Es′, (k + q) discussion of interaction effects at strong magnetic fields, ↑ − ↓ see sec. VIII. s where ˆ 1+sk σ/k, and Es,σ(k)= sv k µ describes For Dirac fermions in 2+1 dimensions, a CDW insta- the twoA branches≡ of· the spectrum near the| |− Dirac points. bility translates into the phenomenon of chiral symmetry Since Πa,a = Πb,b and Πa,b = Πb,a∗ by the honeycomb breaking (CSB), with spontaneous generation of a mass lattice symmetry, the eigenvalues of the spin polariza- term that breaks the sublattice symmetry. The AF state + + tion are Π = Π − Π − , which correspond to is favored by strong on site repulsion and competes with F/AF a,a ± | a,b | ferromagnetic (+) and AF ( ) states. In RPA, the spin the long range part of the Coulomb field, which can fa- −(1) 1 (1) susceptibility isχ ˆ = [1ˆ UΠˆ ]− Πˆ , and the critical vor either strong coupling ferromagnetism (Peres et al., Hubbard coupling required− for a divergence in the spin 2004) or else excitonic CDW instabilities at strong cou- susceptibility in graphene is (Peres et al., 2004) pling (Drut and L¨ahde, 2009a,b; Khveshchenko, 2001a,b; Khveshchenko and Leal, 2004; Liu et al., 2009). F/AF 1 At large N, with N the number of fermionic flavors, Uc = . (5.5) Π(1) (0) the continuum limit of the Hubbard model in the honey- F/AF comb lattice falls in the universality class of the Gross- The ferromagnetic transition translates in the condition Neveu model (Gross and Neveu, 1974) for massless Dirac F 2 fermions in 2+1 dimensions, with four-fermion contact Uc = 2/ρ(µ) D / µ , which is the Stoner criterion, where ρ(E) is the≈ DOS| | and D the band width. The AF interactions. The extended version of this model accom- AF 2 modates the short range piece of the Coulomb interaction transition occurs at Uc D /(D µ ). The application of an in≈ plane magnetic− | | field, B, splits involving the repulsion between nearest neighbor sites, V the spin degeneracy at the Dirac points, creating two (Herbut, 2006). In addition to the Gaussian fixed point, Fermi surface (FS) pockets with opposite spins. In- which controls the semi-metal (SM) phase, the RG flow of the extended model was shown to be controlled by two cluding the Zeeman coupling, HB = σ σBnˆk,σ into the Hamiltonian, the spin polarized energy spectrum is other fixed points at large N: an AF fixed point, and P a CDW fixed point, both unstable towards the Gaussian Es,σ(k) = sv k + σB µ. The nesting between the two Fermi surface| | sheets− can produce a logarithmic di- fixed point at weak coupling, and having a runaway direc- vergence in the spin polarization in the limit B tion to strong coupling when U or V are sufficiently large. max(T, µ ) (Bercx et al., 2009), | | ≫ The two fixed points compete, resulting in the phase di- | | agram shown in Fig. 22. The fact that the AF fixed B point has only one unstable direction to leading order in Π(1) (0) ρ(B) ln | | . (5.6) AF ∼ max(T, µ ) 1/N motivated the conjecture that the SM-I transition to  | |  the AF state is continuous and of the Gross Neveu type This instability brings the possibility of a canted AF state (Herbut, 2006). The symmetry analysis of the possible in graphene. In the presence of Landau level quantiza- quartic terms has been discussed by Herbut et al., 2009. tion due to the application of an out of plane magnetic The 1/N results were confirmed qualitatively by nu- field, electronic interactions may lead to the formation merical renormalization group (NRG) calculations for 34 the extended Hubbard model in the honeycomb lattice whereas the Friedel oscillations decay as 1/r2 for intra- (Raghu et al., 2008). In the presence of next-nearest cone scattering and as 1/r for intercone scattering (Bena, neighbors repulsion, the NRG calculations suggested the 2008). possibility of competition between the CDW and spin Besides defects, zigzag edges also lead to local mag- density wave (SDW) phases with non-trivial topological netism in the presence of interactions (for a more detailed insulating states, such as the quantum spin Hall (QSH) discussion, see Sec. VI). In bulk graphene, a simple way state, where TRS is spontaneously broken (Raghu et al., to generate localized magnetic states is provided by the 2008). Functional renormalization group (FRG) calcula- adsorption of adatoms with inner shell electrons. On tions for the t J model on the honeycomb lattice with the lattice, the adatoms can stay in different locations − on site and nearest neighbor repulsion also suggested the relative to the two sublattices in graphene. Transition possibility of strong coupling CDW and SDW instabili- metals are usually more stable sitting in the hollow site, ties in graphene at half filling (Honerkamp, 2008). In the at the center of the honeycomb hexagon (Chan et al., doped regime, the t J model can favor the formation 2008), whereas simple molecules and atoms such as hy- − of superconducting states for J > 2t, either in the triplet drogen (H) tend to hybridize more strongly with the car- or in the d-wave singlet channels (Honerkamp, 2008). bons, sitting on top of them and generating a large local In the high doping regime, the proximity of the Fermi moment (Yazyev and Helm, 2007). In particular, H ad- level to the Van-Hove singularities, where the graphene sorption creates a midgap state (Boukhvalov et al., 2008; DOS diverges logarithmically, may favor a Pomeranchuck Wehling et al., 2010c) and distorts locally the sp2 carbon instability (PI), rather than a gapped state. In that case, bonds, which acquire sp3 character (Elias et al., 2009). the redistribution of the electronic density generates a de- This distortion can induce a strong local enhancement formation of the Fermi surface, which lowers the lattice of the spin-orbit coupling up to 7 meV, as in dia- ≈ C3v point group, instead of breaking the Z2 sublattice mond, and generate a strong local magnetic anisotropy symmetry. In the extended Hubbard model at high dop- (Castro Neto and Guinea, 2009). Adatoms can also ing, the PI is favored by the repulsion between nearest form local moments from substitutional defects on single neighbor sites, which renormalizes the kinetic energy at and double vacancies in graphene (Krasheninnikov et al., the mean field level, and competes with the on-site repul- 2009; Venezuela et al., 2009). sion, which favors a ferromagnetic state when the Stoner The heuristic criterion that describes the formation of criterion is satisfied (Valenzuela and Vozmediano, 2008). a local magnetic moment is addressed at the mean field When coated with metallic atoms that have a strong level by the Anderson impurity model (Anderson, 1961). tendency to hybridize with the carbon pz orbitals, In the top carbon case, assuming that the adatom sits on graphene can induce strong itinerant ferromagnetism in a carbon (see Fig. 23), say on sublattice B, the hybridiza- the metallic bands (Uchoa et al., 2008b). tion Hamiltonian is HV = V σ[fσ†bσ(0)+h.c.], where fσ (fσ†) annihilates (creates) an electron with spin σ = , at the impurity. In momentumP space, this translates↑ into:↓ C. Local magnetic moments

= V (f †bp + b† f ) . (5.8) For massless Dirac particles, the formation of local- HV σ ,σ p,σ σ p,σ ized states is usually harder than in usual Fermi sys- X tems due to the Klein paradox, in which the fermions can easily tunnel through a barrier regardless of its If nσ = fσ†fσ is the occupation of the localized level for h i height. Defects such as vacancies, where a carbon atom is knocked out from the plane, have been shown a) to generate localized states in graphene (Pereira et al., a) 2006; Vozmediano et al., 2005), and were recently ob- served in STM experiments (Ugeda et al., 2010). Vacan- cies have also been found to host local magnetic states (Chen et al., 2011; Yazyev and Helm, 2007). Short range interacting impurities can generate local resonances, which are quasi-localized states. At half- filling, the energy of the resonance, ε0, is given by (Skrypnyk and Loktev, 2006; Wehling et al., 2007)

D2 U = , (5.7) 0 ε ln ε2/(D2 ε2) b) c)c) 0 | 0 − 0 | where U0 is the scattering potential of the impurity and FIG. 23 (color on line) (a) Honeycomb lattice with an impu- D is the bandwidth. The resonance induces accumula- rity atom. Black: sublattice A; White: sublattice B. Inter- section of the Dirac cone spectrum, E(k) = v k , with the tion of LDOS at the Fermi level around the impurity, ± | | ρ(r, ω), which decays as 1/r (Bena and Kivelson, 2006), localized level Ef = ε0: (b) ε0 > 0, (c) ε0 < 0. 35 a given spin, the effective Hamiltonian of the level is 1 1.5 a) ε b) o < 0 non-magnetic = ε f †f , (5.9) Hf σ σ σ σ X 0.5 1 )/U o

with εσ = ε0 + Un σ, after a proper mean field decom- ε magnetic − - magnetic position of the Hubbard term H = Uf †f f †f , which µ U ( accounts for the charging energy U to↑ doubly↑ ↓ ↓ occupy 0 0.5 the level. The hybridized level becomes magnetic when ε non-magnetic n = n . The occupation is derived self-consistently by o > 0 integrating↑ 6 ↓ the f-electron DOS from the bottom of the -0.5 0 graphene band up to the Fermi level µ, 0 5 10 0 5 10 ∆D/U ∆D/U 1 µ 1 nσ = Im dω , (5.10) −π ω εσ Σff (ω) Z−∞ − − FIG. 24 Boundary between magnetic and non-magnetic im- where Σff (ω) is the self-energy of the localized electrons. purity states in the scaling variables x = ∆D/U and y = (µ ε0)/U for ε0 > 0 (a) and ε0 < 0 (b). ε0 /D = In the cone approximation of the spectrum in graphene, − | | for the top carbon case, 0.029, 0.043, 0.029 and V/D = 0.14, 0.14, 0.04 for circles, squares and triangles, respectively. The upturn close to y = 1 1 and x 0 on panel b) signals a crossover to the Fermi liquid Σff (ω)= ω 1 Z− (ω) i∆ ω θ(D ω ) , (5.11) → − − | | − | | regime µ, U ε0 > 0, where the Dirac points are physically ≫ | | 2 2  irrelevant. This feature is not visible in this scale when V is where ∆ = πV /D is the dimensionless hybridization, very small (triangles) (Uchoa et al., 2008a). D is the effective band width, and

2 2 1 V D The formation of local moments is also affected by the Z− (ω)=1+ ln 1 (5.12) D2 − ω2 specific location of the adatom in the lattice (Fig. 25).

For instance, when the adatom sits in the center of gives the quasiparticle residue, Z(ω ), which vanishes log- the honeycomb hexagon (H-site), the tight-binding hy- arithmically at the Dirac points (ω 0). bridization Hamiltonian is (Uchoa et al., 2009) Because of the vanishing→ DOS, the level R broadening [given by ImΣ (ω)] scales lin- HV = Va,ia† (ai)+ Vb,ib† ( ai) fσ(0)+h.c. , (5.13) ff σ σ − early with the energy around the Dirac σ,i X   points (Gonz´alez-Buxton and Ingersent, 1998; where a (i =1, 2, 3) are the three nearest neighbor vec- Skrypnyk and Loktev, 2006; Uchoa et al., 2008a; i tors of the honeycomb lattice, and Vx,i (x = a,b) is the Zhang et al., 2001). The DOS induced around the bare hybridization strength of the adatom with each of the level, εσ, does not decay like a Lorentzian as in usual nearest surrounding carbon atoms. In momentum repre- metals, but shows a long tail proportional to 1/ω. This sentation, tail induces several peculiar features in the magnetic states. For instance, a local moment is allowed to HV = Va,∗pap† σ + Vb,pbp† σ fσ + h.c. , (5.14) exist when the bare level is empty (ε0 < µ) or doubly pσ occupied (ε + U >µ) (see Fig. 24). The presence of X
0 where the Dirac point also breaks the symmetry around the line µ ε = U/2, and makes the scaling of the curves 3 0 ip ai shown− in Fig. 24 non-universal. Furthermore, there is a Vx,p = Vx,i e · . (5.15) i=1 physical asymmetry between the cases where the level X is above (ε0 > 0) or below (ε0 < 0) the Dirac point. The top carbon case is recovered by setting V p V and a, ≡ When ε0 = 0, as in the case of a vacancy, the level Vb,p = 0 or vice-versa. For s-wave orbitals, Vx,i V , decouples from the bath and becomes magnetic for any whereas for in-plane f-wave orbitals the hybridization≡ is µ> 0, regardless of the value of U (Pereira et al., 2006; anti-symmetric in the two sublattices, Va,i = Vb,i V . Uchoa et al., 2008a). In the case of substitutional impurities (S-sites),− either≡ Since the chemical potential in graphene can be tuned, Va,i = 0 or Vb,i = 0. The quantum interference be- the formation of local magnetic states can be controlled tween the different hybridization paths of the electrons by the application of a gate voltage (Uchoa et al., 2008a). can modify the energy scaling of the level broadening in The low density of states around the localized level also Eq. (5.11) (Uchoa et al., 2009), and can also change the makes the formation of local moments in graphene much shape of the Fano resonances in scanning tunneling spec- easier than in usual metallic hosts. As a result the troscopy (STS) measurements, allowing a clear identifica- adatoms can achieve high magnetic moments at relatively tion of the adatom position with an STS tip (Saha et al., small U (Cornaglia et al., 2009; Uchoa et al., 2008a). 2010; Uchoa et al., 2009; Wehling et al., 2010b). 36

D. Kondo effect T The formation of a Kondo screening cloud around a Critical LM magnetic moment is described by the Anderson Hamil- tonian (5.8) in the strong coupling limit, U , where LM Kondo the valence fluctuations are suppressed and the→ ∞ local mo- ment becomes a good quantum number. In the stan- QCP J dard mean field approach, the spin 1/2 fermionic fields K are replaced by fermionic fields with larger degeneracy, FIG. 26 Schematic phase diagram around the Kondo QCP N >m, which corresponds to an SU(N) extension of at half filling: temperature vs Kondo coupling. LM: local the problem, with a corresponding Kondo Hamiltonian moment phase, where the Kondo cloud is suppressed. In (Coqblin and Schrieffer, 1969) the critical LM phase, quantum critical fluctuations dominate (Ingersent and Si, 2002).

= J ψ† f † ′ f ψk′ ′ , (5.16) HK K k,m m m ,m mm′ kk′ X X c Fermi liquid case in the weak coupling regime, JK boson approach to the Anderson model (Coleman, 1983; 1998; Vojta, 2001) predicted a variety of fixed points. Newns and Read, 1987), or else by an equivalent path At half-filling, in the particle-hole symmetric case, integral approach starting from the Kondo Hamiltonian ε0 = U/2, the Kondo problem has a metallic Kondo − (5.16) (Read and Newns, 1983). screened fixed point at r = 0, which evolves into a strong The application of these methods to semi-metals with coupling fixed point for 0 < r 1/2. In this case, the r c ≤ c a vanishing DOS, ρ(ω)= ρ0 ω , with r > 0, resulted in strong (JK >JK ) and weak coupling (JK 1/2 the local moment remains c r expected for JK > JK = r/(ρ0D ), below the Kondo unscreened for all initial values of the Kondo coupling temperature (Withoff and Fradkin, 1990) (Chen and Jayaprakash, 1995). In the particle-hole asymmetric case (µ = 0, U = 2ε0), for r > r∗ 0.375, T J J c ν , (5.17) 6 − ≈ K ≈ | K − K | the weak and strong coupling regimes are separated by an asymmetric critical point (ACP). For r < r , where ν =1/r, and D is the ultraviolet cut-off. Since the ∗ the particle-hole symmetry is dynamically restored scaling dimension of the hybridization V in the Anderson (Fritz and Vojta, 2004; Gonz´alez-Buxton and Ingersent, model is dim[V ] = (1 r)/2, the case r = 1 acts as an 1998). upper critical scaling− dimension in the problem, where The phase diagram around the QCP is schemati- the scaling is marginal (Vojta and Fritz, 2004). In the cally shown in Fig. 26. The critical local moment fluc- marginal case, the Kondo temperature may have an addi- tuations were studied by Ingersent and Si, 2002, who tional logarithmic scaling with the coupling, upon imple- found linear ω/T scaling of the dynamical spin sus- mentation of an ultraviolet cut-off smoothly connected to ceptibility at the critical point for 0 < r < 1. In the metallic case (r =0)(Cassanello and Fradkin, 1996). the marginal case, r = 1, there are logarithmic cor- Away from half-filling, there is a crossover to the usual rections to scaling (Cassanello and Fradkin, 1997). The Kondo problem for gapless excitations was also exten- a) b) sively studied in the context of magnetic impurities in d-wave superconductors (Borkowski and Hirschfeld, 1992; Cassanello and Fradkin, 1996, 1997; Polkovnikov, 2002; Polkovnikov et al., 2001; Vojta and Bulla, 2001; Zhang et al., 2001; Zhu and Ting, 2000). For a review, see Balatsky et al., 2006. In the graphene case, where r = 1, the Dirac fermions in the bath have an additional pseu- FIG. 25 Two adatom configurations in graphene: a) the dospin structure, which motivated several proposals for adatom (red circle) sits on top of a carbon atom, and b) multichannel Kondo physics (Cassanello and Fradkin, the adatom (blue circle) sits at the center of the honeycomb 1996; Dell’Anna, 2010; Sengupta and Baskaran, 2008; hexagon, hybridizing equally with the two sublattices. Red Zhu et al., 2010). The Kondo resonance in graphene arrows: nearest neighbor vectors. has been calculated with NRG by Cornaglia et al., 2009. 37

At half filling, the local DOS around the impurity can be spontaneously enhanced by the formation of midgap 15 states due to the scattering potential of the impurity (Hentschel and Guinea, 2007), frustrating the Kondo 10 QCP. J (eV) At finite doping, the Kondo temperature has an 5 exponential dependence with the DOS at weak cou- pling, allowing the Kondo cloud to be tuned by gat- 0 ing (Sengupta and Baskaran, 2008). In the crossover 0 0.2 0.4 0.6 regime, at J = Jc, the scaling of the Kondo tempera- µ − ε (eV) x 0 ture with doping becomes power law, TK µ . Recent NRG calculations in graphene have found∝ a particle-hole| | FIG. 27 Kondo coupling vs. chemical potential in graphene asymmetric scaling of the Kondo temperature with dop- for U = 1 eV and V = 1 eV. The Kondo coupling can be x controlled by gate voltage across the weak (J Jc) and ing, TK µ , where x = 1 for µ > 0 and x = 2.6 ≪ for µ < ∝0 | (Vojta| et al., 2010), in contradiction with strong coupling (J & Jc) Kondo regimes, where Jc is the the mean field and poor man scaling analysis for the critical coupling at half-filling. marginal case (Vojta et al., 2010). In the presence of Landau level quantization, the Kondo temperature has reentrant behavior as a function of the chemical poten- where σ = (σ1, σ2, σ3) are the Pauli matrices, S = 1 ′ tial (Dora and Thalmeier, 2007). 2 fσ†σfσ is the localized spin, and Looking at the problem on the lattice, ab initio calcu- lations on Cobalt have found that the interplay of spin V 2U J(µ) < 0 , (5.22) and orbital degrees of freedom can give rise to an SU(4) ≈ (ε0 µ)(ε0 + U µ) Kondo effect in graphene when the spin orbit coupling is − − strong enough (Wehling et al., 2010a). Another ab initio is the exchange coupling defined at the Fermi level, calculation accounting for dynamic correlations, also on µ. Within the tight-binding description, we realize Co, has identified the possibility of a spin 3/2 Kondo that the determinant of the exchange coupling matrix effect, involving multiple orbitals (Jacob and Kotliar, in Eq. (5.21) is identically zero, det[Jˆαα′ ] 0, and 2010). From a tight-binding perspective, for a spin 1/2 hence the exchange Hamiltonian (5.21) can be≡ rotated impurity, the hybridization Hamiltonian (5.14) can be into a new basis where one of the hybridization chan- written in the diagonal basis nels is decoupled from the bath (Pustilnik and Glazman, 2001). The eigenvalues in the new diagonal basis are ′ ′ Ju,k,k = J α Θα,∗ kΘα,k and Jv = 0, implying that = V Θ pc† f + h.c. , (5.19) HV α, α,pσ σ the one-level exchange Hamiltonian (5.21) maps into the α= p,σ problem of aPsingle channel Kondo Hamitonian, = X± X   He 2 J kk′ S sk k′ , where s is the itinerant spin, in − k u, · , where c ,kσ = (1/√2)[bkσ (φk∗ / φk )akσ] are the spite of the implicit valley degeneracy. A multi-channel ± ± | | fermionic operators that diagonalize the graphene Hamil- descriptionP of the one-level problem is nevertheless possi- 3 iai k tonian (2.2), φk = i=1 e · is the tight-binding hop- ble for example in graphene quantum dots, in the contin- ping matrix element defined by Eq. (2.3), and α = uum limit, where valley and angular momentum channels ± labels the conductionP and valence bands. Θ is a phase become good quantum numbers. factor, which accounts for the symmetry and position Unlike the situation in metals, the exchange cou- of the localized orbital with respect to the sublattices pling in graphene can be controlled by gating (Uchoa et al., 2009), (Jacob and Kotliar, 2010; Uchoa et al., 2011), as shown in Fig. 27, in particular when the chemical potential is 1 φp∗ brought to the proximity of the localized level, where Θα,p = Vb,p + αVa,∗p , (5.20) √2V φp the Kondo coupling becomes resonant. This effect opens  | | the possibility of tuning J to the vicinity of the criti- where Vx,p is the hybridization as defined in Eq. (5.15). cal coupling that sets the crossover between the weak As in metals, the Anderson Hamiltonian in graphene and strong coupling regimes. In this region, at finite can be mapped into the spin exchange Hamiltonian by doping, quantum criticality is reminiscent of the frus- a canonical transformation (Schrieffer and Wolff, 1966). trated QCP at µ = 0. Since the width of the Kondo In the large U limit, the spin exchange Hamiltonian be- peak in the spectral function is set by the Kondo tem- tween the magnetic adatom and the graphene electrons perature only, the gating effect permits measuring the is (Uchoa et al., 2011) quantum critical scaling of the Kondo temperature with doping (Uchoa et al., 2011; Vojta et al., 2010) directly

= J Θ∗ Θ ′ k′ S c† ′ ′ ′ σc k , (5.21) with STM probes (Saha et al., 2010; Uchoa et al., 2009; He − α,k α , · α ,σ ,k α,σ, kk′ ′ Wehling et al., 2010b; Zhuang et al., 2009). X Xαα 38

E. RKKY interaction a ripple, the sp2 carbon (C) bonds are spontaneously stretched by the curvature and acquire sp3 character. The Ruderman-Kittel-Kasuya-Yosida (RKKY) inter- Contrary to the perfectly flat case, the adsorption of H action between two local spins is obtained by integrat- atoms on top of the hills helps to stabilize the ripples ing out the itinerant fermions in Eq. (5.21), which gives (Boukhvalov and Katsnelson, 2009). The interplay be- 2 RKKY = J χij Si Sj , where χij is a two point corre- tween the correlations due to the ripples and the RKKY lationH function,− with i,· j indexing the positions of the lo- interaction among the H spins can generate magnetore- cal spins. In momentum space (Brey et al., 2007; Saremi, sistance hysteresis loops and a variety of magnetic spin 2007; Uchoa et al., 2011), textures (Rappoport et al., 2009).

′ xy xy f[Eα (k + q)] f[Eα(k)] χ (q)= ′ − , αα ,k,q ′ ′ M Eα(k) Eα (k + q) kαα − F. Superconductivity X (5.23) where (omitting the αα′ labels for simplicity) The observation of proximity induced supercon- xy x y x y ductivity in graphene junctions has stirred a lot =Θ∗ Θ Θ ′ Θ∗ ′ , (5.24) Mk,q α,k α,k α ,k+q α ,k+q excitement in the field of mesoscopics (Heersche et al., with x, y = A, B, H, S etc, indexing the position of the 2007). The Dirac nature of the quasiparticles gives rise to ballistic transport on a micron scale and allows spins on the lattice, Eα(k) = α φk µ, and f is the AA BB| |− graphene to sustain supercurrents in long junctions, Fermi distribution. k,q = k,q = 1/4 for spins on the same sublattice whereasM M the size of the coherence length in the superconduct- ing metallic leads (Du et al., 2008; Heersche et al., AB 1 φkφk∗+q 2007; Miao et al., 2007; Ojeda-Aristizabal et al., 2009). k,q = αα′ (5.25) M 4 φk φk+q The experimental realization of the proximity ef- | || | fect motivated theoretical studies of the differential for spins on opposite sublattices. In the continuum limit, conductance (DC) in normal-superconductor (NS) where the spectrum is linearized around the Dirac points, interfaces in graphene (Beenakker, 2006; Burset et al., AB 1 iθk,k+q = αα′e , where θ is the angle between k and 2008), graphene nanoribbons (Rainis et al., 2009), and Mk,q 4 k + q (Brey et al., 2007). in graphene normal-insulator-superconductor (NIS) At half-filling, kF = 0, the Fermi surface collapses into junctions (Bhattacharjee and Sengupta, 2006). Due points and the RKKY interaction is mediated by inter- to the Dirac nature of the spectrum, at half-filling, band transitions, which polarize the vacuum as in QED. the Andreev conversion of an electron into a hole at In this case, the Friedel oscillations disappear and the the interface between a normal and a superconduct- sign of the interaction is ferromagnetic for spins on the ing region involves specular reflection rather than same sublattice and anti-ferromagnetic for spins in op- retro reflection (Beenakker, 2006). The specular posite sublattices (Brey et al., 2007; Saremi, 2007). In Andreev reflection leads to the presence of Andreev the overdoped regime, at µ = t, the nesting among modes in SNS junctions that propagate along the the Van Hove singularities in graphene reverses the sign graphene edges at the interface with the superconductor of the RKKY interaction compared to the µ = 0 case (Titov et al., 2007). The Josephson current in graphene (Uchoa et al., 2011). SNS junctions was studied by Titov and Beenakker, At long distances, the spatial decay of the 2006, followed by Bergman and Hur, 2009; 3 RKKY is r− when µ is at the neutrality point Maiti and Sengupta, 2007; Moghaddam and Zareyan, (Brey et al., 2007; Cheianov and Fal’ko, 2006; Saremi, 2006, and Black-Schaffer and Doniach, 2008. Pos- 2007; Vozmediano et al., 2005; Wunsch et al., 2007). sible applications involving the proximity effect Away from half filling, the Friedel oscillations are re- in graphene include proposals for valley sensors stored by the intraband transitions and the RKKY in- (Akhmerov and Beenakker, 2007), current switches 2 teraction decays at r 1/kF as 1/r , similarly to the (Linder et al., 2008; Lutchyn et al., 2008), and a spin 2DEG case (Brey et al.≫, 2007; Wunsch et al., 2007). For current filter (Greenbaum et al., 2007). A review on H or S site spins formed in C3v symmetric orbitals, the Andreev and Klein tunneling processes in graphene can RKKY interaction decays with a fast power law 1/r7 at be found in Beenakker, 2008. half filling (Uchoa et al., 2011). In carbon nanotubes, These experimental developments in transport moti- the RKKY interaction decays as 1/r for top carbon vated a surge of interest in the possibility of making spins and as 1/r5 for H site spins in isotropic orbitals graphene an intrinsic superconductor. Graphene par- (Kirwan et al., 2008). ent compounds, such as the graphite intercalated ma- When distributed regularly on top of graphene, mag- terials CaC6 and KC8, are low temperature supercon- netic adatoms such as hydrogen (H) can form macro- ductors, although neither graphite nor alkaline metals scopic magnetic states at room temperature (Zhou et al., alone superconduct (Csanyi et al., 2005; Hannay et al., 2009). In the disordered case, H atoms in particular can 1965; Weller et al., 2005). Even though intrinsic su- cluster on top of graphene due to rippling. On top of perconductivity has not been observed in the single 39

Regardless of the microscopic origin, the superconduct- ing state in graphene can be analyzed based on the sym- metries of the order parameter in the honeycomb lattice. 1 2 On the lattice, the electrons in graphene carry spin, angu- lar momentum and sublattice quantum numbers. There 3 are four possible pairing channels: singlet/triplet spin channels, and same/opposite sublattices. In the singlet case, if we restrict the analysis to nearest neighbor site interactions only, two competing order parameters can be identified: iθj FIG. 28 Superconducting order parameter ∆1,j = ∆1e (j = 1, 2, 3), with phases along the three different bond di- ∆0 = g0 ai aj = g0 bi bj , (5.26) rections in the lattice. h ↑ ↓i h ↑ ↓i

which corresponds to an s-wave state, and ∆1, defined as layer so far, a few different superconducting mecha- ∆1,ij = g1 ai bj ai bj (5.27) nisms have been proposed. One possibility is a plasmon h ↑ ↓ − ↓ ↑i mediated mechanism in graphene coated with metallic for nearest neighbors and zero otherwise, where g0 and adatoms, in which the plasmons of the metallic band g1 are the coupling strengths. In momentum space, the mediate the attraction between the graphene electrons latter state is described by (Uchoa and Castro Neto, 2007). When isolated islands of metallic atoms are adsorbed on top of graphene, su- 3 iai k perconductivity can also be induced by proximity ef- ∆1,k = ∆1,i e · , (5.28) i=1 fect (Feigel’man et al., 2008). Another possibility is the X Kohn-Luttinger mechanism, which explores the proxim- where ∆ ∆ (a ) are the real space pairing ampli- ity of the Fermi surface to the Van-Hove singularities in 1,i 1 i tudes along≡ the three different bond directions in the the high doping regime (Kohn and Luttinger, 1965). In honeycomb lattice (see Fig. 28). In the simplest case this scenario, the superconductivity can be mediated by the pairing amplitudes are the same, ∆ ∆ , and ∆ a purely electronic mechanism, when the interactions be- 1,i 1 1 is real, giving ≡ come attractive along a specific direction of the BZ near the Van-Hove singularity (Gonz´alez, 2008). The super- ∆1,k = ∆1φk , (5.29) conductivity can also be mediated by in plane or out of plane flexural phonons (Lozovik and Sokolik, 2010). 3 ik ai where φk = i=1 e · gives the hopping matrix el- In graphene, strong doping regimes can be currently ement in the single particle tight-binding spectrum achieved by chemical adsorption of alkaline metals, such (Uchoa and CastroP Neto, 2007). This order parameter as potassium (Gr¨uneis et al., 2009; McChesney et al., represents the pairing between electronic states in oppo- 2010; Uchoa et al., 2008b), or with metal contacts site sides of the BZ, and preserves all the physical sym- (Giovannetti et al., 2008). metries of the honeycomb lattice, including point group Alternative proposals include edge state supercon- and time-reversal symmetry, ∆1,k = ∆1∗, k, where the ductivity, induced by the large DOS at the edges momentum k is measured with respect to− the center of (Sasakia et al., 2007), or strong correlations, which so far the BZ, at the Γ point. In real space, this order param- have not been observed in graphene. As in the cuprates, eter (OP) has extended s-wave symmetry. If expanded the antiferromagnetic attraction between spin singlets around the Fermi surface centered at the Dirac point K, on nearest neighbor sites has been proposed as a pos- from the perspective of the quasiparticle excitations near sible pairing channel in graphene, provided the on site the Fermi level, Hubbard repulsion is strong enough to suppress the lo- iθ cal fluctuations (Pathak et al., 2010). Gonz´alez et al., ∆1,K+p = ∆1e (px + ipy) (5.30) 2001 considered the possible competition between ferro- magnetic and superconducting states in graphene sheets describes a p + ip state in one valley and p ip in the through a renormalization group analysis accounting for opposite one (Uchoa and Castro Neto, 2007).− This state Coulomb interactions. A recent functional renormaliza- is therefore a p + ip state with additional valley degener- tion group calculation has proposed the possibility of a acy. Unlike the case of conventional p + ip superconduc- strongly correlated SDW state that gives way to a singlet tivity, the time reversal operation involves an additional superconducting instability in the d-wave channel, or else exchange of valleys, preserving the TRS of this state, and a CDW solution that allows a triplet pairing instability we shall refer to it as p + ip. in the f-wave channel (Honerkamp, 2008). In two-layer Another possible paring symmetry is the state graphene, the possibility of excitonic pairing of electrons (Black-Schaffer and Doniach, 2007; Jiang et al., 2008) in one layer with holes in the other one has been consid- i(2π/3)j ered (Kharitonov and Efetov, 2008; Min et al., 2008). ∆1,j = ∆1e , (5.31) 40

g g 1 1 g 0 s−wave g s−wave 0

mixed p+ip gapless mixed −wave

FIG. 30 Phase diagram between the s-wave and effective p+ip FIG. 29 (color online) Order parameter (OP) amplitude, phases in the spin singlet channel. On the left: µ = 0 case, which is quantum critical. Right: µ = 0 case. Continuous ∆1,k , in the BZ: (left panel) ∆1,j = ∆1 with j = 1, 2, 3 6 |indexing| the three different bond directions of the crystal [see lines represent second order transitions, and dashed lines rep- i2πj/3 resent first order transitions (Uchoa and Castro Neto, 2007). Eq. (5.28)] and (right panel) ∆1,j = e , which describes a flux phase. Light colors represent higher amplitude. Dirac points are located at the K points, at the edges of the BZ. In all dark spots, the OP has p + ip symmetry around the of the lattice and allows the presence of topological exci- respective high symmetry points. In the three light spots on tations (Roy and Herbut, 2010). ′ the right panel, the OP has s-wave symmetry around the K At the level of nearest neighbor sites, the electron- points. electron interaction can be decomposed into an effective local Hubbard term, j = 1, 2, 3, which describes on the lattice a real space 0 g0 I = aiσ† aiσai† σai σ + biσ† biσbi† σbi σ , pairing wavefunction with dx2 y2 + idxy-wave symme- H 2 − − − − − iσ   try, breaking TRS. This broken symmetry is caused by X (5.33) the circulation of plaquette current loops, which amounts and a non-local part, to global circulation of current along the edges. The low energy description of this state around the Dirac 1 = g a† a b† ′ b ′ . (5.34) points is a combination of s-wave in one valley and HI 1 iσ iσ jσ jσ ij σσ′ p + ip state in the opposite valley (Jiang et al., 2008), Xh i X as shown in Fig. 29. At the mean field level, this state In the singlet pairing channel, the non-local term can was shown to have lower energy than the pure p + ip 1 be decomposed into I = g1 ij ij† ij + ij† ij , state (Black-Schaffer and Doniach, 2007). Due to the H h i −B B D D plus one body terms that can be absorbed into the broken TRS, disorder and quantum fluctuations, which P   are paramount in a 2D system, may strongly inhibit the chemical potential µ. ij = ai bj ai bj is a stan- D ↑ ↓ − ↓ ↑ coherence of the d + id state. Other alternatives are the dard singlet pair operator and = a† b is a Bij σ iσ jσ degenerate states with dx2 y2 and dxy-wave symmetries, bond operator. Decomposition of the interaction at represented by the ∆ pairing− amplitudes (2, 1, 1) the mean field level with = 0P results in the 1,i − − ij and (0, 1, 1), respectively (Black-Schaffer and Doniach, graphene tight-binding HamiltonianhB i for the supercon- 2007). These− states conserve TRS but lower the crystal s ˆs ducting phase, = k Ψk† Ψk + E0, where point group symmetry. H H P 2 2 In the spin triplet channel, the OP is a superposition of E0 = ∆0 /g0 3∆1/g1 , (5.35) Sz = 1, 0, +1 states. Since on-site pairing is forbidden −| | − by the− Pauli principle, for nearest neighbors interaction and t the triplet superconducting states are ∆ij,σσ = aiσ bjσ , t h i µ tφk ∆0 ∆1,k with σ = , for Sz = 1, and ∆0, = ai bj + ai bj , − − in the S ↑=↓ 0 channel.± The OP in↑↓ this caseh ↑ is↓ commonly↓ ↑i ˆS tφk∗ µ ∆1, k ∆0 z k =  − − −  (5.36) H ∆0∗ ∆1∗, k µ tφk defined as a 2 2 tensor, − ×  ∆∗ ∆∗ tφ∗ µ   1,k 0 k  ∆ij = iσ2σ dij , (5.32)   · is the Bogoliubov-de Gennes matrix written in the sub- where the Pauli matrices act in spin space, and dij = lattice and Nambu basis Ψk = (ak ,bk ,a† k ,b† k ). d is an anti-symmetric tensor, violating parity. The ↑ ↑ − ↓ − ↓ − ji The Hamiltonian (5.36) can be diagonalized in a basis case where the OP d has a single vector compo- s B of Bogoliubov quasiparticles: H = Ek,α,snˆ + nent describes the spinless fermionic case, discussed by kαs k,α,s E , wheren ˆB is the quasiparticle number operator and Bergman and Hur, 2009. The possibility of spin triplet 0 P s, α = 1. In the isotropic case, ∆ = ∆ φ , the spec- states beyond nearest neighbors in the S = 0 channel 1,k 1 k z trum is±E = αE , with (Uchoa and Castro Neto, was recently examined in a variational cluster approxi- k,α,s k,s 2007) mation calculation (Sahebsara and S´en´echal, 2009). An- other possibility is a Kekule superconducting state in the 2 2 triplet channel, which breaks the translational symmetry Ek = (t φk + sµ) + ( ∆ + s∆ φk ) , (5.37) ,s | | | 0| 1| | q 41

Away from half-filling, the gap crosses over to (Uchoa et al., 2005) 0.4 ∆ ∆ =2 µ exp [D(1 gc/g )/ µ 1] (5.41) 0 0 | | − 0 0 | |− 0.2 for µ ∆0, which corresponds to the weak coupling BCS| limit,| ≫ where g g , as shown in Fig. 31. The 0 c 0 g g µ /∆ 1 limit corresponds≪ to the strong coupling c | | 0 ≪ regime (g >gc), and the intermediate coupling region FIG. 31 Dependence of the gap, normalized by the band cut- near g gc sets the crossover scale between the two off D, on g in the weak (ggc) regimes∼ at finite µ. Non-equilibrium effects in the pres- sectors for µ = 0 (red line), µ/D = 0.1, and 0.3. The model ence of a dissipative environment may also lead to a dis- has a QCP at half filling (Uchoa et al., 2005). sipation driven quantum phase transition away from half filling (Takei and Kim, 2008). At mean field level, the critical temperature at µ = 0 is where the phase of the OP ∆0 is locked in with ∆1, and Tc = ∆0/2 ln 4, whereas in the opposite limit, µ ∆0, ∆1 is real. The electronic gap described by the spectrum | | ≫ (5.37) is Tc = γ∆0/π, as in the BCS case, where ln γ 0.577 is the Euler constant (Uchoa et al., 2005). Of course,≈ in two dimensions there is no true long range order. The su- E =2 t∆ µ∆ / t2 + ∆2 . (5.38) g | 0 − 1| 1 perconducting transition is of Kosterlitz-Thouless (KT) q type and coherence is actually lost at much lower tem- In the p + ip state (∆ =0, ∆ = 0), E is proportional 0 1 g peratures due to the role of thermal fluctuations, which to the deviation of the chemical potential6 away from half- unbind vortex and anti-vortex pairs above the KT tran- filling, and at µ = 0 this state becomes quantum criti- sition temperature, at T < T . The mean field result cal and gapless. The instability in this case translates KT c indicates the onset of critical fluctuations where the am- into the renormalization of the Fermi velocity, where plitude of the Cooper pairs is completely destroyed, al- t¯ = t 1 + ∆2 is the renormalized hopping amplitude, 1 though the phase coherence is suppressed much earlier, instead of the opening of a gap (Uchoa and Castro Neto, at T . The KT fluctuations of the SC order parame- 2007).p Minimization of the free energy KT ter have been considered by Loktev and Turkowski, 2009, without accounting, nevertheless, for the chiral nature of F = T ln [2 + 2cosh(Ek /T )] + E , (5.39) − ,s 0 the quasiparticles in graphene. k,s X Zero field thermodynamic properties, such as the spe- with respect to ∆ and ∆ gives a set of two cou- cific heat at fixed volume, C = T (∂2F/∂T 2) , can be 0 1 V − V pled BCS-like equations, and leads to the phase diagram extracted from the free energy (5.39). For an isotropic shown in Fig. 30. At half-filling, µ = 0, the emergence condensate of Dirac fermions, the jump of the specific of superconductivity is controlled by quantum critical heat at the phase transition, normalized by the specific lines in the parameters g0 and g1, with critical values heat on the normal side, is (Uchoa et al., 2005) gc = πv2/D and gc = 4πv4/D3, in the linear cone 0 − 1 − 2 approximation, where D is an ultraviolet cut-off and v δCV = 2(ln 4) /[9ζ(3)] 0.35, (5.42) ≈ is the Fermi velocity near the Dirac point (Castro Neto, 2001; Marino and Nunes, 2006; Uchoa and Castro Neto, at half-filling. In the µ /∆0 1 limit, the jump grows to the standard BCS value| | δC≫ = 12/[7ζ(3)] 1.43. 2007; Zhao and Paramekanti, 2006). For finite µ, there V ≈ is a crossover to the standard Fermi liquid case at weak The Meissner effect in graphene, which describes coupling, as shown in Fig. 30. the expulsion of an external magnetic field by the i(2π/3)j circulation of diamagnetic supercurrents, has been When ∆1,j = ∆1e [see Eq. (5.28)], the elec- tronic wavefunctions collect different phases along the recently examined by Kopnin and Sonin, 2008 and different bond links, which gives rise to a current flow, Uchoa and Castro Neto, 2009. In the presence of vor- and the d+id state cannot coexist with an isotropic TRS tices, the Bogoliubov de Gennes equations for Dirac s-wave state. The gap properties of the d + id state and fermions in 2+1 dimensions allow the presence of zero the differential conductance in SN junctions were derived energy modes (Jackiw and Rossi, 1981) which are bound by Jiang et al., 2008. The Josephson current for this to the vortex cores. For a vortex with vorticity n (the winding number of the OP), ∆ = ∆ (r) einφ, with state in SNS junctions was calculated by Linder et al., 0 | n | 2009. (r, φ) as cylindrical coordinates. The physical solutions allowed by the boundary conditions at the center of the In the s-wave state (we assume ∆0 to be real), the gap variation with the coupling at half filling, near the vortex and at infinity result in n zero modes at half filling c 2 (Ghaemi and Wilczek, 2012). The subgap spectrum and quantum critical point g0 = πv /D, is (Castro Neto, 2001) − the wavefunctions in the vortex core have been derived by Bergman and Hur, 2009; Seradjeh, 2008. Away from ∆ = D(1 gc/g ) . (5.40) half filling, for odd vorticity n, there is only one energy 0 − 0 0 42 branch that crosses zero energy for zero angular momen- tum. For n even, no subgap branch intersects zero energy, and no exact zero modes exist (Bergman and Hur, 2009; Khaymovich et al., 2009). Because of the fermionic de- generacy in the valleys, the topological zero modes do not lead to fractional statistics under vortex exchange, as in conventional p+ip superconductors, unless additional in- teractions that lift the fermionic degeneracy are included (Herbut, 2010). Vortex zero modes for excitonic conden- sates in bilayers have been discussed by Seradjeh et al., 2008.

VI. INTERACTIONS AT BOUNDARIES AND LATTICE DEFECTS FIG. 32 (Color online). Sketch of the magnetization at the A. Surface states zigzag edges of a graphene ribbon.

The vanishing density of states of graphene at the neu- trality point implies that localized states can exist at the Dirac energy, much in the same way as localized states et al. appear inside a forbidden energy gap in semiconductors ments (Enoki and Takai, 2009; Joly , 2010) confirm and insulators. In order for these states to be normal- the existence of magnetic moments at graphene edges. izable, special boundary conditions are required. These The effects of the electron-electron interaction on the conditions imply the breaking of the translational sym- midgap states has also been studied beyond the mean metry of the lattice, so that they can only exist near field approximation. The calculations show that the fer- edges or defects. romagnetic phase is stable when the band of localized The most extensively studied examples are the sur- states is half filled. Both a local onsite interaction or face states which exist at graphene zigzag edges, where the long range exchange effect lead to this phase. At the lattice is abruptly terminated (Fujita et al., 1996; very low fillings, electrons tend to form a charge density Nakada et al., 1996). Such edges have been observed wave state, similar to a Wigner crystal (Wunsch et al., in graphene flakes (Girit et al., 2009; Jia et al., 2009), 2008a,b). More complex correlated states are possible and also in graphite (Niimi et al., 2005). As the local- at other fillings. The fact that the midgap states at a ized states form an energy band of zero width, the lo- zigzag edge resemble the wavefunctions of Landau lev- cal density of states at the Dirac energy near a zigzag els, in that the momentum parallel to the edge and the edge changes from zero to infinity, and the electron com- spatial extension are coupled, leads to the intriguing pressibility becomes divergent. Interactions of arbitrarily possibility of states similar to the Laughlin wavefunc- small strength lead to instabilities when the Fermi energy tions which describe the Fractional Quantum Hall Effect lies at the Dirac point. A mean field analysis showed that (Wunsch et al., 2008a). a short range Hubbard interaction can lead to a ferromag- netic ground state (Harigaya, 2001; Harigaya and Enoki, At long distances, straight graphene edges of arbitrary 2002). In zigzag ribbons with two edges, the spins at the orientation other than armchair can support midgap two edges are aligned antiferromagnetically, see Fig. 32. states, as zigzag edges (Akhmerov and Beenakker, 2008). These early theoretical results, based on the tight bind- Hence, local magnetic moments can be a generic prop- ing approximation, were later confirmed by calculations erty of abrupt graphene edges. Zigzag edges and vacan- based on the Local Density Approximation (Pisani et al., cies in bilayer (Bernal) graphene also give rise to midgap 2007; Son et al., 2006). The ferromagnetic order re- states, at least when only the direct nearest neighbor in- mained when the dangling bonds at the zigzag edges terlayer hopping is included (Castro et al., 2008a), and where saturated by hydrogen, which probably is closer to magnetic moments can be formed at the edges of bi- the actual experimental situation. The optimization of layer graphene (Sahu et al., 2008). Models which include the atomic positions at the edges leads to reconstructed other interlayer hoppings lead to sharp resonances near phases with gaps, where the spin up and spin down bands edges and vacancies. These results suggest that moder- do not overlap near the gap, suggesting a half metallic ate interactions can produce local moments in graphene phase (Son et al., 2006). Other phases, with ferroelec- bilayers or in three dimensional graphite. The combi- tric properties (Fern´andez-Rossier, 2008) or canted mo- nation of the Zeeman field associated with magnetic or- ments have been studied (Jung and MacDonald, 2010). dering, and the spin orbit coupling can lead to phases A sketch of the magnetization induced near a zigzag edge characterized by quantized spin currents at the edges of a graphene ribbon is shown in Fig. 32. Recent experi- (Soriano and Fern´andez-Rossier, 2010). 43

trons. Spins at different vacancies interact ferro- or antifer- romagnetically (Brey et al., 2007; Palacios et al., 2008), depending on whether the vacancies occupy the same or different sublattices. At half filling, the RKKY interac- 3 tion mediated by the π band decays as 1/ r r′ , and 2 | − | it goes to the 1/ r r′ dependence typical of a two dimensional electron| − gas| at finite carrier concentrations (Cheianov and Fal’ko, 2006). Voids or cracks can be con- sidered an intermediate case between vacancies and edges (Vozmediano et al., 2005). They also support localized FIG. 33 (Color online). Sketch of the magnetization induced spins at the boundaries. near a vacancy.

C. Midgap States and Random Gauge Fields B. States at vacancies and cracks

Midgap states can occur in other situations where the Midgap states in bulk graphene can also be induced by translational symmetry of the lattice is broken. Similarly magnetic fields (see below), or by strains which mimic et al. to the case of surface states at a zigzag edge, interactions the effect of a magnetic field (Guinea , 2008b). will lead to the spin polarization of these states. The These states have been analyzed using the tight bind- et al. simplest situation where the existence of a partially lo- ing approximation (Guinea , 2008b), or by means et al. calized midgap state can be demonstrated is a lattice of the Local Density Functional method (Wehling , vacancy (Pereira et al., 2006, 2008b). This analysis can 2008a). Corrugations and wrinkles also induce midgap be extended to multilayer samples (Castro et al., 2010). states in graphene (Katsnelson and Prokhorova, 2008; Pereira et al., 2010). The presence of these states en- The existence of these states has been confirmed by hances the effects of the interactions. Mean field calcula- STM spectroscopy on vacancies in irradiated graphite tions suggest the formation of magnetic moments, which (Ugeda et al., 2010). It can be expected that interactions will order ferro- or antiferromagnetically (Guinea et al., lead to the formation of a magnetic moment around the 2008a,b). vacancy. The formation of local moments near vacan- cies is consistent with the observation of ferromagnetism A random strain distribution leads to a random in irradiated graphite (Barzola-Quiquia et al., 2007; gauge field acting on the electrons. The changes Chen et al., 2011; Esquinazi et al., 2003; Ohldag et al., in the electronic density of states induced by a ran- 2007; Ramos et al., 2010). Absorption of hydrogen leads dom gauge field have been studied by RG techniques to similar effects to those of a vacancy, including the (Horowitz and Doussal, 2002; Ludwig et al., 1994). Re- formation of magnetic moments (Yazyev, 2008). Other lated problems arise at the transition between plateaus in the Quantum Hall Effect, and in d-wave supercon- dopants, like carbon atoms and NO2, also lead to the formation of spins (Lehtinen et al., 2003; Wehling et al., ductors. It can be shown that, above a certain disorder 2008b). strength, a random gauge field leads to a divergent den- A sketch of the magnetization induced near a graphene sity of states at the Dirac energy (Horowitz and Doussal, vacancy is shown in Fig. 33. The moment associated 2002; Riu and Hatsugai, 2001). This divergence leads to with the localized level around the vacancy is coupled a vanishing electron compressibility, and enhances the ef- to the extended states, leading to the possibility of the fects of interactions in the same way as the midgap states Kondo effect. Some differences between usual mag- considered earlier. A random gauge field, A(r), can be netic impurities and the situations described here can characterized by a dimensionless number, ∆, be expected: i) The vacancy or adatom modifies sig- (2) A (r)A (r′) = ∆δ δ (r r′). (6.1) nificantly the electronic density of states, rendering in- h µ ν i µν − valid perturbative treatments which relate the magni- tude of the exchange coupling to the unperturbed elec- If the gauge potential is assumed to arise from random tronic structure. The phase shift induced in the conduc- corrugations of average height h and length ℓ, then ∆ tion band remains significant, even near the Dirac energy h4/(a2ℓ2), where a is the lattice constant (Guinea et al.∼, (Hentschel and Guinea, 2007). ii) The localized state is 2008a,b). A similar parameter can be defined if the gauge orthogonal to the extended states. Hence, the coupling potential is due to topological defects, such as disloca- between the local moment and the conduction band does tions (Gonz´alez et al., 2001). The regime ∆ 1 corre- not take place via virtual hops between the two types sponds to ripples large enough to accommodate∼ midgap of states. Instead, it can be expected that the electron- states, leading to a divergence in the density of states. electron interaction favors a ferromagnetic alignment of The changes in the density of states induced by a gauge the local moment and the spins of the conduction elec- field can be written as a logarithmic renormalization of 44

FIG. 34 (Color online). Sketch of the magnetization induced at the edges of a quantum dot.

the Fermi velocity

Λ v v 1 c∆ log , (6.2) → − k  | | where c is a numerical constant, and Λ is a high mo- mentum cutoff of the order of the inverse of the lattice constant. The scaling towards lower Fermi velocities in eq. 6.2 can be combined with the RG analysis of the long range Coulomb interaction (Foster and Aleiner, 2008; FIG. 35 (Color online). Single energy peaks and Coulomb di- amonds in a graphene quantum dot, see (Ponomarenko et al., et al. Foster and Ludwig, 2006a,b; Stauber , 2005). Dis- 2008). order tends to increase the density of states near the Dirac energy, while interactions lead to the opposite ef- fect. To lowest order, this analysis leads to a line of fixed VII. INTERACTION EFFECTS IN MESOSCOPIC points characterized by a finite disorder and finite inter- SYSTEMS actions, as discussed in Sec. III.A.1, see Fig. 9. The tem- perature and frequency dependence of properties such A. Magnetism in quantum dots as the conductivity or the specific heat acquire anoma- lous exponents (Herbut et al., 2008). For high disorder, Mesoscopic samples have a large ratio between the ∆ & 1, it can be shown that a gapped state is more stable perimeter and the area. Midgap states localized at than the gapless density of states expected in the absence the edges can have a significant weight in the to- of interaction effects (Guinea et al., 2008a). tal density of states, and interaction effects are en- Certain strain configurations lead to effects simi- hanced. Early calculations for planar carbon molecules lar to those induced by a constant magnetic field (Stein and Brown, 1987; Tyutyulkov et al., 1998) showed (Guinea et al., 2010). The possible ways in which the gaps associated with the electron-electron interaction, degeneracies of these states are lifted by the interactions and magnetic moments at the edges. A large mag- have been studied (Herbut, 2008), and new phases, with netic moment can be found in triangular graphene flakes properties similar to those of topological insulators may (Fern´andez-Rossier and Palacios, 2007), where the three exist. It is worth noting that STM experiments sug- boundaries have the zigzag orientation, and the carbon gest the existence of very large effective fields due to atoms at the edges belong to the same sublattice. strains, Beff 300T, in small graphene bubbles under As mentioned previously, edges of arbitrary orienta- high strains (Levy∼ et al., 2010). The effects of electron- tions, except the armchair direction, support midgap electron interactions in this regime remain unexplored. states (Akhmerov and Beenakker, 2008). Hence, local moments and magnetism can be expected in graphene 45

FIG. 37 (Color online). Sketch of a graphene ribbon with disordered edges as a series of quantum dots.

FIG. 36 (Color online). Sketch of the extension of edge states in a graphene quantum dot.

quantum dots of any shape, provided that the termi- nation at the edges is abrupt. Model results suggest that this is the case, and the orientation of the mo- ments at the edges depends on the type of sublattice at the edge (Fern´andez-Rossier and Palacios, 2007), as sketched in Fig. 34. Away from half filling, correlated FIG. 38 (Color online). Graphene point contact coupled to a states with unsaturated magnetization, and charge den- quantum dot, see (Stampfer et al., 2009). sity wave states are also possible (Romanovsky et al., 2009; Wunsch et al., 2008a). The charging of a quantum dot leads to a substantial rearrangement of the electronic B. Charging effects. Coulomb blockade levels, in a similar way to the well studied orthogonality catastrophe in metals (Anderson, 1967; Wunsch et al., Graphene quantum dots of many shapes and di- 2008a). The conductance can acquire a non trivial volt- mensions are being extensively studied (Avouris et al., age or temperature dependence, as in a Luttinger liquid 2007; Bunch et al., 2005; Guettinger et al., 2008; (Kane and Fisher, 1992). G¨uttinger et al., 2009; Han et al., 2007; Huard et al., 2007; Molitor et al., 2009a; Moser and Bachtold, 2009; A simple estimate of the number of magnetic moments Ozyilmaz¨ et al., 2007; Ponomarenko et al., 2008; in a quantum dot can be obtained by assuming that the Stampfer et al., 2008; Williams et al., 2007). Sin- average density of edge states is of order ρedge c gle electron effects have been observed in many of [R/(aW )], where c 1 is a numerical constant,≈R ×is ∼ them. Experiments show clear evidence of charg- the radius of the dot, a is the lattice spacing, and W is ing effects in graphene quantum dots, as evidenced the bandwidth of the band of edge states (Wimmer et al., in the diamond patterns formed by the resonances 2010). The Coulomb interaction within each state, which 2 in the conductance through the dot as a func- leads to the formation of local moments is Ec e /R ≈ 2 × tion of gate and bias voltages (Guettinger et al., log(R/a), see below. Naturally, one has to replace e 2008; G¨uttinger et al., 2009; Molitor et al., 2009a,b; e2/ǫ in all formulas, but we do not write the dielectric→ 0 Moriyama et al., 2009; Moser and Bachtold, 2009; constant explicitly in this section. The states which are Ponomarenko et al., 2008; Ritter and Lyding, 2009; spin polarized are those whose distance from the Fermi Schnez et al., 2009; Stampfer et al., 2008), see Fig. 35. energy is less than E . This condition, combined with the c The electrostatic interaction between electrons leads estimate for ρ , gives a maximum number of magnetic edge to Coulomb blockade, which modulates the energy dif- moments within the dot, N E ρ c [e2/(aW )] c edge ference between levels, and induces non Ohmic features log(R/a). This number is not≈ too large.≈ For× W 0.3 × in the conductance through the dot. In a graphene quan- 0.5eV and R 100nm we obtain N 10 20. The∼ total− tum dot of dimension R, the electrostatic energy required magnetic moment∼ of the dot depends∼ on− the sign of the to add a unit of charge scales as e2/R. The mean level couplings between the edge spins, see Fig. 34. spacing between extended states in a ballistic dot scales 2 Experimentally, there is evidence which suggests the as v/R. As the dimensionless parameter α = e /(ǫ0v) in formation of local moments in small graphene flakes, of graphene is of order unity, the energy scales associated dimensions 10 50nm (Sepioni et al., 2010). with charging and confinement effects are comparable. − 46

The edge states discussed earlier can lead to charging ­ K ­ energies larger than those for extended states. Assuming K' ­ that these states are delocalized along the perimeter of the ribbon, over a scale L R, see Fig. 36, and width a ∼ comparable to the lattice spacing, see Fig. 36, the charg- ¯ ¯ 2 K ing energy becomes (e /R) log(R/a) (Wimmer et al., K' ¯ 2010). × aL Charging effects can also modify the transport prop- ­ erties of narrow graphene ribbons. Irregularities in the K K edges may induce the formation of constrictions and K ¯ quantum dots, as sketched in Fig. 37, where charging effects will lead to a transport gap. In a nanoribbon of width W , the typical size of these dots will also be W , K' K' ­ and the transport gap will be of order e2/W . In the K' ¯ absence of charging effects, a ribbon will have confined subbands, separated by gaps of order v/W . Hence, the bL similarity between the energy scales arising from quan- FIG. 39 (Color online). Sketch of the successive splittings of tum confinement and charging effects, which exists in a the Landau levels as the magnetic field is increased. a) Spin quantum dot, also exists in a graphene ribbon. An exper- states are split first, and then the valley degeneracy is broken. imental realization of an all graphene circuit with a point b) Valley degeneracy is lifted first, followed by the breaking contact coupled to a quantum dot (Stampfer et al., 2009) of spin degeneracy. is shown in Fig. 38. This setup can be used to count the passage of charges through the quantum dot. Experiments in graphene nanoribbons are compatible with the relevance of charging effects (Han et al., 2010, 2007; Todd et al., 2009). Some observations can be ex- plained by a model of dots formed in the ribbon con- nected through many channels with the rest of the struc- ture. Such a strongly coupled dot always shows Coulomb blockade effects, unless there is a perfect transmission through one or more of the channels. The effective charging energy, however, is strongly renormalized by the coupling between the dot and the rest of the system 2 g (Sols et al., 2007), Ec e /We− , where g is the con- ductance, in dimensionless≈ units, of the junction between 2 the dot and the electrodes. In general, g T kF W , where T is the transmission amplitude of∼ a given h| | i× channel. The electron-electron interactions can be studied in mesoscopic samples through their effect on the magneto- conductance at low magnetic fields. These experiments probe the phase coherence of electrons at low temper- atures. This quantum effect is suppressed due to the dephasing induced by the interactions. Electronic quan- tum coherence also gives rise to the universal conduc- tance fluctuations observed in disordered metals, which are also reduced by the dephasing due to interactions. FIG. 40 (Color online). Splittings of the Landau levels in The dephasing length shows a temperature dependence graphene as function of magnetic field, see (Zhang et al., consistent with the expected behavior in a dirty metal, 2006). ℓφ (g~v)/[T log(g)], where g is the conductivity in dimensionless∼ units (Tikhonenko et al., 2009) (see also (Chen et al., 2010)). This dependence is replaced by a VIII. INTERACTIONS IN STRONG MAGNETIC FIELDS 2 ℓφ T − in high mobility samples (Tikhonenko et al., 2009∝), as expected in a clean Fermi liquid. Experiments A comprehensive review of graphene in magnetic field that tune the ratio between the dephasing length and has recently appeared (Goerbig, 2011), and here we only the mean free path (Moser et al., 2010) show a variety of mention some of the main effects. The electronic en- regimes, interpolating between weak and strong localiza- ergy bands of graphene in a strong magnetic field col- tion. lapse into Landau levels. In the absence of disorder, 47

spin and valley degeneracy when the Zeeman coupling is neglected gives a new SU(4) symmetry, which may lead to new features, not observable in other two di- mensional electron gases (Goerbig and Regnault, 2007; T¨oke and Jain, 2007). The formation of Landau levels favors the excitonic transition which can also exist in the absence of a magnetic field (Gusynin et al., 2006). The spin split n = 0 level leads to spin polarized edge states (Abanin et al., 2006, 2007; Fertig and Brey, 2006; Shimshoni et al., 2009) where the orientation of the spin depends on the sign of the current, as in topological insulators (Hasan and Kane, 2010; Qi and Zhang, 2011). A magnetic field oriented parallel to the plane does not give rise to Landau levels. In neutral graphene, it leads to metallic states with electrons and holes polarized in opposite directions, providing another route towards an excitonic transition (Aleiner et al., 2007). Experiments show that, indeed, the spin and val- FIG. 41 (Color online). Resistance of a suspended graphene sample as a function of carrier density for two different mag- ley degeneracies of Landau levels in graphene are netic fields. R. V. Gorbachev, D. C. Elias, A. S. Mayorov, A. lifted (Giesbers et al., 2007, 2009; Jiang et al., 2007; A. Zhukov, K. S. Novoselov, A. K. Geim (unpublished). Zhang et al., 2006). The opening of a gap in the n = 0 level in graphene has been extensively studied, and a metal insulator transition with critical features con- the electronic compressibility diverges when the chemi- sistent with a Berezinskii-Kosterlitz-Thouless transition cal potential coincides with the energy of a Landau level, has been reported (Amado et al., 2009; Checkelsky et al., and the effects of the interactions are enhanced, as in 2008, 2009). other two dimensional metallic systems. The typical scale The most striking manifestation of the interac- of the electronic wavefunctions is the magnetic length, tions in the presence of a strong magnetic field is ℓB = ~/(eB) = Φ0/(2πB), where B is the applied the Fractional Quantum Hall Effect. Early theo- field and Φ is the quantum unit of flux. The separation retical calculations showed that the FQHE could be p 0 p between levels is of order v/ℓB, while the relevant scale stable in graphene (Apalkov and Chakraborty, 2006; 2 for interaction effects is e /ℓB. Castro Neto et al., 2006; T¨oke et al., 2006). The condi- There are two sets of Landau levels in graphene, one tions for the FQHE are the existence of sharp Landau lev- for each valley. In addition, graphene has the n = 0 els and sufficiently strong electron electron interactions. level, which combines electron and hole features. Hence, The analysis of FQHE states in graphene can be done interactions can break either the valley degeneracy or the in a similar way to that of a two dimensional electron spin degeneracy. The long range part of the Coulomb in- gas. The main difference is a change in the pseudopoten- teraction is independent of the valley index. The n = 0 tials which describe the interactions between electrons Landau level is localized in a given sublattice, and its in a given Landau level, because the wavefunctions in degeneracy can be lifted by interactions which break the graphene and in a two dimensional electron gas differ. symmetry between sublattices, like the coupling to out of This Fractional Quantum Hall Effect was extensively, plane optical phonons (Fuchs and Lederer, 2007). Hence, but unsuccessfully, sought in samples deposited on SiO2. the removal of the spin and valley degeneracies of the Suspended samples, which showed a much higher elec- Landau levels due to interactions depends on other en- tron mobility, did not exhibit the FQHE, using the stan- ergy scales (Goerbig, 2011), such as the Zeeman splitting, dard experimental four terminal setup. The observa- or the nearest neighbor repulsion, for the case n =0. A tion of the IQHE in suspended bilayer graphene using sketch of the possible symmetry breaking patterns as a a two terminal setup (Feldman et al., 2009) led quickly function of magnetic field is shown in Fig. 39. Early ob- to the discovery of the FQHE in single layer graphene servations of splittings between Landau levels are shown (Bolotin et al., 2009; Du et al., 2009), using the same in Fig. 40, see (Zhang et al., 2006). technique. More recently, four terminal measurements in It is usually assumed that the Zeeman splitting high mobility suspended samples (Ghahari et al., 2011), is much smaller than the other energy scales. Cal- and also samples deposited on a new substrate, boron ni- culations suggest that the spin degeneracy is lifted tride (Dean et al., 2011, 2010), also show the FQHE. In first, leading to excitations with combined spin and two terminal measurements, the existence of the FQHE valley indices (Abanin et al., 2007; Alicea and Fisher, is inferred from plateaus of the longitudinal resistance at 2006; Goerbig et al., 2006; Gusynin et al., 2009; carrier densities which correspond to fractional fillings of Nomura and MacDonald, 2006; Shibata and Nomura, Landau levels, see Fig. 41. The ν =1/3 state turns out to 2008; Wang et al., 2008; Yang et al., 2006). The fourfold be more robust than in other materials, like GaAs, which 48 exhibit the FQHE, and it can be observed at tempera- a) b) +E tures greater than 10K. Fractional plateaus at ν = 2/3 + and ν = 1/2 have also been reported. Theoretical cal- culations suggest that the so called Moore-Read ground −E state at fillings with even denominators, which leads to − the existence of non Abelian anyonic quasiparticles, is +E − not favored in graphene (Wojs et al., 2011).

−E+ IX. INTERACTIONS IN BILAYERS c) K d) E Bilayers are the building blocks for 3D stacks of M graphene, such as graphite. In a bilayer one has two parallel graphene sheets, separated by an equilibrium distance similar to the interlayer distance of graphite Γ (3.35A)˚ (Dresselhaus and Dresselhaus, 1981). The rel- ative position of the two graphene layers is not unique, and this leads to different stacking arrangements of the bilayer, and even more possibilities for multilayers, or graphite. The most stable configuration seems to be the FIG. 42 a) Top view of a graphene bilayer; white and black so-called Bernal AB stacking, in which the two layers circles: top layer carbon atoms; gray and red: bottom layer. o b) four-band spectrum of the bilayer, Eγ (p), with γ = as are rotated by 60 . As a consequence, one of the the ± ± sublattices in the lower layer (say, sublattice A) is ver- shown in Eq. (9.4), near the corner of the Brillouin zone. c) tically aligned with one of the sublattices of the upper Brillouin zone with high symmetry points. d) Illustration of the four band spectrum around the K point. layer (say, sublattice B) [see Fig. 42(a)]. Notice that this particular rotation leads to a breaking of sublattice symmetry between layers. As a first approximation, the figuration is an exception. The twisted bilayer graphene electronic coupling between the layers can be described presents a very rich physics of its own that we will not in terms of the hopping of electrons between the near- cover in this review. Instead, we will focus on the Bernal est neighbor atoms in different layers with an energy t ⊥ configuration which is the most studied case. (also known as γ1 0.39 eV in the graphite literature (Castro Neto et al.,≈2009a). Another possible arrange- We will start from the minimal tight-binding model ment between the layers is the fully aligned configuration, for Bernal bilayers, which includes a basis with two ad- also called AA stacking. In both AB and AA stacking, ditional layer flavors (denoted by an overbar), the unit cell is comprised of 4 atoms, and has the same Ψ = (a ,b , ¯b , a¯ ) , (9.1) 2D extension as the unit cell of a single layer; this im- k,σ k,σ k,σ k,σ k,σ plies that the Brillouin zone is precisely the same as in with σ = , representing the spin. The resulting Bloch monolayer graphene. Hamiltonian↑ ↓ is then a 4 4 matrix with two sublattice, Notice, however, that these configurations are just a and two layer degrees of× freedom, few of an infinite series of commensurate structures be- tween two layers, the so-called twisted bilayer graphene 0 tφk t 0 (Lopes dos Santos et al., 2007). The problem of com- − − ⊥ tφk∗ 0 0 0 mensurate and incommensurate structures always ap- B = Ψk† ,σ − Ψk,σ, H  t 0 0 tφk∗  pears when two crystalline materials are superimposed, kσ − ⊥ − X  0 0 tφk 0  as in the case of bilayers. For commensurate structures,  −  (9.2) the angle between the layers is not arbitrary but follows where t 0.39eV is the interlayer hopping, and t a well defined sequence (Lopes dos Santos et al., 2007). 2.8eV is⊥ the≈ in-plane, nearest neighbor, hopping ampli-≈ Obviously, different angles lead to different broken sym- tude. The momentum dependence is contained in φk, metries and hence to different electronic states. When which is the same as for a monolayer (2.3). The band the angle of rotation is 60 degrees, as in the case of structure associated with eq. (9.2) consists of four non- the Bernal structure, the sublattices are nonequivalent, degenerate bands given by which leads to a broken sublattice symmetry and hence to a putative gap opening. For other angles, there is no 1 2 2 2 broken sublattice symmetry but the unit cell is enlarged E(k)= t t +4t φk . (9.3) ±2 ⊥ ± ⊥ | | as the rotation angle becomes smaller. In this case the  q  massless Dirac dispersion has to be preserved for sym- An expansion k = K + p around the K points of metry reasons (Li et al., 2010; Lopes dos Santos et al., the BZ when v p t shows that the four-band tight- 2007; Mele, 2010). From this perspective, the Bernal con- binding spectrum| | ≪ (9.3) resolves into four hyperbolic 49 bands (Nilsson et al., 2006), as shown in Fig. 42(b), and of next-nearest neighbor hopping matrix elements [which whose form reads: we are neglecting in (9.2)]. Trigonal warping introduces an asymmetry in the conductivity under electron or hole t 2 Eγ (p)= ⊥ 1+ γ 1+4(v p /t ) , (9.4) doping (Li et al., 2009b), and leads to a remarkable Lif- ± ± 2 | | ⊥ h p i shitz transition at low densities, whereby the lowest en- with v 6 eVA˚ being the Fermi velocity (the same Fermi ergy bands split into 4 Dirac cones (Cserti et al., 2007; ≈ velocity of a monolayer), and γ = 1. The Bernal stack- McCann and Fal’ko, 2006). These effects, however, hap- ± ing explicitly breaks the sublattice symmetry in each pen at very low densities (around 1 electron per flake layer, causing an energy split of t between the two for typical 1µm2 samples), and hence are experimen- ⊥ γ = 1 branches, E+ and E , at p = 0 (see Fig. 42). tally very challenging. A detailed description of the ± − Due to a degeneracy at the K points, the two symmetric spectral properties of graphene bilayers can be found in branches +E and E+ touch there, resulting in a gap- Castro Neto et al., 2009a, and Nilsson et al., 2008. − − less spectrum. Just as in a monolayer, the Fermi surface When ∆w < v p t , we recover the so-called clas- of an undoped bilayer reduces to only two points, at K sical limit of the “relativistic”| | ≪ ⊥ problem. This means that and K′; but now the valence and conduction bands have the presence of the uppermost band is not too relevant, a finite curvature and, hence, notwithstanding the ab- and the energy disperses quadratically with momentum sence of a gap, the effective electronic degrees of freedom (the opposite limit of v p t corresponds to the are massive, but still chiral. The degeneracy at K is pro- “ultra-relativistic” regime,| where| ≫ the⊥ bandstructure is es- tected by the Z2 symmetry between the two layers only sentially linear in momentum, like in the monolayer). In (McCann, 2006), and can be lifted with arbitrarily small this case the Hamiltonian (9.2) near the K points can perturbations, such as the ones induced by a bias volt- be projected onto an effective two-band model, written age, by polarizing the two sheets (Zhang et al., 2009), or in terms of the two valleys and a mixed sublattice-layer else by independently changing the carrier concentration basis (McCann and Fal’ko, 2006): in each layer (Ohta et al., 2006). This property opens the exciting prospect of using graphene bilayers as mate- Ψ˜ p,σ = (aK+p,σ, ¯bK+p,σ, ¯b K+p,σ,a K+p,σ) . (9.6) rials with a gate-tunable band gap (Castro et al., 2007; − − Castro Neto et al., 2007; Min et al., 2007). In such a basis, the effective kinetic Hamiltonian is We stress that the low energy effective theory of bilay- 2 ers remains Lorentz invariant, in the following sense. The pα = Ψ˜ † [τ σ ] Ψ˜ p , (9.7) rotation of π/3 between layers breaks the sublattice sym- HB p,σ 2m 0 ⊗ α ,σ pσ α= metry leading to 2 pairs of massive Dirac particles at the X X± K (K′) point. Nevertheless, the system remains metallic where p = px ipy, σ = (σ1 iσ2)/2 operating in the because two bands, belonging to different pairs, touch in sublattice± basis,± and τ ±operates± in the valley space. The a point. More explicitly, the non-interacting bands (9.4) resulting energy spectrum is parabolic, have the form: 2 2 p E1(k)= E (k)= mv + E(k), (9.5a) E(p)= , (9.8) − − − 2 ±2m E2(k)=+E (k)= mv E(k), (9.5b) − − 2 2 with m = t /(2v ) 0.054 me as the effective mass of E3(k)=+E+(k)= mv + E(k), (9.5c) the electron.⊥ From now≈ on we will omit the valley indexes E (k)= E (k)= mv2 E(k), (9.5d) ˜ ¯ 4 − + − − and assume the two component basis Ψp,σ (ap,σ, bp,σ) with a total degeneracy N = 4 in valley and→ spin. where E(k) = (mv2)2 + (vk)2, and m = t /(2v2). ⊥ The electronic Green’s function in this two band Hence, E1(k) and E4(k) [or E2(k) and E3(k)] describe ˆ(0) ˜ ˜ p model, G (k, τ) = T [Ψk(τ)Ψk† (0)] , is given by a massive relativistic dispersion with rest energy given (0) −h1 i 2 Gˆ (k,iω) = (iω ˆ )− or, equivalently, by by mv . Again, the gapless nature of the full spectrum − HB of this problem is due to an accidental degeneracy of 1 1+ sσˆk the simplest tight binding parametrization. Additional Gˆ(0)(k,iω)= (9.9) 2 iω s E(k) hopping terms (Castro Neto et al., 2009a) in the Hamil- s= X± − | | tonian or many body interactions can easily lift this de- generacy. This implies that the Bernal bilayer problem is in the chiral representation, where unstable from the electronic point of view. In contrast, 2 et al. kα the twisted bilayer (Lopes dos Santos , 2007) is sta- σˆk = σα . (9.10) ble because it does not rely on this particular accidental k 2 α= | | degeneracy. Just like in the case of monolayer graphene, X± the introduction of the instantaneous Coulomb interac- Although the fermions are chiral, in bilayers the wave- tion does not preserve this Lorentz invariance. functions of the quasiparticles acquire a 2π phase when At very low energy, below ∆w 1.5 meV, additional winding around the K points, rather than a π-phase, as trigonal warping effects take place≈ due to the influence for Dirac fermions. This property is an admixture of the 50

The polarization function Π(1)(q,ω) at finite density was obtained by Hwang and Das Sarma, 2008a in the T = 0 static limit. The full dynamical case was cal- culated by Sensarma et al., 2010 at T = 0, and by Lv and Wan, 2010 at finite temperature. The finite den- sity result can be obtained in closed analytical form for T = 0; but, in order to avoid reproducing here those lengthy expressions, we simply present Π(1)(q,ω) graph- ically in Figs. 43(a,b). The explicit form of the static limit reads (Hwang and Das Sarma, 2008a; Lv and Wan, 2010)

(1) Π (q, 0) q q = g f θ(q 2kF ) (9.12) − ρ(µ) kF − kF −

at zero temperature, with

2+ x2 x √x2 4 f(x)= x2 4 + ln − − (9.13a) 2x − x + √x2 4! p − 1 2+ √4+ x4 g(x)= 4+ x4 ln . (9.13b) 2 − 4 ! FIG. 43 (Color online) The polarization Π(1)(q,ω) of bi- p layer graphene, obtained within the two-band approxima- The DOS at the Fermi energy, ρ(µ)= Nm/(2π), is con- tion, for finite chemical potential, and zero temperature. All stant and density independent, by virtue of the parabolic panels are normalized to the DOS at the Fermi energy, µ. nature of the low energy approximations (9.7) and (9.8) Panel (a) shows a density plot of the imaginary part and, [note, however, that the consideration of the full 4-band in (b), we have cuts of the same at constant frequency, for spectrum leads to a DOS which is linear in energy; in ω/µ = 0.5, 1.0, 1.5, 2.0, 2.5, 3.0. Panel (c) corresponds to the (1) this sense, the correction to the DOS that arises from static limit Π (q, 0) in eq. (9.12), and includes the intra-band considering the 4 versus the 2 band model is not neg- contribution (dashed), the inter-band contribution (dotted), and the full polarization (solid). In (d) we represent the real ligible (Ando, 2007)]. In this sense the bilayer is simi- and imaginary parts of the polarization in the undoped case lar to the conventional 2DEG. However, just as in the (9.14) as a function of ν = 2mω/q2. monolayer, the existence of two symmetric bands adds an inter-band channel, leading to a rather different quasi- particle spectrum, in comparison with the 2DEG. This behavior of Dirac particles, which are chiral, with con- can be seen by directly comparing Figs. 4(b) and 43(a). (1) ventional electrons, which disperse quadratically. The The behavior of Π (q, 0) is shown in Fig. 43(c), together combination of chirality and a trivial Berry phase has with its decomposition into intra- and inter-band contri- a clear experimental signature in the suppression of the butions, which are respectively associated with the choice ss′ = 1, or ss′ = 1 in eq. (9.11). As intuitively ex- zero-level plateau in the quantum Hall effect of the bi- − layer, whose plateaus are quantized by integer numbers pected, the inter-band contribution dominates at large (McCann and Fal’ko, 2006; Novoselov et al., 2006). momenta/small densities, whereas the intra-band tran- sitions dominate the low momenta/large density regime. Unlike the monolayer, or the 2DEG, the polarization is A. Charge polarization constant for both q kF and q kF . The former limit makes the bilayer similar≪ to the≫ conventional 2DEG and Within the two band model, the one loop polarization monolayer graphene, while the latter is neither akin to function has the generic form given in Eq. (2.12) for the the 2DEG (for which the polarization decreases rapidly single layer. The adaptations for the present case consist with q/kF [Fig. 4(e)]), nor to the monolayer (for which it increases linearly [Fig. 3(e)]). Moreover, at precisely in considering the bilayer spectrum, and a new overlap (1) factor, which, for the bilayer, reads q = 2kF , Π (q, 0) is sharply cusped, which contrasts with the behavior of a monolayer, whose derivative is 1 continuous. According to the standard theories of linear ′ p q = [1 + ss′ cos(2θp p q)]. (9.11) Fs,s , , 2 , + response, this feature at 2kF has important implications for the behavior of the induced charge, the associated de- In this expression θp,p+q is, again, the angle between the cay of the Friedel oscillations around charged impurities, vectors p and p + q. Below we shall focus our discussion the effective RKKY interaction among magnetic impuri- in terms of the effective two-band Hamiltonian (9.7), and ties, Kohn’s anomaly in the phonon dispersion, etc. For dispersion (9.8). example, one expects qualitative differences between the 51 resistivity arising from Coulomb scattering in mono- and reveals that the real part of the RPA dielectric func- (1) bilayer graphene: it should be stronger in the bilayer, tion ǫRPA(q,ω) = ǫ0[1 V (q)Π (q,ω)] will be always and have a more pronounced temperature dependence nonzero. This means that,− although the lack of a Fermi (Hwang and Das Sarma, 2008a; Lv and Wan, 2010). surface does not prevent screening in bilayers (qT F = 0), At long wavelengths, the RPA screened potential reads the formation of zero temperature infrared plasmons6 is RPA (1) 2 V (q)= V (q)/[1 V (q)Π (q)] 2πe /[ǫ0(q + qT F )], suppressed at half-filling. − ≈ 2 with a Thomas-Fermi momentum qT F = Nme /ǫ0. No- The screened Coulomb interaction between the lay- 2 qd qd tice that qT F is the same for the bilayer as in the 2DEG, ers is V (q)=2πe e− /[ǫ0(q + qT F e− )], where d = i.e. it is constant (no density dependence), and also tem- 3.35 A˚ is the interlayer distance. At long wavelengths, 1 perature independent (Lv and Wan, 2010). The temper- q t /v < 1/d 0.3 A˚− , d can be effectively replaced ≪ ⊥ ≈ ature independence of qT F at long wavelengths is an- by zero in first approximation, and the screened interac- other trait that distinguishes this system from both the tion among electrons belonging to the same or different monolayer and the 2DEG. In real space the statically planes can be treated on the same footing. screened potential decays asymptotically as V (r) 1/r3 ∝ At this point we should pause to point out that the (Hwang and Das Sarma, 2008a). behaviors discussed so far at large q have to be inter- At half-filling (undoped situation) and zero tempera- preted within the restrictions regarding the validity of ture, the form of the polarization bubble simplifies fur- the two-band approximation. For example, the fact that ther, and can be cast as in Fig. 43(c) we see the polarization becoming constant at q kF is an artifact of the two-band approximation. Nm 1 1+ ν 1 1+2ν In reality,≫ we should bear in mind that the full dispersion Π(1)(ν)= ln ln − 2π ν 1 ν − 2ν 1 2ν is hyperbolic, and hence becomes linear at high densities.   −   −  1 ν2 We then expect to recover the linear-in-q dependence of + ln − (9.14) Π(1)(q, 0) seen in Fig. 3(e) for the monolayer. 1 ν2  4 −  For this reason, proper caution is needed when consid- ering the extrapolation of these results to highly doped (Barlas and Yang, 2009; Nandkishore and Levitov, bilayers, where the consideration of the four-band hy- et al. 2 2010a; Nilsson , 2006), where ν = 2mω/q is the perbolic dispersion (9.4) is inevitably required. In terms only scaling parameter. This function is plotted in of electronic densities, this corresponds to values above Fig. 43(d). It follows at once that the static limit 12 2 10 cm− , for which the two-band model is no longer (ω 0) is simply ∼ → warranted. The full dynamical response using the spec- trum in eq. (9.4) has been recently derived in closed an- N ln 4 Π(1)(q, 0) = m, (9.15) alytical form by Borghi et al., 2009b. Notwithstanding − 2π the lengthy and cumbersome nature of these analytical results, they afford a more accurate perspective on the consistent with the above discussion when kF = 0. Despite the absence of a Fermi surface at half-filling, screening response of doped bilayer graphene, its col- the Coulomb interaction among the quasiparticles is lective modes, and the crossover between the regimes screened due to the finite density of states at the K of a massive-chiral system at low densities, to a sys- points. However, an important difference here is that tem of weekly coupled monolayers at higher densities. Π(1)(q, 0) is constant for all momenta, unlike traditional Borghi et al., 2009b’s approach is ultimately limited by systems of such high densities that µ t, in which case 2D systems, and stems from the presence of the inter- ≈ band channel. Hence, the Thomas-Fermi wavevector the full tight-binding dispersion (9.3) is needed, but is 2 beyond closed analytical approaches. is exactly qT F = Nm ln(4)e /ǫ0 for all wavelengths, and Friedel oscillations are suppressed at half-filling (Hwang and Das Sarma, 2008a). The additional numer- ical factor ln(4) means a slight increase in the screen- B. Quasiparticles ing strength of undoped bilayer, with respect to the doped situation. One way to interpret this ln(4) en- In the two band model, the structure of perturbation 2 hancement is the following: the factor Nme /ǫ0, being theory for Coulomb interactions is set only by self-energy exactly the same as in a simple 2DEG, is attributable renormalizations in the effective mass of the electrons, m, to the finite DOS, while the extra ln(4) arises from the and in the quasiparticle residue, Z. virtual inter-band transitions. In real space, the stat- From the Hamiltonian (9.7), the renormalized Green’s ically screened potential of undoped bilayer will decay function is as 1/r3, which contrasts with the corresponding behav- 1 ior in the monolayer, where the decay is 1/r (as we saw Gˆ(k,ω)= . (9.16) 2 ˆ before this is due to the fact that, in the RPA, the ef- ω α= kα/(2m)σα Σ(k,ω) − ± − fect of interactions in the monolayer is to simply renor- P malize the background dielectric constant, keeping the Σ(ˆ k,ω) is the quasiparticle self-energy correction, which Coulomb form of the potential). Inspection of Fig. 43(d) is described in the (ak,σ, ¯bk,σ) basis by a matrix in the 52

for the two-band model (Borghi et al., 2009a), and also for the four-band model (Kusminskiy et al., 2009), have found mass renormalization in the bilayer. The mass decreases (m∗/m < 1), and the renormalization grows stronger as the screening is suppressed. In Fig. 44 we show this renormalization within the two-band model, where the parameter λ interpolates between the Thomas- Fermi screened potential (λ = 1) and the unscreened Coulomb potential (λ 0). As a consequence of the re- duced mass, the charge≈ compressibility is also expected to decrease (Borghi et al., 2010; Kusminskiy et al., 2008). More recent calculations that account for the full dynamical screening have found quite differ- ent results. When the dynamical RPA polar- ization bubble, Eq (9.14), is taken into account, the self-energy exhibits a strong ln2 leading diver- FIG. 44 (Color online) Mass renormalization for α = 0.5 in (1) 2 2 2 the bilayer, calculated with a static Thomas-Fermi screened gence, ReΣ+ (k.ω) = 2k+/(Nmπ ) ln (Λ/k), and Coulomb interaction V (q)= e2/[ǫ (q + λq )], as a function (1) 2 2 0 TF ReΣ0 (k,ω) = 4ω/(Nπ ) ln [Λ/(√mω)], at small en- of the electronic density. Blue circles: λ = 1; red squares: λ = − − ergies and momenta (Barlas and Yang, 2009). The ultra- 0.01; green triangles: λ = 10 4. (From Borghi et al., 2009a). ∗ violet momentum scale Λ qT F is related to the effective The inset shows log (m /m) as a function of log (n) for two ∼ 2 10 10 “Bohr radius”, a0 = ǫ0/(me ), and we set Λ=1/a0. At of the λ values; the mass saturates at a finite value for n 0. 2 → leading (ln ) order, the two terms in the self-energy com- pensate each other exactly in Eq. (9.19) and the mass does not renormalize, m∗/m 1 at k 0, while the form → → 2 quasiparticle spectral weight vanishes as Z ln− (Λ/k). ∼ Σ0 Σ+ The RG analysis of the dynamically screened interaction Σ=ˆ , (9.17) Σ Σ0 at large N was carried out by Nandkishore and Levitov, −   2010c, where subleading (single log) contributions were or, equivalently, Σˆ = Σ0σ0 + Σ+σ+ + Σ σ , where collected. These were found to cause a (weak) increase − − σ = (σx iσy)/2. By symmetry, Σ+ = Σ∗ . In a more of the effective mass m∗/m 1 + [0.56/(N2π ln 4)] ln Λ, ± ± − conventional form, and consequently an increase≈ of the compressibility. Z Once again, the validity of a two band model rests ˆ k G( ,ω)= 2 , (9.18) on the assumption that all relevant energy scales are ω Z α= [kα/(2m)+Σα]σα − ± small compared to t 0.4eV. However the Coulomb ⊥ ≈ 2 1 P energy ΛE on the scale of a0 = ǫ0/(me ) is substantial where Z− =1 ∂Σ0/∂ω corresponds to the quasiparticle 2 − for not too strong dielectric screening, ΛE = e /(ǫ0a0) residue, and 2 ≈ 1.47/ǫ0 eV (Nandkishore and Levitov, 2010a,c). Hence, m∗ 1 ∂Σ0/∂ω Coulomb interactions can promote electronic transitions = − 2 (9.19) among the four bands, while the two-band model is only m 1+2m∂Σ+/∂k+ justified in the limit ΛE < t . To what extent the two is the mass renormalization. band model provides a valid⊥ description of the quasipar- We saw in the previous section that, unlike the mono- ticles in the presence of Coulomb interactions is a matter layer, Coulomb interactions in the bilayer are screened. of ongoing discussion. The self-energy is given in terms of the bare Green’s func- tion and the RPA effective interaction by

d2kdε C. Many-body instabilities Σˆ (1)(q,ω)= i V RPA(k,ε)Gˆ(0)(k + q,ε + ω), (2π)3 Z (9.20) The finite DOS in the bilayer enhances the possibility RPA where V (q,ω)= V (q)/ǫRPA(q,ω) is dressed by the of many body instabilities in comparison with the single RPA dielectric function. Even if the ratio between the layer case. For instance, the spin polarization tensor in Coulomb and kinetic energies diverges in the low density the bilayer is defined in leading order by Eq. (5.4), with limit (as in a 2DEG) the validity of RPA can be, in prin- ˆs 2 2 the matrix element (k)=1+ s α= (kα/k )σα. In ciple, justified in the large N limit. If only static screen- matrix form, A ± ing is taken into account (Hartree-Fock-Thomas-Fermi P theory), the self-energy is frequency independent and, to (1)+ (1)+ leading order, the quasiparticle residue Z does not renor- ˆ (1)+ Πaa − Πab − Π − = (1)+ (1)+ , (9.21) malize. Calculations based on the static screening picture Πba − Πbb − ! 53 which leads to one ferromagnetic, and one antiferro- charge between the layers, which dominates the kinetic 2 2 magnetic eigenstate, ΠF/AF = Πaa Πab , by sym- energy at the Hartree level, δEHartree ∆ ln (ΛE/∆) metry under exchange of the a and b±labels. | | In bilay- (McCann et al., 2007). The excitonic∝ instability is in- ers the AF state has a leading logarithmic divergence duced by the exchange term, which is parametrically with the cut-off, Λ, at zero frequency and magnetic field, larger than the Hartree term by the factor (a0/d), where (Nilsson et al., 2006) d is the interlayer distance (Nandkishore and Levitov, 2010a). The existence of a ferroelectric state has nev- m 2Λ Π(1) (q, 0) = ln , (9.22) ertheless been disputed by independent RG calcula- AF π q tions, that also accounted for the dynamically screened  | |  Coulomb interactions, and infrared trigonal warping ef- suggesting (within RPA) a tendency towards an AF in- fects (Lemonik et al., 2010). The spontaneous sym- stability for any value of the Hubbard interaction U. In metry breaking found in this work leads to a Lifshitz addition, at finite U, a first order ferromagnetic tran- transition consistent with the nematic state found by sition can be driven by the Stoner criterion, leading to Vafek and Yang, 2010, rather than the opening of a gap. a ferromagneto-electric state where the layers have dif- In the Quantum Hall (QH) state, two terminal mea- ferent magnetization and polarized charge (Castro et al., surements of the conductivity in clean suspended sam- 2008b). ples have found an insulating state at the ν = 0 fill- Other possibilities include the emergence of CDW ing factor (Feldman et al., 2009), rather than the metal- instabilities induced by the short range part of the lic QH state previously found in supported samples Coulomb interaction, (Dahal et al., 2010) or else, an (Novoselov et al., 2006). Further theoretical works pre- excitonic instability at strong local electronic repul- dicted the possibility of a zero field excitonic QH state, sion (Dillenschneider and Han, 2008). With long which spontaneously breaks time reversal symmetry, and range Coulomb interactions, the inverse electronic com- can evolve into a ferromagnetic QH state at finite mag- pressibility κ 1 becomes negative at small densities − netic field (Nandkishore and Levitov, 2010b). In biased (Kusminskiy et al., 2008), indicating a tendency to bilayers, a chiral anomaly has been predicted in the Wigner crystallization (Dahal et al., 2006), which is com- QHE, splitting the degeneracy of valley quantum num- pensated by the positive compressibility of the lattice. bers (Nakamura et al., 2009). Another predicted effect Bilayers share similar features with one-dimensional resulting from interactions in the QH state is the for- (1D) electron systems, such as the point like Fermi sur- mation of charge 2e skyrmions at even filling factors faces and the parabolic spectrum. In particular, in bi- (Abanin et al., 2009). ased bilayers, the 1D interface between biased regions confines chiral modes that propagate as in a strongly in- teracting Luttinger liquid (Killi et al., 2010). This af- fords the possibility of studying such interacting models X. CONCLUSIONS experimentally in appropriately prepared samples of bi- layer graphene. As we have seen, the understanding of the many-body For short-ranged interactions in 2D, the structure of problem in graphene has evolved quite rapidly in only a the diagrams in bilayers and in 1D electron liquids is quite few years. The case of monolayer graphene in the weak similar, although the diagrams compensate each other in coupling regime (which means, graphene embedded in a rather different way. The dimensionless coupling which an environment with large dielectric constant) is quite determines the strength of the interactions is Ua2m, clear, namely, although Lorentz invariance is explicitly where U is the strength of the local interactions, and a is broken because of the Coulomb interactions, the effec- the lattice constant. Perturbative renormalization group tive low energy theory is still Lorentz invariant with well calculations in the bilayer have identified distinct leading defined quasiparticles. Nevertheless, these quasiparticles instabilities of the electron gas. For different choices of have a renormalized speed of light that grows logarithmi- possible interactions, two different low-temperature bro- cally in the infrared, while their spectral weight decreases ken symmetry phases have been found: in one case, a fer- slowly in the same limit. This situation can be contrasted roelectric gapped phase (Zhang et al., 2010) induced by with the conventional Fermi liquid picture where all the the coupling between the different layers; in the other, physical constants (the so-called Landau parameters) and a nematic phase (Vafek, 2010; Vafek and Yang, 2010), spectral weight are finite in the infrared (that is, at the where each Fermi point splits into two Dirac points. Fermi surface). Hence, these logarithmic renormaliza- The possibility of an excitonic instability has been also tions are weak enough, even in the presence of strong predicted by Nandkishore and Levitov, 2010a, who found Coulomb interactions, and a Dirac liquid picture is pre- that the dynamically screened Coulomb interaction gives served. rise to a ferroelectric state that polarizes the two lay- In the strong coupling regime (that is, graphene in ers. In the ferroelectric state, the kinetic energy inflicts vacuum), many-body instabilities are possible albeit de- 2 an energy cost δEKinetic ∆ ln(ΛE/∆), where ∆ is the pending on a delicate balance of energy scales. This oc- energy gap. Finite separation∝ between the layers gener- curs because the renormalizations of quasiparticle prop- ates an additional electrostatic energy cost to polarize the erties also depend on details of the cut-off procedure in 54 the ultraviolet (as it is shown by the f-sum rule). While discussed in the context of the f-sum rule). Moreover, mean-field theories have predicted instabilities towards the damping by Dirac electrons leads to an anomalously phases with broken chiral symmetry and superconduct- large (and strongly energy dependent) broadening of the ing quasi-long range order (because of the 2D nature adatom energy level. This leads to an unusual situa- of the material), and earlier Monte Carlo studies on a tion as compared to the Anderson impurity problem in a hyper-cubic lattice suggest the presence of instabilities metal, namely, that even when the chemical potential is (Drut and L¨ahde, 2009a,b,c), simulations of interacting above (below) the energy of the doubly (singly) occupied electrons on the honeycomb lattice have still to be per- state, a magnetic moment can emerge. Hence, adatoms formed in order to address these issues, since the strong that may not be magnetic in a metal (hydrogen or fluo- coupling regime cannot be reached by perturbative meth- rine, for instance), might become magnetic in graphene. ods. This remains, at this point in time, as an important On the other hand, the Kondo effect that usually sup- open problem in many-body graphene physics. presses the appearance of magnetic moments in metals The Coulomb impurity problem in graphene shares because of magnetic “screening” (the ultimate conse- many of the issues of the many-body problem but can quence of the so-called “Kondo cloud”), is strongly sup- be studied in much more detail because the 2D hydro- pressed in graphene. This suppression has its roots in gen problem in graphene was solved exactly. In the weak the low density of states and the sublattice structure. In coupling regime (the so-called under-critical regime), the fact, there is a strong dependence of the hybridization Coulomb interaction between a localized charge and the with the position on the lattice (whether it breaks or not electrons leads to only mild changes in the physical prop- the sublattice symmetry). Furthermore, the Kondo effect erties due to the explicitly broken particle-hole symme- is very dependent on the chemical potential (that can be try. In the strong coupling (or super-critical) regime, the easily tuned in graphene by gating). This state of affairs situation is rather different because of the phenomenon of reinforces the conclusion that magnetic states of adatoms fall to the center, that is, the electron states become un- could be more the norm than the exception in graphene, stable, with the generation of resonances near the Dirac in a big contrast with the situation in ordinary metals. point. Just like the many-body problem, the critical lo- Experimentally, there are very few studies of the mag- cal charge depends on the dielectric environment, and in netism of adatoms in graphene. The main problem here vacuum this amazing effect should be observed by local is that most of the experiments done so far are in elec- probes even for a single proton sitting on the graphene tronic transport. Just like the Kondo problem in metals surface. So far, there is no experimental evidence of such and semiconductors, the observation of magnetic effects effect, given that it is difficult to study adatoms in sus- in transport is rather subtle, and requires careful analy- pended samples with local probes, such as scanning tun- sis. At this point in time, this is a rather open field in neling microscopes. In supported samples, because of graphene physics. dielectric screening that brings the system to weak cou- A superconducting state in graphene would have dra- pling, and of the disorder in the substrate, the study of matic consequences given its low dimensionality and un- this problem can be much more elusive. usual electronic spectrum. While true long range order In analogy to the 2DEG problem, the effect of disor- would not be possible because of its 2D nature, quasi- der is rather strong in graphene which again is the ef- long range order would have unusual consequences. For fect of dimensionality. The low dimensionality implies one, because of the sublattice structure, there is room for strong quantum fluctuations that can easily couple to exotic pairing states with even more exotic vortex excita- spatial variations of random scalar (chemical potential) tions. The phase space for pairing is rather large due to and gauge (hopping) fields. Strong localization is the the spin, sublattice, and valley degeneracies. However, ultimate fate of any disordered two dimensional system the low density of states plays a deleterious role here. but because the localization length grows very slowly in One way out of this conundrum would be the enhance- the infrared limit, the finite size of the samples, or the ment of the density of states by either gating or doping finite temperature of the system, ends up cutting off the with adatoms. These two techniques have their own lim- tendency towards Anderson localization and, in practical itations. Gating is limited by the distance from the gate terms, graphene behaves in a metallic way. to the graphene sample, and by the dielectric breakdown The problem of magnetism of adatoms in graphene is of the spacer that separates the two. Doping inevitably rather different from the one found in metallic hosts. Due introduces disorder, or can modify the electronic struc- to the strong energy dependence of the density of states ture of the π band too much leading to extrinsic effects. (that vanishes at the Dirac point), the Anderson impu- There are, however, serious hopes that come from the fact rity problem has features that are unique. Firstly, in that intercalated graphite can be made to superconduct. analogy with the strong coupling regime in the many- An obvious idea would be intercalation of Ca or Yb in body and Coulomb impurity problems, the results are the graphene bilayer. So far, intercalation experiments sensitive to the ultraviolet regularization. In fact, this is in bilayers have not been performed, and very little is a generic feature of the Dirac spectrum, namely, strong known about how to intercalate atoms or molecules in coupling leads to spectral weight transfer from high en- such systems. Again, this is very much an open field of ergies to low energies, that is, to the Dirac point (as research. 55

In addition to the dielectric environment, which has a leading to a huge number of possible many-body states strong influence on many-body effects in bulk graphene, with different quantum numbers. Given this richness one finite size effects are also of great importance. It has been can venture saying that bilayer graphene is the ultimate understood very early on that zig-zag edges are strongly target of many-body theorists in this field. However, it is interacting because of the high density of states they cre- technically a major challenge given the high dimension- ate at the Dirac point. Systems with high density of ality of the problem, with its 24-dimensional spinorial states are prone to many-body states due to Stoner-like structure (spin, valley, sublattice, and plane). Moreover, instabilities. However, the many-body physics of finite from the experimental perspective many details and con- 2D systems is even more sensitive to disorder (either in ditions are still quite uncontrolled, which has led to a the bulk or in the edge) because of the strong bound- few contradictory results, and has so far yielded more ary condition dependence. In graphene this problem is questions than answers. In fact, both theoretically and magnified because the electronic wavefunctions associ- experimentally, the graphene bilayer remains very much ated with impurity states do not decay exponentially, as an open problem. If we now extrapolate from the mono- they would in a semiconductor with a finite gap, or would layer to the bilayer, we see that there are problems that be extended, as in a normal metal, but they are quasi- have not even been addressed theoretically and exper- localized (that is, decay like power law). This implies imentally, like the Anderson impurity problem, or the that evanescent waves play an important role in deter- Kondo effect in bilayers, the problem of magnetism, and mining the physical properties. Experiments in meso- superconductivity, just to mention some. These are top- scopic graphene samples show very clearly these effects ics for the future, for future generations of physicists to through strong oscillations of the electronic conductance address and marvel. and the presence of Coulomb blockade peaks. From theo- retical perspective, such problems are probably the hard- est to solve because they involve the direct interplay Acknowledgments between Anderson localization and interactions. Thus deeper understanding of mesoscopic graphene systems is We are indebted to our collaborators, friends and col- still necessary, and this topic would merit a review of its leagues for their many invaluable contributions, discus- own. sions, comments, and suggestions. In particular we want Magnetic fields also lead to spatial localization due to to explicitly thank E. Andrei, Y. Barlas, S. Das Sarma, the presence of Landau levels with a length scale given V. Fal’ko, M. M. Fogler, E. Fradkin, A. Geim, M. Goer- by the cyclotron length. Hence, this problem shares big, J. Gonzalez, I. Herbut, M. I. Katsnelson, P. Kim, A. many of the difficulties of the previous problems with Lanzara, J. Lopes dos Santos, A. MacDonald, E. Mucci- the added complication that the 2D nature of graphene olo, J. Nilsson, K. Novoselov, N. Peres, S. Sachdev, O. brings a huge degeneracy into play. Once again, the Sushkov, O. Vafek, S. Viola, M. A. H. Vozmediano, and detailed balance between kinetic and Coulomb energies, A. Yacoby. and the details in the ultraviolet, determine the fate of A.H.C.N. acknowledges DOE grant DE-FG02- the many-body ground state. The fractional quantum 08ER46512 and ONR grant MURI N00014-09-1-1063. Hall effect was only observed recently in suspended two- B.U. acknowledges partial support from DOE grant probe experiments (Du et al., 2009), and very little is DE-FG02-91ER45439 at the University of Illinois. F.G. known about the sequence of FQHE fractions and their acknowledges financial support by MICINN (Spain) nature. It is believed that magnetic fields can generate a through grants FIS2008-00124 and CONSOLIDER plethora of new many-body states, with symmetries that CSD2007-00010, and by the Comunidad de Madrid, are rather different from the ones found in the 2DEG. through NANOBIOMAG. V.N.K. acknowledges the But, compared to the 2DEG problem, this field is still in financial support of the University of Vermont. its infancy. While we have demonstrated the complexity of the many-body problem in monolayer graphene, we have not References even touched beyond the surface of the many-body prob- lem in bilayer graphene. There is no doubt, at least from Abanin, D. A., P. A. Lee, and L. S. Levitov, 2006, Phys. Rev. the theoretical perspective, that the many-body problem Lett. 96, 176803. in the bilayer is much richer than in the monolayer. For Abanin, D. A., K. S. Novoselov, U. Zeitler, P. A. Lee, A. K. one, the bilayer has a finite density of states at neutral- Geim, and L. S. Levitov, 2007, Phys. Rev. Lett. 98, 196806. ity, making it similar to the 2DEG problem. However, Abanin, D. A., S. A. Parameswaran, and S. L. Sondhi, 2009, Phys. Rev. Lett. 103, 076802. unlike the 2DEG, the graphene bilayer is a Lorentz in- Abergel, D. S. L., P. Pietil¨ainen, and T. Chakraborty, 2009, variant system with a finite “rest mass” (that is, it has a Phys. Rev. B 80, 081408. hyperbolic dispersion relation) albeit with an accidental Abramowitz, M., and I. A. Stegun, 1964, Handbook of Math- degeneracy that makes it a semi-metal (two of the four ematical Functions (Dover, New York). bands touch at the Dirac point). This accidental degen- Abrikosov, A. A., and S. D. Beneslavskii, 1971, Sov. Phys. eracy can be lifted very easily by hopping or interactions, JETP 32, 699. 56

Adam, S., E. H. Hwang, V. M. Galitski, and S. Das Sarma, Borghi, G., M. Polini, R. Asgari, and A. H. MacDonald, 2007, Proc. Nat. Acad. Sci. 104, 18392. 2009b, Phys. Rev. B 80, 241402. Akhmerov, A. R., and C. W. J. Beenakker, 2007, Phys. Rev. Borkowski, L. S., and P. J. Hirschfeld, 1992, Phys. Rev. B 46, Lett. 98, 157003. 9274. Akhmerov, A. R., and C. W. J. Beenakker, 2008, Phys. Rev. Bostwick, A., T. Ohta, T. Seyller, K. Horn, and E. Rotenberg, B 77, 085423. 2007, Nature Physics 3, 36. Aleiner, I. L., and K. B. Efetov, 2006, Phys. Rev. Lett. 97, Bostwick, A., F. Speck, T. Seyller, K. Horn, M. Polini, R. As- 236801. gari, A. H. MacDonald, and E. Rotenberg, 2010, Science Aleiner, I. L., D. E. Kharzeev, and A. M. Tsvelik, 2007, Phys. 328, 999. Rev. B 76, 195415. Boukhvalov, D. W., and M. I. Katsnelson, 2009, J. Phys. Alicea, J., and M. P. Fisher, 2006, Phys. Rev. B 74, 075422. Chem. C 113, 14176. Amado, M., E. Diez, D. L´opez-Romero, F. Rossella, J. M. Boukhvalov, D. W., M. I. Katsnelson, and A. I. Lichtenstein, Caridad, V. Bellani, and D. K. Maude, 2009, New. J. Phys. 2008, Phys. Rev. B 77, 035427. 12, 053004. Brar, V. W., S. Wickenburg, M. Panlasigui, C.-H. Park, T. O. Anderson, P. W., 1961, Phys. Rev. 124, 41. Wehling, Y. Zhang, R. Decker, C. Girit, A. V. Balatsky, Anderson, P. W., 1967, Phys. Rev. Lett. 18, 1049. S. G. Louie, A. Zettl, and M. F. Crommie, 2010, Phys. Ando, T., 2006, J. Phys. Soc. Jpn. 75, 074716. Rev. Lett. 104, 036805. Ando, T., 2007, J. Phys. Soc. Jpn. 76, 104711. Brey, L., H. A. Fertig, and S. Das Sarma, 2007, Phys. Rev. Ando, T., A. B. Fowler, and F. Stern, 1982, Rev. Mod. Phys. Lett. 99, 116802. 54, 437. Brown, L. S., R. N. Cahn, and L. D. McLerran, 1975, Phys. Apalkov, V. M., and T. Chakraborty, 2006, Phys. Rev. Lett. Rev. D 12, 581. 97, 126801. Bunch, J. S., Y. Yaish, M. Brink, K. Bolotin, and P. L. Appelquist, T., M. Bostwick, D. Karabali, T., and L. R. Wi- McEuen, 2005, Nano Lett. 5, 287. jewardhana, 1988, Phys. Rev. D 60, 2575. Burset, P., A. L. Yeyati, and A. Mart´ın-Rodero, 2008, Phys. Appelquist, T. W., M. Bostwick, D. Karabali, and L. C. R. Rev. B 77, 205425. Wijewardhana, 1986, Phys. Rev. D 33, 3704. Calandra, M., and F. Mauri, 2007, Phys. Rev. B 76, 205111. Avouris, P., Z. Chen, and V. Perebeinos, 2007, Nature Nan- Case, K. M., 1960, Phys. Rev. 80, 797. otechnology 2, 605. Cassanello, C. R., and E. Fradkin, 1996, Phys. Rev. B 53, Balatsky, A. V., I. Vekhter, and J. X. Zhu, 2006, Rev. Mod. 15079. Phys. 78, 373. Cassanello, C. R., and E. Fradkin, 1997, Phys. Rev. B 56, Barlas, Y., T. Pereg-Barnea, M. Polini, R. Asgari, and A. H. 11246. MacDonald, 2007, Phys. Rev. Lett. 98, 236601. Castro, E. V., M. P. L´opez-Sancho, and M. A. H. Vozmediano, Barlas, Y., and K. Yang, 2009, Phys. Rev. B 80, 161408. 2010, Phys. Rev. Lett. 104, 036802. Barraza-Lopez, S., M. Vanevi´c, M. Kindermann, and M. Y. Castro, E. V., K. S. Novoselov, S. V. Morozov, N. R. Peres, Chou, 2010, Phys. Rev. Lett. 104, 076807. J. M. B. Lopes dos Santos, J. Nilsson, F. Guinea, A. K. Barzola-Quiquia, J., P. Esquinazi, M. Rothermel, D. Spe- Geim, and A. H. Castro Neto, 2007, Phys. Rev. Lett. 99, mann, T. Butz, and N. Garc´ıa, 2007, Phys. Rev. B 76, 216802. 161403. Castro, E. V., N. M. R. Peres, J. M. B. Lopes dos Santos, Baym, G., 1969, Lectures on Quantum Mechanics (Addison- A. H. Castro Neto, and F. Guinea, 2008a, Phys. Rev. Lett. Wesley, Reading, MA). 100, 026802. Baym, G., and C. Pethick, 1991, Landau Fermi-Liquid Theory Castro, E. V., N. M. R. Peres, T. Stauber, and N. P. Silva, (John Wiley, New York). 2008b, Phys. Rev. Lett. 100, 186803. Beenakker, C. W. J., 2006, Phys. Rev. Lett. 97, 067007. Castro Neto, A. H., 2001, Phys. Rev. Lett. 86, 4382. Beenakker, C. W. J., 2008, Rev. Mod. Phys. 80, 1337. Castro Neto, A. H., and F. Guinea, 2009, Phys. Rev. Lett. Bena, C., 2008, Phys. Rev. Lett. 100, 076601. 103, 026804. Bena, C., and S. Kivelson, 2006, Phys. Rev. B 72, 125432. Castro Neto, A. H., F. Guinea, and N. M. Peres, 2006, Phys. Bercx, M., T. C. Lang, and F. F. Assaad, 2009, Phys. Rev. B Rev. B 73, 205408. 80, 045412. Castro Neto, A. H., F. Guinea, N. M. R. Peres, K. S. Bergman, D., and K. L. Hur, 2009, Phys. Rev. B 79, 185420. Novoselov, and A. K. Geim, 2009a, Rev. Mod. Phys. 81, Bhattacharjee, S., and K. Sengupta, 2006, Phys. Rev. Lett. 109. 97, 217001. Castro Neto, A. H., V. N. Kotov, J. Nilsson, V. M. Pereira, Biswas, R. R., S. Sachdev, and D. T. Son, 2007, Phys. Rev. N. M. R. Peres, and B. Uchoa, 2009b, Solid State Commun. B 76, 205122. 149, 1094. Black-Schaffer, A. M., and S. Doniach, 2007, Phys. Rev. B Castro Neto, A. H., J. Nilsson, F. Guinea, and N. M. R. Peres, 75, 134512. 2007, Phys. Rev. B 76, 165416. Black-Schaffer, A. M., and S. Doniach, 2008, Phys. Rev. B Ceni, R., 2001, Nucl. Phys. A 696, 605. 78, 024504. Ceperley, D., 1978, Phys. Rev. B 18, 3126. Bloch, F., 1928, Z. Physik 52, 555. Chamon, C., C.-Y. Hou, R. Jackiw, C. Mudry, S.-Y. Pi, and Bolotin, K. I., F. Ghahari, M. D. Shulman, H. L. Stormer, G. Semenoff, 2008a, Phys. Rev. B 77, 235431. and P. Kim, 2009, Nature 462, 196. Chamon, C., C.-Y. Hou, R. Jackiw, C. Mudry, S.-Y.Pi, and Borghi, G., M. Polini, R. Asgari, and A. MacDonald, 2010, A. P. Schnyder, 2008b, Phys. Rev. Lett. 100, 110405. Phys. Rev. B 82, 155403. Chan, K. T., J. B. Neaton, and M. L. Cohen, 2008, Phys. Borghi, G., M. Polini, R. Asgari, and A. H. MacDonald, Rev. B 77, 235430. 2009a, Solid State Commun. 149, 1117. Checkelsky, J. G., L. Li, and N. P. Ong, 2008, Phys. Rev. 57

Lett. 100, 206801. rozov, A. A. Zhukov, P. Blake, L. A. Ponomarenko, I. V. Checkelsky, J. G., L. Li, and N. P. Ong, 2009, Phys. Rev. B Grigorieva, K. S. Novoselov, F. Guinea, and A. K. Geim, 79, 115434. 2011, Nature Physics 7, 701. Cheianov, V. V., and V. I. Fal’ko, 2006, Phys. Rev. Lett. 97, Elias, D. C., R. R. Nair, T. M. G. Mohiuddin, S. Morozov, 226801. P. Blake, M. P. Halsall, A. C. Ferrari, D. W. Boukhvalov, Chen, F., J. Xia, D. K. Ferry, and N. Tao, 2009a, Nano Lett. M. I. Katsnelson, A. K. Geim, and K. S. Novoselov, 2009, 9, 2571. Science 323, 610. Chen, F., J. Xia, and N. Tao, 2009b, Nano Lett. 9, 1621. Enoki, T., and K. Takai, 2009, Solid State Commun. 149, Chen, J.-H., W. G. Cullen, E. D. Williams, and M. S. Fuhrer, 1144. 2011, Nature Physics 7, 535. Esquinazi, P., D. Spemann, R. H¨ohne, A. Setzer, K.-H. Han, Chen, J. H., C. Jang, S. Adam, M. S. Fuhrer, E. D. Williams, and T. Butz, 2003, Phys. Rev. Lett. 91, 227201. and M. Ishigami, 2008, Nature Physics 4, 377. Feigel’man, M. V., M. A. Skvortov, and K. S. Tikhonov, 2008, Chen, K., and C. Jayaprakash, 1995, J. Phys.: Condens. Pris’ma v ZhETP 88, 780. Mattter 7, L491. Feldman, B. E., J. Martin, and A. Yacoby, 2009, Nature Chen, Y.-F., M.-H. Bae, C. Chialvo, T. Dirks, A. Bezryadin, Physics 5, 890. and N. Mason, 2010, J. Phys.: Condens. Matter 22, Fermi, E., 1927, Rend. Accad. Naz. Lincei 6, 602. 205301. Fern´andez-Rossier, J., 2008, Phys. Rev. B 77, 075430. Coleman, P., 1983, Phys. Rev. B 28, 5255. Fern´andez-Rossier, J., and J. J. Palacios, 2007, Phys. Rev. Cooper, L. N., 1956, Phys. Rev. 104, 1189. Lett. 99, 177204. Coqblin, B., and J. R. Schrieffer, 1969, Phys. Rev. 185, 847. Fertig, H. A., and L. Brey, 2006, Phys. Rev. Lett. 97, 116805. Cornaglia, P. S., G. Usaj, and C. A. Balseiro, 2009, Phys. Fetter, A., 1974, Ann. Phys. 88, 1. Rev. Lett. 102, 046801. Fetter, A. L., and J. D. Walecka, 1971, Quantum Theory of Csanyi, G., P. B. Littlewood, A. H. Nevidomskyy, and C. P. Many-Particle systems, International Series in Pure and B. D. Simons, 2005, Nature Physics 1, 42. Applied Physics (McGraw-Hill Book Company). Cserti, J., A. Csord´as, and G. D´avid, 2007, Phys. Rev. Lett. Fogler, M., D. Novikov, and B. Shklovskii, 2007, Phys. Rev. 99, 066802. B 76, 233402. Dahal, H. P., Y. N. Joglekar, K. S. Bedell, and A. V. Balatsky, Foster, M. S., and I. L. Aleiner, 2008, Phys. Rev. B 77, 2006, Phys. Rev. B 74, 233405. 195413. Dahal, H. P., T. O. Wehling, K. S. Bedell, J.-X. Zhu, and Foster, M. S., and A. W. W. Ludwig, 2006a, Phys. Rev. B A. Balatsky, 2010, Physica B 405, 2241. 73, 155104. Das Sarma, S., S. Adam, E. H. Hwang, and E. Rossi, 2011, Foster, M. S., and A. W. W. Ludwig, 2006b, Phys. Rev. B Rev. Mod. Phys. 83, 407. 74, 241102(R). Das Sarma, S., E. H. Hwang, and W.-K. Tse, 2007, Phys. Fradkin, E., 1986, Phys. Rev. B 33, 3263. Rev. B 75, 121406(R). Fritz, L., J.Schmalian, M. M¨uller, and S. Sachdev, 2008, Phys. de Juan, F., A. G. Grushin, and M. A. H. Vozmediano, 2010, Rev. B 78, 085416. Phys. Rev. B 82, 125409. Fritz, L., and M. Vojta, 2004, Phys. Rev. B 70, 214427. Dean, C. R., A. F. Young, P. Cadden-Zimansky, L. Wang, Fuchs, J., and P. Lederer, 2007, Phys. Rev. Lett. 98, 016803. H. Ren, K. Watanabe, T. Taniguchi, P. Kim, J. Hone, and Fujita, M., K. Wakabayashi, K. Nakada, and K. Kusakabe, K. L. Shepard, 2011, Nature Physics 7, 693. 1996, J. Phys. Soc. Jpn. 65, 1920. Dean, C. R., A. F. Young, I. Meric, L. W. C. Lee, S. Sorgen- Furukawa, N., 2001, J. Phys. Soc. Jpn. 70, 1483. frei, K. Watanabe, T. Taniguchi, P. Kim, K. L. Shepard, Gamayun, O. V., E. V. Gorbar, and V. P. Gusynin, 2009, and J. Hone, 2010, Nature Nanotechnology 5, 722. Phys. Rev. B 80, 165429. Dell’Anna, L., 2010, J. Stat. Mech. , P01007. Gamayun, O. V., E. V. Gorbar, and V. P. Gusynin, 2010, Dillenschneider, R., and J. H. Han, 2008, Phys. Rev. B 78, Phys. Rev. B 81, 075429. 045401. Gangadharaiah, S., A. M. Farid, and E. G. Mishchenko, 2008, DiVincenzo, D., and E. Mele, 1984, Phys. Rev. B 29, 1685. Phys. Rev. Lett. 100, 166802. Dora, B., and P. Thalmeier, 2007, Phys. Rev. B 76, 115407. Geim, A. K., and K. S. Novoselov, 2007, Nature Materials 6, Dresselhaus, M. S., and G. Dresselhaus, 1981, Adv. Phys. 30, 183. 139. Ghaemi, P., and F. Wilczek, 2012, Physica Scripta T146, Drude, P., 1900a, Annalen der Physik 1, 566. 014019. Drude, P., 1900b, Annalen der Physik 3, 369. Ghahari, F., Y. Zhao, P. Cadden-Zimansky, K. Bolotin, and Drut, J. E., and T. A. L¨ahde, 2009a, Phys. Rev. B 79, P. Kim, 2011, Phys. Rev. Lett. 106, 046801. 241405(R). Ghosal, A., P. Goswami, and S. Chakravarty, 2007, Phys. Drut, J. E., and T. A. L¨ahde, 2009b, Phys. Rev. Lett. 102, Rev. B 75, 115123. 026802. Giesbers, A. J., U. Zeitler, M. I. Katsnelson, L. A. Pono- Drut, J. E., and T. A. L¨ahde, 2009c, Phys. Rev. B 79, 165425. marenko, T. M. Mohiuddin, and J. C. Maan, 2007, Phys. Du, X., I. Skachko, and E. Y. Andrei, 2008, Phys. Rev. B 77, Rev. Lett. 99, 206803. 184507. Giesbers, A. J. M., U. Zeitler, L. A. Ponomarenko, R. Yang, Du, X., I. Skachko, F. Duerr, A. Luican, and E. Y. Andrei, K. S. Novoselov, A. K. Geim, and J. C. Maan, 2009, Phys. 2009, Nature 462, 192. Rev. B 80, 241411. Eberlein, T., U. Bangert, R. R. Nair, R. Jones, M.Grass, A. L. Giovannetti, G., P. A. Khomyakov, G. Brocks, V. M.Karpan, Bleloch, K. S. Novoselov, A. Geim, and P. R. Briddon, J. van den Brink, and P. J. Kelly, 2008, Phys. Rev. Lett. 2008, Phys. Rev. B 77, 233406. 101, 026803. Elias, D. C., R. V. Gorbachev, A. S. Mayorov, S. V. Mo- Girit, C. O.,¨ J. C. Meyer, R. Erni, M. D. Rossell, 58

C. Kisielowski, L. Yang, C.-H. C.-H. Park, M. F. Crom- Rev. Lett. 98, 206805. mie, M. L. Cohen, S. G. Louie, and A. Zettl, 2009, Science Hands, S., and C. Strouthos, 2008, Phys. Rev. B 78, 165423. 323, 1705. Hanish, T., B. Kleine, A. Ritzl, and E. M¨uller-Hartmann, Goerbig, M. O., 2011, Rev. Mod. Phys. 83, 1193. 1995, Ann. Phys. 4, 303. Goerbig, M. O., R. Moessner, and B. Doucot, 2006, Phys. Hannay, N. B., T. H. Gaballe, B. T. Matthias, K. Andres, Rev. B 74, 161407. P. Schimidt, and D. MacNair, 1965, Phys. Rev. Lett. 14, Goerbig, M. O., and N. Regnault, 2007, Phys. Rev. B 75, 225. 241405. Harigaya, K., 2001, J. Phys. C.: Condens. Matt. 13, 1295. Goldman, S., and G. Drake, 1982, Phys. Rev. A 25, 2877. Harigaya, K., and T. Enoki, 2002, Chem. Phys. Lett. 351, Gonz´alez, J., F. Guinea, and M. A. H. Vozmediano, 1994, 128. Nucl. Phys. B 424, 595. Hasan, M. Z., and C. L. Kane, 2010, Rev. Mod. Phys. 82, Gonz´alez, J., F. Guinea, and M. A. H. Vozmediano, 1996, 3045. Phys. Rev. Lett. 77, 3589. Hawrylak, P., 1987, Phys. Rev. Lett. 59, 485. Gonz´alez, J., F. Guinea, and M. A. H. Vozmediano, 1999, Heersche, H. B., P. J. Herrero, J. B. Oostinga, L. K. Versypen, Phys. Rev. B 59, R2474. and A. Morpurgo, 2007, Nature 446, 56. Gonz´alez, J., 2008, Phys. Rev. B 78, 205431. Hentschel, M., and F. Guinea, 2007, Phys. Rev. B 76, 115407. Gonz´alez, J., 2010, Phys. Rev. B 82, 155404. Herbut, I. F., 2006, Phys. Rev. Lett. 97, 146401. Gonz´alez, J., F. Guinea, and M. A. H. Vozmediano, 2001, Herbut, I. F., 2008, Phys. Rev. B 78, 205433. Phys. Rev. B 63, 134421. Herbut, I. F., 2010, Phys. Rev. Lett. 104, 066404. Gonz´alez-Buxton, C., and K. Ingersent, 1998, Phys. Rev. B Herbut, I. F., V. Juricic, and B. Roy, 2009, Phys. Rev. B 79, 57, 14254. 085116. Gorbar, E. V., V. P. Gusynin, V. A. Miransky, and I. A. Herbut, I. F., V. Juricic, and O. Vafek, 2008, Phys. Rev. Lett. Shovkovy, 2002, Phys. Rev. B 66, 045108. 100, 046403. Greenbaum, D., S. Das, G. Schwiete, and P. G. Silvestrov, Hoddeson, L., G. Baym, and M. Eckert, 1987, Rev. Mod. 2007, Phys. Rev. B 75, 195437. Phys. 59, 287. Greiner, W., B. M¨uller, J. Rafelski, W. Greiner, B. Muller, Honerkamp, C., 2008, Phys. Rev. Lett. 100, 146404. and J. Rafelski, 1985, Quantum Electrodynamics of Strong Horowitz, B., and P. L. Doussal, 2002, Phys. Rev. B 65, Fields (Springer). 125323. Gross, D. J., and A. Neveu, 1974, Phys. Rev. D 10, 3235. Hou, C.-Y., C. Chamon, and C. Mudry, 2007, Phys. Rev. Gr¨uneis, A., C. Attaccalite, A. Rubio, D. V. Vyalikh, S. L. Lett. 98, 186809. Molodtsov, J. Fink, R. Follath, W. Eberhardt, B. B¨uchner, Huard, B., N. Stander, J. A. Sulpizio, and D. Goldhaber- and T. Pichler, 2009, Phys. Rev. B 80, 075431. Gordon, 2008, Phys. Rev. B 78, 121402. Gr¨uneis, A., and D. V. Vyalikh, 2008, Physical Review B 77, Huard, B., J. A. Sulpizio, N. Stander, K. Todd, B. Yang, and 193401. D. Goldhaber-Gordon, 2007, Phys. Rev. Lett. 98, 236803. Gruner, G., 1994, Density waves in solids (Perseus, Cam- Huertas-Herno, D., F. Guinea, and A. Brataas, 2006, Phys. bridge, MA). Rev. B 74, 155426. Grushin, A. G., B. Valenzuela, and M. A. H. Vozmediano, Hwang, E. H., and S. Das Sarma, 2007, Phys. Rev. B 75, 2009, Phys. Rev. B 80, 155417. 205418. Guettinger, J., C. Stampfer, S. Hellmueller, F. Molitor, Hwang, E. H., and S. Das Sarma, 2008a, Phys. Rev. Lett. T. Ihn, and K. Ensslin, 2008, Appl. Phys. Lett. 93, 212102. 101, 156802. Guinea, F., 2007, Phys. Rev. B 75, 235433. Hwang, E. H., and S. Das Sarma, 2008b, Phys. Rev. B 77, Guinea, F., B. Horowitz, and P. L. Doussal, 2008a, Phys. Rev. 081412. B 77, 205421. Hwang, E. H., B. Y.-K. Hu, and S. Das Sarma, 2007, Phys. Guinea, F., M. I. Katsnelson, and A. K. Geim, 2010, Nature Rev. Lett. 99, 226801. Physics 6, 30. Ingersent, K., and Q. Si, 2002, Phys. Rev. Lett. 89, 076403. Guinea, F., M. I. Katsnelson, and M. A. H. Vozmediano, Jackiw, R., 1984, Phys. Rev. D 27, 2375. 2008b, Phys. Rev. B 77, 075422. Jackiw, R., and P. Rossi, 1981, Nucl. Phys. B 190, 681. Gupta, K., and S. Sen, 2008, Phys. Rev. B 78, 205429. Jacob, D., and G. Kotliar, 2010, Phys. Rev. B 82, 085423. Gupta, K. S., A. Samsarov, and S. Sen, 2010, Eur. Phys. J. Jang, C., S. Adam, J.-H. Chen, E. D. Williams, S. Das Sarma, B 73, 389. and M. S. Fuhrer, 2008, Phys. Rev. Lett. 101, 146805. Gupta, K. S., and S. Sen, 2009, Mod. Phys. Lett. A 24, 99. Jia, X., M. Hofmann, V. Meunier, B. G. Sumpter, J. Campos- Gusynin, V. P., V. A. Miransky, S. G. Sharapov, and I. A. Delgado, J.-M. Romo-Herrera, H. Son, Y.-P. Hsieh, Shovkovy, 2006, Phys. Rev. B 74, 195429. A. Reina, J. Kong, M. Terrones, and M. S. Dresselhaus, Gusynin, V. P., V. A. Miransky, S. G. Sharapov, I. A. 2009, Science 323, 1701. Shovkovy, and C. M. Wyenberg, 2009, Phys. Rev. B 79, Jiang, Y., D.-X. Yao, E. W. Carlson, H.-D. Chen, and J.-P. 115431. Hu, 2008, Phys. Rev. B 77, 235420. Gusynin, V. P., S. G. Sharapov, and J. P. Carbotte, 2007, Jiang, Z., Y. Zhang, H. L. Stormer, and P. Kim, 2007, Phys. Int. J. Mod. Phys. B 21, 4611. Rev. Lett. 99, 106802. G¨uttinger, J., C. Stampfer, F. Libisch, T. Frey, J. Burgdorfer, Joly, V. L. J., M. Kiguchi, S.-J. Hao, K. Takai, T. Enoki, T. Ihn, and K. Ensslin, 2009, Phys. Rev. Lett. 103, 046810. R. Sumii, K. Amemiya, H. Muramatsu, T. Hayashi, Y. A. Haldane, F. D. M., 1988, Phys. Rev. Lett. 61, 2015. Kim, M. Endo, J. Campos-Delgado, et al., 2010, Phys. Rev. Han, M. Y., J. C. Brant, and P. Kim, 2010, Phys. Rev. Lett. B 81, 245428. 104, 056801. Jung, J., and A. H. MacDonald, 2010, Phys. Rev. B 81, Han, M. Y., B. Ozyilmaz,¨ Y. Zhang, and P. Kim, 2007, Phys. 195408. 59

Juricic, V., O. Vafek, and I. F. Herbut, 2010, Phys. Rev. B Science 329, 544. 82, 235402. Li, G., A. Luican, and E. Y. Andrei, 2009a, Phys. Rev. Lett. Kane, C., and E. J. Mele, 2005, Phys. Rev. Lett. 95, 226801. 102, 176804. Kane, C. L., and M. P. A. Fisher, 1992, Phys. Rev. Lett. 68, Li, G., A. Luican, J. M. B. Lopes dos Santos, A. H. Castro 1220. Neto, A. Reina, J. Kong, and E. Y. Andrei, 2010, Nature Kashuba, A. B., 2008, Phys. Rev. B 78, 085415. Physics 6, 109. Katsnelson, M., 2006, Phys. Rev. B 74, 201401(R). Li, Z. Q., E. A. Henriksen, Z. Jiang, Z. Hao, M. C. Martin, Katsnelson, M. I., and M. F. Prokhorova, 2008, Phys. Rev. B P. Kim, H. L. Stormer, and D. N. Basov, 2008, Nature 77, 205424. Physics 4, 532. Khalilov, V. R., and C.-L. Ho, 1998, Mod. Phys. Lett. A 13, Li, Z. Q., E. A. Henriksen, Z. Jiang, Z. Hao, M. C. Martin, 615. P. Kim, H. L. Stormer, and D. N. Basov, 2009b, Phys. Rev. Kharitonov, M. Y., and K. B. Efetov, 2008, Phys. Rev. B 78, Lett. 102, 037403. 241401(R). Lieb, E. H., 1981, Rev. Mod. Phys. 53, 603. Khaymovich, M., N. B. Kopnin, A. S. Melnikov, and I. A. Lin, D.-H., 2006, Phys. Rev. A 73, 044701. Shereshevskii, 2009, Phys. Rev. B 79, 224506. Lin, M. F., and K. W. K. Shung, 1996, Phys. Rev. B 53, 1109. Khveshchenko, D. V., 2001a, Phys. Rev. Lett. 87, 206401. Linder, J., A. M. Black-Schaffer, T. Yokoyama, S. Doniach, Khveshchenko, D. V., 2001b, Phys. Rev. Lett. 87, 246802. and A. Sudbø, 2009, Phys. Rev. B 80, 094522. Khveshchenko, D. V., 2006, Phys. Rev. B 74, 161402(R). Linder, J., T. Yokoyama, D. Huertas-Herno, and A. Sudbø, Khveshchenko, D. V., 2009, J. Phys.: Condens. Matter 21, 2008, Phys. Rev. Lett. 100, 187004. 075303. Lindhard, J., 1954, Det Kgl Danske Vid. Selskab, Khveshchenko, D. V., and H. Leal, 2004, Nucl. Phys. B 687, Matematisk-fysiske Meddelelser 28. 323. Liu, G.-Z., W. Li, and G. Cheng, 2009, Phys. Rev. B 79, Killi, M., T. Wei, I. Affleck, and A. Paramekanti, 2010, Phys. 205429(R). Rev. Lett. 104, 216406. Liu, Y., R. F. Willis, K. V. Emtsev, and T. Seyller, 2008, Kirwan, D. F., C. G. Rocha, A. T. Costa, and M. S. Ferreira, Phys. Rev. B 78, 201403(R). 2008, Phys. Rev. B 77, 085432. Loktev, V. M., and V. Turkowski, 2009, Phys. Rev. B 79, Kohn, W., and J. M. Luttinger, 1965, Phys. Rev. Lett. 15, 233402. 524. Lopes dos Santos, J. M. B., N. M. R. Peres, and A. H. Castro Kolezhuk, A., S. Sachdev, R. R. Biswas, and P. Chen, 2006, Neto, 2007, Phys. Rev. Lett. 99, 256802. Phys. Rev. B 74, 165114. Lozovik, Y. E., and A. A. Sokolik, 2010, Phys. Lett. A 374, Kopnin, N. P., and E. B. Sonin, 2008, Phys. Rev. Lett. 100, 2785. 246808. Ludwig, A. W., M. P. A. Fisher, R. Shankar, and G. Grinstein, Kotov, V. N., V. M. Pereira, and B. Uchoa, 2008a, Phys. Rev. 1994, Phys. Rev. B 50, 7526. B 78, 075433. Lundqvist, B., 1967, Zeitschrift fur Physik B Condensed Mat- Kotov, V. N., B. Uchoa, and A. H. Castro Neto, 2008b, Phys. ter 6, 193. Rev. B 78, 035119. Lutchyn, R. M., V. Galitski, G. Refael, and S. Das Sarma, Kotov, V. N., B. Uchoa, and A. H. Castro Neto, 2009, Phys. 2008, Phys. Rev. Lett. 101, 106402. Rev. B 80, 165424. Lv, M., and S. Wan, 2010, Phys. Rev. B 81, 195409. Kovtun, P. K., D. T. Son, and A. O. Starinets, 2005, Phys. Mahan, G. D., 2000, Many-Particle Physics (Plenum, New Rev. Lett. 94, 111601. York). Kramberger, C., R. Hambach, C. Giorgetti, M. H. Rummeli, Maiti, M., and K. Sengupta, 2007, Phys. Rev. B 76, 054513. M. Knupfer, J. Fink, B. Buchner, L. Reining, E. Einarsson, Marino, E. C., and L. H. C. M. Nunes, 2006, Nucl. Phys. B S. Maruyama, F. Sottile, K. Hannewald, et al., 2008, Phys. 741, 404. Rev. Lett. 100, 196803. Marinopoulus, A. G., L. Reining, A. Rubio, and V. Olevano, Krasheninnikov, A. V., P. O. Lehtinen, A. S. Foster, 2004, Phys. Rev. B 69, 245419. P. Pyykko, and R. M. Nieminen, 2009, Phys. Rev. Lett. Martelo, L. M., M. Dzierzawa, L. Siffert, and D. Baeriswyl, 102, 126807. 1997, Z. Phys. B 103, 335. Kusminskiy, S. V., J. Nilsson, D. K. Campbell, and A. H. Martin, J., N. Akerman, G. Ulbricht, T. Lohmann, J. H. Castro Neto, 2008, Phys. Rev. Lett. 100, 106805. Smet, K. von Klitzing, and A. Yacoby, 2008, Nature Kusminskiy, S. V., J. Nilsson, D. K. Campbell, and A. H. Physics 4, 144. Castro Neto, 2009, Europhys. Lett. 85, 58005. Martinazzo, R., S. Casolo, and G. F. Tantardini, 2010, Phys. Laitenberger, P., and R. E. Palmer, 1996, Phys. Rev. Lett. Rev. B 81, 245420. 76, 1952. McCann, E., 2006, Phys. Rev. B 74, 161403(R). Landau, L. D., and E. M. Lifshitz, 1981, Quantum Mechanics: McCann, E., D. S. L. Abergel, and V. I. Falko, 2007, Solid Non-Relativistic Theory (Pergamon Press). State Commun. 143, 110. Laughlin, R. B., 1983, Phys. Rev. Lett. 50, 1395. McCann, E., and V. I. Fal’ko, 2006, Phys. Rev. Lett. 96, Lee, P. A., 1993, Phys. Rev. Lett. 71, 1887. 086805. Lehtinen, P. O., A. S. Foster, A. Ayuela, A. Krasheninnikov, McChesney, J. L., A. Bostwick, T. Ohta, T. Seyller, K. Horn, K. Nordlund, and R. M. Nieminen, 2003, Phys. Rev. Lett. J. Gonzalez, and E. Rotenberg, 2010, Phys. Rev. Let. 104, 91, 017202. 136803. Lemonik, Y., I. L. Aleiner, C. Toke, and V. I. Fal’ko, 2010, Mele, E. J., 2010, Phys. Rev. B 81, 161405(R). Phys. Rev. B 82, 201408. Meng, Z. Y., T. C. Lang, S. Wessel, F. F. Assaad, and A. Mu- Levy, N., S. A. Burke, K. L. Meaker, M. Panlasigui, A. Zettl, ramatsu, 2010, Nature 464, 847. F. Guinea, A. H. Castro Neto, and M. F. Crommie, 2010, Miao, F., S. Wijeratne, Y. Zhang, U. C. Coskun, W.Bao, and 60

C. N. Lau, 2007, Science 317, 1530. 253503. Mil’shtein, A. I., and V. M. Strakhovenko, 1982, Phys. Lett. Novikov, D., 2007a, Phys. Rev. B 76, 245435. A 90, 447. Novikov, D. S., 2007b, Appl. Phys. Lett. 91, 102102. Min, H., R. Bistritzer, J.-J. Su, and A. H. MacDonald, 2008, Novoselov, K. S., A. K. Geim, S. V. Morozov, D. Jiang, M. I. Phys. Rev. B 78, 121401. Katsnelson, I. V. Grigorieva, S. V. Dubonos, and A. A. Min, H., J. E. Hill, N. A. Sinitsyn, B. R. Sahu, L. Kleinman, Firsov, 2005, Nature 438, 197. and A. H. MacDonald, 2006, Phys. Rev. B 74, 165310. Novoselov, K. S., A. K. Geim, S. V. Morozov, D. Jiang, Min, H., B. Sahu, S. K. Banerjee, and A. H. MacDonald, Y. Zhang, S. V. Dubonos, I. V. Grigorieva, and A. A. 2007, Phys. Rev. B 75, 155115. Firsov, 2004a, Science 306, 666. Mishchenko, E. G., 2007, Phys. Rev. Lett. 98, 216801. Novoselov, K. S., D. Jiang, F. Schedin, T. J. Booth, V. V. Mishchenko, E. G., 2008, Europhys. Lett. 83, 17005. Khotkevich, S. V. Morozov, and A. K. Geim, 2004b, Proc. Moghaddam, A. G., and M. Zareyan, 2006, Phys. Rev. B 74, Nat. Acad. Sci. 102, 10451. 241403(R). Novoselov, K. S., E. McCann, S. V. Morozov, V. I. Fal’ko, Molitor, F., S. Droscher, J. G¨uttinger, A. Jacobsen, M. I. Katsnelson, U. Zeitler, D. Jiang, F. Schedin, and C. Stampfer, T. Ihn, and K. Ensslin, 2009a, Appl. Phys. A. K. Geim, 2006, Nature Physics 2, 177. Lett. 94, 222107. Nozi`eres, P., 1964, The theory of interacting Fermi systems Molitor, F., A. Jacobsen, C. Stampfer, J. G¨uttinger, T. Ihn, (Benjamin, NewYork). and K. Ensslin, 2009b, Phys. Rev. B 79, 075426. Ohldag, H., T. Tyliszczak, R. H¨ohne, D. Spemann, P. Es- Monteverde, M., C. Ojeda-Aristizabal, R. Weil, K. Ben- quinazi, M. Ungureanu, and T. Butz, 2007, Phys. Rev. naceur, M. Ferrier, S. Gu´eron, C. Glattli, H. Bouchiat, Lett. 98, 187204. J. N. Fuchs, and D. L. Maslov, 2010, Phys. Rev. Lett. 104, Ohta, T., A. Bostwick, T. Seyller, K. Horn, and E. Rotenberg, 126801. 2006, Science 313, 951. Moriyama, S., D. Tsuya, E. Watanabe, S. Uji, M. Shimizu, Ojeda-Aristizabal, C., M. Ferrier, S. Gu´eron, and H. Bouch- T. Mori, T. Yamaguchi, and K. Ishibashi, 2009, Nano Lett. iat, 2009, Phys. Rev. B 79, 165436. 9, 2891. Ouyang, M., J.-L. Huang, and C. M. Lieber, 2002, Phys. Rev. Moser, J., and A. Bachtold, 2009, Appl. Phys. Lett. 95, Lett. 88, 066804. 173506. Ozyilmaz,¨ B., P. Jarillo-Herrero, D. Efetov, D. A. Abanin, Moser, J., H. Tao, S. Roche, F. Alzina, C. M. Sotomayor L. S. Levitov, and P. Kim, 2007, Phys. Rev. Lett. 99, Torres, and A. Bachtold, 2010, Phys. Rev. B 81, 205445. 166804. Mott, N. F., 1949, Proc. Phys. Soc. London, Ser. A 62, 416. Paiva, M., R. T. Scalettar, W. Zheng, R. R. P. Singh, and M¨uller, B., and J. Rafelski, 1975, Phys. Rev. Lett. 34, 349. J. Oitmaa, 2005, Phys. Rev. B 72, 085123. M¨uller, M., L. Fritz, and S. Sachdev, 2008, Phys. Rev. B 78, Palacios, J. J., J. Fern´andez-Rossier, and L. Brey, 2008, Phys. 115406. Rev. B 77, 195428. M¨uller, M., J. Schmalian, and L. Fritz, 2009, Phys. Rev. Lett. Park, C.-H., F. Giustino, C. D. Spataru, M. L. Cohen, and 103, 025301. S. G. Louie, 2009, Phys. Rev. Lett. 102, 076803. Nair, R. R., P. Blake, A. N. Grigorenko, K. S. Novoselov, T. J. Pathak, S., V. B. Shenoy, and G. Baskaran, 2010, Phys. Rev. Booth, T. Stauber, N. M. R. Peres, and A. K. Geim, 2008, B 81, 085431. Science 320, 1308. Pereira, V. M., A. H. Castro Neto, H. Y. Liang, and L. Ma- Nakada, K., M. Fujita, G. Dresselhaus, and M. S. Dresselhaus, hadevan, 2010, Phys. Rev. Lett. 105, 156603. 1996, Phys. Rev. B 54, 17954. Pereira, V. M., F. Guinea, J. M. B. Lopes dos Santos, N. M. R. Nakamura, M., E. V. Castro, and B. Dora, 2009, Phys. Rev. Peres, and A. H. Castro Neto, 2006, Phys. Rev. Lett. 96, Lett. 103, 266804. 036801. Nandkishore, R., and L. S. Levitov, 2010a, Phys. Rev. Lett. Pereira, V. M., V. N. Kotov, and A. H. Castro Neto, 2008a, 104, 156803. Phys. Rev. B 78, 085101. Nandkishore, R., and L. S. Levitov, 2010b, eprint Pereira, V. M., J. M. B. Lopes dos Santos, and A. H. Castro arXiv:1002.1966. Neto, 2008b, Phys. Rev. B 77, 115109. Nandkishore, R., and L. S. Levitov, 2010c, Phys. Rev. B 82, Pereira, V. M., J. Nilsson, and A. H. Castro Neto, 2007, Phys. 115431. Rev. Lett. 99, 166802. Newns, D. M., and N. Read, 1987, Adv. Phys. 36, 799. Perelomov, A. M., and V. S. Popov, 1970, Theor. Mat. Phys. Ni, Z. H., L. A. Ponomarenko, R. R. Nair, R. Yang, S. Anis- 4, 664. simova, I. V. Grigorieva, F. Schedin, P. Blake, Z. X. Shen, Peres, N. M. R., 2010, Rev. Mod. Phys. 82, 2673. E. H. Hill, K. S. Novoselov, and A. K. Geim, 2010, Nano Peres, N. M. R., M. A. N. Ara´ujo, and D. Bozi, 2004, Phys. Letters 10, 3868. Rev. B 70, 195122. Niimi, Y., T. Matsui, H. Kambara, K. Tagami, M. Tsukada, Peres, N. M. R., F. Guinea, and A. H. Castro Neto, 2005, and H. Fukuyama, 2005, Appl. Phys. Lett. 241, 43. Phys. Rev. B 72, 174406. Nilsson, J., A. H. Castro Neto, F. Guinea, and N. M. R. Peres, Pisani, L., J. A. Chan, B. Montanari, and N. M. Harrison, 2008, Phys. Rev. B 78, 045405. 2007, Phys. Rev. B 75, 064418. Nilsson, J., A. H. Castro Neto, N. M. R. Peres, and F. Guinea, Pisarski, R. D., 1984, Phys. Rev. D 29, 2423. 2006, Phys. Rev. B 73, 214418. Polini, M., R. Asgari, Y. Barlas, T. Pereg-Barnea, and A. H. Nomura, K., and A. MacDonald, 2007, Phys. Rev. Lett. 98, MacDonald, 2007, Solid State Commun. 143, 58. 076602. Polini, M., R. Asgari, G. Borghi, Y. Barlas, T. Pereg-Barnea, Nomura, K., and A. H. MacDonald, 2006, Phys. Rev. Lett. and A. H. MacDonald, 2008a, Phys. Rev. B 77, 081411(R). 96, 256602. Polini, M., A. Tomadin, R. Asgari, and A. H. MacDonald, Nouchi, R., and K. Tanigaki, 2010, Appl. Phys. Lett. 96, 2008b, Phys. Rev. B 78, 115426. 61

Polkovnikov, A., 2002, Phys. Rev. B 65, 064503. Rev. B 82, 195428. Polkovnikov, A., M. Vojta, and S. Sachdev, 2001, Phys. Rev. Sepioni, M., R. R. Nair, S. Rablen, J. Narayanan, F. Tuna, Lett. 86, 296. R. Winpenny, A. K. Geim, and I. V. Grigorieva, 2010, Phys. Ponomarenko, L. A., F. Schedin, M. I. Katsnelson, R. Yang, Rev. Lett. 105, 207205. E. W. Hill, K. S. Novoselov, and A. K. Geim, 2008, Science Seradjeh, B., 2008, Nucl. Phys. B 805, 182. 320, 356. Seradjeh, B., and M. Franz, 2008, Phys. Rev. Lett. 101, Ponomarenko, L. A., R. Yang, T. M. Mohiuddin, S. M. Moro- 146401. zov, A. A. Zhukov, F. Schedin, E. W. Hill, K. S. Novoselov, Seradjeh, B., H. Weber, and M. Franz, 2008, Phys. Rev. Lett. M. I. Katsnelson, and A. K. Geim, 2009, Phys. Rev. Lett. 101, 246404. 102, 206603. Sheehy, D. E., and J. Schmalian, 2007, Phys. Rev. Lett. 99, Popov, V. S., 1971a, Sov. J. Nucl. Phys. 12, 235. 226803. Popov, V. S., 1971b, Sov. Phys. JETP 32, 526. Sheehy, D. E., and J. Schmalian, 2009, Phys. Rev. B 80, Principi, A., M. Polini, G. Vignale, and M. I. Katsnelson, 193411. 2010, Phys. Rev. Lett. 104, 225503. Shibata, N., and K. Nomura, 2008, Phys. Rev. B 77, 235426. Pustilnik, M., and L. I. Glazman, 2001, Phys. Rev. Lett. 87, Shimshoni, E., H. A. Fertig, and G. V. Pai, 2009, Phys. Rev. 216601. Lett. 102, 206408. Qi, X.-L., and S.-C. Zhang, 2011, Rev. Mod. Phys. 83, 1057. Shulke, W., U. Bonse, H. Nagasawa, A. Kaprolat, and Raghu, S., X.-L. Qi, C. Honerkamp, and S.-C. Zhang, 2008, A. Berthold, 1988, Phys. Rev. B 38, 2112. Phys. Rev. Lett. 100, 156401. Shung, K. W.-K., 1986a, Phys. Rev. B 34, 979. Rainis, D., F. Taddei, F. Dolcini, M. Polini, and R. Fazio, Shung, K. W. K., 1986b, Phys. Rev. B 34, 1264. 2009, Phys. Rev. B 79, 115131. Shytov, A., M. Katsnelson, and L. Levitov, 2007a, Phys. Rev. Ramos, M. A., J. Barzola-Quiquia, P. Esquinazi, A. Mu˜noz- Lett. 99, 246802. Martin, A. Climent-Font, and M. Garc´ıa-Hern´andez, 2010, Shytov, A., M. Katsnelson, and L. Levitov, 2007b, Phys. Rev. Phys. Rev. B 81, 214404. Lett. 99, 236801. Rappoport, T. G., B. Uchoa, and A. H. Castro Neto, 2009, Siegel, D. A., C.-H. Park, C. Hwang, J. Deslippe, A. V. Fe- Phys. Rev. B 80, 245408. dorov, S. G. Louie, and A. Lanzara, 2011, Proceedings of Read, N., and D. M. Newns, 1983, J. Phys. C 16, 3273. the National Academy of Sciences 108, 11365. Reed, J. P., B. Uchoa, Y. I. Joe, Y. Gan, D. Casa, E. Fradkin, Skrypnyk, Y. V., and V. M. Loktev, 2006, Phys. Rev. B 73, and P. Abbamonte, 2010, Science 330, 805. 241402(R). Ritter, K. A., and J. W. Lyding, 2009, Nature Materials 8, Sols, F., F. Guinea, and A. H. Castro Neto, 2007, Phys. Rev. 235. Lett. 99, 166803. Riu, S., and Y. Hatsugai, 2001, Phys. Rev. B 65, 033301. Son, D. T., 2007, Phys. Rev. B 75, 235423. Rold´an, R., M. P. L´opez-Sancho, and F. Guinea, 2008, Phys. Son, Y.-W., M. L. Cohen, and S. G. Louie, 2006, Nature 444, Rev. B 77, 115410. 347. Romanovsky, I., C. Yannouleas, and U. Landman, 2009, Phys. Sorella, S., and E. Tosatti, 1992, Europhys. Lett. 19, 699. Rev. B 79, 075311. Soriano, D., and J. Fern´andez-Rossier, 2010, Phys. Rev. B Roy, B., and I. F. Herbut, 2010, Phys. Rev. B 82, 035429. 82, 161302. Ryu, S., C. Mudry, C.-Y. Hou, and C. Chamon, 2009, Phys. Sprinkle, M., D. Siegel, Y. Hu, J. Hicks, A. Tejeda, A. Taleb- Rev. B 80, 205319. Ibrahimi, P. Le F`evre, F. Bertran, S. Vizzini, H. Enriquez, Sabio, J., J. Nilsson, and A. H. Castro Neto, 2008, Phys. Rev. S. Chiang, P. Soukiassian, et al., 2009, Phys. Rev. Lett. B 78, 075410. 103, 226803. Sabio, J., F. Sols, and F. Guinea, 2010a, Phys. Rev. B 82, Stampfer, C., J. G¨uttinger, S. Hellmueller, F. Molitor, K. En- 121413. sslin, and T. Ihn, 2009, Phys. Rev. Lett. 102, 056403. Sabio, J., F. Sols, and F. Guinea, 2010b, Phys. Rev. B 81, Stampfer, C., E. Schurtenberger, F. Molitor, J. G¨uttinger, 045428. T. Ihn, and K. Ensslin, 2008, Nano Lett. 8, 2378. Sachdev, S., 1999, Quantum Phase Transitions (Cambridge Stauber, T., F. Guinea, and M. A. H. Vozmediano, 2005, University Press, Cambridge, UK). Phys. Rev. B 71, 041406. Saha, K., I. Paul, and K. Sengupta, 2010, Phys. Rev. B 81, Stein, S. E., and R. L. Brown, 1987, J. Am. Chem. Soc. 109, 165446. 3721. Sahebsara, P., and D. S´en´echal, 2009, eprint arXiv:0908.0474. Taft, E. A., and H. R. Philipp, 1965, Phys. Rev. 138, A197. Sahu, B., H. Min, A. H. MacDonald, and S. K. Banerjee, Takei, S., and Y. B. Kim, 2008, Phys. Rev. B 78, 165401. 2008, Phys. Rev. B 78, 075404. Tan, Y. W., Y. Zhang, K. Bolotin, Y. Zhao, S. Adam, Saremi, S., 2007, Phys. Rev. B 76, 184430. E. Hwang, S. Das Sarma, H. Stormer, and P. Kim, 2007, Sasakia, K., J. Jiang, R. Saito, S. Onari, and Y. Tanaka, 2007, Phys. Rev. Lett. 99, 246803. J. Phys. Soc. Jpn. 76, 033702. Tediosi, R., N. P. Armitage, E. Giannini, and D. van der Schnez, S., F. Molitor, C. Stampfer, J. Guettinger, I. Sho- Marel, 2007, Phys. Rev. Lett. 99, 016406. rubalko, T. Ihn, and K. Ensslin, 2009, Appl. Phys. Lett. Terekhov, I. S., A. I. Milstein, V. N. Kotov, and O. P. 94, 012107. Sushkov, 2008, Phys. Rev. Lett. 100, 076803. Schrieffer, J. R., and P. A. Wolff, 1966, Phys. Rev. 149, 491. Thomas, L. H., 1927, Mat. Proc. Cambridge Phil. Soc. 23, Schwinger, J., 1951, Phys. Rev. 82, 664. 542. Semenoff, G. W., 1984, Phys. Rev. Lett. 53, 2449. Tikhonenko, F. V., A. A. Kozikov, A. K. Savchenko, and R. V. Sengupta, K., and G. Baskaran, 2008, Phys. Rev. B 77, Gorbachev, 2009, Phys. Rev. Lett. 103, 226801. 045417. Tinkham, M., 1996, Introduction to Superconductivity Sensarma, R., E. H. Hwang, and S. D. Sarma, 2010, Phys. (McGraw-Hill, New York). 62

Titov, M., and C. W. J. Beenakker, 2006, Phys. Rev. B 74, snelson, H. Manoharan, and A. V. Balatsky, 2010b, Phys. 041401(R). Rev. B 81, 085413. Titov, M., A. Ossipov, and C. W. J. Beenakker, 2007, Phys. Wehling, T. O., K. S. Novoselov, S. V. Morozov, E. E. Vdovin, Rev. B 75, 0415417. M. I. Katsnelson, A. K. Geim, and A. I. Lichtenstein, Todd, K., H.-T. Chou, S. Amasha, and D. Goldhaber-Gordon, 2008b, Nano Lett. 8, 173. 2009, Nano Lett. 9, 416. Wehling, T. O., S. Yuan, A. I. Lichtenstein, A. K. Geim, and T¨oke, C., and J. K. Jain, 2007, Phys. Rev. B 75, 245440. M. I. Katsnelson, 2010c, Phys. Rev. Lett. 105, 056802. T¨oke, C., P. E. Lammert, V. H. Crespi, and J. K. Jain, 2006, Weisskopf, V. F., 1939, Phys. Rev. 56, 72. Phys. Rev. B 74, 235417. Weller, T. E., M. Ellerby, S. S. Saxena, R. P. Smith, and N. T. Trevisanutto, P. E., C. Giorgetti, L. Reining, M. Ladisa, and Skipper, 2005, Nature Physics 1, 39. V. Olevano, 2008, Phys. Rev. Lett. 101, 226405. Williams, J. R., L. DiCarlo, and C. M. Marcus, 2007, Science Tyutyulkov, N., G. Madjarova, F. Dietz, and K. Mullen, 1998, 317, 638. J. Phys. Chem. B 102, 10183. Wimmer, M., A. Akhmerov, and F. Guinea, 2010, Phys. Rev. Uchoa, B., G. G. Cabrera, and A. H. Castro Neto, 2005, Phys. B 82, 045409. Rev. B 71, 184509. Withoff, D., and E. Fradkin, 1990, Phys. Rev. Lett. 64, 1835. Uchoa, B., and A. H. Castro Neto, 2007, Phys. Rev. Lett. 98, Wojs, A., G. Moller, and N. R. Cooper, 2011, Acta Phys. 146801. Polon. A119, 592. Uchoa, B., and A. H. Castro Neto, 2009, Phys. Rev. Lett. Wunsch, B., T. Stauber, and F. Guinea, 2008a, Phys. Rev. B 102, 109701. 77, 035316. Uchoa, B., V. N. Kotov, N. M. R. Peres, and A. H. Castro Wunsch, B., T. Stauber, F. Sols, and F. Guinea, 2007, New Neto, 2008a, Phys. Rev. Lett. 101, 026805. J. Phys. 8, 318. Uchoa, B., C.-Y. Lin, and A. H. Castro Neto, 2008b, Phys. Wunsch, B., T. Stauber, F. Sols, and F. Guinea, 2008b, Phys. Rev. B 77, 035420. Rev. Lett. 101, 036803. Uchoa, B., T. G. Rappoport, and A. H. Castro Neto, 2011, Yang, K., S. Das Sarma, and A. H. MacDonald, 2006, Phys. Phys. Rev. Lett. 106, 016801. Rev. B 74, 075423. Uchoa, B., L. Yang, S. W. Tsai, N. M. R. Peres, and A. H. Yao, Y., F. Ye, X.-L. Qi, S.-C. Zhang, and Z. Fang, 2007, Castro Neto, 2009, Phys. Rev. Lett. 103, 206804. Phys. Rev. B 75, 041401(R). Ugeda, M. M., I. Brihuega, F. Guinea, and J. M. G´omez- Yazyev, O. V., 2008, Phys. Rev. Lett. 101, 037203. Rodr´ıguez, 2010, Phys. Rev. Lett. 104, 096804. Yazyev, O. V., and L. Helm, 2007, Phys. Rev. B 75, 125408. Vafek, O., 2007, Phys. Rev. Lett. 98, 216401. Ye, J., and S. Sachdev, 1998, Phys. Rev. Lett. 80, 5409. Vafek, O., 2010, Phys. Rev. B 82, 205106. Zeldovich, Y. B., and V. S. Popov, 1972, Sov. Phys. Uspekhi Vafek, O., and M. J. Case, 2008, Phys. Rev. B 77, 033410. 14, 673. Vafek, O., and K. Yang, 2010, Phys. Rev. B 81, 041401(R). Zhang, F., H. Min, M. Polini, and A. H. MacDonald, 2010, Valenzuela, B., and M. A. H. Vozmediano, 2008, New J. Phys. Phys. Rev. B 81, 041402. 10, 113009. Zhang, G.-M., H. Hu, and L. Yu, 2001, Phys. Rev. Lett. 86, Venezuela, P., R. B. Muniz, A. T. Costa, D. M. Edwards, 704. S. R. Power, and M. S. Ferreira, 2009, Phys. Rev. B. 80, Zhang, Y., Z. Jiang, J. P. Small, M. S. Purewal, Y.-W. Tan, 241413(R). M. Fazlollahi, J. D. Chudow, H. L. Stormer, J. A. Jaszczak, Visscher, P. B., and L. M. Falikov, 1970, Phys. Rev. B 3, and P. Kim, 2006, Phys. Rev. Lett. 96, 136806. 2541. Zhang, Y., Y.-W. Tan, H. L. Stormer, and P. Kim, 2005, Vojta, M., 2001, Phys. Rev. Lett. 87, 097202. Nature 438, 201. Vojta, M., and R. Bulla, 2001, Phys. Rev. B 65, 014511. Zhang, Y., T.-T. Tang, C. Girit, Z. Hao, M. Martin, A. Zettl, Vojta, M., and L. Fritz, 2004, Phys. Rev. B 70, 094502. M. F. Crommie, Y. R. Shen, and F. Wang, 2009, Nature Vojta, M., L. Fritz, and R. Bulla, 2010, Europhys. Lett. 90, 459, 820. 27006. Zhao, E., and A. Paramekanti, 2006, Phys. Rev. Lett. 97, Vozmediano, M. A. H., M. P. L´opez-Sancho, T. Stauber, and 230404. F. Guinea, 2005, Phys. Rev. B 72, 155121. Zhou, J., Q. Wang, Q. Sun, X. S. Chen, Y. Kawazoe, and Wallace, P. R., 1947, Phys. Rev. 71, 622. P. Jena, 2009, Nano Lett. 9, 3867. Wang, J., H. Fertig, and G. Murthy, 2010, Phys. Rev. Lett. Zhou, S. Y., G. H. Gweon, A. V. Fedorov, P. N. First, W. A. 104, 186401. de Heer, D. H. Lee, F. Guinea, A. H. Castro Neto, and Wang, J., A. Iyengar, H. A. Fertig, and L. Brey, 2008, Phys. A. Lanzara, 2007, Nature Materials 6, 770. Rev. B 78, 165416. Zhou, S. Y., D. A. Siegel, A. V. Fedorov, and A. Lanzara, Wehling, T. O., A. V. Balatsky, M. I. Katsnelson, A. I. Licht- 2008, Phys. Rev. B 78, 193404. enstein, and A. Rosch, 2010a, Phys. Rev. B 81, 115427. Zhu, J.-X., and C. S. Ting, 2000, Phys. Rev. B 63, 020506. Wehling, T. O., A. V. Balatsky, M. I. Katsnelson, A. I. Licht- Zhu, W., Z. Wang, Q. Shi, K. Y. Szeto, J. Chen, and J. G. enstein, K. Scharnberg, and R. Wiesendanger, 2007, Phys. Hou, 2009, Phys. Rev. B 79, 155430. Rev. B 75, 125425. Zhu, Z.-G., K.-H. Ding, and J. Berakdar, 2010, Europhys. Wehling, T. O., A. V. Balatsky, A. M. Tsvelik, M. I. Kat- Lett. 90, 67001. snelson, and A. I. Lichtenstein, 2008a, Europhys. Lett. 84, Zhuang, H.-B., Q.-F. Sun, and X. C. Xie, 2009, Europhys. 17003. Lett. 86, 58004. Wehling, T. O., H. P. Dahal, A. I. Lichtenstein, M. I. Kat-