2

arXiv:1004.3396v2 [cond-mat.
mes-hall] 25 Jun 2010
Electronic Properties of Graphene in a Strong Magnetic Field

Mark Oliver GOERBIG Habilitation ` diriger les recherches a Laboratoire de Physique des Solides, CNRS UMR 8502 Universit Paris Sud, Bt. 510, F-91405 ORSAY cedex e a
ii
iii
Preface
This Habilitation Thesis covers the research on the electronic properties of graphene that I have been involved in during the last four years, since the beginning of my CNRS researcher position at the Laboratoire de Physiques des Solides, Orsay. During this time, I have proted from an extremely stimulating working atmosphere in this laboratory. Research is teamwork, often international, that benets from the intellectual exchange between the dierent actors rather than from the sterile individual competition, which is unfortunately favoured at the moment for ideological reasons by numerous irresponsible politicians and even some of our research administrators. It is therefore more than natural that the research presented in the present manuscript has been performed in diverse collaborations, and I would like to emphasise two of them regrouping members from the younger generation of researchers. One fruitful in-house collaboration that I need to mention rst was that with my oce mate Jean-Nol Fuchs on the perturbative treatment of interactions e in the integer quantum Hall regime in graphene. This collaboration, that is reected mainly in Chaps. 3 and 4, was joined from October 2007 to October 2009 by our post-doc Rafael Roldn to whom we owe much insight into the physics a of collective excitations due to electronic interactions (Sec. 3.2). It is noteworth that the collaboration with Jean-Nol was not only concerned with the actual e research in the young eld of graphene, but also with the transmission of the thus obtained knowledge on several levels, from the organisation of a graduateschool class on graphene in 2007 (Ecole doctorale de la Rgion Parisienne) to the e writing of large-audience articles. This more pedagogical approach to graphene is reected in the introductory parts of the present manuscript (Chap. 1). A second major collaboration with my CNRS colleague Nicolas Regnault from the Laboratoire Pierre Aigrain at Ecole Normale Suprieure was cone cerned with the fractional quantum Hall eect in this novel carbon compound (Sec. 5.3), in the framework of a general study of multi-component quantum Hall systems. This research interest stems indeed from my PhD studies, under the joint direction of Cristiane Morais Smith (now Utrecht University, The Netherlands) and Pascal Lederer (Laboratoire de Physique des Solides). The complementarity in our approaches to the physics of the fractional quantum Hall eect, state-of-the-art numerical techniques for Nicolas and analytical ones for myself, made it natural to join our research activities even if we are located in dierent CNRS laboratories of the Paris Region. This collaboration has been joined by two extremely promising students, during a 5-month project in 2007 by Rafal de Gail who is since October 2009 PhD student at the Laboratoire de e Physique des Solides, and since November 2007 by Zlatko Papi within a joint c PhD project (with Milica Milovanovi, University of Belgrade, Serbia) who is c about to terminate his very successful studies (expected defense in autumn 2009). Even if I emphasise the two above-mentioned collaborations, others need to be mentioned as well. Indeed, the collaboration with Jean-Nol was not an exe clusive one inside the Laboratoire de Physique des Solides, but it was embedded
iv in a larger framework centered around a Journal Club on graphene that started in spring 2006 and that regrouped other members of the laboratorys theory group, such as Gilles Montambaux, Frdric Pichon, Pascal Lederer, and later e e e Cristina Bena and Pascal Simon, and of the experimental mesoscopic-physics group (Hl`ne Bouchiat, Alexei Chepelianskii, Richard Deblock, Meydi Feree rier, Sophie Guron, Miguel Monteverde, Claudia Ojeda Aristizabal), as well as e members (Denis Ullmo and Pierre Carmier) of the neighbouring Laboratoire de Physique Thorique et Mod`les Statistiques. Namely the interaction with Gilles e e and Frdric led to a closer collaboration on Dirac-point motion that is briey e e mentioned in Secs. 1.5.1 and 2.4. This collaboration has been joined recently by another promising student, Guangquan Wang, who prepares his PhD studies in a joint project between the University Paris-Sud and the National University of Singapore. Furthermore, the research on electron-phonon coupling in graphene in a magnetic eld was performed in collaboration not only with Jean-Nol, e but also with Vladimir Falko (University of Lancaster, Great Britain; associate researcher at the Laboratoire de Physique des Solides in 2007) and his PhD student Kostyantyn Kechedzhi. Similarly, the interest I shared with Jean-Nol e in the limits of the Dirac equation in the description of high-energy electrons in graphene found another manifestation in an experiment-theory collaboration with the Grenoble High Field Laboratory (namely with Clment Faugeras, e Paulina Plochocka, Marek Potemski, and Claire Berger), where Landau-level spectroscopy is performed on epitaxial graphene (Sec. 2.2). Another collaboration that was extremely important for my research on graphene was the initial one with Roderich Moessner (now Max-Planck-Insititut fr Physik komplexer Systeme, Dresden, Germany) and Beno Douot (Labou t c ratoire de Physique Thorique et Hautes Energies, Jussieu, Paris) on the basic e interaction model and its symmetries (Secs. 3.1 and 5.1). This collaboration was later joined by one of my former PhD advisors (Pascal Lederer) and evolved into a study of the SU(4) quantum Hall ferromagnet parts of which are presented in Sec. 5.2. The aim of the present manuscript is to embed the above-mentioned research in the framework of our present understanding of ultra-relativistic electrons in a mono-layer graphene sheet exposed to a strong magnetic eld. It is therefore not meant to be an extensive presentation of the research I was involved in since the end of my PhD studies. This presentation may be found in a separate manuscript (Notice de travaux, in French) that covers also some other elds of research and that I have decided to join to the present Habilitation Thesis. I would furthermore, in addition to my above-mentioned collaborators, thank Antonio Castro Neto, Christian Glattli and Paco Guinea for their referee reports on my Habilitation Thesis. The comments of their reports are of great value. Many thanks also to Hl`ne Bouchiat, Beno Douot and Anuradha Jaganee t c nathan for having accepted to be members of the jury.
To Anna and Juliette in the hope that they will once understand what their Dad is working on To Mathilde for all her love and for a shared passion also in research
vi
Contents
1 Introduction to Graphene 1.1 The Carbon Atom and its Hybridisations . . . 1.2 Brief History of Carbon-Based Materials . . . . 1.2.1 Fabrication of graphene . . . . . . . . . 1.3 Crystal Structure of Graphene . . . . . . . . . 1.4 Electronic Band Structure of Graphene . . . . 1.4.1 Tight-binding model for electrons on the 1.4.2 Continuum limit . . . . . . . . . . . . . 1.5 Deformed Graphene . . . . . . . . . . . . . . . 1.5.1 Dirac point motion . . . . . . . . . . . . 1.5.2 Tilted Dirac cones . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . honeycomb lattice . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1 2 4 5 7 9 9 16 22 24 26 29 29 30 31 38 41 42 43 45 47 49 51 54 54 58 60
2 Dirac Equation in a Magnetic Field and the Relativistic Quantum Hall Eect 2.1 Massless 2D Fermions in a Strong Magnetic Field . . . . . . . . . 2.1.1 Quantum-mechanical treatment . . . . . . . . . . . . . . . 2.1.2 Relativistic Landau levels . . . . . . . . . . . . . . . . . . 2.2 Limits of the Dirac Equation in the Description of Graphene Landau Levels . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2.3 Landau Level Spectrum in the Presence of an Inplane Electric Field 2.4 Landau Levels in Deformed Graphene . . . . . . . . . . . . . . . 2.4.1 The generalised Weyl Hamiltonian in a magnetic eld . . 2.4.2 LL spectrum in the vicinity of the topological phase transition . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3 Electronic Interactions in Graphene Integer Quantum Hall Regime 3.1 Decomposition of the Coulomb interaction in the Two-Spinor Basis 3.1.1 SU(2) valley symmetry . . . . . . . . . . . . . . . . . . . . 3.2 Particle-Hole Excitation Spectrum . . . . . . . . . . . . . . . . . 3.2.1 Graphene particle-hole excitation spectrum at B = 0 . . . 3.2.2 Polarisability for B = 0 . . . . . . . . . . . . . . . . . . . 3.2.3 Electron-electron interactions in the random-phase approximation . . . . . . . . . . . . . . . . . . . . . . . . . . vii
viii 3.2.4 Dielectric function and static screening . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 62 67 68 69 70 71 72 73
4 Magneto-Phonon Resonance in Graphene 4.1 Electron-Phonon Coupling . . . . . . . . . . . . . . . . . . 4.1.1 Coupling Hamiltonian . . . . . . . . . . . . . . . . 4.1.2 Hamiltonian in terms of magneto-exciton operators 4.2 Phonon Renormalisation and Raman Spectroscopy . . . . 4.2.1 Non-resonant coupling and Kohn anomaly . . . . . 4.2.2 Resonant coupling . . . . . . . . . . . . . . . . . .
5 Electronic Correlations in Partially Filled Landau Levels 5.1 Electrons in a Single Relativistic Landau Level . . . . . . . . 5.1.1 SU(4)-symmetric model . . . . . . . . . . . . . . . . . 5.1.2 Symmetry-breaking long-range terms . . . . . . . . . . 5.1.3 Qualitative expectations for correlated electron phases 5.2 SU(4) Quantum Hall Ferromagnetism in Graphene . . . . . . 5.2.1 Ferromagnetic ground state and Goldstone modes . . 5.2.2 Skyrmions and entanglement . . . . . . . . . . . . . . 5.2.3 SU(4)-symmetry-breaking terms . . . . . . . . . . . . 5.2.4 Comparison with magnetic catalysis . . . . . . . . . . 5.2.5 The quantum Hall eect at = 1 and = 0 . . . . . 5.3 Fractional Quantum Hall Eect in Graphene . . . . . . . . . 5.3.1 Generalised Halperin wave functions . . . . . . . . . . 5.3.2 Polarised vs. unpolarised states . . . . . . . . . . . . . 6 Conclusions and Outlook Appendix Matrix Elements of the Density Operators Bibliography
77 . 78 . 79 . 81 . 82 . 84 . 84 . 86 . 90 . 92 . 94 . 99 . 99 . 101 105 108 111
Chapter 1
Introduction to Graphene
The experimental and theoretical study of graphene, two-dimensional (2D) graphite, has become a major issue of modern condensed matter research. A milestone was the experimental evidence of an unusual quantum Hall eect reported in September 2005 by two dierent groups, the Manchester group led by Andre Geim and a Columbia-Princeton collaboration led by Philip Kim and Horst Stormer [1, 2]. The reasons for this enormous scientic interest are manyfold, but one may highlight some major motivations (see also the recent reviews [3] and [4]). First, one may emphasise its possible technological potential. One of the rst publications on graphene in 2004 by the Geim group reported indeed an electric eld eect in graphene, i.e. the possibility to control the carrier density in the graphene sheet by simple application of a gate voltage [5]. This eect is a fundamental element for the design of electronic devices. Todays silicon-based electronics reaches its limits in miniaturisation, which is on the order of 50 nm for an electric channel, whereas it has been shown that a narrow graphene strip with a width of only a few nanometers may be used as a transistor [6], i.e. as the basic electronics component. One may therefore hope to improve the miniaturisation by one order of magnitude when using graphene-based electronics. Apart from these promising technological applications, two major motivations for fundamental research may be emphasised. Graphene is the rst truely 2D crystal ever observed in nature and possess remarkable mechanical properties. Furthermore, electrons in graphene show relativistic behaviour, and the system is therefore an ideal candidate for the test of quantum-eld theoretical models which have been developed in high-energy physics. Most promenently, electrons in graphene may be viewed as massless charged fermions living in 2D space, particles one usually does not encounter in our three-dimensional world. Indeed, all massless elementary particles happen to be electrically neutral, such as photons or neutrinos.1 Graphene is therefore an exciting bridge between
1 The neutrino example is only partially correct. The observed oscillation between dierent neutrino avors ( ) requires indeed a tiny non-zero mass [7].
ground state
excited state ( ~ 4 eV) ~ 4 eV 2s
Energy
2px 2py 2s
2pz
2px 2py
2pz
1s
1s
Figure 1.1:
state (right).
Electronic congurations for carbon in the ground state (left) and in the excited
condensed-matter and high-energy physics, and the research on its electronic properties unites scientists with various thematic backgrounds.
1.1
The Carbon Atom and its Hybridisations
In order to understand the crystallographic structure of graphene and carbonbased materials in general, it is useful to review the basic chemical bonding properties of carbon atoms. The carbon atom possesses 6 electrons, which, in the atomic ground state, are in the conguration 1s2 2s2 2p2 , i.e. 2 electrons ll the inner shell 1s, which is close to the nucleus and which is irrelevant for chemical reactions, whereas 4 electrons occupy the outer shell of 2s and 2p orbitals. Because the 2p orbitals (2px , 2py , and 2pz ) are roughly 4 eV higher in energy than the 2s orbital, it is energetically favourable to put 2 electrons in the 2s orbital and only 2 of them in the 2p orbitals (Fig 1.1). It turns out, however, that in the presence of other atoms, such as e.g. H, O, or other C atoms, it is favourable to excite one electron from the 2s to the third 2p orbital, in order to form covalent bonds with the other atoms. In the excited state, we therefore have four equivalent quantum-mechanical states, |2s , |2px , |2py , and |2pz . A quantum-mechanical superposition of the state |2s with n |2pj states is called spn hybridisation. The sp1 hybridisation plays, e.g., an important role in the context of organic chemistry (such as the formation of acetylene) and the sp3 hybridisation gives rise to the formation of diamonds, a particular 3D form of carbon. Here, however, we are interested in the planar sp2 hybridisation, which is the basic ingredient for the graphitic allotopes which are introduced in Sec. 1.2. As shown in Fig. 1.2, the three sp2 -hybridised orbitals are oriented in the xy-plane and have mutual 120 angles. The remaining unhybridised 2pz orbital is perpendicular to the plane. A prominent chemical example for such hybridisation is the benzene molecule
The Carbon Atom and its Hybridisations

(b) (a)
11111111111 00000000000 11111111111 00000000000 111111111111 000000000000 11111111111 00000000000 111111111111 000000000000 11111111111 00000000000 111111111111 000000000000 11111111111 00000000000 111111111111 000000000000 11111111111 00000000000 111111111111 000000000000 11111111111 00000000000 111111111111 000000000000 11111111111 00000000000 111111111111 000000000000 11111111111 00000000000 111111111111 000000000000 11111111111 00000000000 111111111111 000000000000 11111111111 00000000000 111111111111 000000000000 11111111111 00000000000 111111111111 000000000000 11111111111 00000000000 111111111111 000000000000 11111111111 00000000000 111111111111 000000000000 11111111111 00000000000 111111111111 000000000000 11111111111 00000000000 111111111111 000000000000 11111111111 00000000000 111111111111 000000000000 11111111111 00000000000 111111111111 000000000000 11111 00000 11111 00000 11111 00000 0 11111 00000 11111 00000 11111 00000 11111 00000 11111 00000 11111 00000 11111 00000 11111 00000 11111 00000 11111 00000 11111 00000 11111 00000 11111 00000 11111 00000 11111 00000 11111 00000 11111 00000 11111 00000 11111 00000 11111 00000 11111 00000 11111 00000 11111 00000
3
H
H C
C C
120
C H C
C H
(c)
(d)
Figure 1.2: (a) Schematic view of the sp2 hybridisation. The orbitals form angles of 120o . (b) Benzene molecule (C6 H6 ). The 6 carbon atoms are situated at the corners of a hexagon and form covalent bonds with the H atoms. In addition to the 6 covalent bonds between the C atoms, there are three bonds indicated by the doubled line. (c) The quantum-mechanical ground state of the benzene ring is a superposition of the two congurations which dier by the position of the bonds. The electrons are, thus, delocalised over the ring. (d) Graphene may be viewed as a tiling of benzene hexagons, where the H atoms are replaced by C atoms of neighbouring hexagons and where the electrons are delocalised over the whole structure.
the chemical structure of which has been analysed by August Kekul in 1865 e [8, 9]. The molecule consists of a hexagon with carbon atoms at the corners linked by bonds [Fig. 1.2 (b)]. Each carbon atom has, furthermore, a covalent bond with one of the hydrogen atoms which stick out from the hexagon in a star-like manner. In addition to the six bonds, the remaining 2pz orbitals form three bonds, and the resulting double bonds alternate with single bonds around the hexagon. Because a double bond is stronger than a single bond, one may expect that the hexagon is not perfect. A double bond (C=C) yields indeed a carbon-carbon distance of 0.135 nm, whereas it is 0.147 nm for a single bond (CC). However, the measured carbon-carbon distance in benzene is 0.142 nm for all bonds, which is roughly the average length of a single and a double bond. This equivalence of all bonds in benzene was explained by Linus Pauling in 1931 within a quantum-mechanical treatment of the benzene ring [10]. The ground state is indeed a quantum-mechanical superposition of the two possible congurations for the double bonds, as shown schematically in Fig. 1.2 (c). These chemical considerations indicate the way towards carbon-based condensed matter physics any graphitic compound has indeed a sheet of graphene as its basic constituent. Such a graphene sheet may be viewed simply as a tiling of benzene hexagons, where the hydrogen are replaced by carbon atoms to form a neighbouring carbon hexagon [Fig. 1.2 (d)]. However, graphene has remained the basic constituent of graphitic systems during a long time only on the the-
(a)
(b)
(c)
(d)
Figure 1.3:
Graphitic allotopes (a) Piece of natural graphite. (b) Layered structure of graphite (stacking of graphene layers). (c) 0D allotope: C60 molecule. (d) 1D allotope: single-wall carbon nanotube.
oretical level. From an experimental point of view, graphene is the youngest allotope and accessible to physical measurements only since 2004.
1.2
Brief History of Carbon-Based Materials
Historically, the longest known allotope is 3D graphite [Fig. 1.3 (a)]. Graphite was discovered in a mine near Borrowdale in Cumbria, England in the 16th century, and its use for marking and graphical purposes was almost immediately noticed. Indeed, the nearby farmers used graphite blocks from the mine for marking their sheep. Due to its softness and dark color, graphite was considered during a long time as some particular type of lead. The full name lead pencil still witnesses this historical error.2 That graphite was formed from carbon atoms was discovered by the Swedish-German pharmacist Carl Wilhelm Scheele in the middle of the 18th century. But it was the German chemist Abraham Gottlob Werner in 1789 who coined the material by its current name graphite, thus emphasising its main use for graphical purposes.3 Graphite may be viewed as a stacking of graphene sheets [Fig. 1.3 (b)] that stick together due to the van der Waals interaction, which is much weaker
2 The graphitic core of a pencil is still called lead, and the German name for pencil is Bleistift, Blei being the German name for lead. 3 The term is derived from the Greek word graphein (to draw, to write).
Brief History of Carbon-Based Materials
than the in-plane covalent bonds. This physical property explains the graphic utility of the material: when one writes with a piece of graphite, i.e. when it is scratched over a suciently rough surface, such as a piece of paper, the van der Waals force is overcome and thin stacks of graphene sheets are exfoliated from bulk graphite and now stick to the surface. The 0D graphitic allotope (fullerenes) has been discovered in 1985 by Robert Curl, Harold Kroto, and Richard Smalley [11]. Its most prominent representative is the C60 molecule which has the form of a football. It consists of a graphene sheet, where some hexagons are replaced by pentagons, which cause a crumbling of the sheet and the nal formation of a graphene sphere [Fig. 1.3 (c)]. Its existance had been predicted before, in 1970, by the Japanese theoretician Eiji Ozawa [12]. Carbon nanotubes, the 1D allotope, may be viewed as graphene sheets which are rolled up [Fig. 1.3 (d)], with a diameter of several nano-meters. One distinguishes single-wall from multi-wall nanotubes, according to the number of rolled up graphene sheets. The discovery of carbon nanotubes is most often attributed to Sumio Iijima and his 1991 publication in Nature [13]. Recently, doubts about this attribution have been evoked because it seems that carbon nanotubes had a longer history in the community of material scientists [14]. Indeed, a publication by the Soviet scientists Radushkevich and Lukyanovich in 1952 contained a transmission electron microscope image showing carbon nanotubes [15]. It is, however, the merit of the 1991 Iijima paper to have attracted the interest of the condensed matter physics community to carbon nanotubes and to have initiated an intense research on this compound, also in the prospect of nanotechnological applications.
1.2.1
Fabrication of graphene
Graphene research is intimitely related to the fabrication of single-layer graphite samples. We review, in this section, briey the two most popular fabrication techniques, exfoliation and thermal surface decomposition of SiC crystals. Exfoliated graphene The exfoliation technique in the fabrication of graphene [16] makes use of the above-mentioned weak van der Waals force between the graphene sheets in graphite as compared to the strong covalent in-plane carbon-carbon bonds. The technique consists of peeling thin layers o the graphite crystal that are transfered to an insulating substrate (typically 300 nm thick SiO2 that yields a good optical contrast). It has been shown that this technique may also be used in the fabrication of 2D crystals of a chemical composition dierent from carbon similarly to graphene, single layers of BN, NbSe2 or Bi2 Sr2 CaCo2 Ox have been obtained [16]. One may, furthermore, control the carrier density in the metallic 2D crystals, such as graphene, by the electric eld eect; the 300 nm thick insulating SiO2 layer is indeed on top of a positively doped metallic Si substrate, which serves
as a backgate. The combined system graphene-SiO2 -backgate may, thus, be viewed as a capacitor the capacity of which is C = Q/Vg = 0 A/d, where Q = en2D A is the capacitor charge, in terms of the total surface A, Vg is the gate voltage, d = 300 nm is the thickness of the SiO2 layer with the dielectric constant = 3.7. The eld-eect induced 2D carrier density is thus given by n2D = Vg with cm2 0 7.2 1010 . ed V (1.1)
The gate voltage may vary roughly between 100 and 100 V, such that the induced maximal carrier densities is on the order of 1012 cm2 , on top of the intrinsic carrier density which turns out to be zero in graphene, as is discussed below. At gate voltages above 100 V, the capacitor breaks down (electrical breakdown). Epitaxial graphene An alternative fabrication of graphene consists of exposing an epitaxially grown hexagonal (4H or 6H-) SiC crystal to temperatures of about 1300o C in order to evaporate the less tightly bound Si atoms from the surface [17, 18]. The remaining carbon atoms on the surface form a graphitic layer (graphitsation) the physical properties of which depend on the the chosen SiC surface. In the case of the Si-terminated (0001) surface, the graphitisation process is slow, and one may thus control the number of formed graphene layers (usually one or two). The resulting electron mobility, however, turns out to be rather low, such that the Si-terminated surface is less popular for the fabrication of samples used in transport measurements. For a C-terminated (000 surface, the graphitisation 1) process is very fast, and a large number of graphene layers are formed (up to 100). In contrast to epitaxial graphene on the Si-terminated surface, the electron mobility is, here, rather high. In contrast to exfoliated graphene, the SiC substrate must be considered as an integral part of the whole system of epitaxial graphene. It is indeed the mother compound, and the rst graphitic layer formed during the graphitisation process is tightly bound to the SiC substrate. X-ray diraction measurements have revealed a certain amount of rotational disorder in the stacking of the graphene layers [19]. This, together with the rather large spacing between the graphene sheets on top of the buer layer, corroborates the view that the graphitic layer on the SiC substrate consists indeed of almost independent graphene layers rather than of a thin graphite ake. However, the charge is not homogeneously distributed between the dierent graphene layers it turns out that the graphene layer,4 which is closest to the substrate, is electron-doped due to a charge transfer from the bulk SiC, whereas the subsequent layers are hardly charged. As a consequence of this inhomogeneous charge distribution between the dierent layers, transport measurements in epitaxial graphene are often more
4 We refer to the layers on top of the buer layer when using the term graphene. The buer layer may not be counted due to its tight bonding to the SiC substrate.
Crystal Structure of Graphene

(a) a2 a1 y x
: A sublattice : B sublattice
7
(b)
1111111111111111 0000000000000000 1111111111111111 0000000000000000 1111111111111111 0000000000000000 1111111111111111 0000000000000000 1111111111111111 0000000000000000 * 1111111111111111 0000000000000000 1111111111111111 0000000000000000 2 1111111111111111 0000000000000000 1111111111111111 0000000000000000 1111111111111111 0000000000000000 1111111111111111 0000000000000000 1111111111111111 0000000000000000 1111111111111111 0000000000000000 1111111111111111 0000000000000000 1111111111111111 0000000000000000 1111111111111111 0000000000000000 * 1111111111111111 0000000000000000 1 1111111111111111 0000000000000000 1111111111111111 0000000000000000 1111111111111111 0000000000000000 1111111111111111 0000000000000000 1111111111111111 0000000000000000 1111111111111111 0000000000000000 1111111111111111 0000000000000000 1111111111111111 0000000000000000 1111111111111111 0000000000000000 1111111111111111 0000000000000000
2 3
M K
1 a=0.142 nm
Figure 1.4: (a) Honeycomb lattice. The vectors 1 , 2 , and 3 connect nn carbon atoms, separated by a distance a = 0.142 nm. The vectors a1 and a2 are basis vectors of the triangular Bravais lattice. (b) Reciprocal lattice of the triangular lattice. Its primitive lattice vectors are a and a . The shaded region represents the rst Brillouin zone (BZ), with its centre 1 2 and the two inequivalent corners K (black squares) and K (white squares). The thick part of the border of the rst BZ represents those points which are counted in its denition such that no points are doubly counted. The rst BZ, dened in a strict manner, is, thus, the shaded region plus the thick part of the border. For completeness, we have also shown the three inequivalent cristallographic points M , M , and M (white triangles). delicate to interpret than those of exfoliated graphene. Indeed, the quantum Hall eect has only recently been observed in epitaxial graphene with a single active layer obtained from graphitisation of C-terminated SiC [20, 21, 22], whereas the absence of the quantum Hall eect in epitaxial graphene on the Si-terminated surface has erroneously been attributed to to their high mobility [18] instead of possible short circuits in the metallic layers [23].
1.3
Crystal Structure of Graphene
As already mentioned in the last section, the carbon atoms in graphene condense in a honeycomb lattice due to their sp2 hybridisation. The honeycomb lattice is not a Bravais lattice because two neighbouring sites are inequivalent from a crystallographic point of view.5 Fig. 1.4 (a) illustrates indeed that a site on the A sublattice has nearest neighbours (nn) in the directions north-east, north-west, and south, whereas a site on the B sublattice has nns in the directions north, south-west, and south-east. Both A and B sublattices, however, are triangular6 Bravais lattices, and one may view the honeycomb lattice as a triangular Bravais lattice with a two-atom basis (A and B). The distance between nn carbon atoms is a = 0.142 nm, which is the average of the single (CC) and double (C=C) covalent bonds, as in the case of benzene.
5 This needs to be clearly distinguished from a chemical point of view according to which they may be equivalent as in the case of graphene where both types of sites consist of carbon atoms. 6 The triangular lattice is sometimes also called hexagonal lattice.
The three vectors which connect a site on the A sublattice with a nn on the B sublattice are given by 1 = a 3ex + ey , 2 2 = a 3ex + ey , 2 3 = aey , (1.2)
and the triangular Bravais lattice is spanned by the basis vectors 3a a1 = 3aex and a2 = ex + 3ey . 2
(1.3)
The modulus of the basis vectors yields the lattice spacing, a = 3a = 0.24 2 nm, and the area of the unit cell is Auc = 3 /2 = 0.051 nm2 . The density of a carbon atoms is, therefore, nC = 2/Auc = 39 nm2 = 3.9 1015 cm2 . Because there is one electron per carbon atom that is not involved in a covalent bond, there are as many valence electrons as carbon atoms, and their density is, thus, n = nC = 3.9 1015 cm2 . As discussed in the following section, this density is not equal to the carrier density in graphene, which one measures in electrical transport measurements. The reciprocal lattice, which is dened with respect to the triangular Bravais lattice, is depicted in Fig. 1.4 (b). It spanned by the vectors 2 a = 1 3a ey ex 3 and a = 2 4 ey . 3a (1.4)
Physically, all sites of the reciprocal lattice represent equivalent wave vectors. Any wave be it a vibrational lattice excitation or a quantum-mechanical electronic wave packet propagating on the lattice with a wave vector diering by a reciprocal lattice vector has indeed the same phase up to a multiple of 2, due to the relation ai a = 2ij (1.5) j (for i, j = 1, 2) between direct and reciprocal lattice vectors. The rst Brillouin zone [BZ, shaded region and thick part of the border of the hexagon in Fig. 1.4 (b)] represents a set of inequivalent points in the reciprocal space, i.e. of points which may not be connected to one another by a reciprocal lattice vector, or else of physically distinguishable lattice excitations. The long wavelength excitations are situated in the vicinity of the point, in the centre of the rst BZ. Furthermore, one distinguishes the six corners of the rst BZ, which consist of the inequivalent points K and K represented by the vectors 4 K = ex . 3 3a (1.6)
The four remaining corners [shown in gray in Fig. 1.4 (b)] may indeed be connected to one of these points via a translation by a reciprocal lattice vector. These cristallographic points play an essential role in the electronic properties of graphene because their low-energy excitations are centered around the two points K and K , as is discussed in detail in the following section. We emphasise,
Electronic Band Structure of Graphene
because of some confusion in the literature on this point, that the inequivalence of the two BZ corners, K and K , has nothing to do with the presence of two sublattices, A and B, in the honeycomb lattice. The form of the BZ is an intrinsic property of the Bravais lattice, independent of the possible presence of more than one atom in the unit cell. For completeness, we have also shown, in Fig. 1.4 (b), the three crystallographically inequivalent M points in the middle of the BZ edges.
1.4
As we have discussed in the previous section, three electrons per carbon atom in graphene are involved in the formation of strong covalent bonds, and one electron per atom yields the bonds. The electrons happen to be those responsible for the electronic properties at low energies, whereas the electrons form energy bands far away from the Fermi energy [24]. This section of the introduction is, thus, devoted to a brief discussion of the energy bands of electrons within the tight-binding approximation, which was originally calculated for the honeycomb lattice by P. R. Wallace in 1947 [25].
1.4.1
Tight-binding model for electrons on the honeycomb lattice
In the case of two atoms per unit cell, we may write down a trial wave function k (r) = ak k (r) + bk k (r),
(A) (B)
(1.7)
(A)
where ak and bk are complex functions of the quasi-momentum k. Both k (r) (B) and k (r) are Bloch functions with k (r) =
Rl (j)
eikRl (j) (r + j Rl ),
(1.8)
where j = A/B labels the atoms on the two sublattices A and B, and j is the vector which connects the sites of the underlying Bravais lattice with the site of the j atom within the unit cell. The (j) (r + j Rl ) are atomic orbital wave functions for electrons that are in the vicinity of the j atom situated at the position Rl j at the (Bravais) lattice site Rl . Typically one chooses the sites of one of the sublattices, e.g. the A sublattice, to coincide with the sites of the Bravais lattice. Notice furthermore that there is some arbitrariness in the choice of the phase in Eq. (1.8) instead of choosing exp(ik Rl ), one may also have chosen exp[ik (Rl j )], for the atomic wave functions. The choice, however, does not aect the physical properties of the system because it simply leads to a redenition of the weights ak and bk which aquire a dierent relative phase [26].
10
With the help of these wave functions, we may now search the solutions of the Schrdinger equation o Hk = k k . Here, we have chosen an arbitrary representation, which is not necessarily that in real space.7 Multiplication of the Schrdinger equation by k from the left o yields the equation k Hk = k k k , which may be rewritten in matrix form with the help of Eqs. (1.7) and (1.8) (a , b ) Hk k k ak bk = k (a , b ) Sk k k ak bk . (1.9)
Here, the Hamiltonian matrix is dened as Hk and the overlap matrix Sk k k (B) (A) k k
(A) (A)
k Hk (B) (A) k Hk
(A)
(A)
k Hk (B) (B) k Hk
(A)
(B)
= Hk ,
(1.10)
k k (B) (B) k k
(A)
(B)
= Sk
(1.11)
accounts for the non-orthogonality of the trial wave functions. The eigenvalues k of the Schrdinger equation yield the electronic bands, and they may be o obtained from the secular equation det Hk Sk = 0, k (1.12)
which needs to be satised for a non-zero solution of the wave functions, i.e. for ak = 0 and bk = 0. The label denotes the energy bands, and it is clear that there are as many energy bands as solutions of the secular equation (1.12), i.e. two bands for the case of two atoms per unit cell. Formal solution Before turning to the specic case of graphene and its energy bands, we solve formally the secular equation for an arbitrary lattice with several atoms per unit cell. The Hamiltonian matrix (1.10) may be written, with the help of Eq. (1.8), as
ij Hk
=
Rl ,Rm
eik(Rl Rm ) eikRl
kl
d2 r (i) (r + i Rl )H(j) (r + j Rm )
= N
d2 r (i) (r) [H a + V ] (j) (r + ij Rl ) (1.13)
= N (i) sij + tij k k

7 The
wave function k (r) is, thus, the real space representation of the Hilbert vector k .
Electronic Band Structure of Graphene where ij j i , sij k eikRl

Rl
11
d2 r (i) (r Rk )(j) (r + ij Rm ) =
ij Sk N
(1.14)
and we have dened the hopping matrix tij k eikRl

Rl
d2 r (i) (r Rk )V (j) (r + ij Rm ) .
(1.15)
Here, we have separated the Hamiltonian H into an atomic orbital part Ha =

2
2m
l + V (rl Rl + j ),
which satises the eigenvalue equation H a (j) (r + j Rm ) = (j) (j) (r + j Rm ) and a perturbative part V which takes into account the potential term that arise from all other atoms dierent from that in the atomic orbital Hamiltonian. The last line in Eq. (1.13) has been obtained from the fact that the atomic wave functions (i) (r) are eigenstates of the atomic Hamiltonian H a with the atomic energy (i) for an orbital of type i. This atomic energy plays the role of an onsite energy. The secular equation now reads det tij (i) k k = 0. (1.16)
Notice that, if the the atoms on the dierent sublattices are all of the same electronic conguration, one has (i) = 0 for all i, and one may omit this onsite energy, which yields only a constant, physically irrelevant shift of the energy bands. Solution for graphene with nearest-neighbour and next-nearest-neighour hopping After these formal considerations, we now study the particular case of the tightbinding model on the honeycomb lattice, which yields, to great accuracy, the energy bands of graphene. Because all atomic orbitals are pz orbitals of carbon atoms, we may omit the onsite energy 0 , as discussed in the last paragraph. We choose the Bravais lattice vectors to be those of the A sublattice, i.e. A = 0, and the equivalent site on the B sublattice is obtained by the displacement B = AB = 3 (see Fig. 1.5). The nn hopping amplitude is given by the expression t d2 r A (r)V B (r + 3 ), (1.17)
and we also take into account next-nearest neighbour (nnn) hopping which connects neighbouring sites on the same sublattice tnnn d2 r A (r)V A (r + a1 ) = d2 r B (r)V B (r + a1 ). (1.18)
12
a3
B2 A
a2
B1
3
B3
a1
Figure 1.5:
Tight-binding model for the honeycomb lattice.
Notice that one may have chosen any other vector j or a2 , respectively, in the calculation of the hopping amplitudes. Because of the normalisation of the atomic wave functions, we have d2 r(j) (r)(j) (r) = 1, and we consider furthermore the overlap correction between orbitals on nn sites, s d2 r A (r)B (r + 3 ). (1.19)
We neglect overlap corrections between all other orbitals which are not nn, as well as hopping amplitudes for larger distances than nnn. If we now consider an arbitrary site A on the A sublattice (Fig. 1.5), we may see that the o-diagonal terms of the hopping matrix (1.15) consist of three terms corresponding to the nn B1 , B2 , and B3 , all of which have the same hopping amplitude t. However, only the site B3 is described by the same lattice vector (shifted by 3 ) as the site A and thus yields a zero phase to the hopping matrix. The sites B1 and B2 correspond to lattice vectors shifted by a1 and a3 a2 a1 , respectively. Therefore, they contribute a phase factor exp(ik a1 ) and exp(ik a3 ), respectively. The o-diagonal elements of the hopping matrix may then be written as8 tAB = tk = tBA , k k as well as those of the overlap matrix
sAB = sk = sBA k k
(sAA = sBB = 1, due to the above-mentioned normalisation of the atomic wave k k functions), where we have dened the sum of the nn phase factors k 1 + eika1 + eika3 .
3
(1.20)
The nnn hopping amplitudes yield the diagonal elements of the hopping matrix, tAA = tBB = 2tnnn k k
i=1
8 The
cos(k ai ) = tnnn |k |2 3 ,
hopping matrix element tAB corresponds to a hopping from the B to the A sublattice. k
Electronic Band Structure of Graphene and one obtains, thus, the secular equation det tAA k k (t sk )k
(t sk )k AA tk k
13
=0
(1.21)
with the two solutions ( = ) = k tAA + t|k | k . 1 + s|k | (1.22)
This expression may be expanded under the reasonable assumptions s 1 and tnnn t, which we further justify at the end of the paragraph, k tAA + t|k | st|k |2 = t |k |2 + t|k | k nnn
3 3
= 2t nnn
i=1
cos(k ai ) + t
3+2
i=1
cos(k ai ),
(1.23)
where we have dened the eective nnn hopping amplitude t nnn tnnn st , (1.24)
and we have omitted the unimportant constant 3tnnn in the last equation. One, therefore, notices that the overlap corrections simply yield a renormalisation of the nnn hopping amplitudes. The hopping amplitudes may be determined by tting the energy dispersion (1.23) obtained within the tightbinding approximation to those calculated numerically in more sophisticated band-structure calculations [27] or to spectroscopic measurements [28]. These yield a value of t 3 eV for the nn hopping amplitude and t nnn 0.1t, which justies the above-mentioned expansion for t /t 1. Notice that this nnn tting procedure does not allow for a distinction between the true nnn hopping amplitude tnnn and the contribution from the overlap correction st. We, therefore, omit this distinction in the following discussion and drop the prime at the eective nnn hopping amplitude, but one should keep in mind that it is an eective parameter with a contribution from nn overlap corrections. Energy dispersion of electrons in graphene The energy dispersion (1.23) is plotted in Fig. 1.6 for tnnn /t = 0.1. It consists of two bands, labeled by the index = , each of which contains the same number of states. Because each carbon atom contributes one electron and each electron may occupy either a spin-up or a spin-down state, the lower band with = (the or valence band) is completely lled and that with = + (the or conduction band) completely empty. The Fermi level is, therefore, situated at the points where the band touches the band. Notice that only if tnnn = 0 the energy dispersion (1.23) is electron-hole symmetric, i.e. = . This means that nnn hopping and nn overlap corrections break the k k electron-hole symmetry. The points, where the band touches the band,
14
(a) (b) Energy [in units of t]
4 4
3 3 2 2 1 1 wave vector 3 3
Energy
K K
K K
K K K
-2 2
-1 1 -1 1
2 -2
1 1
2 2
ky kx
Energy dispersion obtained within the tight-binding approximation, for tnnn /t = 0.1. One distinguishes the valence () band from the conduction ( ) band. The Fermi level is situated at the points where the band touches the band. (a) Energy dispersion as a function of the wave-vector components kx and ky . (b) Cut throught the energy dispersion along characteristic lines (connecting the points K M K. The energy is measured in units of t and the wave vectors in units of 1/a.
Figure 1.6:
are called Dirac points. They are situated at the points kD where the energy dispersion (1.23) is zero, D = 0. (1.25) k Eq. (1.25) is satised when kD = 0, i.e. when RekD = = and, equally, ImkD = sin D 3a D D 3a D (kx + 3ky ) + sin (kx + 3ky ) = 0. 2 2 (1.27) 1 + cos(kD a2 ) + cos(kD a3 ) (1.26) D 3a D D 3a D (kx + 3ky ) + cos (kx + 3ky ) = 0 1 + cos 2 2
D The last equation may be satised by the choice ky = 0, and Eq. (1.26), thus, when 4 3a D D =0 kx = . k 1 + 2 cos 2 x 3 3a
Comparison with Eq. (1.6) shows that there are, thus, two inequivalent Dirac points D and D , which are situated at the points K and K , respectively, 4 kD = K = ex . 3 3a (1.28)
15
Although situated at the same position in the rst BZ, it is useful to make a clear conceptual distinction between the Dirac points D and D , which are dened as the points where the two bands and touch each other, and the purely crystallographic points K and K , which are dened as the corners of the rst BZ. There are, indeed, situations where the Dirac points move away from the points K and K , as we will discuss in Sec. 1.5. Notice that because of the symmetry k = k , which is a consequence of time-reversal symmetry, Dirac points occur necessarily in pairs if kD is a solution of k = 0, so is kD . In graphene, there is one pair of Dirac points, and the zero-energy states are, therefore, doubly degenerate. One speaks of a twofold valley degeneracy, which survives when we consider low-energy electronic excitations that are restricted to the vicinity of the Dirac points, as is discussed in Sec. 1.4.2.
Eective tight-binding Hamiltonian Before considering the low-energy excitations and the continuum limit, it is useful to dene an eective tight-binding Hamiltonian, Hk tnnn |k |2 + t Here, 0 k
k 0
(1.29)
represents the 2 2 one-matrix =

1 0 0 1 .
This Hamiltonian eectively omits the problem of non-orthogonality of the wave functions by a simple renormalisation of the nnn hopping amplitude, as alluded to above. The eigenstates of the eective Hamiltonian (1.29) are the spinors = k a k b k , (1.30)
the components of which are the probability amplitudes of the Bloch wave function (1.7) on the two dierent sublattices A and B. They may be determined by considering the eigenvalue equation Hk (tnnn = 0) = t|k | , which does k k not take into account the nnn hopping correction. Indeed, these eigenstates are also those of the Hamiltonian with tnnn = 0 because the nnn term is proportional to the one-matrix . The solution of the eigenvalue equation (1.30) yields a = k
k b = eik b k |k | k
16 and, thus, the eigenstates9 1 = k 2 where we have dened the angle k = arctan Imk Rek . 1 eik ,
(1.31)
(1.32)
As one may have expected, the spinor represents an equal probability to nd an electron in the state on the A as on the B sublattice because both k sublattices are built from carbon atoms with the same onsite energy (i) .10
1.4.2
Continuum limit
In order to describe the low-energy excitations, i.e. electronic excitations the characteristic energy of which is much smaller than the band width |t|, one may concentrate on excitations at the Fermi level. This amounts to restricting the excitations to quantum states in the vicinity of the Dirac points, and one may expand the energy dispersion around K. The wave vector is, thus, decomposed as k = K + q, where |q| |K| 1/a. The small parameter, which governs the expansion of the energy dispersion, is, therefore, |q|a 1. It is evident from the form of the energy dispersion (1.23) and the eective Hamiltonian that the basic entity to be expanded is the sum of the phase factors
9 The eigenstates are dened up to a global (but k-dependent) phase, and one may also choose 1 1 eik eik /2 = , or = k k eik /2 2 2
which is also found in the literature (see e.g. Ref. [4]). Notice, however, that the second choice yields wave functions that are not single-valued when the phase k changes by 2. 10 This is not the case for other lattices such as 2D boron nitride (BN), which also forms a honeycomb lattice. However, in the case of BN, one sublattice consists of boron atoms with an atomic (onsite) energy A and the other one of nitrogen atoms with an onsite energy B = A . The dierence in the onsite energy = A B may be accounted for in the eective Hamiltonian (1.29) if one adds a term 2 1 0 0 1 .
This term opens a gap in the dispersion relation at the points K and K , = tnnn |k |2 + k t2 |k |2 + 2 , 4
and it is energetically favourable to ll preferentially the sublattice with lower onsite energy.
Continuum Limit
17
k . We need to distinguish the sum at the K point from that at the K point,
q k=K+q
1 + eiKa2 eiqa2 + eiKa3 eiqa3 1 1 + ei2/3 1 + iq a2 (q a2 )2 2 1 +ei2/3 1 + iq a3 (q a3 )2 2

(0) (1) (2) q + q + q
By denition of the Dirac points and their position at the BZ corners K and (0) K , we have q = K = 0. We limit the expansion to second order in |q|a. First order in |q|a The rst order term is given by 3a (1) q = i (qx + 3qy )ei2/3 + (qx + 3qy )ei2/3 2 3a = (qx iqy ), (1.33) 2 which is obtained with the help of sin(2/3) = 3/2 and cos(2/3) = 1/2. This yields the eective low-energy Hamiltonian
e, Hq = vF (qx x + qy y ),
(1.34)
where we have dened the Fermi velocity11 vF and the Pauli matrices x = 0 1 1 0 and y = 0 i i 0 . 3|t|a 3ta = 2 2 (1.35)
Furthermore, we have introduced the valley isospin = , where = + denotes the K point at +K and = the K point at K modulo a reciprocal lattice vector. The low-energy Hamiltonian (1.34) does not take into account nnn hopping corrections, which are proportional to |k |2 and, thus, occur only in the second-order expansion of the energy dispersion [at order O(|q|a)2 ]. The energy dispersion (1.23) therefore reads q,= = vF |q|, (1.36)
11 The minus sign in the denition is added to render the Fermi velocity positive because the hopping parameter t 3 eV happens to be negative, as mentioned in the last section.
18
independent of the the valley isospin . We have already alluded to this twofold valley degeneracy in Sec. 1.4.1, in the framework of the discussion of the zeroenergy states at the BZ corners. From Eq. (1.36) it is apparent that the continuum limit |q|a 1 coincides with the limit || |t|, as described above, because |q | = 3ta|q|/2 |t|. It is convenient to swap the spinor components at the K point (for = ), k,=+ =
A k,+ B k,+
k,= =
B k, A k,
i.e. to invert the role of the two sublattices. In this case, the eective low-energy Hamiltonian may be represented as
e, Hq = vF (qx x + qy y ) = vF z q ,
(1.37)
in the last line via the 4 4 matrices z =
i.e. as two copies of the 2D Dirac Hamiltonian HD = vF p (with the momentum p = q), where we have introduced the four-spinor representation A q,+ B q,+ q = B q, A q, 0 0 ,
where we have, now,
and ( x , y ). In this four-spinor representation, the rst two components represent the lattice components at the K point and the last two components those at the K point. We emphasise that one must clearly distinguish both types of isospin: (a) the sublattice isospin is represented by the Pauli matrices j , where spin up corresponds to the component on one sublattice and spin down to that on the other one. A rotation within the SU(2) sublattice isospin space yields the band indices = , and the band index (or band isospin) is, thus, intimitely related to the sublattice isospin. (b) The valley isospin, which is described by a second set of Pauli matrices j , the z-component of which appears in the Hamiltonian (1.37), is due to the twofold valley degeneracy and is only indirectly related to the presence of two sublattices. The eigenstates of the Hamiltonian (1.37) are the four-spinors 0 1 1 1 eiq 0 = , (1.38) and q, = =+ = q, 0 1 2 2 eiq 0 q = arctan qy qx . (1.39)
Continuum Limit
19
energy
conduction band ( = +) 0 valence band ( = )
=+
= K ( = +)
=+ K ( = )
momentum
Figure 1.7: Chirality In high-energy physics, one denes the helicity of a particle as the projection of its spin onto the direction of propagation [29], q = q , |q| (1.40)
Relation between band index , valley isospin , and chirality in graphene.
which is a Hermitian and unitary operator with the eigenvalues = , q | = = | = . (1.41) Notice that describes, in this case, the true physical spin of the particle. In the absence of a mass term, the helicity operator commutes with the Dirac Hamiltonian, and the helicity is, therefore, a good quantum number, e.g. in the description of neutrinos, which have approximately zero mass. One nds indeed, in nature, that all neutrinos are left-handed ( = ), i.e. their spin is antiparallel to their momentum, whereas all anti-neutrinos are right-handed ( = +). For the case of graphene, one may use the same denition (1.40), but the Pauli matrices no longer dene the true spin, but the sublattice isospin. Because of this dierence, one also calls q the chirality operator. It clearly commutes with the massless 2D Dirac Hamiltonian (1.37), and one may even express it in terms of the chirality operator
e, Hq = vF |q|q ,
which takes into account the two-fold valley degeneracy, in terms of the valley isospin = . The band index , which describes the valence and the conduction band, is, therefore, entirely determined by the chirality and the valley isospin, and one nds = , (1.42)
20
which is depicted in Fig. 1.7. We notice nally that the chirality is a preserved quantum number in elastic scattering processes induced by impurity potentials Vimp = V (r) that vary smoothly on the lattice scale. In this case, inter-valley scattering is suppressed, and the chirality thus conserved, as a consequence of Eq. (1.42). This eect gives rise to the absence of backscattering in graphene [30] and is related to the Klein paradox [31] in the quantum theory of fermionic elds. Higher orders in |q|a Although most of the fundamental properties of graphene are captured within the eective model obtained at rst order in the expansion of the energy dispersion, it is useful to take into account second-order terms. These corrections include nnn hopping corrections and o-diagonal second-order contributions from the expansion of k . The latter yield the so-called trigonal warping, which consist of an anisotropy of the energy dispersion around the Dirac points. The diagonal second-order term, which stems from the nnn hopping, is readily obtained from Eq. (1.33),
(1) Hnnn = tnnn |q |2 tnnn |q |2 =
9a2 tnnn |q|2 , 4
(1.43)
independent of the valley index . The o-diagonal second-order terms are a (2) tq = vF (qx iqy )2 . 4 Notice that there is a natural energy hierarchy between the diagonal and odiagonal second-order terms when compared to the leading linear term; whereas the o-diagonal terms are on the order O(|q|a) as compared to the energy scale vF |q|, the diagonal term is on the order O((tnnn /t)|q|a) and thus roughly an order of magnitude smaller. We therefore take into account also the o-diagonal third order term a2 (3) tq = vF (qx + iqy )|q|2 , 8 which also needs to be taken into account when calculating the high-energy corrections of the energy levels in a magnetic eld (see Sec. 2.2). Up to third order, the o-diagonal terms therefore read a a2 tq = vF (qx + iqy ) (qx iqy )2 |q|2 (qx + iqy ) , 4 8 (1.44)
where one may omit the valley-dependent sign before the y-components of the wave vector by sweeping the sublattice components in the spinors when changing the valley. In order to appreciate their inuence on the energy bands, we need to cal culate the modulus of q ,
(1) (1) (2) |q | |q |2 + 2Req q 1/2
3a |q|a |q| 1 cos(3q ) , 2 4
21
(a)
ky
3 2 1
(b)
0.4
2eV 1.5eV 1eV
qy
0.2
K
-3 -2 -1 -1
K kx
2 3
-0.4 -0.2
K
0.2
qx
0.4
-0.2
-2
-0.4
-3
Figure 1.8: Contours of constant (positive) energy in the wave-vector space. (a) Contours obtained from the full dispersion relation (1.23). The dashed line corresponds to the energy t + tnnn , which separates closed orbits around the K and K points (black lines, with energy < t + tnnn ) from those around the point (gray line, with energy < t + tnnn ). (b) Comparison of the contours at energy = 1 eV, 1.5 eV, and 2 eV around the K point. The black lines correspond to the energies calculated from the full dispersion relation (1.23) and the gray ones to those calculated to second order within the continuum limit (1.45). where we have used the parametrisation qx = |q| cos q and qy = |q| sin q , and where we have restricted the expansion to second order. Finally, the energy dispersion (1.23) expanded to second order in |q|a reads = q, 9a2 |q|a tnnn |q|2 + vF |q| 1 cos(3q ) 4 4 . (1.45)
As we have already mentioned in Sec. 1.4.1, it is apparent from Eq. (1.45) that the nnn correction (which also accounts for nn overlap corrections) breaks the electron-hole symmetry = . This correction is, however, a rather q, q, small correction, of order |q|atnnn /t, to the rst-order eective Hamiltonian (1.37). The second-order expansion of the phase factor sum q yields a more relevant correction the third term in Eq. (1.45), of order |q|a |q|atnnn /t to the linear theory. It depends explicitly on the valley isospin and renders the energy dispersion anisotropic in q around the K and K point. The tripling of the period, due to the term cos(3q ), is a consequence of the symmetry of the underlying lattice and is also called trigonal warping, as mentioned above. The trigonal warping of the dispersion relation is visualised in Fig. 1.8, where we have plotted the contours of constant (positive) energy in Fourier space. The closed energy contours around the K and K points at low energy are sparated by the high-energy contours around the point by the dashed lines in Fig. 1.8 (a) at energy |t + tnnn | the crossing points of which correspond to the M points. As mentioned above, the dispersion relation has saddle points at these points at the border of the rst BZ, which yield van Hove singularities in the density of states. In Fig. 1.8 (b), we compare constant-energy contours of
22
1111111111111111111 0000000000000000000 1111111111111111111 0000000000000000000 1111111111111111111 0000000000000000000 1111111111111111111 0000000000000000000 1111111111111111111 0000000000000000000 1111111111111111111 0000000000000000000 1111111111111111111 0000000000000000000 1111111111111111111 0000000000000000000 tnnn 1111111111111111111 0000000000000000000 1111111111111111111 0000000000000000000 1111111111111111111 0000000000000000000 1111111111111111111 0000000000000000000 a2 1111111111111111111 0000000000000000000 1111111111111111111 0000000000000000000 1111111111111111111 0000000000000000000 1111111111111111111 0000000000000000000 a1 t nnn 1111111111111111111 0000000000000000000 1111111111111111111 0000000000000000000 1111111111111111111 0000000000000000000
tnnn t nnn
a tnnn tnnn
a t axis of deformation
Quinoid-type deformation of the honeycomb lattice the bonds parallel to the deformation axis (double arrow) are modied. The shaded region indicates the unit cell of the oblique lattice, spanned by the lattice vectors a1 and a2 . Dashed and dashed-dotted lines indicate next-nearest neigbors, with characteristic hopping integrals tnnn and t nnn , respectively, which are dierent due to the lattice deformation.
Figure 1.9:
the full dispersion relation to those obtained from Eq. (1.45) calculated within a second-order expansion. The contours are indistinguishable for an energy of = |t|/3 1 eV, and the continuum limit yields rather accurate results up to energies as large as 2 eV. Notice that, in todays exfoliated graphene samples on a SiO2 substrates, one may probe, by eld-eect doping of the graphene sheet, energies which are on the order of 100 meV. Above these energies the capacitor breaks down, and Fig. 1.8 (a) indicates that the continuum limit (1.45) yields extremely accurate results at these energies. We nally mention that, when higher-order terms in |q|a are taken into account, the chirality operator (1.40) no longer commutes with the Hamiltonian. Chirality is therefore only a good quantum number in the vicinity of the Dirac points.
1.5
Deformed Graphene
In the previous section, we have considered a perfect honeycomb lattice which is invariant under a 2/3 rotation. As a consequence, all hopping parameters along the nn bonds j were equal. An interesting situation arises when the graphene sheet is deformed, such that the rotation symmetry is broken. In order to illustrate the consequences, we may apply a uniaxial strain in the y-direction,12 a a = a + , in which case one obtains a quinoid-type deformation (Fig. 1.9) the hopping t along 3 turns out to be dierent from that t along 1 and 2 [33, 34, 35, 36, 37, 38], t t t = t + a. (1.46) a
12 In our simplied model, we only consider one bond length changed by the strain. The more general case has been considered by Peirera et al. [32]. However, the main eects are fully visible in the simplied model.
Deformed Graphene
23
Figure 1.10: Band dispersion of the quinoid-type deformed the honeycomb lattice, for a lattice distortion of a/a = 0.4, with t = 3 eV, tnnn /t = 0.1, t/a = 5 eV/, and A A tnnn /a = 0.7 eV/. The inset shows a zoom on one of the Dirac points, D . Furthermore, also four of six nnn hopping integrals are aected by the strain (see Fig. 1.9), tnnn tnnn t a. (1.47) nnn = tnnn + a If one considers a moderate deformation a/a 1, the eect on the hopping amplitudes may be estimated with the help of Harrisons law [39], according to which t = C 2 /ma2 , where C is a numerical prefactor of order 1. One therefore nds 2t t = 4.3 eV/ A a a and t = t(1 2) (1.48)
which coincides well with the value t/a 5 eV/, which may be found in the A literature [24, 40]. The estimation of the modied nnn hopping integral t nnn is slightly more involved. One may use a law tnnn (b, a) t(a)e(ba)/d(a) familiar in the context of the extended Hckel model [41], where b is the nnn u distance, and d a/3.5 0.4 is a caracteristic distance related to the A overlap of atomic orbitals. In undeformed graphene one has b = a 3, whereas in quinoid-type graphene b = b(1 + /2), which gives t nnn = tnnn (1 2 + b/2d). (1.49)
24
The electronic properties of quinoid-type graphene may then be described in terms of an eective Hamiltonian of the type (1.29), Hk = tnnn hk + t with [37] hk = 3kx a t nnn 2 cos 3kx a + 2 cos + ky a tnnn 2 3 3kx a + ky a + , + cos 2 2 3 + 2 (1.51) 0 k k 0 , (1.50)
and the o-diagonal elements k = 2eiky a(3/2+) cos The resulting energy dispersion = tnnn hk + t|k | k (1.53) 3 kx a 2 + (1 2). (1.52)
is plotted in Fig. 1.10 for an unphysically large deformation, = 0.4, for reasons of visibility. Notice that the physically possible deformations are limited by a value of 0.1...0.2 beyond which the graphene sheet cracks [42]. One notices, in Fig. 1.10, two eects of the deformation: i) the Dirac points no longer coincide with the corners of the rst BZ, the form of which is naturally also modied by the deformation; and ii) the cones, i.e. the dispersion relation in the vicinity of the Dirac points, are tilted, i.e. the nnn hopping term (1.51) breaks the electron-hole symmetry already at linear order in |q|a. These two points are discussed in more detail in the following two subsections.
1.5.1
Dirac point motion
In order to evaluate quantitatively the position of the Dirac points, which are dened as the contact points between the valence ( = ) and the conduction ( = +) bands, one needs to solve the equation kD = 0, in analogy with the case of undeformed graphene discussed in Sec. 1.4.1. One then nds
D ky = 0
and
2 t D kx a = arccos 2t 3
(1.54)
where the valley index = denotes again the two inequivalent Dirac points D and D , respectively. As already mentioned, the Dirac points D and D coincide, for undistorted graphene, with the crystallographic points K and K , respectively, at the corners of the rst BZ. The distortion makes both pairs of
Deformed Graphene
(a) (b)
25
(c)
(d)
Topological semi-metal insulator transition in the model (1.55) driven by the gap parameter . (a) Two well-separated Dirac cones for 0, as for graphene. (b) When lowering the modulus of the (negative) gap parameter, the Dirac points move towards a single point. (c) The two Dirac points merge into a single point at the transition ( = 0). The band dispersion remains linear in the qy -direction while it becomes parabolic in the qx -direction. (d) Beyond the transition ( > 0), the (parabolic) bands are separated by a band gap (insulating phase).
Figure 1.11:
points move in the same direction due to the negative value of t/a. However, unless the parameters are ne-tuned, this motion is dierent, and the two pairs of points no longer coincide. One further notices that Eq. (1.54) has (two) solutions only for t 2t. Indeed, the two Dirac points merge at the characteristic point M at the border of the rst BZ (see Fig. 1.4). The point t = 2t is special insofar as it characterises a topological phase transition between a semi-metallic phase (for t < 2t) with a pair of Dirac cones and a band insulator (for t > 2t) [32, 35, 36, 43, 44, 45]. In the vicinity of the transition, one may expand the Hamiltonian (1.50) around the merging point M [43, 44], and one nds13
M Hq
0 +
2 2 qx 2m
+ + i cqy
2 2 qx 2m
i cqy
(1.55)
in terms of the mass m = 2 2 /3ta2 and the velocity c = 3ta/ [44]. The gap parameter = t 2t changes its sign at the transition it is negative in the semi-metallic and positive in the insulating phase, where it describes a true gap. The Hamiltonian (1.55) has quite a particular form in the vicinity of the merging points: it is linear in the qy -direction, as one would expect for Dirac points, but it is quadratic in the qx -direction [35]. This is a general feature of merging points, which may only occur at the point or else at half a reciprocal lattice vector G/2, i.e. in the centre of a BZ border line (such as the M
13 We do not consider the diagonal part of the Hamiltonian, here, i.e. we choose t nnn = 0, because it does not aect the position of the Dirac points.
26
points) [43]. Indeed, one may show that in the case of a time-reversal symmetric Hamiltonian, the Fermi velocity in the x-direction then vanishes such that one must take into account the quadratic order in qx in the energy band. Notice that such hybrid semi-Dirac points, with a linear-parabolic dispersion relation, are unaccessible in graphene because unphysically large strains would be required [32, 42]. However, such points may exist in other physical systems such as cold atoms in optical lattices [34, 36, 46, 47, 48], the quasi-2D organic material (BEDT-TTF)2 I3 [49, 50] or VO2 /TiO2 heterostructures [51].
1.5.2
Tilted Dirac cones
Another aspect of the quinoid-type deformation of the graphene sheet and a consequence of the fact that the Dirac points no longer coincide with the BZ corners K and K of high crystallographic symmetry is the tilt of the Dirac cones. This may be appreciated when expanding the Hamiltonian (1.50) to linear order around the Dirac points kD , instead of an expansion around the point M as in the last subsection. In contrast to the undeformed case (1.43), the diagonal components hk now yield a linear contribution [37] tnnn hkD +q w0 q , in terms of the tilt velocity 2 3 (tnnn a sin 2 + t a sin ) w0x = nnn
and
w0y = 0,
(1.56)
where we have dened arccos(t /2t). The linear model is therefore described by the Hamiltonian,14
Hq = (w0 + wx kx x + wy ky y ),
(1.57)
with the renormalised anisotropic velocities 3 t a 3ta sin and wy = wx = 2
2 1+ . 3
Diagonalising the Hamiltonian (1.57) yields the dispersion relation (q) = w0 q +

2 2 2 2 wx qx + wy qy ,
(1.58)
and one notices that the rst term ( w0 q) breaks indeed the symmetry (q) = (q) in each valley, i.e. it tilts the Dirac cones in the direction opposite to
14 This model may be viewed as the minimal form of the generalised Weyl Hamiltonian (with 0 ) v k , HW = =0,...,3
which is the most general 2 2 matrix Hamiltonian that yields a linear dispersion relation.
Deformed Graphene
27
w0 , as well as the electron-hole symmetry (q) = (q) at the same wave vector.15 Indeed, the linearity in q of the generalised Weyl Hamiltonian (1.57) satises only the symmetry Hq = Hq inside each valley. Furthermore, one notices that the chiral symmetry is preserved even in the presence of the tilt term if one redenes the chirality operator (1.40) as q = wx qx x + wy qy y
2 2 2 2 wx qx + wy qy
which naturally commutes with the Hamiltonian (1.57). The eigenstates of the chirality operator are still given by 1 = 2 1 eiq ,
with tan k wy qy /wx qx , and one notices that these states are also the natural eigenstates of the Hamiltonian (1.57). One nally notices that not all values of the tilt parameter w0 are indeed physical. In order to be able to associate = + to a positive and = to a negative energy state, one must full the condition w0 < 1, in terms of the tilt parameter w0 w0x wx
2
(1.59)
w0y wy
(1.60)
Unless this condition is fullled, the iso-energetic lines are no longer ellipses but hyperbolas. In quinoid-type deformed graphene, the tilt parameter may be evaluated as [37] w0 = 2 tnnn sin 2 t + nnn t sin t 2 (tt t tnnn ) 0.6, t2 nnn (1.61)
where we have used Eqs. (1.48) and (1.49). Even at moderate deformations ( < 0.1), the tilt of the Dirac cones is on the order of 5%, and one may therefore hope to observe the eect, e.g. in angle-resolved photoemission spectroscopy (ARPES) measurements [52] that have been successfully applied to epitaxial graphene [53] and graphitic samples [54]. Notice that the Dirac cones are naturally tilted in (BEDT-TTF)2 I3 [49, 50], where the Dirac points occur at positions of low crystallographic symmetry within the rst BZ.
15 In the absence of the tilt term w q, this is a consequence of the symmetry z H z = 0 H, which is satised both by the eective Hamiltonian (1.29) for tnnn = 0 and the linearised version (1.37) in each valley for undeformed graphene.
28
Chapter 2
Dirac Equation in a Magnetic Field and the Relativistic Quantum Hall Eect
As already mentioned in the introduction, a key experiment in graphene research was the discovery of a particular quantum Hall eect [1, 2], which unveiled the relativistic nature of low-energy electrons in graphene. For a deeper understanding of this eect and as a basis for the following parts, we discuss here relativistic massless 2D fermions in a strong quantising magnetic eld (Sec. 2.1). The limits of the Dirac equation in the treatment of the high-eld properties of graphene are discussed in Sec. 2.2, and we terminate this chapter with a discussion of the relativistic Landau level spectrum in the presence of an in-plane electric eld (Sec. 2.3) and that of deformed graphene (Sec. 2.4).
2.1
Massless 2D Fermions in a Strong Magnetic Field
Remember that, in order to describe free electrons in a magnetic eld, one needs to replace the momentum by its gauge-invariant form [55] p = p + eA(r), (2.1)
where A(r) is the vector potential that generates the magnetic eld B = A(r). This gauge-invariant momentum is proportional to the electron velocity v, which must naturally be gauge-invariant because it is a physical quantity. In the case of electrons on a lattice, the substitution (2.1) is more tricky because of the competition between dierent length scales and the presence of 29
30
Dirac Equation in a Magnetic Field and the RQHE
several bands. Furthermore, the vector potential is unbound, even for a nite magnetic eld; this becomes clear if one chooses a particular gauge, such as e.g. the Landau gauge AL (r) = B(y, 0, 0), in which case the value of the vector potential may become as large as B Ly , where Ly is the macroscopic extension of the system in the y-direction. However, it may be shown that the substitution (2.1), which is called Peierls substitution in the context of electrons on a lattice, remains correct as long as the lattice spacing a is much smaller than the magnetic length , (2.2) eB which is the fundamental length scale in the presence of a magnetic eld. Because a = 0.24 nm and lB 26 nm/ B[T], this condition is fullled in graphene for the magnetic elds, which may be achieved in todays high-eld laboratories ( 45 T in the continuous regime and 80 T in the pulsed regime). With the help of the (Peierls) substitution (2.1), one may thus immediately write down the Hamiltonian for charged particles in a magnetic eld if one knows the Hamiltonian in the absence of the eld, lB = H(p) H() = H(p + eA) = H B (p, r). Notice that because of the spatial dependence of the vector potential, the resulting Hamiltonian is no longer translation invariant, and the (gauge-dependent) momentum p = q is no longer a conserved quantity. For the Dirac Hamiltonian (1.37), which we have derived in the preceding chapter to lowest order in |q|a, the Peierls substitution yields
e, HB = vF (qx x + qy y ) HB = vF (x x + y y ).
(2.3)
We further notice that, because electrons do not only possess a charge but also a spin, each energy level resulting from the diagonalisation of the Hamiltonian (2.3) is split into two spin branches separated by the Zeeman eect Z = gB B, where g is the g-factor of the host material (g 2 for graphene [56]) and B = e /2m0 is the Bohr magneton, in terms of the bare electron mass m0 . However, in the following chapter, we concentrate on the orbital degrees of freedom which yield the characteristic level structure of electrons in a magnetic eld and therefore neglect the spin degree of freedom, i.e. we consider spinless fermions. Eects related to the internal degrees of freedom are discussed in a separate chapter (Chap. 5) in the framework of the quantum-Hall ferromagnet.
2.1.1
Quantum-mechanical treatment
One may easily treat the Hamiltonian (2.3) quantum-mechanically with the help of the standard canonical quantisation [57], according to which the components of the position r = (x, y) and the associated canonical momentum p = (px , py ) satisfy the commutation relations [x, px ] = i , [y, py ] = i and [x, y] = [px , py ] = [x, py ] = [y, px ] = 0. (2.4)
31
As a consequence of these relations, the components of the gauge-invariant momentum no longer commute, and, with the help of the commutator relation [57] df [O1 , O2 ] (2.5) [O1 , f (O2 )] = dO2 between two arbitrary operators, the commutator of which is an operator that commutes itself with both O1 and O2 , one nds [x , y ] = ie Ax Ay x y
2
= i
2 lB
(2.6)
in terms of the magnetic length (2.2). For the quantum-mechanical solution of the Hamiltonian (2.3), it is convenient to use the pair of conjugate operators x and y to introduce ladder operators in the same manner as in the quantum-mechanical treatment of the one-dimensional harmonic oscillator. These ladder operators play the role of a complex gauge-invariant momentum (or velocity), and they read1 lB a = (x iy ) 2 and lB a = (x + iy ) , 2 (2.7)
where we have chosen the appropriate normalisation such as to obtain the usual commutation relation [, a ] = 1. a (2.8) It turns out to be helpful for practical calculations to invert the expression for the ladder operators (2.7), x = a + a 2lB and y = a a . i 2lB (2.9)
2.1.2
Relativistic Landau levels

vF = 2 lB
In terms of the ladder operators (2.7), the Hamiltonian (2.3) becomes

HB = vF
0 x + iy
x iy 0
0 a
a 0
(2.10)
One remarks the occurence of a characteristic frequency = 2vF /lB , which plays the role of the cyclotron frequency in the relativistic case. Notice, however, that this frequency may not be written in the form eB/mb because the band mass is strictly zero in graphene, such that the frequency would diverge.2
1 In order to distinguish the ladder operators a() clearly from the carbon-carbon distance a, we add hats on top of the ladder operators. 2 Sometimes, a cyclotron mass m C is formally introduced via the equality eB/mC . However, this mass is a somewhat articial quantity, which turns out to depend on the carrier density.
32
The eigenvalues and the eigenstates of the Hamiltonian (2.10) are readily obtained by solving the eigenvalue equation HB n = n n , in terms of the 2-spinors, un . n = vn We thus need to solve the system of equations a vn = n un which yields the equation a a vn = n
2
and
a un = n vn ,
(2.11)
vn
(2.12)
for the second spinor component. One may therefore identify, up to a numerical factor, the second spinor component vn with the eigenstate |n of the usual number operator a a, with a a|n = n|n in terms of the integer n 0. Further more, one observes that the square of the energy is proportional to this quantum number, 2 = ( )2 n. This equation has two solutions, a positive and a negn ative one, and one needs to introduce another quantum number = , which labels the states of positive and negative energy, respectively. This quantum number plays the same role as the band index ( = + for the conduction and = for the valence band) in the zero-B-eld case discussed in the preceding chapter. One thus obtains the spectrum [58] ,n = vF 2n lB (2.13)
of relativistic Landau levels (LLs), which disperse as Bn as a function of the magnetic eld (see Fig. 2.1). Notice that, as in the B = 0 case, the level spectrum is independent of the valley index and therefore two-fold valleydegenerate. Once we know the second spinor component, the rst component is obtained from Eq. (2.11), which reads un a vn a|n |n 1 because of the usual equations and a|n = n|n 1 (2.14) a |n = n + 1|n + 1 for the ladder operators, where the last equation is valid for n > 0. One then needs to distinguish the zero-energy LL (n = 0) from all other levels. Indeed, for n = 0, the rst component is zero because a|n = 0 = 0 In this case one obtains the spinor n=0 = 0 |n = 0 . (2.16) (2.15)
33
energy
+,n=4 +,n=3 +,n=2 +,n=1
2 3 magnetic
field 4 B
n=0
-2
,n=1 ,n=2 ,n=3 ,n=4
-4
Figure 2.1:
Relativistic Landau levels as a function of the magnetic eld.
In all other cases (n = 0), one has positive and negative energy solutions, which dier among each other by a relative sign in one of the components. A convenient representation of the associated spinors is given by 1 ,n=0 = 2 |n 1 |n
(2.17)
The particular form of the n = 0 spinor (2.16) associated with zero-energy states merits a more detailed comment. One notices that only the second spinor component is non-zero. Remember that this component corresponds to the Bsublattice in the K-valley ( = +) and to the A-sublattice in the K -valley ( = ) the valley isospin therefore coincides with the sublattice isospin, and the two sublattices are decoupled at zero energy. Notice that this is also the case in the absence of a magnetic eld, where the relation (1.30) between the chirality, the band index and the valley isospin is only valid at non-zero values of the wave vector, i.e. not exactly at zero energy. Indeed, the chirality can no longer be dened as the projection of the sublattice isospin on the direction of propagation q/|q|, which is singular at q = 0. It is therefore useful to identify the chirality with the valley isospin. Notice, however, that this particularity concerns, in the absence of a magnetic eld, only a non-extensive number of states (only two) because of the vanishing density of states at zero energy, whereas the zero-energy LL n = 0 is macroscopically degenerate, as discussed in the following paragraphs.
34 LL degeneracy
A particular feature of LLs, both relativistic and non-relativistic ones, consists of their large degeneracy, which equals the number of ux quanta NB = A B/(h/e) threading the 2D surface A occupied by the electron gas. From the classical point of view, this degeneracy is related to the existence of a constant of motion, namely the position of the guiding centre, i.e. the centre of the classical cyclotron motion. Indeed, due to translation invariance in a uniform magnetic eld, the energy of an electron does not depend on the position of this guiding centre. Translated to quantum mechanics, this means that the operator corresponding to this guiding centre R = (X, Y ) commutes with the Hamiltonian H(p + eA) which corresponds to the kinetic energy. In order to understand how the LL degeneracy is related to the guidingcentre operator, we formally decompose the position operator r=R+ (2.18)
into its guiding centre R and the cyclotron variable = (x , y ). Whereas the guiding centre is a constant of motion, as mentioned above, the cyclotron variable describes the dynamics of the electron in a magnetic eld and is, classically, the time-dependent component of the position. Indeed, the cyclotron variable is perpendicular to the electrons velocity and thus related to the gauge-invariant momentum by [57] x = y eB and y = x , eB (2.19)
which, as a consequence of the commutation relations (2.6), satisfy [x , y ] = [x , y ] 2 = ilB , (eB)2 (2.20)
whereas they commute naturally with the guiding-centre components X and Y . Equation (2.20) thus induces the commutation relation
2 [X, Y ] = [x , y ] = ilB ,
(2.21)
in order to satisfy [x, y] = 0. These commutation relations indicate that the components of the guidingcentre operator form a pair of conjugate variables, and one may introduce, in the same manner as for the gauge-invariant momentum operator , the ladder operators = 1 (X + iY ) b 2lB and = 1 (X iY ), b 2lB (2.22)
which again satisfy the usual commutation relations [ ] = 1 and which natub, b rally commute with the Hamiltonian. One may then introduce a number operator b b associated with these ladder operators, the eigenstates of which satisfy the eigenvalue equation b b|m = m|m .
35
One thus obtains a second quantum number, an integer m 0, which is necessary to describe the full quantum states in addition to the LL quantum number n, and the completed quantum states (2.16) and (2.17) then read
n=0,m = n=0 |m =
0 |n = 0, m |n 1, m |n, m ,
(2.23)
and 1 n,m = n |m = 2 (2.24)
respectively. One may furthermore use the commutation relation (2.21) for counting the number of states, i.e. the degeneracy, in each LL. Indeed, this relation indicates that one may not measure both components of the guiding centre simultaneously, which is therefore smeared out over a surface
2 XY = 2lB .
(2.25)
This minimal surface plays the same role as the surface (action) h in phase space and therefore allows us to count the number of possible quantum states of a given (macroscopic) surface A, NB = A A = 2 = nB A, XY 2lB
where we have introduced the ux density nB = B 1 2 = h/e , 2lB (2.26)
which is nothing other than the magnetic eld measured in units of the ux quantum h/e, as already mentioned above. The ratio between the electronic density nel and this ux density then denes the lling factor = nel hnel = , nB eB (2.27)
which characterises the lling of the dierent LLs. The relativistic quantum Hall eect The integer quantum Hall eect (IQHE) in 2D electron systems [59] is a manifestation of the LL quantisation and the macroscopic degeneracy (2.26) of each level. In a nutshell, this energy quantisation yields a quantisation of the Hall resistance h (2.28) RH = 2 , e N
36
where N = [] is the integer part of the lling factor (2.27), while the longitudinal resistance vanishes.3 The resistance quantisation reects the presence of an incompressible quantum liquid with gapped single-particle and density excitations. In the case of the IQHE, at integer lling factors, the gap is simply given by the energy dierence between adjacent LLs which must be overcome by an electron that one adds to the system. Notice that if one takes into account the electron spin and a vanishing Zeeman eect, the condition for the occurence of the IQHE is satised when both spin branches of the last LL n are completely lled, and one thus obtains the Hall-resistance quantisation at the lling factors IQHE = 2n, (2.29)
i.e. for even integers. Odd integers may principally be observed at higher magnetic elds when the Zeeman eect becomes prominent, and the energy gap is then no longer given by the inter-LL spacing but by the Zeeman gap. This picture is naturally simplistic and needs to be modied if one takes into account electronic interactions their consequences, such as the fractional quantum Hall eect or ferromagnetic states are discussed, in the context of graphene, in Chap. 5. The phenomenology of the relativistic quantum Hall eect (RQHE) in graphene is quite similar to that of the IQHE. Notice, however, that one is confronted not only with the two-fold spin degeneracy of electrons in graphene (in the absence of a strong Zeeman eect), but also with the two-fold valley degeneracy due to the presence of the K and K points in the rst BZ, which govern the low-energy electronic properties. The lling factor therefore changes by steps of 4 between adjacent plateaus in the Hall resistance. Furthermore, the lling factor (2.27) is dened in terms of the carrier density which vanishes at the Dirac point. This particle-hole symmetric situation naturally corresponds to a half-lled zero-energy LL n = 0, whereas all levels with = are completely lled and all = + levels are unoccupied. In the absence of a Zeeman eect and electronic interactions, there is thus no quantum Hall eect at = 0, and the condition of a completely lled (or empty) n = 0 LL is found for = 2 ( = 2). As a consequence, the signature of the RQHE is a Hall-resistance quantisation at the lling factors [61, 62, 63] RQHE = 2(2n + 1), (2.30)
which needs to be contrasted to the series (2.29) of the IQHE in non-relativistic 2D electron systems. The series (2.30) has indeed been observed in 2005 within the quantum Hall measurements [1, 2], which thus revealed the relativistic character of electrons in exfoliated graphene. More recently, the RQHE has been observed also in epitaxial graphene [20, 21, 22] obtained from graphitisation of the Si-terminated surface of the SiC crystal.
3 A simultaneous measurement of the Hall and the longitudinal resistance requires a particular geometry with at least four electric contacts (for a recent review on the quantum Hall eect, see Ref. [60]).
37
(a)
relative transmission
Relative transmission
1.00
(C) (B) (D)
(b)
relative transmission
A
E1 E1
Relative transmission
10
20
30
40
50
60
70
80
90
E
0.98
1.00
L3 L2
L1
~ E1 c 2e B
(A)
B D C
0.2T
L0
B
1 2
3
0.96 0.4 T 1.9 K
L L
L
transition C
0.3T 0.99 0.5T 0.7T
10
20
30
40
50
60
70
80
(c)
transmission energy [meV]
80 70
nergie [meV]
Energy (meV)
L L
3 2
L L
2 3
( (
D) D)
1.00
L L
L L
C) (C )
(
0.98 0.96 0.94 0.92 1 T 0.90 0.4T
Transition energy (meV)
60
50
0
1
1
0
B)
B)
40
30
20
A)
0.88 0.86
transition B
20 30 40
2T 4T
10
0 0.0 0.5 1.0
10
50
60
70
80
90
sqrt(B)
1/2
1.5
2.0
energy [meV]
Energy (meV)
LL spectroscopy in graphene (from Sadowski Ref. [64]). (a) For a xed magnetic eld (0.4 T), one observes resonances in the transmission spectrum as a function of the irradiation energy. The resonances are associated with allowed dipolar transitions between relativistic LLs. (b) These resonances are shifted as a function of the magnetic eld. If (c) one plots the resonance energies as a function of the square root of the magnetic eld, B, a linear dependence is observed as one would expect for relativistic LLs.
Figure 2.2:
Experimental observation of relativistic Landau levels The Bn dispersion of relativistic LLs has been observed experimentally in transmission spectroscopy, where one shines monochromatic light on the sample and measures the intensity of the transmitted light. Such experiments have been performed both on epitaxial [64] and exfoliated graphene [65]. When the monochromatic light is in resonance with a dipole-allowed transition from the (partially) lled (, n) to the (partially) unoccupied LL ( , n 1), it is absorbed due to an electronic excitation between the two levels [see Fig. 2.2(a)]. Notice that, in a non-relativistic 2D electron gas, the only allowed dipolar transition is that from the last occupied LL n to the rst unoccupied one n + 1. The transition energy is C , independently of n, and one therefore observes a single absorbtion line (cyclotron resonance) that is robust to electron-electron interactions, as a consequence of Kohns theorem [66]. In graphene, however, there are many more allowed transitions due to the presence of two electronic bands, the conduction and the valence band, and the
38
transitions have the energies n, = v lB 2(n + 1) 2n , (2.31)
where = + denotes an intraband and = an interband transition [64, 67, 68]. One therefore obtains families of resonances the energy of which disperses as n, B, as it has been observed in the experiments (see Fig. 2.2(c), where we show the results from Sadowski et al. [64]). Notice that the dashed lines in Fig. 2.2(c) are ts with a single tting parameter (the Fermi velocity v), which matches well all experimental points for dierent values of n.
2.2
Limits of the Dirac Equation in the Description of Graphene Landau Levels
Transmission spectroscopy is an ideal tool for the study of the high-energy part of the LL spectrum when considering the transitions ( = , n) ( = +, n 1), for n 1. As discussed in Sec. 1.4.2, one expects deviations [of order O(|q|2 a2 )] from the linear dispersion in this limit. These deviations renormalise the energy of the LLs and thus the transition energies. In order to quantify the eect [69], we may use the Peierls substitution (2.1) and the expressions (2.9) in the terms (1.43) and (1.44) corresponding to the higher-order diagonal and o-diagonal band terms, respectively. This yields the Hamiltonian h h , (2.32) HB = h h where the diagonal elements read 3tnnn a h = a a, 2tlB and the o-diagonal ones are
2 aw1 2 a2 w2 2 a h = a 2 a a . 4lB 2 2lB
(2.33)
(2.34)
Naturally, to lowest order in a/lB , one obtains again the Hamiltonian (2.10). The dimensionless parameters w1 and w2 are articially added to the expressions and play the role of tting parameters in the comparison with experimental measurements, as will be discussed below. They measure the deviation from the tight-binding-model expectation, w1 = w2 = 1. Notice furthermore that, since we are interested in the n 1 limit, we do not care about corrections related to the ordering of the ladder operators. In the calculation of the LL spectrum, one may proceed in the same manner as in Sec. 2.1.2 the eigenvalue equation (2.12) for the second spinor component now becomes h h vn = (n h )2 vn . (2.35)
Limits of the Dirac Equation in the Description of Graphene LLs
39
Here, the combination H h h is interpreted as some fake Hamiltonian which needs to be diagonalised in order to obtain the modied LLs. Notice that n remains a good quantum number if one considers h on the right-hand side of the eigenvalue equation. The left-hand side consists of a term
2 2 4w2 w1 H0 = ( )2 a a 8
a lB
a a
that contains powers of a a and thus respects the quantum number n, but in addition it contains the trigonal-warping term w1 ( )2 3 ( + a3 ), a Ht.w. = 2 2 (2.36)
which does not commute with a a and which needs to be treated apart. If we neglect this trigonal-warping term for a moment, the LL energies are obtained from the quadratic equation ( )2 n
2 2 4w2 w1 8
a lB
n2 =
3tnnn a n n 2tlB
(2.37)
In order to account for the trigonal-warping term in the eigenvalue equation (2.35), we may use a perturbative treatment, which is justied because of the small parameter a/lB . There is no contribution at rst order since n|()3 |n = a 0 due to the orthogonality of the eigenstates n|n = n,n . At second order, one obtains 2 ( )2 a n = [3n(n + 1) + 2] , 8 lB which needs to be added to the right-hand side in Eq. (2.37). Interestingly, trigonal warping thus yields the same correction to the energies of the relativistic LLs as the third-order term in the expansion of the band dispersion, although trigonal warping occurs at second order in the absence of a magnetic eld, as we have discussed in Sec. 1.4.2. This eect is due to the anisotropy of the band correction; in the presence of a magnetic eld, the cos(3q ) term in Eq. (1.45) is averaged over the angle q , and therefore only contributes at second order in the perturbation theory described above. This eventually yields a correction of order (a/lB )2 to the LL energy, as does the third-order term in the correction of the band dispersion. One nally obtains, in the large-n limit, where these corrections become relevant, the energies of the relativistic LLs [69] n = 3w2 vF vF 3tnnn a 2n 1 n+ lB t lB lB 8 a lB
2
n [1 + O(1/n)] , (2.38)
independent of the valley index , where O(1/n) stands for corrections of order 1/n. Furthermore, we have dened the eective dimensionless parameter w2
40
(a)
1
(b)
0.5
energy
-0.5
-1 0 0.005 0.01 flux per hexagon 0.015 0.02
Higher-order corrections to the LL quantisation in graphene. (a) Comparison with numerical calculations obtained from the lattice model (curtesy of P. Dietl). The energy (arbitrary units) is shown as a function of the ux threading each hexagon. The green line shows the energy levels (2.13) calculated from the Dirac equation, whereas the red curve represents the LLs (2.38) with higher-order band corrections. (b) Transmission spectroscopy on epitaxial multilayer graphene [69]. The inset shows a representative transmission spectrum. The main gure represents the positions of the absorption lines as a function of (the square root of) the magnetic eld. The dashed lines correspond to transitions calculated at linear order.
Figure 2.3:
LL Spectrum in the Presence of an Inplane Electric Field
41
2 2 (w1 + 2w2 )/3, which may be evaluated from a comparison with spectroscopic measurements. Equation (2.38) generalises a calculation for the relativistic LLs when only nnn hopping is taken into account [63]. Figure (2.3)(a) shows a comparison of the LL energies (2.38) with numerical calculations on the honeycomb-lattice model. At (unphysically) large magnetic elds, the LL spectrum evolves into the so-called Hofstadter buttery when the energy is plotted as a function of the magnetic ux through a hexagon in units of the ux quantum [70, 71]. Whereas already at moderate magnetic elds, one notices a discrepancy between the numerical results and the relativistic LLs (green curve) calculated from the linear Dirac equation [Eq. (2.13)], the 2 expression (2.38) which takes into account corrections on the order O(a2 n/lB ) remains valid over a larger range of (moderate) magnetic elds (red curve). In Fig. (2.3), we show experimental results obtained from high-eld transmission spectroscopy on multi-layer epitaxial graphene [69]. Qualitatively, one notices a downward renormalisation of the transition energies
n = =+,n =,n in the interband regime for large values of n. Notice that because transmission spectroscopy is sensitive to energy-level dierences, the nnn correction in < Eq. (2.38) yields only a correction on the order of (tnnn /t)(a/lB )/n 1%/n at B 25 T as compared to the energy scale t(a/lB )n of the transition, whereas the other term yields a correction on the order of (a/lB )2 n 0.5% n. The latter corrections thus become more relevant in the large-n limit than the nnn correction. Indeed, the experiment [69] was not capable of probing the electronhole symmetry breaking associated with the nnn term, whereas a quantitative study of the high-energy transitions revealed a good semi-quantitative agreement with the calculated LL spectrum (2.38). However, it has been shown that the simple-minded tight-binding approach (with w = 1) underestimates the higher-order band corrections and that the best t to Eq. (2.38) is obtained for a value of w = 2.8. The origin of this discrepancy is yet unexplained, and it may be interesting to perform high-eld transmission spectroscopy measurement also on single-layer exfoliated graphene in order to understand whether the stronger downward renormalisation of the LLs is due to interlayer couplings in the epitaxial multi-layer sample.
2.3
Landau Level Spectrum in the Presence of an Inplane Electric Field
A remarkable consequence of the relativistic character of electrons in graphene and the Lorentz invariance of the Dirac equation is their behaviour in crossed magnetic and electric elds, where the magnetic eld remains perpendicular to the graphene sheet and the electric eld is applied in the plane. Remember that in a non-relativistic 2D electron systems, the electric eld E = Eey (in the y-direction) simply lifts the LL degeneracy and adds a term (E/B)k to the LL
42
energies, where k is the wave vector in the x-direction. At a xed wave vector k, the LL spacing is unaected by the in-plane eld. The situation is dierent for relativistic electrons in graphene, where as a consequence of the Lorentz invariance of the Dirac equation, one may choose a reference frame in which the electric eld vanishes as long as the drift velocity vD = E/B is smaller than the Fermi velocity, which plays the role of an upper bound for the physically signicant velocities in the same manner as the velocity of light in relativity (magnetic regime).4 In addition to the lifted LL degeneracy, the LL spacing is reduced [72, 73], as may be seen from a Lorentz boost into the reference frame which moves at the drift velocity and in which the electric eld vanishes. In this reference frame, the magnetic eld is reduced by the factor 1 (E/vF B)2 , such that the LLs (2.13), which scale as magnetic eld, read = ,n (2.39)
B = B[1 (E/vF B)2 ]1/4 with the
vF [1 (E/vF B)2 ]1/4 2n, lB
where the primes indicate the physical quantities in the moving frame of reference. When measuring the energy in the original (lab) frame of reference, the above energy spectrum also needs to be transformed into this frame of reference, which amounts to being multiplied by another factor (2.39), such that the spectrum of relativistic LLs in the presence of an in-plane electric eld becomes [72] E vF [1 (E/vF B)2 ]3/4 2n + k. (2.40) ,n;k = lB B The quantum-mechanical derivation of this result will be discussed in more detail in Sec. 2.4.1 in the context of the generalised Weyl Hamiltonian in a magnetic eld.
2.4
Landau Levels in Deformed Graphene
As we have discussed in Sec. 1.5, a uniaxial strain deforms the graphene sheet and modies the electronic structure. The induced anisotropy of the Fermi velocity wx = wy is essentially washed out by the magnetic eld, which yields an eective averaging over the Fermi surface, vF vF = wx wy . More spectacular are the two further consequences of the deformation; (a) the tilt of the Dirac cones accounted for in the generalised Weyl Hamiltonian (1.57) and (b) the topological phase transition due to the Dirac point motion. The implication for the LL spectrum are briey reviewed in the following sections.
4 In the opposite case, v D > vF , one may choose a reference system in which the magnetic eld vanishes (electric regime) [55].
LLs in Deformed Graphene
43
2.4.1
The generalised Weyl Hamiltonian in a magnetic eld
With the help of the Peierls substitution (2.1) and the expression of the gaugeinvariant momentum in terms of ladder operators (2.9), the generalised Weyl Hamiltonian (1.57) may be cast into the form
HB =
2wx wy lB
w0 a i 2 (e
+ a ei ) a
w0 a i 2 ( e
a + a ei )
(2.41)
where w0 ei
w0y w0x +i , wx wy
in terms of the eective tilt parameter (1.60) and the angle between the xaxis and the direction of the eective tilt (w0x /wx , w0y /wy ), renormalized by the Fermi velocities wx and wy in the x- and y-direction, respectively. The Hamiltonian (2.41) may be solved quantum-mechanically in a straightforward, but lengthy manner [73, 74]. Instead, one may also obtain the result in a simpler semi-classical treatment [37], with the help of the Onsager relation [75, 76] according to which the surface S() enclosed by a trajectory of constant energy in reciprocal space is quantized as
2 S()lB = (2)2 0
d ( ) = 2(n + ),
where n is an integer denoting the energy level which coincides with the Landau level in the full quantum treatment. The additional contribution is related to a Berry phase acquired by an electron during its cyclotron orbit. Usually, one has = 1/2 except if there is an extra Berry phase of , which in our case yields = 0, as in the case of graphene with no tilt [77]. If one considers a density of states which scales as () , the energy levels thus scale as n [B(n + )]1/(1+) , (2.42)
in the large-n limit. Because the density of states vanishes linearly at the Dirac point, as in the case of no tilt, i.e. = 0, the scaling argument (2.42) yields the energy levels, v ,n 2 F n, lB (2.43)
as for unconstrained graphene, apart from a renormalisation of the Fermi velocity. The renormalisation of the Fermi velocity is readily obtained from the calculation of the total number of states below the energy within the positive energy cone is given by N+ () = 1 (2)2
2w w x y + () q
dx dy = q q
2 , 2 2 2 vF 2 1
(2.44)
44
where we have dened qx/y wx/y qx/y , and the renormalised Fermi velocity is written in integral form, 1 1 2 = w w vF x y
2 0
1 d = wx wy (1 w0 )3/2 2 2 (1 + w0 cos )2
(2.45)
in terms of the eective tilt parameter (1.60). This yields the result wx wy ,n = (1 w0 )3/4 2n 2 lB
(2.46)
for the LL spectrum in the presence of a tilt, which coincides with the one obtained from the full quantum treatment [73, 74].5 One nally notices that the LL spacing becomes zero for w0 = 1, which corresponds to the condition (1.59) of maximal tilt for the Dirac cones, as discussed in Sec. 1.5 indeed for values of w0 larger than 1, the isoenergetic lines are no longer closed elliptic orbits but open hyperbolas, for which the energy is not quantised. Tilted Dirac cones in a crossed magnetic and electric eld One notices that the form (2.46) of LLs for tilted Dirac cones is the same as that of the LL spectrum (2.40) if one interprets the drift velocity vD = E/vF B as an eective electric-eld induced tilt. The magnetic regime E/B < vF corresponds then to the regime of closed orbits (w0 < 1) and the open hyperbolic orbits may be identied with the electric regime E/B > vF . Mathematically, the generalised Weyl Hamiltonian with an in-plane electric eld may still be cast into the form (2.41) E HB HE/B = HB + k , B
where HB is the same as that of Eq. (2.41) if one replaces the tilt parameter w0 exp(i) by [78] wy wx +i . (2.47) w (E)ei (E) wx wy
Here, the renormalized tilt velocity is given by w = (wx , wy ) w0 EB , B2 (2.48)
and the angle is the angle between this velocity and the x-axis. The resulting energy spectrum is given by wx wy E 3/4 (E) = 2n + 1 w (E)2 k. ,n;k lB B
(2.49)
5 The semi-classical calculation needs to be handled with care in the case of the zero-energy level n = 0, where a full quantum treatment is required. One may show, however, that the zero-energy level is protected by space-inversion symmetry, HB (r, ) = HB (r, ), which yields a particle-hole symmetric LL spectrum [37]. This is noteworthy because the generalised Weyl Hamiltonian (1.57) in the absence of a magnetic eld does not respect particle-hole symmetry.
LLs in Deformed Graphene

Energy [a.u.]
45
+,3 +,2 +,1
+,4 +,3 +,2 +,1 0 ,1 ,2 ,3 ,4 =
0 ,1 ,2 ,3 =+
Figure 2.4: Sketch of the valley-dependent LL spectrum for tilted Dirac cones in the presence of an electric eld (thick lines). We omitted the inclination of the LLs due to the term (E/B)k, which lifts the LL degeneracy. The dashed lines schematically represent the tilted cones in the two valleys = + and in the absence of an electric eld. The cones in the two valleys are tilted in opposite directions in the momentum-energy space, whereas the electric eld acts in the same direction. The LL spectrum in the presence of an electric eld in = + ( = ) is that of a cone with a decreased (increased) tilt (full lines). Naturally one obtains the result (2.40) for undeformed graphene in an in-plane electric eld, for wx = wy = vF and w0 = 0, as well as the LL spectrum (2.46) for the generalised Weyl Hamiltonian with tilted Dirac cones for zero in-plane eld (E = 0). However, the most interesting situation arises when both the tilt and an in-plane eld are present, in which case one observes a lifting of the valley degeneracy that is maximal when the electric eld is applied perpendicular to the tilt velocity, E w0 [78]. The resulting LL spectrum is schematically depicted in Fig. 2.4, and one notices that the electric eld redresses the cone in one valley whereas it enhances the tilt in the other one. Notice that, in order to obtain an eect on the order of 1%, extremely large electric elds would be required (on the order of 106 V/m) for a 10% deformation of the lattice [78]. It seems therefore dicult to observe the eect in graphene, e.g. in high-eld transmission spectroscopy or transport measurments, whereas the eect may be more visible in (BEDT-TTF)2 I3 , where the Dirac cones are naturally tilted [49, 50] and where lower electric elds would be required for a comparable eect due to a roughly ten times smaller eective Fermi velocity.
2.4.2
LL spectrum in the vicinity of the topological phase transition
We terminate the chapter with a short discussion of the LL spectrum in the vicinity of the topological phase transition when varying one of the hopping parameters t with respect to the other two, as described in Sec. 1.5.1. The spectrum may be obtained from the eective Hamiltonian (1.55) with the help
46
Figure 2.5: From Ref. [43]. LL spectrum as a function of the dimensionless gap parameter , in units of ( 2 c2 e2 B 2 /2m )1/3 . The blue dots on the = 0 axis indicate the values at the topological phase transition obtained in Ref. [35]. of the Peierls substitution (2.1). The resulting LLs are plotted in Fig. 2.5 as a function of the dimensionless gap parameter = /( 2 c2 e2 B 2 /2m )1/3 , in terms of the parameters introduced in Sec. 1.5.1 [43, 44]. In the semi-metallic phase < 0, one retrieves the typical Bn low-energy behaviour of the relativistic LLs (2.13). The two-fold valley degeneracy is lifted at larger energies when the LL energies become comparable to that of the van Hove singularity at the M points in the rst BZ. Because the energy corresponding to the van Hove singularity is lowered when approaching the topological phase transition, the valley degeneracy of the LLs is lifted at even lower energies (see Fig. 2.5). At the transition for = 0, the valley degeneracy is completely lifted even in the n = 0 LL. As a consequence of the linear-parabolic dispersion relation, in the absence of a magnetic eld, the density of states varies as , and the scaling argument (2.42) then yields (with = 1/2) a behaviour ,n [B(n + 1/2)]2/3 (2.50)
for the LLs at the phase transition [35]. The contribution = 1/2 arises from the annihilation of the Berry phases associated with the two valleys = due to the merging of the Dirac points [44]. Above the transition ( > 0), one obtains an insulating phase with parabolic bands, and the LL spectrum retrieves therefore the usual non-relativistic =+,n B(n + 1/2) behaviour for the conduction band and =,n + B(n + 1/2) for the valence band, where the oset reects simply the band gap between the two energy bands.
Chapter 3
Electronic Interactions in Graphene Integer Quantum Hall Regime

In the preceding chapters, we have discussed the electronic properties of graphene within a one-particle model, i.e. we have neglected the Coulomb interaction between the electrons. In many materials, the one-particle picture yields the correct qualitative description of the electronic properties and is modied only quantitatively if one includes the electron-electron interactions within perturbation theory [79, 80]. Notice, however, that there exists a class of materials strongly correlated electron systems the electronic properties of which may not be described correctly, not even on the qualitative level, within a one-particle picture. In order to quantify the role of the electronic interactions, i.e. the correlations, in graphene, one needs to compare the characteristic Coulomb energy Eint = e2 / at the average inter-electronic distance ( is the dielectric constant describing the environment the graphene sheet is embedded in) to the kinetic one Ekin at the same length scale,1 rs = Eint . Ekin (3.1)
If this dimensionless interaction parameter becomes very large, rs 1, the electrons are strongly correlated. In non-relativistic 2D metals with a parabolic band dispersion, the dimensionless parameter reads rs = me2 , 2 a0 (3.2)
1 The kinetic energy is then the Fermi energy, i.e. the kinetic energy at the Fermi wave vector kF 1/.
47
48
Electronic Interactions in Graphene IQHE Regime
in terms of the eective Bohr radius a = a0 m0 /m, where a0 = 0.5 A is the 0 Bohr radius in vacuum and m/m0 the ratio between the band and the bare electron mass. The relevance of electronic correlations therefore increases in the dilute limit when a . This result seems quite counterintuitive at the 0 rst sight because the Coulomb interaction becomes weaker in the dilute limite. However, the kinetic energy decreases faster than the Coulomb energy, such that the latter becomes relatively more relevant. The same argument applied to graphene yields a completely dierent result. Whereas the scaling of the Coulomb energy remains the same, that of the kinetic energy is changed due to the linearity of the band dispersion. As a consequence the dimensionless interaction parameter in graphene reads
G rs =
2.2 e2 , vF
(3.3)
independent of the carrier density. The correlations are therefore in an intermediate regime but may be decreased if the graphene sheet is embedded in an environment with a large dielectric constant. Notice that the expression (3.3) is the same as that of the ne structure constant = e2 / c = 1/137 in quantum electrodynamics [29] if one replaces the Fermi velocity by the velocity of G light, which is roughly 300 times larger. One therefore calls rs alternatively the graphene ne structure constant.
Long-range versus short-range interactions

Another important aspect of interacting electrons is the range of the interaction potential. Whereas the underlying Coulomb potential e2 /r is long-range, shortrange interaction models such as the Hubbard model are often and successfully used in the description of correlated metals. The use of such a short-range interaction may be justied by the screening properties of interacting electrons, which are correctly captured in a Thomas-Fermi approach [79, 80] according to which the Coulomb interaction potential is screened above a characteristic screening length T F 1/kT F . In 2D, the Thomas-Fermi wave vector kT F rs kF (3.4) is given in terms of the dimensionless interaction parameter (3.1) and the Fermi wave vector kF .2 One notices that, for metals with a parabolic dispersion relation, the ThomasFermi wave vector is simply given in terms of the inverse eective Bohr radius, kT F 1/a , independent of the electronic density. Unless the band mass is very 0 small as compared to the bare electron mass or the dielectric constant of the host material very large, the Coulomb interaction is therefore screened on the atomic scale. One may then describe the electronic correlations in the framework of short-range interaction models, such as the Hubbard model. Typical examples are heavy-fermion compounds (for a review see Ref. [81]).
2 In 2 2 three space dimensions, the relation reads kT F rs kF .
Decomposition of the Coulomb interaction in the Two-Spinor Basis
49
The situation is again drastically dierent in graphene where the ThomasFermi wave vector (3.4) becomes
G kT F
2.2 kF nel ,
(3.5)
i.e. it vanishes at the Dirac points where the carrier density goes to zero, and the screening length then diverges.3 Notice that even for doped graphene, where one may typically induce carrier densities on the order of 1012 cm2 , the screening > length is T F 10 nm, i.e. much larger than the lattice scale. One thus comes to the conclusion that the relevant electronic interactions in graphene are long-range Coulomb interactions that may not be captured, in contrast to other materials with a parabolic band dispersion, within models such as the Hubbard model [83, 84]. We therefore investigate, in this chapter, the fate of the long-range Coulomb interaction in a strong magnetic eld. In Sec. 3.1, we decompose the Coulomb interaction Hamiltonian in the two-spinor basis of the low-energy electronic wave functions in graphene and comment on its symmetry with respect to the valley isospin. The role of these interactions in the particle-hole excitation spectrum is studied in Sec. 3.2, where we discuss the resulting collective excitations in the IQHE regime, which allows for perturbative treatment. The strong-correlation regime of partially lled LLs (regime of the fractional quantum Hall eect) is presented separately in Chap. 5.
3.1
Generally, the Coulomb interactions for 2D electrons may be accounted for by the Hamiltonian 1 v(q)(q)(q), (3.6) Hint = 2 q in terms of the Fourier components (q) = d2 r exp(iq r) (r)(r) of the electronic density (r)(r) and the 2D Fourier transformation of the 1/r Coulomb potential, v(q) = 2e2 /|q|.4 The density operators may be decomposed in the basis of the spinor wave functions (2.23) and (2.24) for relativistic electrons in graphene, (q) =
n,m; n ,m ;
3 Due to this divergence of the screening length, one principally needs to describe screening beyond the level of linear-response theory [82]. 4 If one takes into account the electronic spin =, , the Coulomb interaction respects the associated SU(2) symmetry, and the Fourier components are then simply the sum of the densities (q) in both spin orientations, (q) = (q) + (q). For notational convenience, we neglect the spin index in the following discussion keeping in mind that the spin SU(2) symmetry is respected.
n,m; eiqr n ,m ; c n,m; c n ,m ; ,
(3.7)
50
()
where cn,m; are fermion operators in second quantisation that annihilate (create) an electron in the quantum state5 n,m;=+ and n,m;= = = 1 |n 1, m n 2 |n, m n eiKr eiKr . (3.8)
2 |n, m n 1 |n 1, m n
Contrary to the expressions (2.23) and (2.24), the state (3.8) is valid for both n = 0 and n = 0 by using the short-hand notation 1 (1 n,0 )/2 and n 2 (1 + n,0 )/2. Furthermore, we have explicitly taken into account the n rapidly oscillating part exp(iK r) due to the two dierent valleys, whereas the expressions (2.23) and (2.24) are only concerned with the slowly varying envelope function. Explicitly, the Fourier components of the density operator (3.7) then read , (q) = Fn, n (q), n (q), n, (3.9)
n, n ,
in terms of the reduced density operators , n (q) = n,

m,m
m ei[q+( )K]R m c n,m; c n ,m ; ,
(3.10)
which may also be interpreted as magneto-exciton operators associated with a particular inter-LL transition (see Sec. 4.1), and the form factors Fn, n (q)
, Fn, n (q)
(3.11)
iq
= for intra-valley and

+, Fn, n (q)
1 1 n n
n1 e
n 1 +
2 2 n n
n e
iq
1 2 n ei(q+2K) n 1 n n
,+ 1 2 n 1 ei(q+2K) n = F n ,n (q) n n
(3.12)
for inter-valley processes. Here, we have used the decomposition r = R + of the position operator into its guiding centre and cyclotron coordinate (see Sec. 2.1.1) and the fact that f1 ()f2 (R)|n, m = f1 ()|n f2 (R)|m , for two arbitrary functions f1 and f2 . The full expressions for the matrix elements in Eqs. (3.10) and (3.12) may be found in Appendix 6. In terms of the reduced density operators (3.10), the interaction Hamiltonian (3.6) reads Hint = 1 2
4 2 n4 1 n3 v1 ......4 n4 (q)1 ,1 ,3 n3 (q)2 ,2 ,4 n4 (q), 1 n1 q 1 n1 ...4 n4 1 ...4
(3.13)
5 In order to avoid confusion in the case of inter-valley coupling, we use a representation in which the rst spinor component represents the amplitude on the A sublattice and the second on the B sublattice for both valleys.

(a)
3 , 1 4 , 2
51
4 ,
(b)
3 ,
1 , 1
2 , 2 4 ,
1 ,
2 , 4 , 2
(c)
3 ,
(d)
3 , 1
1 ,
2 ,
1 , 1
2 , 2
Figure 3.1:
1 1
Diagrammatic representation of the interaction vertex (we use the short-hand notation i = (i ni , mi ) for the quantum numbers); (a) vertex associated with terms of the

,, , ,, , ,,, form v n ... n (q) or v n ... n (q), (b) vertex of Umklapp type, v n ... n (q), (c) ,,, vertex of backscattering type, v n ... n (q) and (d) vertex respecting the SU(2) valley,, , isospin symmetry v n ... n (q).
1 1 4 4 1 1 4 4 4 4 1 1 4 4 1 1 4 4
where the interaction vertex is dened as

4 v1 ......4 n4 (q) = 1 n1
2e2 1 ,3 4 (q)F2 ,2 ,4 n4 (q). F 2n |q| 1 n1 ,3 n3
(3.14)
3.1.1
SU(2) valley symmetry
One notices that, in contrast to the SU(2) symmetry associated with the physical spin, the Hamiltonian (3.13) does not respect a similar valley-isospin symmetry due to possible intervalley couplings. An SU(2) valley-isospin symmetry would be respected for the case 1 = 3 and 2 = 4 , i.e. if the interaction vertex (3.14)
4 v1 ......4 n4 (q) 1 ,3 2 ,4 . 1 n1
(3.15)
One may show, however, that the SU(2) valley-isospin symmetry is approximately respected when considering the dierent classes of interaction vertices depicted in Fig. 3.1. Consider the diagram in Fig. 3.1(a), which represents a vertex of the ,, , ,, , type v1 n1 ...4 n4 (q) or v1 n1 ...4 n4 (q). In this case, the particle on the left remains in the same valley whereas that on the right changes the valley. Such a process would require a momentum transfer of K, i.e. of the wave vector connecting the two valleys, and therefore does not respect momentum conservation, in the absence of a magnetic eld. Naturally,
52
Electronic Interactions in Graphene IQHE Regime momentum is not a good quantum number here due to to the magnetic eld, but momentum conservation manifests itself by an exponential suppression of such processes. In order to appreciate this point, we need to consider the Gaussian in the form factors (3.11) and (3.12),
, Fn, n (q) e|q+( )K|
2 2 lB /4
as discussed in Appendix 6 [see Eq. (2)]. One therefore sees that the interaction vertex contains a Gaussian term
,, , v1 n1 ...4 n4 (q)

e(q
+|qK|2 )l2 /4 B
2 2 lB /8
e|K|
e#lB /a ,
e(|q |
2 2
+|K|2 /4)l2 /2 B
where # represents an unimportant numerical factor and where we have shifted the momentum q = q K/2 in the second step. The processes associated with the diagram in Fig. 3.1(a) are thus exponentially sup2 pressed in lB /a2 104 /B[T] and may safely be neglected in the range of physically accessible magnetic elds. The same fate is reserved for the diagram in Fig. 3.1(b), which represents a process of Umklapp type. In this case, the vertex reads
,,, v1 n1 ...4 n4 (q) e(|q+K|
2
+|qK|2 )l2 /4 B
e|K|
2 2 lB /2
e#lB /a ,
2 which is again exponentially small in lB /a2 .
The situation is dierent for backscattering-type diagrams [Fig. 3.1(c)], in which case the interaction vertex is
,,, v1 n1 ...4 n4 (q) e(|qK|
2
+|qK|2 )l2 /4 B
One may then redene the wave vector q = q K, which is eventually an integration variable in the interaction Hamiltonian (3.13), and the interaction vertex becomes
,,, v1 n1 ...4 n4 (q )
2 2 2e2 q 2 l2 /2 2e2 B . eq lB /2 e K| |q |K|
As an order of magnitude, with |K| 1/a, one then notices that the backscattering interaction vertex is suppressed by a factor of a/lB 0.005 B[T] as compared to the leading energy scale e2 /lB . The leading interaction vertex is therefore the SU(2) valley-isospin symmetric one depicted in Fig. 3.1(d), for which the rapidly oscillating contribution at K vanishes, as may be seen directly from the form factors (3.12).
53
The above argument, which generalises symmetry considerations for the interactions in a single relativistic LL [85, 86, 87, 88], shows that although the valley SU(2) symmetry is not an exact symmetry, such as the SU(2) symmetry associated with the physical spin, it is approximately respected by the longrange Coulomb interaction. Valley-symmetry breaking terms are due to lattice eects beyond the continuum limit and therefore suppressed by the small factor a/lB , which quanties precisely corrections due to eects on the lattice scale. If one takes into account the additional spin degree of freedom, the resulting four-fold spin-valley degeneracy may then be described within the larger SU(4) symmetry group, which turns out to be relevant in the description of strongcorrelation eects in partially lled LLs (Chap. 5). The SU(4)-symmetric part of the interaction Hamiltonian (3.13) nally reads
sym Hint =
1 2
sym v1 n1 ...4 n4 (q)1 n1 ,3 n3 (q)2 n1 ,4 n4 (q), q 1 n1 ...4 n4
(3.16)
where the symmetric interaction vertex is

sym v1 n1 ...4 n4 (q) =
2e2 F1 n1 ,3 n3 (q)F2 n2 ,4 n4 (q), |q|
(3.17)
in terms of the reduced density operators n, n (q) =

= =, m,m
, n (q) n,
=
(3.18) m eiqR m c n,m;, c n ,m ;, ,
where we have explicitly taken into account the spin index =, in the last line. We nally notice that the graphene form factors (3.11) may also be rewritten in terms of the LL form factors Fn,n (q) = n eiq n , (3.19)
which arise in a similar decomposition of the Coulomb interaction in Landau states in the non-relativistic 2D electron gas, as Fn, n (q) = 1 1 Fn1,n 1 (q) + 2 2 Fn,n (q). n n n n To summarise the dierences and the similarities between the interaction Hamiltonians in graphene and the non-relativistic 2D electron system, one rst realises that its structure is the same if one replaces the LL form factor (3.19) by the graphene form factors (3.11) and if one takes into account the larger (approximate) internal symmetry SU(4), due to the spin-valley degeneracy, instead of the spin SU(2) symmetry. In the remainder of this chapter, we neglect the symmetry-breaking part of the Hamiltonian and consider the Coulomb interaction to respect the SU(2) valley symmetry.
54
(a) CB
Ib Ia
(b)
EF IIa qF
IIb
qF
VB
Figure 3.2: Zero-eld particle-hole exctiation spectrum for doped graphene. (a) Possible intraband (I) and interband (II) single-pair excitations in doped graphene. The exctiations close to the Fermi energy may have a wave-vector transfer comprised between q = 0 (Ia) and q = 2qF (Ib), in terms of the Fermi wave vector qF . (b) Spectral function Im 0 (q, ) in the wave-vector/energy plane. The regions corresponding to intra- and interband excitations are denoted by (I) and (II), respectively.
3.2
Particle-Hole Excitation Spectrum
The considerations of the previous section allow us to discuss the role of the Coulomb interaction within a perturbative approach in the IQHE regime for = 2(2n + 1), where the (non-interacting) ground state is non-degenerate and separated by the cyclotron gap 2( vF /lB )( n + 1 n) from its excited states. Quite generally, the inter-LL transitions evolve into coherent collective excitations, as a consequence of these Coulomb interactions. Prominent examples in the non-relativistic 2D electron gas are the upper-hybrid mode (sometimes also called magneto-plasmon), which is the magnetic-eld counterpart of the usual 2D plasmon [80], and magneto-excitons [89]. In the present section, we discuss how these modes manifest themselves in graphene in comparison with the non-relativistic 2D electron gas.
3.2.1
Graphene particle-hole excitation spectrum at B = 0
Before discussing the particle-hole excitation spectrum (PHES) for graphene in the IQHE regime, we briey review the one for B = 0 as well as its associated collective modes [90, 91, 92, 93]. Quite generally, the PHES is determined by the spectral function 1 (3.20) S(q, ) = Im (q, ), which may be viewed as the spectral weight of the allowed particle-hole excitations, in terms of the polarisability (q, ), which plays the role of a densitydensity repsonse function [79, 80].
55
The particle-hole excitations for non-interacting electrons in doped graphene are depicted in Fig. 3.2.6 In contrast to the PHES of electrons in a single parabolic band (the non-relativistic 2D electron gas), there are two dierent types of excitations: intraband excitations [labeled by I in Fig. 3.2(a)], where both the electron and the hole reside in the conduction band (CB), and interband excitations [labeled by II in Fig. 3.2(a)], where an electron is promoted from the valence band (VB) to the CB. In undoped graphene, there exist naturally only interband excitations (II). If the electron and the hole have an energy close to the Fermi energy, the allowed excitations imply a wave-vector transfer that lies in between q = 0 (Ia) and q = 2qF (Ib). At non-zero values of the transfered energy, one needs to search for available quantum states at larger wave vectors, and the particle-hole pair wave vector is then restricted to / vF < q < 2qF + / vF , as a consequence of the linear dispersion relation in graphene. This gives rise to the region I, which describes the intraband particlehole continuum, and its linear boundaries in the PHES described by the spectral function in Fig. 3.2(b). In addition to intraband excitations, one notices that interband excitations become possible above a threshold energy of F , where an electron at the top of the VB (at q = 0) may be promoted to an empty state slightly above the Fermi energy. The associated wave-vector transfer is naturally q = qF . The point (qF , F ) marks the bottom of the region II in Fig. 3.2(b), which determines the region of allowed interband excitations (interband particle-hole continuum). Direct interband excitations with zero wave-vector transfer are possible above an energy of 2F . Another aspect of the PHES in Fig. 3.2 is the strong concentration of spectral weight around the central diagonal = vF |q|. This concentration is a particularity of graphene due to the electrons chirality [94]. Indeed, if one considers a 2qF backscattering process in the vicinity of the Fermi energy in the CB, Eq. (1.42) indicates that the chirality, i.e. the projection of the sublatticeisospin on the direction of the wave vector, is preserved. The inversion of the direction of propagation in the 2qF process would therefore require an inversion of the A and B sublattices that is not supported by most of the scattering or interaction processes. This eect is reected by a strong suppression of the spectral weight when approaching the right boundary of the region I in the PHES associated with processes of the type Ib in Fig. 3.2(a). Similarly, the conservation of the electrons chirality (1.42) favours 2qF processes in the interband region (II) and the suppression of direct q = 0 interband excitations of the type IIa in Fig. 3.2(a). Notice that, although the direction of the wave vector is inverted in a 2qF process, this indicates still the absence of backscattering because the group velocity v = q / = vF q/|q| remains unchanged the q change in the sign due to the inversion of the wave vector is indeed canceled by the one associated with the change of the band index.
6 We
consider here only the case of a Fermi energy F in the conduction band, for simplicity.
56
(a)
q+q, +
(b)
II
q,
Figure 3.3:
(a) Particle-hole bubble diagram (polarisability), in terms of Greens functions G(q, ) (lines). (b) Spectral function Im RP A (q, ) for doped graphene in the wavevector/energy plane. The electron-electron interactions are taken into account within the RPA. We have chosen rs = 1 here.
Formal calculation of the spectral function In order to obtain the spectral function, it is apparent from Eq. (3.20) that one needs to calculate the polarisability (q, ) of the 2D system, which may be found with the help of the Greens functions G(q, ), (q, ) = iTr d 2 G(q , )G(q + q , + ),
q
(3.21)
where Tr means the trace since the Greens functions are 2 2 matrices as a consequence of the matrix character of the kinetic Hamiltonian. Diagrammatically, the polarisability may be represented by the so-called bubble diagram shown in Fig. 3.3(a), and one nds for non-interacting electrons in graphene [90, 91, 92, 93, 94] g 0 (q, ) = A n n +q q q

q ,,
+q + + i q q
C (q , q + q),
(3.22)
where = vF |q| F is the energy of the quantum state (q) measured q from the Fermi energy , g = 4 takes into account the four-fold spin-valley degeneracy, and n( ) is the Fermi-Dirac distribution function that reduces to a q Heavyside step function n( ) = ( ) at zero temperature. Equation (3.22) q q is nothing other than the Lindhard function [79, 80], apart from the factor C (q , q + q) 1 + cos q ,q +q , 2
in terms of the angle q ,q +q between q and q + q, which takes into account the particular chirality properties of graphene as already mentioned above,
57
this chirality factor vanishes for backscattering processes, i.e. for intraband ( = ) 2qF processes with q ,q +q = as well as for interband ( = ) q = 0 processes with q ,q +q = 0 or 2. Notice nally that the quantity in Eq. (3.22) is an innitesimal energy in the case of pure graphene and may be used (for nite values) as a phenomenological measure of the impurity broadening / , in terms of a life time of the excitations. RPA polarisability The diagrammatic approach is particularly adapted for taking into account the electronic interactions on the level of the random-phase approximation (RPA), which amounts to calculating a geometric series of bubble diagrams and which has shown to yield reliable results for doped graphene [92, 93, 95].7 The RPA polarisability then becomes RP A (q, ) = 0 (q, ) RP A (q, ) , (3.23)
in terms of the polarisability (3.22) for non-interacting electrons and the dielectric function 2e2 0 RP A (q, ) = 1 (q, ). (3.24) |q| The spectral function associated with the RPA polarisability (3.23), which is shown in Fig. 3.3(b), reveals the characteristic coherent 2D plasmon mode, which corresponds to the solution of the implicit equation RP A (q, pl ) and the dispersion relation of which reads8 pl (q) 3 2 2e2 F q + vF q 2 . 2 4 (3.25)
> The expression (3.25) suggests that the asymptotic dispersion is pl (q) 3vF q/2 and that the mode could evolve into the intraband particle-hole continuum. However, a numerical solution of the poles of the polarisability indicates that the prefactor is 3/2 only for intermediate values of the wave vector and that > the asymptotic dependence is indeed given by the central diagonal uh (q) vF q, as suggested by the numerical solution presented in Fig. 3.3 [90, 92]. Therefore, contrary to the plasmon in 2D metals with a parabolic dispersion relation, the plasmon in graphene does not enter region I, but only the interband particlehole continuum (region II). In this region, the Landau damping is less ecient,
7 The RPA has also been applied to undoped graphene [96, 97], but its validity has been questioned [98, 99] because of the vanishing density of states, which would require to take into account diagrams beyond the RPA [82]. 8 Interestingly, this equation is valid both for non-relativistic electrons in conventional 2D electron systems [100] and for relativistic electrons in graphene [90] if one takes into account the dierence in the density dependence of the Fermi energy (F = nel /m for non-relativistic 2D electrons and F = vF nel ) as well as that in the Fermi velocity (vF = 2F /m for non-relativistic electrons and constant vF in graphene).
58
and the modes coherence thus survives to a certain extent without decaying into incoherent particle-hole excitations.
3.2.2
Polarisability for B = 0
In the case of a strong magnetic eld applied perpendicular to the graphene sheet, one needs to take into account the quantisation of the kinetic energy into relativistic LLs described in Sec. 2.1.1, as well as the spinorial eigenfunctions n,m in the calculation of the polarisability. One nds a similar expression for the zero-temperature polarisability of non-interacting electrons as in Eq. (3.22), (F n) (F n ) |Fn, n (q)|2 , 0 (q, ) = g B n n + + i n, n (3.26) in terms of the graphene form factors (3.11) and the characteristic frequency = 2vF /lB introduced in Sec. 2.1.1. One notices that the rst part is nothing other than a Lindhard function [79, 80] for relativistic LLs lled up to the Fermi energy F = (vF /lB ) 2NF , which is chosen to be situated between a completely lled (NF ) and a completely empty (NF + 1) LL in the CB (IQHE regime). The second factor is the modulus square of the graphene form factors which plays the role of the chirality factor C, (q , q + q) in the absence of a magnetic eld [101, 102, 103].9 As for the zero-eld case, one may distinguish two contributions to the polarisability, one that may be viewed as a vacuum polarisability vac (q, ) and that stems from interband excitations when the Fermi level is at the Dirac point, and a second one that comes from intraband excitations in the case of doped graphene. Because undoped graphene with zero carrier density does not correspond to an IQHE situation as we have already discussed in Sec. 2.1, the zero-energy LL n = 0 is only half-lled then , we dene, here, the vacuum polarisability with respect to the completely lled zero-energy level. In order to describe more explicitly the dierent contributions to the polarisability, we dene the auxiliary quantities [101] n, n (q, ) = |Fn,n (q)| + ( + ) vF n vF n + + i
2
where + indicates the replacement + i i and

n1 Nc
n (q, ) =
n =0
n, n (q, ) +
n =n+1
n, n (q, ) + n,n (q, )
(3.27) which verify n (q, ) = n (q, ). The vacuum polarization may then be dened as
9 A similar expression for the polarisability has also been obtained in Refs. [104, 105] though with approximate form factors.
59
(a)
II
(b)
Figure 3.4:
Particle-hole excitation spectrum for graphene in a perpendicular magnetic eld. We have chosen NF = 3 in the CB and a LL broadening of = 0.05 vF /lB (a) and = 0.2 vF /lB (b). The ultaviolet cuto is chosen such as Nc = 70.
Nc
vac (q, ) =
+n (q, )
n=1
(3.28)
where Nc is a cuto that delimits the validity of the continuum approximation. Notice that, already in the absence of a magnetic eld, the validity of the continuum approximation is delimited by t. One may then introduce an upper level cuto with the help of Nc = (vF /lB ) 2Nc t, that leads to Nc 104 /B[T ], which is a rather high value even for strong magnetic elds. However, due to the fact that the separation between LLs in graphene decreases with n, it is always possible to obtain reliable semi-quantitative results from smaller values of Nc . The spectral function S(q, ) = Im 0 (q, )/ is shown in Fig. 3.4 for B NF = 3 and for two dierent values of the phenomenologically introduced LL broadening . One notices rst that the spectral weight is restricted to the two regions I and II corresponding to the intraband and interband particlehole continuum, respectively, in the zero-eld limit. This is not astonishing because the electron-hole pair wave vector remains a good quantum number also in the presence of a magnetic eld; if one considers the pair with its overall charge neutrality, its motion is unaected by the magnetic eld. Indeed, the pair momentum may be viewed as the sum of the pseudomomenta associated 2 with the guiding-centre variable for the electron, K = R ez /lB , and the hole 2 K = R ez /lB , respectively. Each of the pseudomomenta is naturally a constant of motion because so is the guiding centre, as we have discussed in Sec. 2.1.1. One therefore obtains the relation
2 q = K K = R ez /lB
or
2 R = |q|lB ,
(3.29)
where R is the distance between the guiding centre of the electron and that of the hole. The boundaries of the PHES in Fig. 3.4 may then be obtained from
60
the decomposition (2.18), which yields R + , with the help of the average values || = lB 2n + 1 and = lB 2n + 1, 2n + 1 2n + 1 qlB 2n + 1 + 2n + 1. (3.30) Because the energy scales also with n, one obtains the linear boundaries of the particle-hole continua as in the zero-eld case mentioned above. In contrast to these similarities with the zero-eld PHES, one notices a structure in the spectral weight that is due to the strong magnetic eld. As a consequence of the relativistic LL quantisation, the spectral weight corresponds to inter-LL transitions at energies = 2 (vF /lB )( n n ), where n > NF and n NF (for = +) or n > 0 (for = ). For larger values of NF , or quite generally when increasing the energy, the level density increases due to the n scaling of the LLs and the transitions. The LL structure is therefore only visible in the lower part of PHES, in the clean limit = 0.05 vF /lB [Fig. 3.4(a)], whereas the inter-LL transitions are blurred at larger energies or even for the lower transitions in the case of less clean samples [Fig. 3.4(b), for = 0.2 vF /lB ].10 In addition to the (blurred) LL structure in the PHES, one notices another structure of the spectral weight, which is organised in lines parallel to the central diagonal = vF |q|. This weight is again decreased when approaching the right boundary of the intraband continuum (region I) and the left one of the interband continuum (region II), due to the above-mentioned chirality properties of electrons in graphene. The emergence of diagonal lines is a consequence of the graphene form factors (3.11) the modulus square of which intervenes in the polarisation function. Indeed, these form factors F(n+m), n (q) are (associated) Laguerre polynomials with n + 1 zeros [107] due to the overlap between the wave function of the hole in the level n and that of the electron in the LL (n+m) [103]. These zeros in the inter-LL transitions are organised in lines that disperse parallel to the central diagonal and thus give rise to zones of vanishing spectral weight. Interestingly, it is this structure of diagonal lines that survives in more disordered samples in which the horizontal lines associated with interLL transitions start to overlap, i.e. once the LL spacing is smaller than the level broadening .11
3.2.3
Electron-electron interactions in the random-phase approximation
The PHES of non-interacting electrons in graphene gives already insight into the collective modes which one may expect once electron-electron interactions
10 The value = 0.2 v /l is a reasonable estimate for todays exfoliated graphene samples F B on an SiO2 substrate [106]. 11 This behaviour is in stark contrast to that of non-relativistic 2D electrons, where the LL spacing is constant and given by the cyclotron energy eB/m. If this quantity is larger than the level broadening , there is no qualitative dierence between low and high energies, and the horizontal lines associated with the inter-LL excitations (multiples of the cyclotron energy) remain well separated.
61
(a)
II
(b)
II
Figure 3.5:
Particle-hole excitation spectrum for graphene in a perpendicular magnetic eld. The Coulomb interaction is taken into account within the RPA. We have chosen NF = 3 in the CB and a LL broadening of = 0.05 vF /lB (a) and = 0.2 vF /lB (b). The ultaviolett cuto is chosen such that Nc = 70.
are taken into account. Indeed the regions of large spectral weight evolve into coherent collective excitations as a consequence of these interactions. Because the regions of large spectral weight are organised in lines parallel to the central diagonal = vF |q|, as mentioned above, one may expect that the dominant collective excitations are roughly linearly dispersing modes instead of the more conventional weakly dispersing magneto-excitons, which emerge from the interLL transitions [89]. It has though been argued that such magneto-excitons may play a role at low energies in clean samples with low doping [67] and that they may renormalise the cyclotron energy at zero wave vector [108].12 As in the zero-eld case, we take into account the Coulomb interaction within the RPA [see Eq. (3.23)]. The resulting spectral function is shown in Fig. 3.5 for the same choice of parameters as in the non-interacting case (Fig. 3.4). Furthermore, we have chosen a dimensionless interaction parameter rs = 1, here, which corresponds to a dielectric constant of 2. Upper-hybrid mode and linear magneto-plasmons One notices the prominent mode that evolves in the originally forbidden region for particle-hole excitations. This mode, which is called upper-hybrid mode, is the magnetic-eld descendent of the 2D plasmon mode (3.25), which acquires 2 a density-dependent cyclotron gap C = eBvF /F = eBvF / vF nel . Its dispersion relation is then given by [109] uh (q) =
2 2 pl (q) + C =
2 e2 vF nel 3 2 q + vF q 2 + 2 4
2 eBvF vF nel
, (3.31)
12 Notice that such a renormalisation is forbidden in the case of non-relativistic 2D electrons, where the cyclotron resonance is protected due to Kohns theorem [66].
62
(a)
(b)
Figure 3.6: From Ref. [103]. Static polarisation function 0 (q, = 0) for non-interacting electrons in graphene without (a) and with (b) a magnetic eld. To compare both cases, we have chosen a Fermi wave vector qF = 2NF + 1/lB = 7/lB that corresponds to NF = 3. The full black line represents the total polarisability, whereas the red dotted and the blue dashed lines show the intraband and the interband contributions, respectively. as may be shown easily within a hyrdodynamic approach that has been successfully applied to the upper-hybrid mode in non-relativistic 2D electron systems [110]. It is apparent from Fig. 3.5 that this mode remains coherent also within the region II, which corresponds to the zero-eld interband particle-hole continuum. In addition to the upper-hybrid mode, one notices linearly dispersing coherent modes in the regions I and II that emerge from the lines of large spectral weight in the non-interacting PHES, as expected from the qualitative discussion above. We may call these modes linear magneto-plasmons [101, 103] in order to distinguish them clearly from the upper-hybrid mode (3.31) and the weakly dispersing magneto-excitons at low doping [67]. These modes are more prominent in the interband than in the intraband region although they are also visible there. They are genuine to graphene with its characteristic Bn LLs which inevitably overlap in energy, above a critical LL n, if level broadening is taken into account, and they may not be captured in the usual magneto-exciton approximation where the collective modes are adiabatically connected to the inter-LL excitations [67, 89, 108].
3.2.4
Dielectric function and static screening
We terminate this chapter with a brief discussion of the dielectric function (3.24) in the static limit RP A (q) RP A (q, = 0) = 1 2e2 0 (q, = 0), |q| (3.32)
by comparing the B = 0 to the zero-eld case, as shown in Fig. 3.6. As mentioned above, one may distinguish two separate contributions to the static polarisability, the vacuum polarisability vac (q) vac (q, = 0) due to interband excitations and the intraband contribution dop (q) dop (q, = 0),
Particle-Hole Excitation Spectrum which is only present in doped graphene, 0 (q, = 0) = vac (q) + dop (q).
63
(3.33)
One notices that up to 2qF , the zero-eld static polarisability [Fig. 3.6(a)] remains constant. Indeed, the interband contribution (blue dashed line) increases linearly with the wave vector [91, 93] vac (q) = q 4 vF (3.34)
and thus cancels the linear decrease of the intraband contribution (red dotted line), q < dop (|q| 2qF ) (F ) 1 , (3.35) 2qF where (F ) = 2 F
2v2 F
(3.36)
is the density of states per unit area at the Fermi energy. At wave vectors larger than 2qF , the intraband contribution dies out, and the polarisability is dominated by interband excitations. With the help of these two contributions, we may rewrite the static dielectric function (3.32) as RP A (q) = 1 + where we have dened RP A (|q| ) = 1 + rs , 2 (3.37) rs dop (q) 2 vac (q) ,
i.e. the value that the static dielectric function approaches at large wave vectors. One notices that the above form of the static dielectric function may be cast into a Thomas-Fermi form,
T F (q) 1 + rs
qT F q
(3.38)
in terms of the eective coupling constant

rs =
rs rs = , 1 + rs /2
(3.39)
which is plotted in Fig. 3.7. One notices that the interband excitations yield a contribution to the dielectric constant that originally takes into account the dielectric environment in which the graphene sheet is embedded, = . This is a direct consequence of the linear behaviour of the vacuum polarisation (3.34) as a function of the wave vector and thus specic to graphene. Furthermore, one may also
64
1.0

r* s
0.8
2/
Out[22]=
0.6
0.4
0.2
rs
2 4 6 8 10
Figure 3.7: Eective coupling constant rs as a function of the bare coupling rs . The dashed line indicates the asymptotic value 2/ obtained for large values of the bare coupling.
dene an eective Thomas-Fermi wave vector qT F = qT F / = rs qF , which describes the screening length in the presence of the vacuum polarisation. As a consequence of the saturation of the eective coupling constant (3.39) at large values of rs , the eective Thomas-Fermi vector is thus always on the order of the Fermi wave vector unless rs 1, where rs rs . Finally, the screened Coulomb interaction potential may be approximated as 2e2 vscr (q) . (3.40) (1 + rs qF /q)q
One notices from this expression that processes at wave vectors q qF , where the interband excitations play a negligible role [see Fig. 3.6(a)], are still governed by the bare coupling constant rs vscr (q qF )/ vF q. However, processes at or above the Fermi wave vector, such as those that are relevant in the electronic > transport, are characterised by the eective coupling constant rs vscr (q qF )/ vF q, which saturates at a value of 2/ as mentioned above. If we consider the value (3.3), rs 2.2, for the bare coupling constant of graphene in vacuum, the eective coupling is roughly four times smaller rs 0.5, such that the electrons in doped graphene approach the weak-coupling limit. The situation is dierent in undoped graphene, where recent renormalisation-group [111, 112, 113] and lattice gauge theoretical calculations [114, 115] indicate a ow towards strong coupling at moderate values of rs . In Fig. 3.6(b), we have plotted the static polarisability for graphene in the IQHE regime. Qualitatively, the result agrees with the zero-eld behaviour, with a (roughly) linearly increasing vacuum polarisability and a decreasing intraband contribution, apart from some superimposed oscillations due to the overlap functions that are reected by the form factors (3.11). An important dierence is manifest in the small wave-vector limit of the polarisability. In contrast to the zero-eld case, where the polarisability saturates at a non-zero density of states, the system is gapped in the IQHE regime with a resulting vanishing density of states at the Fermi energy. This gives rise to a q 2 behaviour of the polarisability at small wave vectors. Furthermore, the static dielectric
65
Figure 3.8: Static dielectric function for graphene in the IQHE regime for NF = 1, 2 and 3 (blue, red and green curves). function, which is shown in Fig. 3.8 [102, 103], does no longer diverge in this limit, contrary to the zero-eld Thomas-Fermi result (3.38). Indeed, the small-q behaviour may be approximated as RP A (q) 1 rs NF qlB , which is the same as for non-relativistic 2D electrons [116].13 The maximum of the static dielectric function is obtained at qlB 1/qF lB 1/ 2NF + 1, at the value max RP A (q 1/lB 2NF + 1) rs NF . It therefore scales as max NF in contrast to a NF scaling in non-relativistic 2D systems [116]. At large wave vectors, the static dielectric function saturates at the same value as in zero magnetic eld.
3/2
13 Notice, however, that the expression becomes exact only in the large-N F limit and that in non-relativistic 2D electron systems, the coupling constant rs also depends on NF , rs 1/2 NF .
66
Chapter 4
Magneto-Phonon Resonance in Graphene

In the previous chapter, we have discussed the role of electron-electron interactions in the IQHE regime, where a perturbative (diagrammatic) approach may be applied. Similarly, one may treat the electron-phonon interaction in a perturbative manner in this regime. This is the subject of the present chapter, before discussing again electron-electron interactions in the strong coupling limit of partially lled LLs (Chap. 5). As a consequence of the honeycomb-lattice structure of graphene, with two inequivalent sublattices, there are four in-plane phonons, two acoustic and two optical ones. Each phonon type occurs in a longitudinal (longitudinal acoustic, LA, and longitudinal optical, LO) and a transverse (transverse acoustic, TA, and transverse optical, TO) mode. In addition to these four phonons, one nds two out-of-plane phonons, one acoustic (ZA) and one optical (ZO) (for a review of phonons in graphene, see Refs. [24] and [117]). Here, we concentrate on the in-plane optical phonons LO and TO, which couple to the electronic degrees of freedom. More specically, we discuss these phonons at the point (E2g modes) in the centre of the rst BZ, which are attributed to the G-band at ph 0.2 eV in the Raman spectra (see e.g. Refs [118, 119, 120, 121, 122]). One of the most prominent eects of electron-phonon coupling in metals and semiconductors is the so-called Kohn anomaly [123], which consists of a singularity in the phonon dispersion due to a singularity in the electronic densitydensity response function. The analog of the Kohn anomaly in graphene yields a logarithmic divergence of the phonon frequency when the bare phonon frequency coincides with twice the Fermi energy [124, 125, 126]. We investigate, in this chapter, how this renormalisation manifests itself in graphene in a strong magnetic eld [127, 128]. In Sec. 4.1, we consider the specic form of the electron-phonon coupling and discuss its consequences for the renormalisation of the optical phonons at the point in Sec. 4.2. More specically, we consider both non-resonant (Sec. 4.2.1) and resonant coupling (Sec. 4.2.2), the latter 67
68
(a)
dt/da TO phonon LO phonon
Electron-Phonon Coupling
(b)
q
q u(r) electron on A k+q g ~ dt/da
(c)
(q~0)
phonon k electron on B
optical phonons
electronphonon coupling
Brillouin zone
Electron-phonon coupling in graphene. (a) Optical phonons in graphene, with a wave vector q in the vicinity of the point at the centre of the rst BZ [see part (c)]. The amplitude of the LO phonon is in the direction of propagation (black arrows), that of the TO phonon perpendicular (red arrows). The optical phonons modify the bond lengths of the honeycomb lattice. (b) As a consequence of the modied bond lengths, the electronic hopping is varied, and the electron-phonon coupling is o-diagonal in the sublattice index.
Figure 4.1:
being specic to graphene in a magnetic eld when the phonon is in resonance with an allowed inter-LL transition (magneto-phonon resonance) [128].
4.1
The LO and TO phonons in graphene are schematically represented in Fig. 4.1(a). As already mentioned above, we concentrate on phonons at small wave vectors, in the vicinity of the point. The origin of the electron-phonon coupling may easily be understood from the variation of the bond length caused by the phonon, which aects the electronic hopping amplitude between nn carbon atoms. As we have discussed in Sec. 1.5, the eect may be quantied with the help of Harrisons law [39], which yields t/a 4.3 eV/ [see Eq. (1.48)]. A The order of magnitude for the bare electron-phonon energy is then obtained by multiplying this variation with the typical length scale /M , which characterises the amplitude of a lattice vibration of frequency ph within the harmonic approximation. The intervening mass M is that of the carbon atom. Indeed, a tight-binding calculation [124, 129] corroborates this argument, apart from a numerical prefactor 3/2, and yields a bare electron-phonon coupling g= 3 t 2 a 0.26 eV. (4.1)
M ph
This value agrees well with ab-initio calculations [130], although it is slightly lower than the value determined experimentally, which ranges from g 0.3 eV [118, 131] to g 0.36 eV [119]. Furthermore, one notices that, because the electron-phonon coupling is mediated by a bond variation between sites that belong to two dierent sublattices, the coupling constant is o-diagonal in the sublattice basis. This is diagrammatically depicted in Fig. 4.1(b).
69
4.1.1
Coupling Hamiltonian
The above considerations help us to understand the dierent terms in the Hamiltonian, H = Hel + Hph + Hcoupl which serves as the basis for the analysis of the electron-phonon coupling. The Hamiltonian for 2D electrons in a magnetic eld, Hel =
e, d2 r (r)HB (r) = n,m;
,n c n,m; cn,m;
(4.2)
may be written, in second quantisation, in terms of the one-particle Hamiltonian (2.3) and the fermionic elds (r) =
n,m
n,m; (r)cn,m; ,
where n,m; (r) is the wave function in position space associated with the spinor (3.8). The lattice vibration is characterised by the relative displacement u(r) between the two sublattices, which may be decomposed in terms of the bosonic operators b,q and b , ,q u(r) =
,q iqr , b,q + b ,q e,q e
2Nuc M (q)
(4.3)
where e,q denotes the two possible linear polarisations ( = LO,TO) at the wave vector q and Nuc = A/(3 3a2 /2) is the number of unit cells in the system. The phonon Hamiltonian then reads, on the level of the harmonic approximation, Hph = (q)b b,q , (4.4) ,q
,q
in terms of the phonon dispersion (q). Notice that, at the point, the frequencies of the LO and TO phonons coincide, and one has ph (q = 0). It is then convenient to describe the phonon modes in terms of circularly polarised modes, u (r) = [ux (r) + iuy (r)]/ 2 and u (r) = u (r). Finally, taking into account the above considerations on the electron-phonon coupling, the coupling Hamiltonian reads [124, 126, 129] Hcoupl = g 2M ph
d2 r (r) [ u(r)] (r),
(4.5)
where u(r) = x uy (r) y ux (r) is the 2D cross product between the Pauli matrices and the displacement eld.
70
4.1.2
Hamiltonian in terms of magneto-exciton operators
As a consequence of the o-diagonal character of the electron-phonon coupling (4.5), one notices that the intervening matrix elements are proportional to n,n1 , and one thus obtains the selection rules n (n 1), in analogy with the magneto-optical selection rules discussed in Sec. 2.1.1. Furthermore, if we x the energy of the dipole transition (2.31) to be1 n n,= = 2 (vF /lB )( n + 1 + n), there are two possible transitions, which may be distinguished by the circular polarisation of the phonon they are coupled to. Indeed, the form of the electron-phonon coupling (4.5) indicates that the -polarised phonon is coupled to the transition (n + 1) +n, whereas the -polarised phonon couples to the n +(n + 1) interband transition [128]. It is then convenient to introduce magneto-exciton operators, associated with the above-mentioned inter-LL transitions (n, ) = (n, ) = i 1 + n,0 Nn i 1 + n,0 Nn c +n,m; c(n+1),m; , c +(n+1),m; cn,m; , (4.6)
where the index A = , characterizes the angular momentum of the excitation and where the normalization factors Nn and Nn = = 2(1 + n,0 )NB (n+1) +n 2(1 + n,0 )NB n +(n+1)
are used to ensure the bosonic commutation relations of the exciton operators, [A (n, ), (n , )] = A,A , n,n , including the two-fold spin degeneracy. A These commutation relations are obtained within the mean-eld approximation with c n,m; c n ,m ; = , , n,n m,m (, + ,+ n ), where 0 n 1 is the partial lling factor of the n-th LL, normalised to 1. One notices that the magneto-exciton operators are, apart from a normalisation constant, nothing other than the reduced density operators (3.10), (n, ) , +n,(n+1) (q = 0) and (n, ) , +(n+1),n (q = 0), respectively, at zero wave vector. Notice furthermore that, because of the relative sign between the kinetic part (4.2) and the electron-phonon coupling Hamiltonian (4.5), the optical phonons couple to the valleyanti-symmetric magneto-exciton combination A,as (n) = [A (n, = +) A (n, = )]/ 2. This needs to be contrasted to the magneto-optical coupling [64, 67, 68], where the photon couples to the valley-symmetric magneto-exciton mode A,s (n) = [A (n, = +) + A (n, = )]/ 2.
1 We consider mainly interband transitions here, which may have a chance to be in resonance with the phonon of energy ph 0.2 eV.

= +
11111111111 00000000000 11111111111 00000000000 11111111111 00000000000 11111111111 00000000000
71
11111111111 00000000000 11111111111 00000000000 11111111111 00000000000 11111111111 00000000000
~ =
Figure 4.2:
Diagrammatic representation of Dysons equation. The dressed phonon propagator (thick twiggled line) may be written in terms of the bare phonon propagator (thin twiggled line) and the magneto-exciton propagator (thick gray line), which may be viewed as an electron-hole polarisation bubble (below). The coupling is indicated by the black circles.
The magneto-exciton operators (4.6) allow one to rewrite the electron-phonon Hamiltonian at the point (q = 0) in a bosonic form as [128] H =
=s,as A,n
n (n)A, (n) + A, gA (n) b A;as (n) A
ph b bA A (4.7)
+
A,n
+ bA (n) , A;as
in terms of the eective coupling constants g (n) = and g (n) = g g (1 + n,0 ) (1 + n,0 ) (n+1) +n , n +(n+1) , (4.8)
2 with the constant 3 3a2 /2lB . One therefore remarks that, although the bare coupling constant g is rather large [see Eq. (4.1)], the eective coupling is reduced by a factor of a/lB , gA (n) g a 1 . . . 2 meV lB B[T]. (4.9)
4.2
Phonon Renormalisation and Raman Spectroscopy
The Hamiltonian (4.7) shows that a phonon may be destroyed by exciting a magneto-exciton, and the associated Dyson equation, represented diagrammatically in Fig. 4.2, for the dressed phonon propagator D() reads DA () = D0 () + D0 ()A ()DA (), in terms of the bare bosonic phonon propagator D0 () = and A () = 1 2 2 2 ph (4.11) (4.10)
Nc n=NF +1
2 2 2NF gA (n) 2n gA (n) + . 2 2 2 2 2 2 n NF
(4.12)
72
The form of the last expression is transparent; the magneto-exciton is a boson, and its propagator is therefore of the same form as that of the bare phonon. It is equivalent to a particle-hole propagation associated with a polarisation bubble [see Fig. 4.2 and also Fig. 3.3(a) in the preceding chapter], but the expression (4.12) also takes into account the square of the eective coupling constant which is due to the double occurence of the electron-phonon coupling rst when the phonon is converted into a magneto-exciton and the second time when the magneto-exciton is destroyed by creating a phonon. The last term in Eq. (4.12) takes into account the (only) possible intra-band magneto-exciton from the last lled LL NF to NF + 1 with energy NF = 2( vF /lB )( NF + 1 NF ), which we have omitted in the Hamiltonian (4.7) because it is irrelevant for resonant coupling. The parameter Nc is the same high-energy cuto, dened by Nc = (vF /lB ) 2Nc t, as in Sec. 3.2 of the preceding chapter. The renormalised phonon frequencies A may be obtained from the Dyson equation (4.10), by searching the poles of the dressed phonon propagator DA ()1 = 0 = DA ( A )1 A ( A ), and one nds [127, 128]
2 A ph = 2
4ph
Nc n=NF +1
2 2 NF gA (NF ) n gA (n) + . 2 2 2 2 2 2 A A NF n
(4.13)
4.2.1
Non-resonant coupling and Kohn anomaly
where = (2/ 3)(g/t)2 3.3 103 is the dimensionless electron-phonon coupling constant introduced in Refs. [124, 127], and 0 ph + 2
0 Nc
Before discussing resonant coupling, i.e. when the phonon frequency is in resonance with a possible inter-LL excitation, in a strong magnetic eld, we comment on the relation between Eq. (4.13) and the (non-resonant) renormalisation of the phonon frequency in zero magnetic eld. The zero-eld limit may indeed be obtained from Eq. (4.13) if one replaces the sum n by an integral dn, i.e. if the spacing between the LLs vanishes, NF 0. Linearising Eq. (4.13), if replaces A ph in the denominators, and using the approximation one n + n + 1 2 n yields vF ph ph + 2 2NF vF /lB 2NF 0 + , (4.14) ln lB 4 ph 2 2NF vF /lB
dn
2 n gA (n) 2 2 2 n ph
is the physical phonon frequency at zero doping. Indeed, the frequency ph is not relevant in a physical measurement in graphene even if it occurs in the Hamiltonian, but one measures 0 at zero doping and B = 0. Equation (4.14)

~ ~ renormalised phonon frequency 0
73
4 2 50
1/2
-2
100 chemical potential / 150 200 ~ 0
-4 -6
-8
Figure 4.3: Renormalised phonon frequency as a function of the lling controled by the chemical potential . The dashed line indicates the asymptotic behaviour. coincides precisely with the zero-eld result [124, 125, 126] one identies the if chemical potential with the energy of the last lled LL, = 2NF vF /lB [128]. The renormalised phonon frequency is plotted in Fig. 4.3 as a function of the chemical potential . One notices the logarithmic divergence, which is at the origin of the Kohn anomaly, when the chemical potential is half the (bare) phonon frequency [124, 125, 126]. Qualitatively, one may understand the behaviour of the phonon frequency in terms of a simple level repulsion indeed, inter-LL excitations with an energy below that of the phonon have a tendency to push the phonon frequency upwards whereas those with a larger energy yield a decrease. If one increases the band lling from the Dirac point, the low-energy magneto-excitons, which pushed the phonon frequency upwards, get successively blocked due to the Pauli principle, and as a consequence the phonon energy decreases. Beyond = 0 /2, the high-energy magneto-excitons become blocked, and the phonon frequency increases again. The increase is roughly linear, as one would expect from the adiabatic approximation according to which the renormalisation of the phonon frequency is proportional to the density of states () times the square of the coupling constant, 0 ( = 0) g 2 () . Notice, however, that the adiabatic approximation, which is valid only for , misses the logarithmic divergence which occurs when the q = 0 interband excitation with energy ph becomes blocked, i.e. at = 0 /2 [125]. Experimentally, the logarithmic phonon softening has been observed in bilayer graphene [132], which is governed by similar physics, whereas in monolayer graphene only the increase of the phonon frequency as a function of doping has been found [119], probably due to important inhomogeneities in the charge density of the samples.
4.2.2
Resonant coupling
Apart from the non-resonant coupling discussed in the preceding section, the high-eld electron-phonon coupling reveals a linear eect when the phonon is
74

(b)
2 2
Energy in units of ph
1.75 1.5 1 1 1.25 0.75 0.5 0.25
(a) n=2 n=1 n=0
B0 0
10 10
1520 20
25 30
30 4035
Magnetic Field [T]
(a) From Ref. [128]. Anticrossing of coupled phonon-magneto-exciton modes as a function of the magnetic eld. (b) From Ref. [131]. Colour plot of the Raman spectra as a function of the magnetic eld. The continuous white lines indicate the magnetic eld for which the phonon is in resonance with an inter-LL excitation of energy n . Top: data of the full B-eld range. Bottom: zoom on the range from 0 to 10 T.
Figure 4.4:
in resonance with a particular magneto-exciton, ph n . In this case, the sum on the right-hand side in Eq. (4.13) is dominated by a single term and 2 may be approximated by 2(ph / )gA (n)/( A n ). This results in a ne structure of mixed phononmagneto-exciton modes, A,as (n) cos +bA sin with frequency A and A,as (n) sin bA cos with frequency A [where cot 2 = + (n 0 )/2gA ]. The frequencies of these mixed boson modes read [127, 128] A (n) = 1 2 n + 0 1 4 n
2
2 + gA (n),
(4.15)
and the resulting phononmagneto-exciton anticrossing is depicted in Fig. 4.4(a). The above-mentioned anticrossing of the coupled phononmagneto-exciton modes has been observed in recent Raman experiments on epitaxial graphene. Remember that Raman spectroscopy is sensitive to the phonon component of the mixed modes [131]. The results are shown in Fig. 4.4(b) and corroborate the theoretically expected behaviour [127, 128]. Indeed, one may obtain the oscillating behaviour from a numerical solution of Eq. (4.13) if one expresses the equation in terms of 0 instead of ph and if one takes into account a nite broadening of the levels. If the phonon is out of resonance with an inter-LL transition, its frequency is essentially eld-independent and coincides with the energy of the E2g line at 1586 cm1 0.2 eV. When it approaches the resonance (by increasing the magnetic eld), its energy is shifted upwards as a consequence of the anticrossing but rapidly dies out in intensity once the magneto-exciton component becomes dominant in the A mode. Upon further increase of the +

(a) B=B0 2g I =0
75
Oscillator strength [a.u.]
mode splitting 2g 2g
II 0 < || < 2
(b) B=B0 2g
g 2
2g 2g
6 (c) B=Bn>0
mode splitting
III || = 2
g 2 2g 4n2 4n+2 2g
4n+6
~ 0
Fine structure of the resonant electron-phonon coupling as a function of the lling factor (From Ref. [128]). (a) Mode splitting as a function of the lling factor with the resonance condition n=0 0 , for = 0 in (I), 0 < || < 2 (in II), and = 2 (in III). The absolute intensity of the modes is in arbitrary units, but the height and the width reect the expected relative intensities. (b) Mode splitting for n = 0, as a function of the lling factor . (c) Same as in (b) for n 1.
Figure 4.5:
magnetic eld, the A mode becomes more phonon-like and therefore visible in the Raman spectra. Expected Raman spectra at resonance when varying the lling factor The ne structure of the high-eld resonant electron-phonon coupling may furthermore be investigated by sweeping the chemical potential when the magnetic eld is held xed at resonance. The eect is most pronounced for the resonance ph n=0 , which is expected at B 30 T [see Fig. 4.4(a)]. In this case, the mode consists of an equal-weight superposition of the phonon and the magneto-exciton (cos = sin = 1/ 2), and the E2g band would appear as two lines [see part I in Fig. 4.5(a)], at the energies = 0 gA , for the case of undoped graphene.2 With the above estimation (4.1) for the bare electron-phonon coupling constant, one obtains for the line splitting 2gA 16 meV ( 130 cm1 ), which largely exceeds the G-band width observed in Refs. [118, 119, 120, 121, 122]. It is apparent from the expressions (4.9) for the eective coupling constants g and g that the splitting may be controled by the LL lling factor, as may be seen in Fig. 4.5(b), where we have plotted the mode splitting as a function of in the n = 0 LL [128]. Exactly at zero doping, both coupling constants coincide, g = g , but if one increases the doping the transition 1 0 associated with
2 Notice, however, that only an oscillation of the phonon mode, and not a splitting, has been observed in the experiment by Faugeras et al. [131].
76
the -polarisation becomes weaker due to the reduced number of nal states in n = 0, whereas the 0 +1 transition (with polarisation ) is strengthened. As a consequence, the associated coupling constants are increased and decreased, respectively [see Fig. 4.5(b)], until the coupling constant g vanishes at = 2. The above-mentioned lling-factor dependence has a direct impact on the Raman lines [128]. Whereas at = 0, one expects two lines separated by the energy 2g = 2g , the degeneracy in the circular polarisation is lifted between 0 < < 2.3 One therefore expects to observe four lines instead of two, where the inner ones are associated with the polarisation , whereas the outer ones with increased splitting correspond to the opposite polarisation [Fig. 4.5(a), part II]. The separation between the inner lines vanishes then at = 2, where the splitting of the outer lines is maximal and where one expects to observe three lines [Fig. 4.5(a), part III]. We nally mention that the situation is similar for other resonances (n + 1) +n n +(n + 1)], which occur at lower magnetic elds, Bn = [and B0 /( n+ n + 1)2 [see the lines Tn in Fig. 4.4(b)] However, the mode splitting is less pronounced due to the B-eld dependence of the coupling constants (4.9). The two polarisations are degenerate, g (n) = g (n), for < 2(2n1), whereas above = 2(2n 1), the LL n starts to be populated and thus to block the transition (n + 1) +n. As a consequence, the coupling constant g (n) is decreased and vanishes eventually at = 2(2n + 1). In contrast to the n = 0 transition described above, the -polarisation remains constant in this llingfactor interval, and g (n) decreases for > 2(2n + 1) [see Fig. 4.5(c)].
3 We present the argument for a Fermi energy in the CB, i.e. > 0, but the situation is generic, and the argument also applies in the VB if one interchanges the polarisations.
Chapter 5
Electronic Correlations in Partially Filled Landau Levels

This last chapter is devoted to the physics of interacting electrons in the strongcorrelation limit of a partially lled LL. The motivation stems from non-relativistic quantum Hall systems in GaAs heterostructures, where these correlations lead to the formation of incompressible quantum-liquid phases, which display the fractional quantum Hall eect (FQHE) [133], as well as of exotic electron-solid phases, such as the high-eld Wigner crystal [134, 135] or the theoretically predicted bubble and stripe phases [136, 137, 138]. The latter are likely to be at the origin of highly anisotropic transport properties at half-lled higher LLs [139, 140], particular electron transport under microwave irradiation [141, 142, 143], and an intriguing reentrance of the IQHE in n = 1 and n = 2 [144, 145]. It is therefore natural to ask whether such strongly correlated phases exist also in graphene and if so what the dierences are with respect to non-relativistic 2D electrons. Moreover, the fact that the electrons reside at the surface opens up the possibility to probe these phases by spectroscopic means, such as scanning tunneling spectroscopy, which has already been applied successfully in the analysis of the electron density distribution of exfoliated [146] and epitaxial graphene [147], as well as graphene on graphite substrates [148]. After a short discussion of the role of the Coulomb interaction in graphene as compared to non-relativistic 2D electrons, we introduce the basic model of interacting electrons in a partially lled relativistic LL. This model allows us to make qualitative predictions for the above-mentioned correlated electronic phases in the context of graphene, as compared to non-relativistic electrons. In Sec. 5.2, we apply this model to the quantum Hall ferromagnetism with an internal SU(4) symmetry that is the relevant symmetry in graphene LLs and discuss its relation with the experimentally observed degeneracy lifting of the zero-energy LL n = 0 [56]. We terminate this chapter with a review of the 77
78
Electronic Correlations in Partially Filled LL
specic FQHE in graphene, which has recently been observed in a two-terminal geometry [149, 150].
5.1
Electrons in a Single Relativistic Landau Level
Quite generally, the origin of strongly correlated electron phases stems from a quenched kinetic energy, where the partially lled LL is separated by the cyclotron gap from the neighbouring ones such that inter-LL excitations constitute high-energy degrees of freedom. The Coulomb interaction, which may though be small with respect to the cyclotron gap, remains then as the only relevant energy scale which dominates the low-energy degrees of freedom if we can neglect disorder eects. This leads to the seemingly counter-intuitive nding of strongly correlated phases in weakly correlated matter. In order to quantify the degree of separation between the energy scales, one may use a similar argument as the one that led us to the denition of the dimensionless interaction parameter (3.1), introduced at the beginning of Chap. 3. Indeed, one needs to compare the Coulomb interaction energy Eint = e2 /RC at the characteristic length scale1 RC = lB 2n + 1 to the LL spacing C = eB/m, where we concentrate on non-relativistic electrons rst,
B rs =
e2 , vF (n, B)
with
vF (n, B) RC C .
(5.1)
If one identies the Fermi wave vector kF 2n/lB , one obtains the same expression as for the zero-eld coupling constant (3.2),
B rs = rs =
me2 1 k , 2 F a0
where F is the average distance between electrons. This means that the degree of LL mixing is still governed by rs , and the inter-LL excitations are well separated from the low-energy intra-LL degrees of freedom unless rs becomes very large. Notice, however, that rs 1 in most 2D electron systems. In the case of partially lled relativistic LLs in graphene, one is tempted to apply the same argument if the Coulomb interaction e2 /RC is suciently small as compared to the LL spacing n , the relevant degrees of freedom are those which couple quantum states in the same LL, whereas inter-LL excitations may be considered as frozen out (see Fig. 5.1). Whereas this seems a reasonable assumption for the lowest LLs, one is confronted with the apparent problem that the LL spacing rapidly decreases once the Fermi level resides in higher LLs, n = vF vF ( n + 1 n) . 2 lB lB 2n (5.2)
1 In contrast to what one may naively expect, the intra-LL Coulomb ineractions are governed by the cyclotron radius and not by the magnetic length, as one may show with the help of scaling arguments [151]. This point is also briey discussed below, in the framework of the SU(4) model of interacting electrons.
79
(a)
LL separation
(b)
X
Figure 5.1:
X
x x x x
(a) Completely lled topmost LL. Due to the Pauli principle, the only possible excitations are inter-LL transitions. (b) Partially lled LL. For suciently small values of rs , the inter-LL excitations constitute high-energy degrees of freedom that may be omitted at low energies, where the relevant degrees couple states within the same LL.
Notice, however, that this decrease is balanced by the 1/ 2n scaling of the characteristic Coulomb interaction, such that even in higher LLs the separation between low- and high-energy degrees of freedom is governed by the dimensionless coupling constant e2 e2 /lB 2n G B = = rs , (5.3) rs = vF vF /lB 2n which coincides with the scale-invariant zero-eld coupling constant (3.3). From the interaction point of view, the restriction of the electron dynamics to a single partially lled LL in the large-n limit is therefore as justied as for the lowest relativistic LLs. Naturally, this statement only holds true in the absence of disorder that induces stronger LL mixing for n 1 than in n = 0 or 1.
5.1.1
SU(4)-symmetric model
Formally, the above-mentioned separation into high- and low-energy degrees of freedom may be realised with the help of the reduced density operators (3.10). For the moment, we only consider the case where = , i.e. we concentrate on the valley-symmetric model, in which case the reduced (intra-valley) density operators (3.18) fall into two distinct classes: for n = n or = , the operators n, n (q) describe density uctuations corresponding to inter-LL transitions of an energy equal to or larger than the LL separation n , whereas the projected density operators (q) n,n (q) =
m eiqR m c n,m;, cn,m ;,
(5.4)
= =,
m,m
describe the density uctuations inside the LL n that interest us here. Notice that we have dropped the index n in the denition of the projected density operators; they satisfy the quantum-mechanical commutation relations [152] [(q), (q )] = 2i sin
2 q q lB 2
(q + q ),
(5.5)
where q q = (q q)z = qx qy qy qx is the 2D vector product between q and q, and these commutation relations are independent of the LL index n. The
80
information about the LL is indeed waved to the eective interaction potential vn (q) = 2e2 2 [Fn (q)] , q (5.6)
in terms of the LL form factors [see Eq. (3.11) and their explicit form (2) discussed in Appendix 6] Fn (q) = 1 (1 n,0 )Ln1 2
2 q 2 lB 2
+ (1 + n,0 )Ln
2 q 2 lB 2
eq
2 2 lB /4
, (5.7)
independent of the band index [85, 153]. The Hamiltonian resulting from Eq. (3.16) 1 Hn = vn (q)(q)(q) (5.8) 2 q therefore denes, together with the commutation relation (5.5) the model of strongly correlated electrons restricted to a single relativistic LL. The model respects the SU(4) spin-valley symmetry, and naturally, there is no kinetic energy scale because all processes involve states within the same LL. The SU(4) spin-valley symmetry is formally described with the help of the spin and valley-isospin operators S (q) = (S ) (q) and I (q) = ( I ) (q), (5.9)
respectively, that are tensor products between the projected density operators (5.4) and the operators S and I , which are (up to a factor 1/2) Pauli matrices. The operators (S ) and ( I ) are the generators of the SU(2)SU(2) subgroup of SU(4). However, once combined in a tensor product with the projected density operators (q), the SU(2)SU(2)-extended magnetic translation group is no longer closed due to the non-commutativity of the Fourier compo nents of the projected density operators. The commutators [S (q), I (q)] yield the remaining generators of the SU(4)-extended magnetic translation group [154, 155, 156]. Validity of the model With the help of the Hamiltonian (5.8), we may render more transparent the model assumption of electrons restricted to a single relativistic LL. We need to show that the energy scale that governs the model (5.8) and its resulting phases is indeed given by e2 /RC and not e2 /lB . As an upper bound for the energy scale, one may use the energy of a completely lled LL described by c n,m;, cn,m ;, = m,m , the mean-eld energy Hn /N of which is simply the exchange energy,2
n EX =
2 The
1 2
vn (q) =
e2 2
dq [Fn (q)] .
(5.10)
direct energy is compensated by the positively charged background (jellium model)
[79]
81
In order to estimate the integral in the large-n limit, one may use the scaling form [107, 157] of the Laguerre polynomials Ln
2 q 2 lB 2
eq
2 2 lB /4
J0 (qlB 2n + 1),
(5.11)
in terms of the Bessel function J0 (x), such that one obtains by a simple change of the integration variable 0 dq[Fn (q)]2 (lB 2n)1 0 dx[J0 (x)]2 = c/lB 2n, where c is a numerical factor of order one. The exchange energy of a completely lled LL n therefore scales with n 1 as
n EX c
e2 e2 c , RC lB 2n
in agreement with the model assumption of a separation between high- and lowenergy degrees of freedom and the denition (5.3) of the dimensionless coupling B constant rs .
5.1.2
Symmetry-breaking long-range terms
When decomposing the Coulomb interaction in the two-spinor basis (Sec. 3.1), we have seen that the SU(4)-symmetric model yields the leading energy scale, whereas the only relevant symmetry-breaking term is associated with backscattering processes at an energy scale roughly a/lB times smaller than the leading one. When restricted to a single relativistic LL n, these backscattering terms yield a contribution
sb Hn =
1 2
sb vn (q), (q), (q), = q
(5.12)
in terms of the eective backscattering potential

sb vn (q)
= =
2e2 +, Fn,n (q) q
(5.13)
2 |q K|2 lB 2 2
2e2 (1 0,n ) 2 (qy Ky )2 lB L1 n1 q 2n
e|qK|
2 2 lB /2
where we have made use of Eq. (3.12) and of the explicit expressions for the intervening matrix elements (4). The eect of this symmetry-breaking term will be discussed in more detail in Sec. 5.2 in the context of the SU(4) quantum Hall ferromagnetism. The term (5.12) is only relevant in relativistic LLs n = 0 as a consequence of the factor (1 n,0 ) in the expression (5.13) for the backscattering potential [85]. This is a consequence of the chiral symmetry of the zero-energy LL [158] where the sublattice index is confunded with the valley isospin, as may be seen from the expression (2.16) for the associated wave functions. Notice, however, that there may occur other symmetry-breaking terms in n = 0 as a consequence of short-range interactions on the lattice scale [86, 87, 88] or from a strain-induced gauge eld [159].
82
(a) effective interaction potential
12
(b)
n pseudopotentials V m
n=0 (relativistic and nonrelativistic)

8 6 5 4 2
10 10
0.8
n=1 (relativistic) n=1 (nonrelativistic) n=5 (relativistic)
0.8 0.6 0.4

1
n=0 n=1 (rel.) n=1 (nonrel.)
0.6 0.4
n=5 (nonrelativistic)
2 4 4 6 6 8 8 10 10
r/l B
5 4
Figure 5.2: (a) From Ref. [85]; comparaison between the relativistic (black curves) and nonrelativistic (grey curves) potentials for the LLs n = 0, 1, and 5. The dashed line shows the potential in n = 0, which is the same in both the relativistic and the non-relativistic case. (b) Pseudopotentials for n = 0 (black circles), n = 1 relativistic (blue) and n = 1 non-relativistic (green). The lines are a guide to the eye. The open circles represent pseudopotentials with even relative pair angular momentum that are irrelevant in the case of completely spin-valley polarised electronic states. The energies are given in units of e2 /lB .
5.1.3
Qualitative expectations for correlated electron phases
The model of interacting electrons in a single relativistic LL has the same structure as the one for non-relativistic LLs in both cases, one has an interaction Hamiltonian that is quadratic in the projected density operators (5.4) which satisfy the commutation relations (5.5). This is a noticeable result in the sense that, whereas non-relativistic 2D electron systems are governed by Galileian invariance, the electrons in graphene are embedded in a Lorentz-invariant spacetime. However, once restricted to a single LL, the electrons forget about their original spatial symmetry properties and are governed by the magnetic translation algebra, which is at the origin of the commutation relations (5.5). As a consequence and in contrast to the IQHE, the dierences between strongly correlated electrons in graphene and non-relativistic 2D electrons do not stem from their respective space-time properties, as one would expect from a mean-eld Chern-Simons approach [63, 160]. The dierence between graphene and non-relativistic 2D electrons are rather to be seeked in the larger internal symmetry instead of an SU(2) spin symmetry, one has an SU(4) spin-valley symmetry if one neglects the small symmetrybreaking term (5.12) in the interactions. Another dierence arises from the dierent eective interaction potential (5.6), instead of
nonrel vn (q) =
2e2 2e2 2 Ln [Fn (q)] = q q
2 q 2 lB 2
eq
2 2 lB /2
(5.14)
for the usual 2D electron gas, in terms of the form factors (3.19). As one may see from the graphene form factors (5.7), the eective interaction potential in graphene for n = 0 is the geometric average of the non-relativistic ones in the adjacent LLs n and n 1, whereas for n = 0 there is no dierence between the relativistic and the non-relativistic case (see Fig. 5.2), as a consequence of the above-mentioned chiral properties.
83
One notices, furthermore, that the dierence between the relativistic and non-relativistic eective interaction potentials become less prominent in the large-n limit [see n = 5 in Fig. 5.2(a)]. This may be understood from the approximate expression (5.11) of the form factors, which yields Fn (q) [J0 (qlB 2n + 1)+ J0 (qlB 2n 1)]/2 J(qlB 2n) + O(1/n). This result agrees indeed to leading order in 1/n with the scaling expression of the form factors (3.19) for usual non-relativistic 2D electrons. The strongest dierence in the interaction potentials is found for n = 1, which in graphene is quite reminiscent to that in n = 0, apart from a reduced repulsion at very short distances, whereas for non-relativistic 2D electrons it has an additional structure [see Fig. 5.2(a)]. The behaviour of the eective interaction potential may also be analysed with the help of Haldanes pseudopotentials [161] 2 2 1 2 Vn = vn (q)L (q 2 lB )eq lB /2 , (5.15) 2 q which represent the interaction between pairs of electrons in a magnetic eld, in a relative angular momentum state with quantum number . This quantum number is related to the average distance lB 2 + 1 between the two particles constituting the pair and is a good quantum number for any two-particle interaction potential v(ri rj ). The pseudopotentials for graphene have been calculated in Refs. [85, 162, 163] and are shown in Fig. 5.2(b) for n = 0 and 1. Haldanes pseudopotentials are an extremely helpful quantity in the understanding of the possible FQHE states which one may expect in 2D electron systems. One notices rst that as a consequence of the anti-symmetry of a twoparticle wave function under fermion exchange, the relative angular-momentum quantum number must be an odd integer, i.e. only the pseudopotentials with odd values of play a physical role in the description of two interacting electrons of the same type (spin or valley). Even- pseudopotentials become relevant if the SU(4) spin-valley isospin is not completely polarised, in the treatment of two electrons with dierent internal quantum number or . One then notices that the n = 1 pseudopotentials, apart from a the dierence in V=0 , are much more reminiscent of those in n = 0 than of those for non-relativistic 2D electrons in the same LL n = 1 [see Fig. 5.2(b)]. If one considers polarised electrons, one therefore expects essentially the same strongly correlated electronic phases in graphene for n = 1 as for n = 0 [85], as also corroborated by numerical studies for FQHE states [162, 163, 164, 165, 166] and electron-solid phases [167, 168, 169]. Because the pseudopotentials (5.15) are systematically larger in n = 1 than in n = 0 (apart from the short-range component for = 0), the gaps of the FQHE states in n = 1 are larger than the corresponding ones in n = 0, as one may also see from numerical calculations [162, 163]. Most saliently, one would expect the absence of the 5/2 FQHE state [170], with its non-Abelian quasiparticles [171, 172], in graphene. Similarly, the particular competition between FQHE states and electron-solid phases, which is characteristic of the non-relativistic n = 1 LL [173, 174] and which is at the origin of the reentrance phenomena observed in transport measurements [143, 145], is
84
expected to be absent in the n = 1 graphene LL.
5.2
SU(4) Quantum Hall Ferromagnetism in Graphene
Before considering possible FQHE states in a more quantitative manner, beyond the above-mentioned qualitative arguments, we discuss the physics of generalised quantum Hall ferromagnets. These phases arise in systems with a discrete internal degree of freedom described by an SU(N ) symmetry, such that each single-particle quantum state n,m occurs in N copies. Prominent examples are the non-relativistic quantum Hall systems when taking into account the electronic spin =, (N = 2) or bilayer quantum Hall systems that consist of two parallel 2D electron gases, where the layer index may be viewed as a spin 1/2 (N = 2, or N = 4 if one also takes into account the physical spin [154, 155]).3 In this sense, graphene may be viewed as an SU(4) quantum Hall system as a consequence of its fourfold spin-valley degeneracy.
5.2.1
Ferromagnetic ground state and Goldstone modes
Quite generally, quantum Hall ferromagnetism arises when the lling factor, dened from the bottom of the LL,4 is an integer that is not a multiple of N [177]. In this case one is confronted, in the absence of interactions, with a macroscopic ground-state degeneracy. Even if one has an integer lling factor, the situation is thus much more reminiscent of the FQHE, i.e. the relevant energy scale is the Coulomb interaction, and the system may be described in the framework of the model (5.8) of interacting electrons in a single (relativistic) LL. For the moment, we consider that there are no symmetry-breaking terms, such as the backscattering term (5.12) or Zeeman-type terms that are discussed below in Sec. 5.2.3. Qualitatively, one may understand the formation of a ferromagnetic ground state as a consequence of the repulsive Coulomb interaction. In order to minimise this interaction, the electrons prefer to form a state described by a maximally anti-symmetric orbital wave function that must then be accompanied by a fully symmetric SU(N ) spin wave function to satisfy an overall fermionic (antisymmetric) wave function. In a usual metal with a nite band dispersion, this ferromagnetic ordering (e.g. all electrons in the spin- states) is accompanied by a cost in kinetic energy indeed, the Fermi energy for spin- electrons is increased whereas that of spin- electrons is lowered. The competition between the gain in interaction and the cost in kinetic energy denes the degree of polarisation, i.e. how ferromagnetic the electrons eectively are. In the quantum Hall eect, however, we are confronted with a highly degenerate LL that may be
a review on non-relativistic multi-component systems see Ref. [175, 176]. that the lling factor in graphene is dened with respect to the centre of the n = 0 LL. There is thus a shift of 2 in the lling factor as compared to the non-relativistic case.
4 Remember 3 For
85
viewed as an innitely at band, such that the kinetic-energy cost for complete spin polarisation is zero. As an example of an SU(N ) quantum Hall ferromagnet, one may consider the state |FM =
k NB 1 i=1 m=0
c |vac , m,i
(5.16)
that consists of k < N arbitrarily chosen completely lled subbranches [i {(K, ), (K , ), (K, ), (K , )}, for the SU(N = 4) symmetry in graphene LLs], where we have omitted the LL index n at the fermion operators to simplify the notation. The arbirariness in the choice of the SU(N ) spin subbranches may be viewed as a spontaneous symmetry breaking that accompanies the ferromagnetism. Indeed, the state (5.16) is no longer invariant under an SU(N ) rotation, but only under a rotation described by the subgroup SU(k)SU(N k), where the rst factor indicates a symmetry transformation in the fully occupied subbranches i = 1, ..., k and the second factor one in the empty subbranches i = k + 1, ..., N . Therefore the quantum Hall ferromagnet (5.16) is associated with an order parameter with a spontaneous symmetry breaking described by the coset space SU(N )/SU(k) SU(N k) U(1) U(N )/U(k) U(N k), where the additional U(1) is due to the phase dierence between the occupied and the unoccupied subbranches [177, 178]. The coset space, with its N 2 k 2 (N k)2 = 2k(N k) complex generators, denes also the Goldstone modes, which are nothing other than the k(N k) spin-wave excitations of the ferromagnetic ground state (5.16).5 The number of the spin-wave modes may also have been obtained from a simple inspection into the LL-subbranch spectrum. Indeed, a spin wave can be described with the help of the components of the projected density operators (5.4), ij (q) =
m,m
m eiqR m c cm ,j , m,i
which represent coherent superpositions at wave vector q of excitations from the occupied subbranch j to the empty subbranch i. One has then k possibilities for the choice of the initial subbranch j and N k for the nal one, and one obtains therefore k(N k) dierent spin-wave excitations, in agreement with the above group-theoretical analysis. Notice that all spin-wave excitations have the same dispersion, which may be calculated within a mean-eld approximation [86, 88, 89, 178], Eq = FM|ij (q)Hn ij (q) Hn |FM = which saturates at large values of q,
n Eq = 2EX = k
5 The complex generators come in by pairs of conjugate operators, and each pair corresponds to one mode.
1 2
vn (k) sin2
k
2 q k lB 2
, (5.17)
vn (k),
(5.18)
86
i.e. at twice the value of the exchange energy (5.10). This result is not astonishing insofar as the large-q limit corresponds, as we have discussed in Sec. 3.2.2 [see Eq. (3.29)], to an electron-hole pair where the electron is situated far away from the hole. The energy (5.18) is therefore nothing other than the cost in exchange energy to create a spin-ip excitation, i.e. an electron with reversed spin and a hole in the ferromagnetic ground state. Because of the large distance between the electron and the hole in such an excitation and the resulting decoupled dynamics, one may be tempted to view this energy as the activation gap of the quantum-Hall state at = k, but we shall see in Sec. 5.2.2 that there exist elementary charged excitations (skyrmions) that have in some LLs a lower energy than these spin-ip excitations. In the opposite limit of small wave vectors (qlB 1), one may not understand the excitation in terms of decoupled holes and electrons, and the excitation can therefore not contribute to the charge transport. A Taylor expansion of the sine in the spin-wave dispersion (5.17) yields the usual q 2 dispersion of the spin-wave Goldstone modes, Eq0 = in terms of the spin stiness n = s 1 4
2 vn (k)|k|2 lB . k
n 2 2 s q lB , 2
(5.19)
(5.20)
One notices that the above results for the excitation energies do not depend on the size of the internal symmetry group, but they can be derived within the SU(2) model of the quantum Hall ferromagnetism [175, 179] the enhanced internal symmetry of graphene (or of a general N -component system) aects only the degeneracies of the dierent modes.
5.2.2
Skyrmions and entanglement
In addition to the above-mentioned spin-wave modes, the SU(N ) ferromagnetic ground state (5.16) is characterised by a particular elementary excitation that consists of a topological spin texture, the so-called skyrmion [179]. Similarly to a spin-wave in the limit qlB 1, the variation of the spin texture in a skyrmion excitation is small on the scale of the magnetic length lB , such that its energy is determined by the spin stiness (5.20) in the small-q limit. Indeed, one may show that its energy is given by [175, 176, 179] Esk = 4n |Qtop |, s (5.21)
in terms of the topological charge Qtop , which may be viewed as the number of times the Bloch sphere is covered by the spin texture [see Fig. 5.3(b)] and which we discuss in more detail below. Skyrmions are the relevant elementary excitations of the quantum Hall ferromagnetism if the energy (5.21) is lower than that of an added electron (or hole) with reversed spin, which is nothing

z
87
(a)
y x
(b)
z
y x
Excitations of the SU(2) ferromagnetic state. (a) Spin waves. Such an excitation can be continuously deformed into the ferromagnetic ground state [spin represented on the Bloch sphere (right) by fat arrow] the grey curve can be shrinked into a single point (b) Skyrmion with non-zero topological charge (from Ref. [180]). The excitation consists of a a reversed spin at the origin z = 0, and the ferromagnetic state is recovered at large distances |z/| . Contrary to spin-wave excitations, the spin explores the whole surface of the Bloch sphere and cannot be transformed by a continuous deformation into the majority spin (fat arrow).
Figure 5.3:
n other than the exchange energy (5.10), i.e. if Esk < EX . Whereas this condition is fullled in non-relativistic LLs only in the lowest one n = 0, skyrmions are the lowest-energy elementary excitations in the graphene LLs n = 0, 1 and 2 [163, 178], as a consequence of the dierence in the form factors. As in the case of the spin waves discussed above, the skyrmion energy is independent of the size of the internal symmetry group, and we will rst illustrate the skyrmion texture in an eective SU(2) model, where the texture is formed only from states within the last occupied (k) and the rst unoccupied (k + 1) LL subbranch. The skyrmion may then be described with the help of the wave function, in terms of the complex coordinate z = (x iy)/lB ,
|Sk,k+1 =
||2
1 z k (z) + k+1 (z) + |z|2
(5.22)
where | k (z) corresponds to states in the subbranch k and | k+1 (z) to those in k + 1, at the position z. One notices that at the origin z = 0 the spin associated with these two components is because the rst component of Eq. (5.22) vanishes, whereas the spins are at |z/| (see Fig. 5.3), where the ferromagnetic ground state is then recovered. The parameter plays the role of the skyrmion size, measured in units of lB indeed, for |z| = ||, both components are of the same weight and the spin is therefore oriented in the xy-plane.
88
The skyrmion excitation (5.22) can also be illustrated on the so-called Bloch sphere on the surface of which the (normalised) spin moves (see Fig. 5.3). The angles ( for the azimuthal and for the polar angle) of the spin orientation on the Bloch sphere correspond to the SU(2) parametrisation | = cos(/2)| + sin(/2) exp(i)| , and the spin orientation at the circle |z| = || in the complex plane describes the equator of the Bloch sphere. The topology of the skyrmion excitation becomes apparent by the number of full circles the spin draws when going around the origin of the xy-plane on the circle |z| = ||. More precisely, the topological charge Qtop is not dened in terms of such closed paths, but it is the number of full coverings of the Bloch sphere in a skyrmion excitation [Qtop = 1 in the example (5.22)]. Notice that a spin-wave excitation has a topological charge Qtop = 0 and corresponds to an excursion of the spin on the Bloch sphere that is not fully covered and that can then be reduced continuously to a single point describing the ferromagnetic ground state [Fig. 5.3(a)]. The above considerations may be generalised to systems with larger internal symmetries, i.e. to SU(N ) quantum Hall ferromagnets. The state (5.22) is invariant under the SU(N ) subgroup SU(k 1) SU(N k 1), where the rst factor describes a rotation of the occupied subbranches that do not take part in the skyrmion excitation and the second factor is associated with a symmetry transformation of the corresponding unoccupied subbranches k + 2, ..., N . A similar group-theoretical analysis as the one presented in Sec. 5.2.1 yields the number of residual symmetry transformations [178] 2k(N k)+ 2(N 1), where the rst term describes the Goldstone modes of the ferromagnetic ground state, and the second one corresponds to the N 1 internal modes of the skyrmion excitation. In addition to the topological charge, skyrmions in quantum Hall systems carry an electric charge that coincides, for = k with the topological charge. Indeed, the skyrmion state (5.22) describes an electron that is expelled from the origin z = 0 in the -component, and its net electric charge is therefore that of a hole. This means that skyrmions are excited when sweeping the lling factor away from = k, and the net topological charge is then given by Qtot = | k|NB . The number of internal modes is then Qtot (N 1), in addition to one mode per charge that corresponds to a simple translation z z + a of the excitation [156]. As a consequence of the Coulomb repulsion, it is energetically favourable to form a state in which Qtot skyrmions of charge 1 are homogeneously distributed over the 2D plane than a single defect with charge Qtot [175]. A natural (semi-classical) candidate for the ground state of Qtot skyrmions is then a skyrmion crystal [181] that has recently been revisited in the framework of the SU(4) symmetry in graphene [182, 183]. In this case, the Qtot (N 1) internal modes, which are dispersionless zero-energy modes in the absence of electronic interactions or Zeeman-type symmetry-breaking terms, are expected to yield N 1 Bloch bands of Goldstone-type, in addition to the Qtot translation modes that form a magnetic-eld phonon mode of the skyrmion crystal with a characteristic q 3/2 dispersion [184].
SU(4) Quantum Hall Ferromagnetism in Graphene Spin-valley entanglement in graphene
89
In an experimental measurement, one typically has not direct access to the full SU(4) spin that describes the internal degrees of freedom in graphene LLs, but only to the SU(2) part associated with the physical spin, e.g. in a magnetisation measurement. It is therefore useful to parametrise the SU(4) spin in a manner such as to keep track of the two SU(2) copies associated with the physical spin and with the valley isospin, respectively. This may be achieved with the socalled Schmidt decomposition of the four-spinor |(z) = cos |S |I + sin ei |S |I , 2 2 (5.23)
where and are functions of the complex position z, and the local twocomponent spinors |S , |S , |I , and |I are constructed according to | =
cos 2 sin 2 ei
and
| =
sin ei 2 cos 2
The angles and dene the usual unit vector n(, ) = (sin cos , sin sin , cos ) , which explores the surface of the Bloch sphere depicted in Fig. 5.3. Notice that one has two Bloch spheres, one for the unit vector n(S , S ) associated with the spin angles S and S and a second one for n(I , I ) for the valley-isospin angles I and I (see Fig. 5.4). In addition, one may associate a third Bloch sphere with the angles and that describe the degree of factorisability of the wave functions and thus the degree of entanglement between the spin and the valley isospin [156]. With the help of the Schmidt decomposition (5.23), one obtains immediately the reduced density matrices for the spin and the valley-isospin sectors S I = = |S S | + sin2 |S S | , 2 2 TrS (| |) = cos2 |I I | + sin2 |I I | , 2 2 TrI (| |) = cos2
respectively, and the local spin and valley-isospin densities are simply ma = Tr(S S a ) = cos S |S a |S = cos na (S , S ) S and m = Tr(I I ) = cos I |I |I = cos n (I , I ), I (5.24) (5.25)
where S a and I represent the components of the spin and valley-isospin operators, respectively [see Eq. (5.9)]. One notices from these expressions that, for the case = 0 or (i.e. cos2 < 1), the local (iso)spin densities are no longer normalized, but they are of length |mS/I |2 = cos2 . Thus, in a semiclassical picture, the (iso)spin dynamics is no longer restricted to the surface of
90
(a) z S y S

(b) z
I y x I x
spin
pseudospin
(c)
=0
=0 y x
entanglement
From Ref. [156]; Bloch spheres for entangled spin-pseudospin systems. Bloch sphere for the spin (a), pseudospin (b), and a third type of spin representing the entanglement (c). In the case of spin-pseudospin entanglement (| cos | = 1), the (pseudo)spinmagnetizations explore the interior of their spheres, respectively (black arrows).
Figure 5.4:
the Bloch sphere, but explores the entire volume enclosed by the sphere (Fig. 5.4) [156]. This result indicates that one may be confronted, in the case of full entanglement (e.g. = /2), with an SU(4) quantum Hall ferromagnet the (spin) magnetisation of which completely vanishes, as one would naively expect for an unpolarised state.
5.2.3
SU(4)-symmetry-breaking terms
The considerations presented in the previous sections 5.2.1 and 5.2.2 were concerned only with the fully symmetric model and neglected all types of symmetrybreaking terms, such as the one due to valley backscattering (5.13) or Zeemantype terms that couple to the spin or valley-isospin degrees of freedom. This approach may be justied by considering the orders of magnitude for the dierent energy scales. As already mentioned above, the SU(4)-symmetric interaction is characterised by an energy scale e2 /RC 250 B[T]/n K (for graphene on a SiO2 substrate with a typical dielectric constant 2.5),6 whereas the symmetry-breaking interaction term (5.13) in n 1 (or similar terms in n = 0 [86, 87, 88]) is suppressed by a factor a/lB . The Zeeman eect yields an energy scale that is on the order of Z 1.2B[T] K, for a g-factor that has been experimentally determined as g 2 [56]. Notice that this energy scale is only slightly larger than the one set by symmetry-breaking interaction terms on the
6 Notice that this value does not account for the screening due to the electrons in the completely lled valence band. As we have argued in Sec. 3.2.4, this yields an additional factor of = 1 + rs /2 [see Eq. (3.37)] to the dielectric constant, and one then nds 4.8, which agrees well with the value of the dielectric constant used in Ref. [86].
91
lattice scale. There have also been proposals for Zeeman-type terms that couple to the valley isospin, such as ripples of a longer wave length [185]. The latter may be described by a strain-induced disordered gauge eld the magnetic eld of which yields an easy-plane anisotropy in n = 0 [159], similarly to the backscattering term (5.13) in higher LLs [85]. A related symmetry-breaking mechanism is due to frozen phonon uctuations that allow for an energy reduction that is proportional to the applied magnetic eld and thus scales in the same manner as the Zeeman eect. Whereas it is argued in Ref. [186] that a spontaneous crumbling of the graphene sheet due to out-of-plane optical phonons breaks the inversion symmetry of the lattice and yield a relativistic mass term, more recent studies [187, 188] concentrate on in-plane phonons that yield a kekule type distortion. Both mechanisms open a gap in the central n = 0 LL, but simply renormalise the energy of the LLs n = 0 while keeping the four-fold spin-valley degeneracy [189]. Also these terms are expected to be small when compared to the leading SU(4)-symmetric interaction term, although they have been estimated to be slightly larger than the Zeeman eect, kek 2B[T] K, [188, 190]. Another class of terms is concerned with the breaking of time-reversal symmetry. In an original work, Haldane argued that such terms may arise in a honeycomb lattice with zero net magnetic eld but with an inhomogeneous ux distribution inside each hexagon and may yield a quantum Hall eect in the absence of an external magnetic eld [189]. Apart from this theoretical situation, it has been shown that time-reversal symmetry breaking is also provided by spin-orbit interactions [191], which, however, turn out to be vanishingly small (on the order of 10 mK), whereas an extrinsic Rashba-type spin-orbit coupling in graphene can be on the order of 1 K [192]. The dierent energy scales are summarised in the table below for dierent values of the magnetic eld, where we present the results with and without screening due to the completely lled valence band.
energy e2 /lB e2 / lB Z kek sb < (e2 /lB )(a/lB ) sb < (e2 / lB )(a/lB ) value for arbitrary B 250 B[T] K 130 B[T] K 1.2B[T] K 2B[T] K < 1B[T] K < 0.5B[T] K for B = 6 T 620 K 330 K 7K 12 K <6K <3K for B = 25 T 1250 K 660 K 30 K 50 K < 25 K < 13 K
These considerations concerning the dierent energy scales yield the conclusion that at = k (i.e. at = 1 and = 0 in graphene) the electronic properties are governed by the SU(4)-symmetric part of the interaction, which yields an SU(4) ferromagnetic ordering. The symmetry-breaking terms may then favour a particular orientation of the SU(4)-spin, e.g. in the direction of the magnetic eld in the presence of a Zeeman eect or else in the xy-plane for the valley isospin as a consequence of the symmetry-breaking term (5.13) [85]. In the n = 0 LL, an easy-axis ferromagnetism in the valley-isospin channel,
92
with a residual Z2 symmetry, has furthermore been proposed [86, 88], whereas a Kekul distortion forces the valley-isospin polarisation to be oriented in the xye plane [187, 188]. The dominating energy scale of the states is therefore e2 /RC , and symmetry-breaking terms yield an additional contribution, e.g. in the case of the Zeeman eect, the spin-wave mode (5.17) acquires a gap even at small wave vectors, Eq Eq + Z . The eect of symmetry-breaking terms on the quantum Hall states at = k and their activation gaps is discussed in Sec. 5.2.5 in the context of the available experiments at these lling factors.
5.2.4
Comparison with magnetic catalysis
An alternative scenario for the degeneracy lifting in n = 0 is that of the magnetic catalysis [87, 193, 194, 195, 196], which was discussed even before the discovery of graphene [197, 198]. According to this scheme, the Coulomb interaction yields a spontaneous generation of a mass term for the (originally massless) 2D electrons once the magnetic eld increases the density of states at zero energy by the formation of the highly degenerate n = 0 LL. As a consequence of this mass generation, the particles condense in a state of coherent particlehole pairs (excitonic condensation). This excitonic condensation is at rst sight reminiscent of the excitonic condensation at = 1 in non-relativistic bilayer quantum Hall systems [199, 200, 201]. Its superuid behaviour gives rise to a zero-bias anomaly in the tunneling conductance between the two layers [202] as well as to a simultaneous suppression of the longitudinal and the Hall resistance in a counterow experiment [203, 204]. Indeed, this bilayer excitonic condensate may be described as an easy-plane quantum Hall ferromagnet [175], where the spin mimics the layer index. The origin of this easy-plane anisotropy stems from the dierence in the interactions between electrons in the same layer as compared to the weaker one for electron pairs in dierent layers. This comparison with non-relativistic 2D electrons in bilayer systems indicates that there may exist a close relation between the quantum Hall ferromagnetism and the scenario of the magnetic catalysis also in graphene in a strong magnetic eld. Notice, however, that the excitonic state in graphene is not in the same universality class as that of the quantum-Hall bilayer in the latter case the symmetry of the (interaction) Hamiltonian is U(1) as a consequence of the easy-plane anisotropy, and the symmetry breaking is associated with a superuid mode that disperses linearly with the wave vector, q. In contrast to this system, the interaction Hamiltonian (5.8) has the full SU(4) symmetry, and even for a suciently strong Zeeman eect, the symmetry is quite large with SU(2) SU(2) , i.e. each spin projection and is governed by the residual SU(2) valley symmetry and has the characteristic q 2 isospin-wave modes. In order to make a closer connection between the two scenarios, we may consider the symmetry of the mass terms, which play the role of order parameters in the magnetic catalysis and which are generated on the mean-eld level within the Hartree-Fock approximation of the relativistic model when both valleys are
SU(4) Quantum Hall Ferromagnetism in Graphene taken into account [195]. In addition to the Zeeman-type terms spin Z
z valley AB spin
93
(5.26)
A A A A A A A A K, K, K, K, + K , K , K , K , + (A B)
and
z valley valley AB spin Z
(5.27)
for the spin and valley channels, respectively,7 there are the mass terms [193, 195]
z z Ms valley AB spin
(5.28)
and [189]
z Mt (valley AB spin )
(5.29)
for the singlet and triplet channels of the SU(2) groups, respectively.8 A B Notice that in the zero-energy LL the components K, and K , vanish as a consequence of the chiral properties which identify the sublattice and the valley isospins in n = 0, as we have discussed in Sec. 2.1.2. The singlet mass term (5.28) then renormalises simply the overall chemical potential,
z z n=0 Ms valley AB spin (valley AB spin )
whereas the triplet mass term (5.29) is no longer distinguishable from the order parameter associated with the valley Zeeman term (5.27),
z z Mtn=0 (valley AB spin ) valley AB spin .
This scenario is reminiscent of the degeneracy lifting in n = 0 as a consequence of a Peierls type instability due to the out-of-plane phonon [186]. Furthermore, in higher LLs, the triplet mass term (5.29) is not capable of lifting the level degeneracy because it only renormalises the LL energy [189], whereas the level splitting due to a singlet mass term is a consequence only of the z valley component in the expression (5.28), as in the case of the valley-Zeeman term (5.27). For the SU(2) SU(2) symmetry, the residual ferromagnetic properties are revealed by the arbitrariness in the choice of the z-quantisation z axis for the valley isospin valley , and one should therefore view the magnetic catalysis and the quantum Hall ferromagnet as intimitely related [195].
7 We have used a compact notation where is an eight-spinor associated with the tensor product of the valley isospin (Pauli matrices valley and valley ), the sublattice isospin (AB and AB ) and the true spin (spin and spin ). For a better understanding, we have given the explicit expression in terms of spinor components in Eq. (5.26). 8 For a classication of all possible dierent order parameters that may arise in graphene, see Ref. [205].
94
5.2.5
Before discussing the experimental results on the quantum Hall eect, which may be understood in the framework of the quantum Hall ferromagnet, a clarication on the lling factor is required. In the preceding sections of this chapter, the lling factor = k has been dened with respect to the bottom of the partially lled LL. However, in graphene, the lling factor G of the n = 0 LL is naturally dened with respect to its centre, which corresponds to the chargeneutral point in the undoped case. For the remaining parts of this chapter, we thus make a clear distinction in the notation for the lling factors and introduce explicitly the relation = G + 2. (5.30) Experimentally, quantum Hall eects that require a lifted spin-valley degeneracy have been observed in exfoliated graphene at G = 0, 1 and 4 [56, 206], where the latter corresponds to the LLs 1. The Hall conductance measured in the experiment of Ref. [56] is shown in Fig. 5.5(a). The experiments indicate that the quantum Hall states at G = 1 are sensitive only to the perpendicular component of the magnetic eld [206] such that a spin Zeeman eect is unlikely to play a role in the physical properties of these states, in contrast to those at G = 4. Furthermore, activation-gap measurements at G = 1, though not fully conclusive, seem to indicate a B dependence [206], as one would expect for the scaling of the interaction energy e2 /lB B [Fig. 5.5(b)]. In the framework of the quantum Hall ferromagnet, the activation gap may be viewed as the energy to create a skyrmion of topological charge Q = 1 and an anti-skyrmion of charge Q = 1 that are well-separated from each other such that one may neglect their residual interaction. The energy of such a skyrmion anti-skyrmion pair is then given, in the absence of symmetry-breaking terms, by twice the energy (5.21) sym = 8n . (5.31) a s The energies of the theoretical activation gaps for n = 0 and n = 1 are shown in the table below.
activation gap n = 0 for = 2.5 n = 0 for 5 n = 1 for = 2.5 n = 1 for 5
e2 2 lB e2 1 2 2 lB 7 e2 32 2 lB e2 7 32 2 lB 1 2
The quantum Hall eect at = 1 and = 0
arbitrary value of B 160 80 70 40 B[T] K B[T] K B[T] K B[T] K
B=6T 390 K 190 K 170 K 90 K
B = 25 T 780 K 390 K 340 K 170 K
Consider the scenario in which the Zeeman eect is the only SU(4)-symmetry breaking term.9 This is indeed consistent with the experimental situation because the Zeeman eect would give a small energetic advantage to spin- electrons, and the electrons at G = 1 then preferentially form a valley-spin
9 The energetic argument remains valid if the dominant term is a valley-Zeeman eect if one interchanges the role of the spin and the valley isospin.
95
(a)
25T
T=1.4K
(b)
42T 45T 37T 9T (30mK)
30T 25T
Quantum Hall states at G = 0, 1, and 4. (a) From Ref. [56]; Hall conductance of a graphene sheet at T = 1.4 K fo several values of the magnetic eld up to 45 T. The upper inset shows the Hall and longitudinal resistances at 25 T. The level spectrum is shown on the left-hand side and associated with the dierent Hall plateaus (blue for G = 0 and red for G = 1). (b) From Ref. [206]; magnetic-eld dependence of the activation gap. The (red) continuous line shows the expected B behaviour for an interation-dominated gap at = 1 (red squares), whereas those at = 4 (blue and black squares) seem to scale linearly with the magnetic eld (dashed lines).
Figure 5.5:
96
ferromagnet.10 The activation gap would then be dominated by valley (anti)skyrmions with no reversed physical spin such that one would not expect any dependence of the gap on the in-plane component of the magnetic eld, in agreement with the experimental ndings [206]. The situation is dierent when both valley states of the spin- branch are occupied, i.e. at G = 0; an excitation of the SU(4) ferromagnet with a full spin polarisation would then necessarily comprise reversed spins, and the corresponding Zeeman energy must be taken into account in the energy of the (spin) skyrmionanti-skyrmion pair (5.31), Z = 8n + 2Nrs Z , a s (5.32)
where Nrs ||2 is the number of reversed spins in a single (anti-)skyrmion. Notice that this number depends on the competition between the Zeeman eect itself, which tries to reduce the skyrmion size , and the cost in exchange energy due to the strong variation in small textures [179, 175].11 The energy of a skyrmionanti-skyrmion pair in the spin channel at G is therefore larger than that (5.31) of a pair in the valley channel this energy increase may even be signicant for large skyrmions because of the larger number of reversed spins. As a thumbrule, the stability of a quantum Hall state is proportional to the activation gap, which we have identied in the present case with the skyrmion anti-skyrmion energy. The above arguments lead then to the conclusion that the state at G = 0 is more stable, as a consequence of the additional Zeeman eect, than at G = 1, where the eect is absent. This is indeed in agreement with the experimental observations of Ref. [56], where the G = 0 state is found at elds B 10 T, whereas the less stable states at G = 1 require a minimal eld of B 17 T. In principle, this scenario may also apply in the LLs with n = 1, where the quantum Hall ferromagnetism supports skyrmion excitations [163, 178], and one would then expect quantum Hall states at G = 3, 4 and 5. Notice, however, that only the states at G = 4 have yet been observed. This observation may nd a natural explanation in the framework of a sublattice degeneracy lifting, such as the Peierls-type instability [186] or the spontaneous generation of a triplet mass discussed in the previous subsection. Indeed, as mentioned above, such a scenario would not yield a valley degeneracy lifting in LLs dierent from n = 0 [189]. A possible explanation is nevertheless also found in the framework of quantum Hall ferromagnetism whereas the symmetric activation gaps (5.31) may be too small to resolve these states, in view of impurity-broadened LLs, the Zeeman eect may help in its stability. The roughly linear dependence of the gaps at G = 4 [206], expected for the Zeeman eect and shown in Fig. 5.5(b), hints at such a possibility. Notice that, if one considers the activation energy at 17 T above which the G = 1 states are found [56], one would expect, in the
G = +1, the same argument applies due to particle-hole symmetry. energy cost may be evaluated from a gradient expansion of the energy in the magnetisation elds. At leading order, one obtains, however, a non-linear sigma model that is scale-invariant, such that the energy cost must be calculated at higher orders [175].
11 This 10 For

(a)
K, Energy Energy v Z Z K, K, K, y Z
v Z
97
(b)
K, K, K, K, y
Figure 5.6: Possible scenarios for the lifted spin-valley degeneracy at G = 0. (a) Z > v Z in the bulk. When approaching the edge, the energy dierence between the two valleys increases drastically, and two levels (K , ) and (K, ) cross the Fermi energy at the edge depicted by the dashed line. (b) v > Z in the bulk (Quantum Hall state). The K Z subbranches are already located above the Fermi energy, and those of K below, such that the energy dierence is simply increased when approaching the edge with no states crossing the Fermi energy (Insulator). same sample, a valley splitting in n = 1 above 80 T [85], as a consequence of the reduced Coulomb energy scales in higher LLs (see table above). The = 0 quantum Hall eect is insofar special as the plateau is only visible in the Hall conductance, and not in the resistance, as a consequence of the vanishing carrier density and particle-hole symmetry. One may nevertheless speak of a quantum Hall state, and not an insulator, because the longitudinal resisitivity does not diverge when lowering the temperature [207]. Furthermore, it corroborates the above-mentioned scenario of a Zeeman eect as the leading symmetry-breaking term at G = 0, as may be seen from Fig. 5.6. A quantum Hall state may be viewed as a bulk insulator with (chiral) conducting electrons at the sample edges, which cross the Fermi energy. A simplistic model for the edge connement of relativistic carriers in graphene is obtained by an additional term z M (y)AB in the Hamiltonian, which has the symmetry of a triplet mass (5.29) or else, in n = 0, that of a valley-Zeeman term (5.27), as argued in the preceding subsection. The parameter M (y) is zero in the bulk and increases drastically at the edge at a certain value of the coordinate y.12 Although the model is a simplication to treat the graphene edges in the continuum description of the Dirac equation, a more sophisticated treatment that takes into account the geometry of the edges yields, apart from a ne structure of the levels at the edge, qualitatively similar results [208]. If, in addition to the Zeeman eld Z that couples to the physical spin, one has in the bulk a valley-Zeeman eect z v AB (as for example in the above-mentioned Peierls-type instability), the Z connement term simply adds up to the latter. We may then consider two scenarios for the bulk: (a) the Zeeman eect is larger than the valley-Zeeman eect, Z > v , or (b) the inverse situation Z
12
For the present argument, we consider translation invariance in the x-direction.
98
Z > v . In the rst case depicted in Fig. 5.6(a), the spin- branches are Z occupied at G = 0 and the spin- branches empty. This is the scenario exposed above, where we have considered v = 0. When approaching the edge, however, Z the valley-Zeeman term is enhanced by the rapidly increasing contibution from M (y), and the (K, ) level eventually crosses the (K , ) one at the edge at the Fermi energy [207]. In view of the above-mentioned criterion, this is a quantum Hall state with a bulk insulator and (two) conducting channels. The edge states are not chiral, but each spin orientation yields a chiral state (helical edge). The quantum Hall state therefore remains stable unless magnetic impurities couple the two chiralities [209]. One notices furthermore a change in the spin polarisation at the edge; whereas the spin polarisation in the bulk is complete, the system becomes spin unpolarised at the edge. If one takes into account the exchange interaction, the change in the polarisation takes place over a certain distance, and the conducting properties may be described in terms of spincarrying one-dimensional edge excitations [209]. In the opposite limit, where the valley Zeeman eect is larger than the Zeeman eect in the bulk, the system at G = 0 is already valley-polarised, and an increase of the valley-Zeeman term when approaching the edge does not induce a level crossing at the Fermi energy. Therefore, there are no zero-energy states at the edge, and the system would be insulating both in the bulk and at the edge. Although this is in contradiction with the experimental ndings of Refs. [56] and [206], it has been shown experimentally that the system at G = 0 becomes insulating at very large magnetic elds [210, 211]. Interestingly, the transition between the quantum Hall and the insulating state is governed by a scaling of the resistivity that is reminiscent of a Kosterlitz-Thouless (KT) phase transition [212] if one replaces the temperature by the magnetic eld as the parameter driving the transition. It has been argued that this eect may be understood within the above scenario of a Zeeman-dominated quantum Hall ferromagnet in the bulk, in the framework of a Luttinger-liquid description of the domain wall seperating the polarised from the unpolarised region at the edge [209]. Another appealing picture for this unusual behaviour of the resisitivity has been provided in Refs. [187, 188] if the valley gap due to a Kekul-type e distortion is the leading symmetry-breaking mechanism, both spin branches are lled and the valley-isospin magnetisation at G = 0 is forced to be in the xyplane, such that the system is naturally in the KT universality class. The KT transition may then not only be triggered by temperature but eventually also by disorder when lowering the magnetic eld. The high-eld insulating phase found in Refs. [210, 211] would then correspond to an easy-plane valley ferromagnet, whereas the low-eld metallic phase is identied with a disordered state. Further experimental and theoretical studies are necessary to clearly identify the leading symmetry-breaking mechanism in the zero-energy LL at G = 0 and 1. We nally mention scanning-tunneling spectroscopic results for the level splitting at G = 0 that were performed on graphene on a graphite substrate [213]. Although a gap has been observed as one may expect in the framework of the above scenario, saturates as a function of the magnetic eld. This is in it contrast to both the B-behaviour of an interaction-dominated gap as well as
Fractional Quantum Hall Eect in Graphene
99
to the linear dependence of the Zeeman eect. A probable origin of this gap is the commensurability of the graphene lattice with the graphite substrate that may lift the inversion symmetry between the two sublattices by a term of the type (5.29). The coupling with the substrate being essentially electrostatic, one would then expect no or only a weak magnetic-eld dependence of the splitting.
5.3
The most salient aspect of strongly correlated 2D electrons in partially lled LLs is certainly the FQHE, which is due to the formation of incompressible liquid phases at certain magic values of the lling factor. The observation of the FQHE, though theoretically expected in graphene (for a recent review see Ref. [166]), requires high-quality samples with low disorder, as in the case of non-relativistic 2D electron systems in semiconductor heterostructures. High-quality suspended graphene samples may indeed be obtained by current annealing that increases the electronic mobility to up to 200 000 cm2 /Vs in the case of moderately doped graphene [214, 215]. A FQHE at = 1/3 has indeed very recently been observed in currentannealed suspended graphene samples in the two-terminal conguration, where the voltage (and thus the resistance) is measured between the same two contacts used to drive the electric current through the sample [149, 150].13 Although this conguration allows for an identication of the FQHE, it does not allow for a simultaneous measurement of the Hall and the longitudinal resistance the signature of a quantum Hall eect is indeed the quantisation of the Hall resistance, which yields the characteristic Hall plateau, while the longitudinal resistance shows activated behaviour. Such a simultaneous measurement requires, however, a four- or six-terminal conguration for which a graphene FQHE has not yet been observed. The resistance measured in a two-terminal conguration is thus a superposition of the longitudinal and the Hall resistance. As a consequence of the above considerations concerning the model of electrons restricted to a single relativistic LL (Sec. 5.1) and the experimental situation, we concentrate in this section on a theoretical discussion of the graphene FQHE from the perspective of an SU(4) quantum Hall system.
5.3.1
Generalised Halperin wave functions
The theoretical study of the FQHE is intimitely related to trial N -particle wave functions. In 1983, Laughlin proposed a one-component wave function [216],
N
L ({zk }) = m
k<l
(zk zl ) e
N k
|zk |2 /2
(5.33)
which allows for an understanding of incompressible FQHE states at the lling factors = 1/m that are determined by the exponent m for the particle pairs
13 There are also some weak indications for FQHE states at other lling factors than = 1/3, but they are not yet conclusive at the moment where these lines are written.
100
k, l in Eq. (5.33). The variable zk = (xk iyk )/lB is the complex position of the k-th particle, and the form of the Laughlin wave function (5.33) is dictated by the analyticity condition for wave functions in the lowest LL.14 Furthermore, the exponent m must be an odd integer as a consequence of the fermionic statistics imposed on the electronic wave function. Even if Eq. (5.33) describes only a trial wave function, one can show that it is the exact ground state for a class of model interactions that yield, with the help of Eq. (5.15), the pseudopotentials [161] V > 0 for < m and V = 0 for m. (5.34) Although the Coulomb interaction does not full such strong conditions, the pseudopotentials decrease as 1/ m for large values of m. Because the incompressible ground state is protected by a gap that is on the order of V1 , one may view the pseudopotentials Vm as an irrelevant perturbation that does not change the nature of the ground state. Indeed, exact-diagonalisation calculations have shown that, for the most prominent FQHE at = 1/3, the overlap between the true ground state and the Laughlin state (5.33) is extremely large (> 99%) [217, 218]. Soon after Laughlins original proposal, Halperin generalised the wave function (5.33) to the SU(2) case of electrons with spin, in the absence of a Zeeman eect [219] one has then two classes of particles, N spin- and N spin- par() () ticles, which are described by the complex positions zk and zk , respectively. In the (theoretical) absence of interactions between electrons with dierent spin orientation, the most natural ground-state candidate would then be a simple product of two Laughlin wave functions (5.33), L m zk
()
L m
zk
()
one for each spin component with the exponents m and m , respectively, that need not necessarily be identical. Inter-component correlations may be taken into account by an additional factor
N N k k
zk zk
()
()
where the exponent n may now also be an even integer because the fermionic anti-symmetry condition is concerned only with electrons in the same spin state. Halperins idea is easily generalised to the case of more than two components, and the corresponding trial wave function for an SU(N ) quantum Hall system with N components reads [165]
SU(N ) m1 ,...,mK ;nij = L 1 ,...,mN inter , nij m
(5.35)
14 The lowest-LL condition of analytic wave functions may seem a very strong restriction when discussing FQHE states in higher LLs. However, the model (5.8) indicates that all LLs may be treated as the lowest one, n = 0, if the interaction potential is accordingly modied. We adopt this point of view here.
Fractional Quantum Hall Eect in Graphene in terms of the product L 1 ,...,mN m =

N Nj
101
j=1 kj <lj
zkj zlj
(j)
(j)
mj
N j=1
Nj kj =1
|zk |2 /4
j
(j)
of N Laughlin wave functions and the term inter nij =

N Ni Nj i<j ki kj
zki zkj
(i)
(j)
nij
which describes inter-component correlations. As in the case of Halperins twocomponent wave function [219], the exponents mj must be odd integers for fermionic particles whereas the exponents nij may also be even integers. These exponents dene a symmetric N N matrix M = nij , where the diagonal elements are nii mi . This exponent matrix encodes the statistical properties of the quasi-particle excitations, such as their (fractional) charge and their statistical angle [200, 220]. Moreover, the exponent matrix M determines the component densities j or, equivalently, the component lling factors j = j /nB , 1 1 . . = M 1 . , (5.36) . . . N 1
where = 1 +. . .+N is the total lling factor measured from the bottom of the lowest LL. Notice that Eq. (5.36) is only well-dened if the exponent matrix M is invertible. In this case, all component lling factors j are completely determined, whereas otherwise some of the component llings remain unxed, e.g. 1 and 2 for the sake of illustration, although the sum of them (1 + 2 ) is xed. This is nothing other than a consequence of the underlying ferromagnetic properties of the FQHE state that, similarly to the states at = k discussed in Sec. 5.2, are described by subgroups of SU(N ). Finally, we notice that not all SU(N ) wave functions (5.35) describe incompressible quantum liquids with a homogeneous charge density for all components. A generalisation of Laughlins plasma picture, according to which the modulus square of the trial wave function corresponds to the Boltzmann weight of a classical 2D plasma [216], shows that all eigenvalues of the exponent matrix M must be positive (or zero for states with ferromagnetic order). Otherwise, some of the dierent components phase-separate in the 2D plane because the inter-component repulsion between them exceeds the intra-component repulsion [221].
5.3.2
Polarised vs. unpolarised states
After these general considerations, we concentrate on the case of graphene with N = 4, where we attribute the four spin-valley components as 1 = (, K),
102
2 = (, K ), 3 = (, K), and 4 = (, K ). We discuss wave functions, where all intra-component exponents are identical mi = m, i.e. we consider the same interaction potential for any of the components, as it is the case in graphene. Furthermore, we restrict the discussion to states with n13 = n24 na and n12 = n14 = n23 = n34 ne , which makes an explicit distinction between intercomponent correlations in the same valley (na ) and those in dierent valleys (ne ). This distinction may occur somewhat arbitrary indeed, it does no longer treat the spin on the same footing as the valley isospin as stipulated by the SU(4) symmetry of the interaction model (5.8). However, this distinction is useful in some cases if one intends to describe states with intermediate polarisation, such as for a moderate Zeeman eld. The equivalence between spin and valley isospin is naturally restored for ne = na . We use the notation [m; ne , na ] to describe these subclasses of trial wave functions (5.35). [m; m, m] wave functions If all exponents are identical odd integers m = ne = na , one obtains a completely anti-symmetric orbital wave function that is accompanied by a fully symmetric SU(4) spin-valley wave function. As we have argued in Sec. 5.2.1, this situation represents precisely a perfect SU(4) quantum Hall ferromagnet indeed, for m = 1, the generalised Halperin wave function (5.35) is nothing other than the orbital wave function of the state at = 1, i.e. when one of the subbranches is completely lled. The SU(4) symmetry is then spontaneously broken, and the group-theoretical analysis presented in Sec. 5.2.1 yields 3 degenerate Goldstone modes that are generalised spin waves. The situation is exactly the same for any other odd exponent m, but the orbital wave function (5.35) is then a Laughlin wave function (5.33) in terms of the particle positions zk regardless of their internal index j = 1, ..., 4. The ferromagnetic properties of these wave functions may be described by the same equations as the spin-wave and skyrmion modes derived in Sec. 5.2 if one takes into account a renormalisation of the spin stiness, as it has been discussed extensively in the literature for SU(2) quantum Hall ferromagnets (see Refs. [175, 176, 179]). States described by such a wave function are ground-state candidates for the lling factors = 1/m, which correspond to the graphene lling factors [see Eq. (5.30)] G = 2 + 1/m or hole states at G = 2 1/m. One notices that these lling factors do not correspond to the FQHE at G = 1/3 observed in the experiments [149, 150]. Such states may, however, be described in a similar framework if the spin-valley degeneracy is (at least partially) lifted. Consider for example a large Zeeman eect, in which case one may view the G = 1/3 FQHE as a valley state described by an SU(2) Halperin wave function.15 Although this is an unlikely situation because of the smallness of the (bare) Zeeman eect [56], there is a cooperative eect where the Zeeman eect is enhanced by the formation of a quantum Hall ferromagnetism as the
15 Naturally, one may also invoke a lifted valley degeneracy with an SU(2) spin state or else a simple one-component Laughlin state if the spin-valley degeneracy is fully lifted by external elds.
103
background state at G = 0 the FQHE at G = 1/3 may then be interpreted as quasi-particles that condense into a Laughlin-type 1/3 state [222]. [m; m 1, m] wave functions As already mentioned in the discussion of Laughlins trial wave function, the [m; m, m] state screens all pseudopotentials (5.15) with pair angular momentum < m and consequently has a large overlap with the true ground state for the SU(4)-symmetric Coulomb potential (5.6) in n = 0 [164]. A similarly good wave function is [m; m 1, m], where the intervalley-component exponents are decreased by one. It also screens all pseudopotentials V<m in any pair of electrons within the same valley, but an electron pair in two dierent valleys is aected by the pseudopotential V=m1 . The lling factor, where this wavefunction occurs, is16 = 2 2m 1 or G = 2 + 2 2m 1
i.e. at slightly larger densities than the [m; m, m] state discussed above. The exponent matrix M is not invertible but of rank 2, instead of rank 1 for the [m; m, m] state. As a consequence, the lling factors in each of the two valleys, K = 1 + 3 and K = 2 + 4 , respectively, are xed, K = K = 1/(2m 1), and one may view the wave function as a state with ferromagnetic spin ordering, but that is valley-isospin unpolarised. Alternatively, the [m; m 1, m] wave function may be interpreted as a tensor product of an SU(2) Halperin (m, m, m 1) isospin-singlet wavefunction [219] and a completely symmetric (ferromagnetic) two-spinor that describes the physical spin. The relevance of a [m; m 1, m] state with m = 3 in the description of the ground state at = 2/5 (G = 2 + 2/5) for the n = 0 and n = 1 graphene LLs has been corroborated by exact-diagonalisation calculations [164, 166]. Whereas the overlaps are extremely high in n = 0 (above 97% for the largest system size with N = 10 particles), they are slightly smaller in n = 1 (roughly 88%) [166]. However, the rather large overlap indicates that also in n = 1 the putative FQHE at = 2/5 is governed by a valley-isospin singlet accompanied by SU(2) spin physics, in contrast to results obtained in the framework of recent densitymatrix renormalisation group calculations [223]. It is useful to compare the situation of the [3; 2, 3] state in graphene with that of non-relativistic 2D electrons with SU(2) spin symmetry at = 2/5. It has been shown that, in the latter case, the 2/5 state is indeed well described by an unpolarised (3, 3, 2) Halperin wave function [224], which may also be viewed as an SU(2) composite-fermion wave function [225]. The energy dierence between the polarised and the unpolarised 2/5 states is, however, relatively small as compared to the Zeeman eect at the corresponding magnetic elds, such that a polarised state is usually favoured. Intriguing spin transitions, that were
16 We only discuss electronic states here, but the arguments are equally valid for the particlehole symmetric states at G = 2 2/(2m 1).
104
observed experimentally at = 2/5, indicate even more complex physical properties of the 2/5 FQHE [226, 227]. Notice that the situation of the [3; 2, 3] state in graphene is remarkably dierent from that of non-relativistic 2D electrons; even in the presence of a strong Zeeman eect, only the ferromagnetically ordered physical spin is polarised, while the state remains a valley-isospin singlet. Whether such a valley-isospin singlet state is realised in graphene depends sensitively on the valley-symmetry breaking terms; whereas a possible easy-axis ferromagnetism that has been proposed for the zero-energy LL n = 0 [86] may destroy the [3; 2, 3] state, it is favoured in the case of an easy-plane valleyisospin anisotropy, which may occur in the n = 1 graphene LL due to the valley-backscattering term (5.12) discussed in Sec. 5.1. [m; m 1, m 1] wave functions In contrast to the above-mentioned states, the [m; m 1, m 1] wave function, which is a candidate for a FQHE at = 4 4m 3 or G = 2 + 4 , 4m 3
describes an SU(4) singlet with an invertible exponent matrix M of rank 4. The lling factor in each of the spin-valley components is then 1/(4m3). As for the [m; m, m] and [m; m 1, m] wave functions, all intra-component correlations are such that the pseudopotentials V<m are screened, but V=m1 is relevant for all inter-component interactions. Exact-diagonalisation calculations for N = 4 and 8 particles have shown that the [m; m 1, m 1] wave function with m = 3 (at = 4/9) describes to great accuracy the ground state for a Coulomb interaction (5.6), with overlaps ON =8 = 0.992 in n = 0 and ON =8 = 0.944 and in the n = 1 graphene LL [166]. These results indicate that a possible 4/9 FQHE state in graphene is, remarkably, of a completely dierent nature than the composite-fermion state at = 4/9 in a one-component system, such as for non-relativistic 2D electrons with complete spin polarisation. However, the SU(4) singlet state may be destroyed if one takes into account the Zeeman eect at high magnetic elds, which favours spin polarisation. A complementary composite-fermion calculation with SU(4) symmetry has revealed that, at = 4/9, states with intermediate SU(4) spin-valley polarisation such as a valley-isospin singlet with full spin polarisation may exist, with a slightly higher energy than the composite-fermion SU(4) singlet [164], which is indeed identical to the [3; 2, 2] wavefunction. One may, therefore, expect a transition between two 4/9 states with dierent polarisation when the Zeeman energy outcasts the energy dierence between the two states. This is similar to the above-mentioned 2/5 state in a conventional 2DEG [224].
Chapter 6
Conclusions and Outlook

During the last years, graphene research has become an important eld of condensed-matter physics. The interest stems both from fundamental issues, such as the realisation of a 2D relativistic quantum system governed by a Dirac rather than a Schrdinger equation, as well as from the exciting prospect that o graphene may be used in future nano-electronical devices. Several special journal issues have been devoted to graphene research, and excellent reviews are already available (see e.g. Ref. [4] or Ref. [3] for a review on more experimental aspects of graphene). The aim of the present manuscript is to review more specically the quantummechanical properties of relativistic 2D electrons exposed to a perpendicular magnetic eld. The most salient phenomenon is clearly the RQHE which was observed in 2005 [1, 2] and which was the rst experimental evidence for relativistic massless carriers in graphene. The basic understanding of this phenomenon stems from the theory of non-interacting (relativistic) electrons described by the 2D Dirac equation, which governs the low-energy properties of undoped and moderately doped graphene and which yields a characteristic LL spectrum when the perpendicular magnetic eld is taken into account. The theory of non-interacting electrons in graphene has been introduced in the rst two chapters of this review. Moreover, we have briey discussed the model with anisotropic nn hopping which makes the Dirac points move in the rst BZ and eventually yields a topological (semi)metal-insulator transition. The continuum limit of this model, which describes -electrons in a graphene sheet under uniaxial strain (even if the topological phase transition is beyond the limit of maximal physical strain), is a generalised Weyl Hamiltonian that yields an anisotropic linear band dispersion with an additional tilt of the Dirac cones. This generalised Weyl Hamiltonian has also been discussed in the framework of the electronic properties of (BEDT-TTF)2 I3 under pressure. The main parts of this review are concerned with the role of electronic interactions in graphene LLs. Whereas we have argued that these interactions may be treated perturbatively in the regime of the (integer) RQHE, they constitute the relevant energy scale in partially lled graphene LLs due to the quenching of 105
106
the kinetic energy. This is reminiscent of partially lled LLs in non-relativistic 2D electron systems, and the most prominent consequence of this quenched kinetic energy and the macroscopic LL degeneracy is certainly the FQHE. The graphene FQHE, rst experimental manifestations of which have recently been reported, is expected to be reminiscent of that of non-relativistic 2D electrons but it is governed by a larger internal degeneracy described to great accuracy by the SU(4) group. In the perturbative regime of the RQHE, the theoretical study of electronelectron interactions indicates the presence of fascinating novel collective modes, such as linear magneto-plasmons, that are particular to graphene and do not have a counterpart in non-relativistic 2D electron systems in a perpendicular magnetic eld. Also the upper hybrid mode, which is the magnetic-eld counterpart of the usual 2D plasmon, is expected to behave in a particular manner in graphene as a consequence of the linear disperison relation and the vanishing band mass. Whereas these studies are at present only theoretical, these collective modes may nd an experimental verication in inelastic light-scattering measurements. Similar to the role of electron-electron interactions in the RQHE regime, the electron-phonon coupling yields exciting resonance phenomena in graphene in a strong magnetic eld. The electron-phonon interaction in graphene LLs has been discussed in the framework of a perturbative approach. Indications for the magneto-phonon resonance, e.g., have recently been found in Raman spectroscopy of epitaxial graphene. As a perspective for future research on relativistic 2D electrons in a strong magnetic eld, one may mention two major aspects. One is concerned with a more profound theoretical understanding of the consequences of the SU(4) spin-valley symmetry in the FQHE. We have discussed, in the last chapter, several states with particular polarisations, and one may speculate that such states have an experimental signature in the magnetisation. Indeed, the magnetic polarisation has been studied experimentally for non-relativistic 2D electrons in GaAs heterostructures, and the behaviour of the spin relaxation rate in nuclear magnetic resonance (NMR) experiments has been interpreted in terms of skyrmions both at = 1 [228] and = 1/3 [229]. Naturally, such experiments are sensitive only to the physical spin part of the SU(4) isospin, and, as we have discussed in the context of the SU(4) quantum Hall ferromagnetism, the magnetisation associated with the physical spin may vanish as a consequence of the spin-valley entanglement even if a state is ferromagnetically ordered. A difculty that one is confronted with in the study of the spin polarisation of FQHE states in graphene is the 12 C isotope, which is dominantly found among the carbon atoms (roughly 99%) and which is NMR inactive because of a zero nuclear spin. Even if one enriches the graphene sheet with NMR active 13 C atoms, the density remains too low for usual NMR experiments as a consequence of the 2D character of graphene. An experimental technique that circumvents this problem is resistively detected NMR [230] that has been successfully applied in the study of quantum Hall states in GaAs heterostructures [231, 232], and one may speculate that this technique could also be used in 13 C enriched graphene.
107
A second aspect that may be mentioned for a future research is a deeper understanding of the interplay between relativistic 2D electrons and strong correlations. The Coulomb interactions in undoped graphene seem to be situated in the low-coupling regime, and we have argued in Sec. 3.2 that in the case of doped graphene the eective coupling constant saturates at rather low values. The situation is drastically dierent in (BEDT-TTF)2 I3 , where the bare coupling constant is roughly ten times larger than in graphene as a consequence of a lower average Fermi velocity, and this organic compound may therefore be a good candidate for a deeper theoretical and experimental investigation of strongly-correlated 2D electrons. However, due to its layered structure, one may expect drastically dierent screening properties.
108
Matrix Elements of the Density Operators

The matrix elements that intervene in the expression for the density operators (3.9) are of the form n, m| exp(iq r)|n , m and may be calculated with the help of the decomposition of the cyclotron variable and the guiding centre R into the ladder operators a and respectively [see Eqs. (2.7), (2.19) and b, (2.22)]. We futhermore dene the complex wave vectors q (qx + iqy )lB and q = (qx iqy )lB ,1 One nds n, m|eiqr |n , m = = m|eiqR |m n|eiq |n m|e
i 2 (q + b q b)
(1) |n .
|m n|e
i 2 (a +q) q a
The two matrix elements may be simplied with the help of the Baker-Hausdor formula exp(A) exp(B) = exp(A + B) exp([A, B]/2), for the case [A, [A, B]] = [B, [A, B]] = 0 [57]. The second matrix element thus becomes, for n n n|eiq |n = n|e
i q a 2 (a +q ) 2
= e|q| = e|q|
/4 /4
n|e
j
i i a 2 qa 2 q
|n
n|e n ! n! n ! n!
i 2 qa
|n
|j j|e
i 2 q a
|n |q|2 2
n j
= e|q| = e|q| where we have used n|e

i 2 q a
/4
i q 2 i q 2
nn n j=0 nn
n! (n j)!(n j)!j!
/4
nn Ln
|q|2 2
(2)
|j =
0
n! 1 j! (nj)! i 2 q nj
for j > n for j n
1 We use this notation solely in the present appendix. Throughout the main text, q denotes the modulus of the wave vector q, q = |q|.
109
110
Appendix A
in the third line and the denition of the associated Laguerre polynomials [107],
nn Ln (x) n
=
m=0
(n
(x)m n! . + m)! m)!(n n m!
In the same manner, one obtains for m m m|eiqR |m = m|e

i b q b) 2 (q +
= e
|q|2 /4
m ! m!
|m iq 2
mm
Lmm m
|q|2 2
(3)
With the help of the denition G

n,n
(q)
n ! n!
iq 2
nn
nn Ln
|q|2 2
one may rewrite the expressions without the conditions n n and m m , n|eiq |n = [(n n )Gn,n () + (n n 1)Gn ,n (q)] e|q| q and q m|eiqR |m = [(m m )Gm,m (q) + (m m 1)Gm ,m ()] e|q|
2 2
/4
(4)
/4
. (5)
Bibliography
[1] Novoselov, K. S., Geim, A. K., Morosov, S. V., Jiang, D., Katsnelson, M. I., Grigorieva, I. V., Dubonos, S. V., and Firsov, A. A. (2005) Nature, 438, 197. [2] Zhang, Y., Tan, Y.-W., Stormer, H. L., and Kim, P. (2005) Nature, 438, 201. [3] Geim, A. K. and Novoselov, K. S. (2007) Nat. Materials, 6, 183. [4] Castro Neto, A. H., Guinea, F., Peres, N. M. R., Novoselov, K. S., and Geim, A. K. (2009) Rev. Mod. Phys., 81, 109. [5] Novoselov, K. S., Geim, A. K., Morosov, S. V., Jiang, D., Zhang, Y., Dubonos, S. V., Grigorieva, I. V., and Firsov, A. A. (2004) Science, 306, 666. [6] Ponomarenko, L. A., Schedin, F., Katsnelson, M. I., Yang, R., Hill, E. W., Novoselov, K. S., and Geim, A. K. (2008) Science, 320, 356. [7] Fukuda, Y. et al. (Super-Kamiokande Collaboration) (1998) Phys. Rev. Lett., 81, 1562. [8] Kekul, A. (1865) Bulletin de la Societe Chimique de Paris, 3, (2) 98. e [9] Kekul, A. (1866) Annalen der Chemie und Pharmazie, 137, (2) 129. e [10] Pauling, L. (1960) The Nature of Chemical Bonds. Cornell UP. [11] Kroto, H. W., Heat, J. R., OBrien, S. C., Curl, R. F., and Smalley, R. E. (1985) Nature, 318, 162. [12] Ozawa, E., Kroto, H. W., Fowler, P. W., and Wassermann, E. (1993) Phil. Trans. R. Soc. (London), A 343, 1. [13] Iijima, S. (1991) Nature, 354, 56. [14] Monthioux, M. and Kuznetsov, V. L. (2006) Carbon, 44, 1621. [15] Radushkevich, L. V. and Lukyanovich, V. M. (1952) Zurn. Fisic. Chim., 26, 88. 111
112 [16] Novoselov, K. S., Jiang, D., Booth, T., Khotkevich, V. V., Morozov, S. M., and Geim, A. K. (2005) PNAS , 102, 10451. [17] Berger, C., Song, Z., Li, T., Ogbazghi, A. Y., Feng, R., Dai, Z., Marchenkov, A. N., Conrad, E. H., First, P. N., and de Heer, W. A. (2004) J. Phys. Chem., 108, 19912. [18] de Heer, W. A., et al. (2007) Solid State Comm., 143, 92. [19] Hass, J., Varchon, F., Millan-Otoya, J. E., Sprinkle, M., de Heer, W. A., Berger, C., First, P. N., Magaud, L., and Conrad, E. H. (2008) Phys. Rev. Lett., 100, 125504. [20] Shen, T., Gu, J. J., Xu, M., Wu, Y. Q., Bolen, M. L., Capano, M. A., Engel, L. W., and Ye, P. D. (2009) Appl. Phys. Lett., 95, 172105. [21] Wu, X., Hu, Y., Ruan, M., Madiomanana, N. K., Hankinson, J., Sprinkle, M., Berger, C., and de Heer, W. A. (2009) Appl. Phys. Lett., 95, 223108. [22] Jobst, J., Waldmann, D., Speck, F., Hirner, R., Maude, D. K., Seyller, T., and Weber, H. B. (2009) arXiv:0908.1900 . [23] Darancet, P., Wipf, N., Berger, C., de Heer, W. A., and Mayou, D. (2008) Phys. Rev. Lett., 101, 116806. [24] Saito, R., Dresselhaus, G., and Dresselhaus, M. S. (1998) Physical Properties of Carbon Nanotubes. Imperial College Press. [25] Wallace, P. R. (1947) Phys. Rev., 71, 622. [26] Bena, C. and Montambaux, G. (2009) New J. Phys., 11, 095003. [27] Partoens, B. and Peeters, F. (2006) Phys. Rev. B , 74, 075407. [28] Mucha-Kruczyski, M., Tsyplyatyev, O., Grishin, A., McCann, E., Falko, n V. I., Bostwick, A., and Rotenberg, E. (2008) Phys. Rev. B , 77, 195403. [29] Weinberg, S. (1995) The Quantum Theory of Fields. Cambridge UP. [30] Shon, N. H. and Ando, T. (1998) J. Phys. Soc. Jpn., 67, 2421. [31] Klein, O. (1929) Z. Phys., 53, 157. [32] Pereira, V. M., Castro Neto, A. H., and Peres, N. M. R. (2009) Phys. Rev. B , 80, 045401. [33] Hasegawa, Y., Konno, R., Nakano, H., and Kohmoto, M. (2006) Phys. Rev. B , 74, 033413. [34] Zhu, S.-L., Wang, B., and Duan, L.-M. (2007) Phys. Rev. Lett., 98, 260402.
113 [35] Dietl, P., Pichon, F., and Montambaux, G. (2008) Phys. Rev. Lett , 98, e 236405. [36] Wunsch, B., Sols, F., and Guinea, F. (2008) New Journal of Physics, 10, 103027. [37] Goerbig, M. O., Fuchs, J.-N., Montambaux, G., and Pichon, F. (2008) e Phys. Rev. B , 78, 045415. [38] Farjam, M. and Rai-Tabar, H. (2009) Phys. Rev. B , 80, 167401. [39] Harrison, W. A. (1981) Phys. Rev. B , 24, 5835. [40] Dillon, R. O., Spain, I. L., and McClure, J. W. (1977) J. Phys. Chem. Solids, 38, 635. [41] Salem, L. (1966) Molecular Orbital Theory of Conjugated Systems. Benjamin. [42] Lee, C., Wei, X., Kysar, J. K., and Hone, J. (2008) Science, 321, 385. [43] Montambaux, G., Pichon, F., Fuchs, J.-N., and Goerbig, M. O. (2009) e Phys. Rev. B , 80, 153412. [44] Montambaux, G., Pichon, F., Fuchs, J.-N., and Goerbig, M. O. (2009) e Europhys. J. B , 72, 509. [45] Esaki, K., Sato, M., Kohmoto, M., and Halperin, B. I. (2009) Phys. Rev. B , 80, 125405. [46] Zhao, E. and Paramekanti, A. (2006) Phys. Rev. Lett., 97, 230404. [47] Hou, J.-M., Yang, W.-X., and Liu, X.-J. (2009) Phys. Rev. A, 79, 043621. [48] Lee, K. L., Grmaud, B., Han, R., Englert, B.-G., and Miniatura, C. e (2009) Phys. Rev. A, 80, 043411. [49] Katayama, S., Kobayashi, A., and Suzumura, Y. (2006) J. Phys. Soc. Jpn., 75, 054705. [50] Kobayashi, A., Katayama, S., Y.Suzumura, and Fukuyama, H. (2007) J. Phys. Soc. Jpn., 76, 034711. [51] Banerjee, S., Singh, R. R. P., Pardo, V., and Pickett, W. E. (2009) Phys. Rev. Lett., 103, 016402. [52] Damascelli, A. (2004) Phys. Scripta, T109, 61. [53] Bostwick, A., Ohta, T., Seyller, T., Horn, K., and Rotenberg, E. (2007) Nat. Phys., 3, 36.
114 [54] Zhou, S. Y., Gweon, G.-H., Graf, J., Fedorov, A. V., Spataru, C. D., Diehl, R. D., Kopelevich, K., Lee, D.-H., Louie, S. G., and Lanzara, A. (2006) Nat. Phys., 2, 595. [55] Jackson, J. D. (1999) Classical Electrondynamics. Wiley, 3rd ed. [56] Zhang, Y., Jiang, Z., Small, J. P., Purewal, M. S., Tan, Y.-W., Fazlollahi, M., Chudow, J. D., Jaszczak, J. A., Stormer, H. L., and Kim, P. (2006) Phys. Rev. Lett., 98, 197403. [57] Cohen-Tannoudji, C., Diu, B., and Lalo, F. (1973) Quantum Mechanics. e Hermann. [58] McClure, J. W. (1956) Phys. Rev., 104, 666. [59] v. Klitzing, K., Dorda, G., and Pepper, M. (1980) Phys. Rev. Lett., 45, 494. [60] Goerbig, M. O. (2009) Quantum Hall Eects. Lecture notes of the Les Houches Summer School 2009 (Singapore Session). [61] Gusynin, V. P. and Sharapov, S. G. (2005) Phys. Rev. Lett., 95, 146801. [62] Gusynin, V. P. and Sharapov, S. G. (2006) Phys. Rev. B , 73, 245411. [63] Peres, N. M., Guinea, F., and Castro Neto, A. H. (2006) Phys. Rev. B , 73, 125411. [64] Sadowski, M. L., Martinez, G., Potemski, M., Berger, C., and de Heer, W. A. (2006) Phys. Rev. Lett., 97, 266405. [65] Jiang, Z., Henriksen, E. A., L. C. Tung, Y.-J. W., Schwartz, M. E., Han, M. Y., Kim, P., and Stormer, H. L. (2007) Phys. Rev. Lett., 98, 197403. [66] Kohn, W. (1961) Phys. Rev., 123, 1242. [67] Iyengar, A., Wang, J., Fertig, H. A., and Brey, L. (2007) Phys. Rev. B , 75, 125430. [68] Abergel, D. S. L. and Falko, V. I. (2007) Phys. Rev. B , 75, 155430. [69] Plochocka, P., Faugeras, C., Orlita, M., Sadowski, M. L., Martinez, G., Potemski, M., Goerbig, M. O., Fuchs, J.-N., Berger, C., and de Heer, W. A. (2008) Phys. Rev. Lett , 100, 087401. [70] Hofstadter, D. R. (1976) Phys. Rev. B , 14, 2239. [71] Rammal, R. (1985) J. Physique, 46, 1345. [72] Lukose, V., Shankar, R., and Baskaran, G. (2007) Phys. Rev. Lett., 98, 116802.
115 [73] Peres, N. M. R. and Castro, E. V. (2007) J. Phys.:Condensed Matter , 19, 406231. [74] Morinari, T., Himura, T., and Tohyama, T. (2009) J. Phys. Soc. Jpn., 78, 023704. [75] Onsager, L. (1952) Philos. Mag., 43, 1006. [76] Lifshitz, I. M. and Kosevich, A. M. (1956) Sov. Phys. JETP , 2, 636. [77] Mikitik, G. P. and Sharlai, Y. V. (1999) Phys. Rev. Lett., 82, 2147. [78] Goerbig, M. O., Fuchs, J.-N., Montambaux, G., and Pichon, F. (2009) e EPL, 85, 57005. [79] Mahan, G. D. (1993) Many-Particle Physics. Plenum Press, 2nd Ed. [80] Giuliani, G. F. and Vignale, G. (2005) Quantum Theory of Electron Liquids. Cambridge UP. [81] Coleman, P. (2003) An.. Henri Poincar , 4, 1. e [82] Katsnelson, M. I. (2006) Phys. Rev. B , 74, 201401. [83] Herbut, I. F. (2006) Phys. Rev. Lett., 97, 146401. [84] Herbut, I. F. (2007) Phys. Rev. B , 76, 085432. [85] Goerbig, M. O., Douot, B., and Moessner, R. (2006) Phys. Rev. B , 74, c 161407. [86] Alicea, J. and Fisher, M. P. A. (2006) Phys. Rev. B , 74, 075422. [87] Herbut, I. F. (2007) Phys. Rev. B , 75, 165411. [88] Doretto, R. L. and Morais Smith, C. (2007) Phys. Rev. B , 76, 195431. [89] Kallin, C. and Halperin, B. I. (1984) Phys. Rev. B , 30, 5655. [90] Shung, K. W. K. (1986) Phys. Rev. B , 34, 979. [91] Ando, T. (2006) J. Phys. Soc. Jpn., 75, 074716. [92] Wunsch, B., Stauber, T., Sols, F., and Guinea, F. (2006) New Journal of Physics, 8, 318. [93] Hwang, E. H. and Das Sarma, S. (2007) Phys. Rev. B , 75, 205418. [94] Polini, M., Asgari, R., Borghi, G., Barlas, Y., Pereg-Barnea, T., and MacDonald, A. H. (2008) Phys. Rev. B , 77, 081411. [95] Sabio, J., Nilsson, J., and Castro Neto, A. H. (2008) Phys. Rev. B , 78, 075410.
116 [96] Gonzlez, J., Guinea, F., and Vozmediano, M. A. H. (1994) Nucl. Phys. a B , 424, 595. [97] Gonzlez, J., Guinea, F., and Vozmediano, M. A. H. (1999) Phys. Rev. a B , 59, R2474. [98] Gangadharaiah, S., Farid, A. M., and Mishchenko, E. G. (2008) Phys. Rev. Lett., 100, 166802. [99] Kotov, V. N., Uchoa, B., and Castro Neto, A. H. (2007) Phys. Rev. B , 78, 035119. [100] Stern, F. (1967) Phys. Rev. Lett., 18, 546. [101] Roldn, R., Fuchs, J.-N., and Goerbig, M. O. (2009) Phys. Rev. B , 80, a 085408. [102] Shizuya, K. (2007) Phys. Rev. B , 75, 245417. [103] Roldn, R., Goerbig, M. O., and Fuchs, J.-N. (2010) Semicond. Sci. Techa nol., 25, 034005. [104] Berman, O., Gumbs, G., and Lezovik, Y. E. (2008) Phys. Rev. B , 78, 085401. [105] Tahir, M. and Sabeeh, K. (2008) J. Phys.:Condens. Matter , 20, 425202. [106] Ando, T. (2007) J. Phys. Soc. Jpn., 76, 024712. [107] Gradshteyn, I. S. and Ryzhik, I. M. (2000) Table of Integrals, Series and Products (6th Ed.). Academic Press. [108] Bychkov, Y. A. and Martinez, G. (2008) Phys. Rev. B , 77, 125417. [109] Roldn, R., Fuchs, J.-N., and Goerbig, M. O. unpublished . a [110] Chiu, K. W. and Quinn, J. J. (1974) Phys. Rev. B , 9, 4724. [111] Herbut, I. F., Jurii, V., and Roy, B. (2009) Phys. Rev. B , 79, 085116. cc [112] Herbut, I. F., Jurii, V., and Vafek, O. (2009) Phys. Rev. B , 80, 075432. cc [113] Jurii, V., Herbut, I. F., and Semeno, G. W. (2009) Phys. Rev. B , 80, cc 081405(R). [114] Drut, J. E. and Lhde, T. A. (2009) Phys. Rev. Lett., 102, 026802. a [115] Drut, J. E. and Lhde, T. A. (2009) Phys. Rev. B , 79, 165425. a [116] Aleiner, I. L. and Glazman, L. I. (1995) Phys. Rev. B , 52, 11296. [117] Wirtz, L. and Rubio, A. (2004) Solid Stat. Comm., 131, 141.
117 [118] Pisana, S., Lazzeri, M., Casiraghi, C., Novoselov, K. S., Geim, A. K., Ferrari, A. C., and Mauri, F. (2007) Nature Mat., 6, 198. [119] Yan, J., Zhang, Y., Kim, P., and Pinczuk, A. (2007) Phys. Rev. Lett., 98, 166802. [120] Ferrari, A. C., et al. (2006) Phys. Rev. Lett., 97, 187401. [121] Gupta, A., Chen, G., Joshi, P., Tadigadapa, S., and Eklund, P. (2006) Nano Lett., 6, 2667. [122] Graf, D., Molitor, F., Ensslin, K., Stampfer, C., Jungen, A., Hierold, C., and Wirtz, L. (2007) Nano Lett., 7, 238. [123] Kohn, W. (1959) Phys. Rev. Lett., 2, 393. [124] Ando, T. (2006) J. Phys. Soc. Jpn, 75, 124701. [125] Lazzeri, M. and Mauri, F. (2006) Phys. Rev. Lett., 97, 266407. [126] Castro Neto, A. H. and Guinea, F. (2007) Phys. Rev. B , 75, 045404. [127] Ando, T. (2007) J. Phys. Soc. Jpn., 76, 024712. [128] Goerbig, M. O., Fuchs, J.-N., Kechedzhi, K., and Falko, V. I. (2007) Phys. Rev. Lett., 99, 087402. [129] Ishikawa, K. and Ando, T. (2006) J. Phys. Soc. Jpn., 75, 084713. [130] Piscanec, S., Lazzeri, M., Mauri, F., Ferrari, A. C., and Robertson, J. (2004) Phys. Rev. Lett., 93, 185503. [131] Faugeras, C., Amado, M., Kossacki, P., Orlita, M., Sprinkle, M., Berger, C., de Heer, W. A., and Potemski, M. (2009) Phys. Rev. Lett., 103, 186803. [132] Yan, J., Henriksen, E., Kim, P., and Pinczuk, A. (2008) Phys. Rev. Lett., 101, 136804. [133] Tsui, D. C., Strmer, H., and Gossard, A. C. (1982) Phys. Rev. Lett., 48, o 1559. [134] Andrei, E. Y., Deville, G., Glattli, D. C., Williams, F. I. B., Paris, E., and Etienne, B. (1988) Phys. Rev. Lett., 60, 2765. [135] Williams, F. I. B., et al. (1991) Phys. Rev. Lett., 66, 3285. [136] Koulakov, A. A., Fogler, M. M., and Shklovskii, B. I. (1996) Phys. Rev. Lett., 76, 499. [137] Fogler, M. M., Koulakov, A. A., and Shklovskii, B. I. (1996) Phys. Rev. B , 54, 1853.
118 [138] Moessner, R. and Chalker, J. T. (1996) Phys. Rev. B , 54, 5006. [139] Lilly, M. P., Cooper, K. B., Eisenstein, J. P., Pfeier, L. N., and West, K. W. (1999) Phys. Rev. Lett., 82, 394. [140] Du, R. R., Tsui, D. C., Stormer, H. L., Pfeier, L. N., Baldwin, K. W., and West, K. W. (1999) Solid State Comm., 109, 389. [141] Lewis, R. M., Ye, P. D., Engel, L. W., Tsui, D. C., Pfeier, L. N., and West, K. W. (2002) Phys. Rev. Lett., 89, 136804. [142] Lewis, R. M., Chen, Y., Engel, L. W., Tsui, D. C., Ye, P. D., Pfeier, L. N., and West, K. W. (2004) Phys. Rev. Lett., 93, 176808. [143] Lewis, R. M., Chen, Y., Engel, L. W., Tsui, D. C., Pfeier, L. N., and West, K. W. (2005) Phys. Rev. B , 71, 081301(R). [144] Cooper, K. B., Lilly, M. P., Eisenstein, J. P., Pfeier, L. N., and West, K. W. (1999) Phys. Rev. B , 60, R11285. [145] Eisenstein, J. P., Cooper, K. B., Pfeier, L. N., and West, K. W. (2002) Phys. Rev. Lett., 88, 076801. [146] Martin, J., Akerman, N., Ulbricht, G., Lohmann, T., Smet, J. H., von Klitzing, K., and Yacobi, A. (2008) Nat. Phys., 4, 144. [147] Mallet, P., Varchon, F., Naud, C., Magaud, L., Berger, C., and Veuillen, J.-Y. (2007) Phys. Rev. B , 76, 041403(R). [148] Li, G., Luican, A., and Andrei, E. Y. (2009) Phys. Rev. Lett., 102, 176804. [149] Du, X., Skachko, I., Duerr, F., Luican, A., and Andrei, E. Y. (2009) Nature, 462, 192. [150] Bolotin, K. I., Ghahari, F., Shulman, M. D., Stormer, H. L., and Kim, P. (2009) Nature, 462, 196. [151] Goerbig, M. O. and Morais Smith, C. (2003) Europhys. Lett., 63, 736. [152] Girvin, S. M., MacDonald, A. H., and Platzman, P. M. (1986) Phys. Rev. B , 33, 2481. [153] Nomura, K. and MacDonald, A. H. (2006) Phys. Rev. Lett., 96, 256602. [154] Ezawa, Z. F. and Hasebe, K. (2002) Phys. Rev. B , 65, 075311. [155] Ezawa, Z. F., Tsitsishvili, G., and Hasebe, K. (2003) Phys. Rev. B , 67, 125314. [156] Douot, B., Goerbig, M. O., Lederer, P., and Moessner, R. (2008) Phys. c Rev. B , 78, 195327.
119 [157] Abramowitz, M. and Stegun, I. (1970) Handbook of Mathematical Functions (9th Ed.). Dover Publications. [158] Arikawa, M., Hatsugai, Y., and Aoki, H. (2008) Phys. Rev. B , 78, 205401. [159] Abanin, D. A., Lee, P. A., and Levitov, L. S. (2007) Phys. Rev. Lett., 98, 156801. [160] Khveshchenko, D. V. (2007) Phys. Rev. B , 75, 153405. [161] Haldane, F. D. M. (1983) Phys. Rev. Lett., 51, 605. [162] Apalkov, V. M. and Chakraborty, T. (2006) Phys. Rev. Lett., 97, 126801. [163] Tke, C., Lammert, P. E., Crespi, V. H., and Jain, J. K. (2006) Phys. o Rev. B , 74, 235417. [164] Tke, C. and Jain, J. K. (2007) Phys. Rev. B , 75, 245440. o [165] Goerbig, M. O. and Regnault, N. (2007) Phys. Rev. B , 75, 241405. [166] Papi, Z., Goerbig, M. O., and Regnault, N. (2009) Solid State Comm., c 149, 1056. [167] Zhang, C.-H. and Joglekar, Y. N. (2007) Phys. Rev. B , 75, 245414. [168] Zhang, C.-H. and Joglekar, Y. N. (2008) Phys. Rev. B , 77, 205426. [169] Poplavskyy, O., Goerbig, M. O., and Morais Smith, C. (2009) Phys. Rev. B , 80, 195414. [170] Willett, R. L., Eisenstein, J. P., Stormer, H. L., Tsui, D. C., Gossard, A. C., and English, J. H. (1987) Phys. Rev. Lett., 59, 1776. [171] Moore, G. and Read, N. (1991) Nucl. Phys. B , 360, 362. [172] Greiter, M., Wen, X.-G., and Wilczek, F. (1991) Phys. Rev. Lett., 66, 3205. [173] Goerbig, M. O., Lederer, P., and Morais Smith, C. (2003) Phys. Rev. B , 68, 241302. [174] Goerbig, M. O., Lederer, P., and Morais Smith, C. (2004) Phys. Rev. B , 69, 115327. [175] Moon, K., Mori, H., Yang, K., Girvin, S. M., MacDonald, A. H., Zheng, I., Yoshioka, D., and Zhang, S.-C. (1995) Phys. Rev. B , 51, 5143. [176] Ezawa, Z. F. (2000) Quantum Hall Eects Field Theoretical Approach and Related Topics. World Scientic. [177] Arovas, D. P., Karlhede, A., and Lilliehok, D. (1999) Phys. Rev. B , 59, o 13147.
120 [178] Yang, K., Das Sarma, S., and MacDonald, A. H. (2006) Phys. Rev. B , 74, 075423. [179] Sondhi, S. L., Karlhede, A., Kivelson, S. A., and Rezayi, E. H. (1993) Phys. Rev. B , 47, 16419. [180] Girvin, S. M. (1999) The Quantum Hall Eect: Novel Excitations and Broken Symmetries, in A. Comptet, T. Jolicoeur, S. Ouvry and F. David (Eds.) Topological Aspects of Low-Dimensional Systems Ecole dEte de Physique Thorique LXIX . Springer. e [181] Brey, L., Fertig, H. A., Ct, R., and MacDonald, A. H. (1995) Phys. Rev. o e Lett., 75, 2562. [182] Ct, R., Boisvert, D. B., Bourassa, J., Boissonneault, M., and Fertig, o e H. A. (2007) Phys. Rev. B , 76, 125320. [183] Ct, R., Jobidon, J.-F., and Fertig, H. A. (2008) Phys. Rev. B , 78, o e 085309. [184] Fukuyama, H. (1975) Solid State Commun., 17, 1323. [185] Meyer, J. C., Geim, A. K., Katsnelson, M. I., Novoselov, K. S., Booth, T. J., and Roth, S. (2007) Nature, 446, 60. [186] Fuchs, J.-N. and Lederer, P. (2007) Phys. Rev. Lett., 98, 016803. [187] Nomura, K., Ryu, S., and Lee, D.-H. (2009) Phys. Rev. Lett., 103, 216801. [188] Hou, C.-Y., Chamon, C., and Mudry, C. (2010) Phys. Rev. B , 81, 075427. [189] Haldane, F. D. M. (1988) Phys. Rev. Lett., 61, 2015. [190] Ajiki, H. and Ando, T. (1995) J. Phys. Soc. Jpn., 64, 260. [191] Kane, C. and Mele, E. J. (2005) Phys. Rev. Lett., 95, 146802. [192] Min, H., Hill, J. E., Sinitsyn, N. A., Sahu, B. R., Kleinman, L., and MacDonald, A. H. (2006) Phys. Rev. B , 74, 165310. [193] Gusynin, V. P., Miransky, V. A., Sharapov, S. G., and Shovkovy, I. A. (2006) Phys. Rev. B , 74, 195429. [194] Herbut, I. F. (2008) Phys. Rev. B , 78, 205433. [195] Gorbar, E. V., Gusynin, V. P., Miransky, V. A., and Shovkovy, I. A. (2008) Phys. Rev. B , 78, 085437. [196] Ezawa, M. (2007) J. Phys. Soc. Jpn., 76, 094701. [197] Khveshchenko, D. V. (2001) Phys. Rev. Lett., 87, 206401.
121 [198] Gorbar, E. V., Gusynin, V. P., Miransky, V. A., and Shovkovy, I. A. (2002) Phys. Rev. B , 66, 045108. [199] Fertig, H. A. (1989) Phys. Rev. B , 40, 1087. [200] Wen, X.-G. and Zee, A. (1992) Phys. Rev. Lett , 69, 1811. [201] Ezawa, Z. F. and Iwazaki, A. (1993) Phys. Rev. B , 47, 7295. [202] Spielman, I. B., Eisenstein, J. P., Pfeier, L. N., and West, K. W. (2000) Phys. Rev. Lett., 84, 5808. [203] Kellogg, M., Eisenstein, J. P., and an K. W. West, L. N. P. (2004) Phys. Rev. Lett., 036801. [204] Tutuc, E., Shayegan, M., and Huse, D. A. (2004) Phys. Rev. Lett , 93, 036802. [205] Ryu, S., Mudry, C., Hou, C.-Y., and Chamon, C. (2009) Phys. Rev. B , 80, 205319. [206] Jiang, Z., Zhang, Y., Stormer, H. L., and Kim, P. (2007) Phys. Rev. Lett., 99, 106802. [207] Abanin, D. A., Novoselov, K. S., Zeitler, U., Lee, P. A., Geim, A. K., and Levitov, L. S. (2007) Phys. Rev. Lett., 98, 196806. [208] Brey, L. and Fertig, H. (2006) Phys. Rev. B , 73, 195408. [209] Shimshoni, E., Fertig, H. A., and Vanketeswara Pai, G. (2009) Phys. Rev. Lett., 102, 206408. [210] Checkelsky, J. G., Li, L., and Ong, N. P. (2008) Phys. Rev. Lett., 100, 206801. [211] Checkelsky, J. G., Li, L., and Ong, N. P. (2009) Phys. Rev. B , 79, 115434. [212] Kosterlitz, J. M. and Thouless, D. J. (1973) J. Phys. C , 6, 1181. [213] Li, G., Luican, A., and Andrei, E. Y. (2009) Phys. Rev. Lett., 102, 176804. [214] Bolotin, K. I., Sikes, K. J., Jiang, Z., Klima, M., Fudenberg, G., Hone, J., Kim, P., and Stormer, H. L. (2008) Solid State Commun., 146, 351. [215] Du, X., Skachko, I., Barker, A., and Andrei, E. Y. (2008) Nat. Nanotech., 3, 491. [216] Laughlin, R. B. (1983) Phys. Rev. Lett., 50, 1395. [217] Haldane, F. D. M. and Rezayi, E. H. (1985) Phys. Rev. Lett., 54, 237. [218] Fano, G., Ortolani, F., and Colombo, E. (1986) Phys. Rev. B , 34, 2670.
122 [219] Halperin, B. I. (1983) Helv. Phys. Acta, 56, 75. [220] Wen, X.-G. and Zee, A. (1992) Phys. Rev. B , 46, 2290. [221] de Gail, R., Regnault, N., and Goerbig, M. O. (2008) Phys. Rev. B , 77, 165310. [222] Papi, Z., Goerbig, M. O., and Regnault, N. (2010) arXiv:1005.5121 . c [223] Shibata, N. and Nomura, K. (2009) J. Phys. Soc. Jpn, 78, 104708. [224] Chakraborty, T. and Zhang, F. C. (1984) Phys. Rev. B , 29, 7032. [225] Jain, J. K. (2007) Composite Fermions. Cambridge UP. [226] Kang, W., Young, J. B., Hannahs, S. T., Palm, E., Campman, K. L., and Gossard, A. C. (1997) Phys. Rev. B , 56, R12776. [227] Kukushkin, I. K., v. Klitzing, K., and Eberl, K. (1999) Phys. Rev. Lett., 82, 3665. [228] Barrett, S. E., Dabbagh, G., Pfeier, L. N., West, K. W., and Tycko, R. (1995) Phys. Rev. Lett., 74, 5112. [229] Khandelwal, P., Kuzma, N. N., Barrett, S. E., Pfeier, L. N., and West, K. W. (1998) Phys. Rev. Lett., 81, 673. [230] Desrat, W., Maude, D. K., Potemski, M., Portal, J. C., Wasilewski, Z. R., and Hill, G. (2002) Phys. Rev. Lett., 88, 256807. [231] Spielman, I. B., Tracy, L. A., Eisenstein, J. P., Pfeier, L. N., and West, K. W. (2005) Phys. Rev. Lett., 94, 076803. [232] Kumada, N., Muraki, K., Hashimoto, K., and Hirayama, Y. (2005) Phys. Rev. Lett., 94, 096802.
Zhang et al. PRL 96, 236806 (2006)

25T
T=1.4K
42T 45T 37T 9T (30mK)
30T 25T
1.0
12
0.8
10 8 6
0.6
Out[22]=
0.4
4
0.2
2
2 4
10
10
effective interaction potential
(a) 12
10 10 8 6 5 4 2
n=0 (relativistic and nonrelativistic) n=1 (relativistic) n=1 (nonrelativistic) n=5 (relativistic)
n=5 (non relativistic)
2 2
4 4
r/l B
6 6
8 8
10 10
0.8 0.6 0.4
mode splitting 2g 2g g 2 2g
(b) B=B0
6 mode splitting g 2 2g 4n2 4n+2 2g 4n+6 4 2 0 2 4 6 (c) B=Bn>0
0.5
energy
-0.5
-1 0 0.005 0.01 flux per hexagon 0.015 0.02
(a) B=B0 2g I =0
Oscillator strength [a.u.]
II 0 < || < 2 2g 2g III || = 2
~ 0
-2
-4
(a)
(b)
y z x
541 500 400 300 200 100 25

2 1.75 1.5 1.25 1 0.75 0.5 0.25
398
154
before 2006
5 10 15 20 25
2006
2007
2008
2009*: during first three months

30 35

2

Transféré par

Informations du document

Description originale:

Copyright

Formats disponibles

Partager ce document

Partager ou intégrer le document

Options de partage

Avez-vous trouvé ce document utile ?

Ce contenu est-il inapproprié ?

Droits d'auteur :

Formats disponibles

2

Transféré par

Droits d'auteur :

Formats disponibles

arXiv:1004.3396v2 [cond-mat.

mes-hall] 25 Jun 2010

Electronic Properties of Graphene in a Strong Magnetic Field

viii 3.2.4 Dielectric function and static screening . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 62 67 68 69 70 71 72 73

77 . 78 . 79 . 81 . 82 . 84 . 84 . 86 . 90 . 92 . 94 . 99 . 99 . 101 105 108 111

excited state ( ~ 4 eV) ~ 4 eV 2s

The Carbon Atom and its Hybridisations

The Carbon Atom and its Hybridisations

Brief History of Carbon-Based Materials

Brief History of Carbon-Based Materials

Crystal Structure of Graphene

Crystal Structure of Graphene

Electronic Band Structure of Graphene

Electronic Band Structure of Graphene

Tight-binding model for electrons on the honeycomb lattice

d2 r (i) (r) [H a + V ] (j) (r + ij Rl ) (1.13)

= N (i) sij + tij k k

Electronic Band Structure of Graphene where ij j i , sij k eikRl

and we have dened the hopping matrix tij k eikRl

Here, we have separated the Hamiltonian H into an atomic orbital part Ha =

Tight-binding model for the honeycomb lattice.

with the two solutions ( = ) = k tAA + t|k | k . 1 + s|k | (1.22)

Electronic Band Structure of Graphene

represents the 2 2 one-matrix =

1 + eiKa2 eiqa2 + eiKa3 eiqa3 1 1 + ei2/3 1 + iq a2 (q a2 )2 2 1 +ei2/3 1 + iq a3 (q a3 )2 2

Electronic Band Structure of Graphene

in the last line via the 4 4 matrices z =

where we have, now,

conduction band ( = +) 0 valence band ( = )

9a2 tnnn |q|2 , 4

3a |q|a |q| 1 cos(3q ) , 2 4

Electronic Band Structure of Graphene

2eV 1.5eV 1eV

Dirac point motion

Tilted Dirac cones

with the renormalised anisotropic velocities 3 t a 3ta sin and wy = wx = 2

Diagonalising the Hamiltonian (1.57) yields the dispersion relation (q) = w0 q +

Massless 2D Fermions in a Strong Magnetic Field

Dirac Equation in a Magnetic Field and the RQHE

Massless 2D Fermions in a Strong Magnetic Field

Relativistic Landau levels

In terms of the ladder operators (2.7), the Hamiltonian (2.3) becomes

Dirac Equation in a Magnetic Field and the RQHE

Massless 2D Fermions in a Strong Magnetic Field

+,n=4 +,n=3 +,n=2 +,n=1

,n=1 ,n=2 ,n=3 ,n=4

Relativistic Landau levels as a function of the magnetic eld.

Dirac Equation in a Magnetic Field and the RQHE

Massless 2D Fermions in a Strong Magnetic Field

and 1 n,m = n |m = 2 (2.24)

where we have introduced the ux density nB = B 1 2 = h/e , 2lB (2.26)

Dirac Equation in a Magnetic Field and the RQHE

Massless 2D Fermions in a Strong Magnetic Field

(C) (B) (D)

0.96 0.4 T 1.9 K

0.98 0.96 0.94 0.92 1 T 0.90 0.4T

Transition energy (meV)

0 0.0 0.5 1.0

Dirac Equation in a Magnetic Field and the RQHE

transitions have the energies n, = v lB 2(n + 1) 2n , (2.31)

Limits of the Dirac Equation in the Description of Graphene Landau Levels

Limits of the Dirac Equation in the Description of Graphene LLs

Dirac Equation in a Magnetic Field and the RQHE