Interacting Dirac Materials

S. Banerjee1,2,3, D.S.L. Abergel2, H. Ågren3,4, G. Aeppli5,6,7, A.V. Balatsky2,8

1 Theoretical Physics III, Center for Electronic Correlations and Magnetism, Institute of Physics, University of Augsburg, 86135 Augsburg Germany 2Nordita, KTH Royal Institute of Technology and Stockholm University, Roslagstullsbacken 23, 10691 Stockholm, Sweden 3Division of Theoretical Chemistry and Biology, Royal Institute of Technology, SE-10691 Stockholm, Sweden 4 Department of Physics and Astronomy, Uppsala University, Box 516, SE-751 20 Uppsala, Sweden 5 Paul Scherrer Institute, CH-5232 Villigen PSI, Switzerland 6 Laboratory for Solid State Physics, ETH Zurich, Zurich, CH-8093, Switzerland 7 Institut de Physique, EPF Lausanne, Lausanne, CH-1015, Switzerland 8 Department of Physics, University of Connecticut, Storrs, Connecticut 06269, USA E-mail: [email protected]

26 May 2020

Abstract. We investigate the extent to which the class of Dirac materials in two-dimensions provides general statements about the behavior of both fermionic and bosonic Dirac in the interacting regime. For both types, we find common features for the interaction induced renormalization of the conical Dirac spectrum. We perform the perturbative renormalization analysis and compute the self-energy for both quasiparticle types with different interactions and collate previous results from the literature whenever necessary. Guided by the systematic presentation of our results in Table 1, we conclude that long- range interactions generically lead to an increase of the slope of the single-particle , whereas short-range interactions lead to a decrease. The quasiparticle statistics does not qualitatively impact the self-energy correction for long-range repulsion but does affect the behavior of short-range coupled systems, giving rise to different thermal power-law contributions. The possibility of a universal description of the Dirac materials based on these features is also mentioned.

Keywords: Fermi liquid, Dirac , perturbation theory, quasiparticle

arXiv:1803.11480v3 [cond-mat.str-el] 23 May 2020 Submitted to: J. Phys.: Condens. Matter Interacting Dirac Materials 2

1. Introduction materials bear interesting similarities: both have symmetry protected band crossing points around which The discovery of [1, 2, 3] and other materials the excitations have linear dispersion parameterized by with Dirac nodes has led to a strong interest in a velocity. However, for , these are typically materials with Dirac dispersion [4, 5, 6] in condensed low-energy excitations around the at the matter physics. In a typical d-dimensional metal, the Dirac nodes; whereas the bosonic Dirac excitations lie is a (d − 1)-dimensional manifold. In at a higher energy from the ground state (see Fig. 1). Dirac materials, on the other hand, the locus of points Several important questions about this new class with zero quasiparticle energy shrinks to d − 2 or of materials remain unanswered: is there a universal d − 3 dimensions due to some additional symmetries behavior for bosonic and fermionic Dirac excitations present in the system [4, 6]. Specifically, for two- when interactions are taken into account? What dimensional (2D) Dirac materials, Dirac nodes have role does the form of the interaction play? And to be point objects. The low-energy quasiparticles how are these results modified at finite temperature? around this Fermi point have linear dispersion and are The first question is particularly important because described by a massless Dirac-like equation. Because the quasiparticle statistics will have a strong impact of the linear dispersion, the quasiparticles behave like on interacting behavior. Our results in Sec. 3, relativistic particles with an effective speed of light taken together with the existing literature in the replaced by the Fermi velocity. Many realizations of field of Dirac materials, allows us to extract certain Dirac fermions have been discussed in, for example, striking similarities between the renormalization of superfluid phases of 3He [7], the surface states of fermionic and bosonic Dirac materials. On a topological insulators [8, 9, 10], high-temperature qualitative level, long-range interactions (Coulomb d-wave superconductors [11], graphene [12] and type) lead to an increase in the slope of the several artificial electronic systems [13, 14]. The Dirac cone, whereas short-range interactions decrease resulting Dirac cone in the spectrum is protected by this slope. Furthermore, both the fermionic and symmetries [15] and is robust under disorder and bosonic excitations acquire logarithmic temperature- interaction. For example, Dirac nodes are protected dependent self-energy corrections (see Table 1). Such by parity and time-reversal symmetries in graphene [3], temperature dependence arises from Dirac dispersions by time-reversal and mirror symmetries in topological in the fermionic illustration, whereas for bosons it insulators [8] and by a combination of the C4 rotation is associated with the typical parabolic dispersions. and gauge transformation in the case of d-wave On the other hand, different power-law corrections superconductors in two-dimensional square lattices [4]. in temperature and momentum are obtained for the The Dirac nodes lead to commonly observed universal renormalized dispersions for short-range interactions. features of excitations such as the scaling√ of the Table 1 summarizes the trends in renormalization Landau levels with the magnetic field as B [3], of Dirac node velocity. These trends indicate common universal metallic and thermal conductivities related features in velocity dependence in short and long range to the nodal structure of carriers, suppressed back- interacting cases. One indeed might suspect there scattering [4], and Klein tunneling [16]. The common are universal trends in renormalzations of Dirac nodes properties shared by these systems bring about the depending on dimensionality and nature of interaction. unified concept of Dirac materials as a useful category To determine if indeed these trends are universal one for explaining their behavior. would need to investigate a broader class of models in Recently, we are witnessing an upsurge of D = 1,2,3. This investigation of broader class of models theoretical and experimental investigations in bosonic is left for the future work. systems possessing Dirac-like spectra. Plausible In this paper, we analyze the renormalized physical realizations of such bosonic Dirac materials dispersions and the induced quasiparticle lifetimes range from artificial honeycomb lattices made out for the weakly interacting two-dimenisonal Dirac of superconducting grains [17] to magnets [18], excitations. We therefore set up our models for : photonic and acoustic crystals [19, 20], and plasmonic (a) Dirac fermions interacting via an onsite Hubbard devices [21]. At the level of the single-particle physics, repulsion and (b) Dirac bosons subject to a long-range the band structures of bosonic and fermionic Dirac Coulomb interaction. We purposefully recollect the Interacting Dirac Materials 3

(a) (b) various fermionic Dirac excitations; whereas various Refs. [17, 18, 19, 20, 21] (and more) investigated the bosonic counterparts. Band-structures like the ones shown in Fig. 1 containing Dirac-like nodes typically arise from a tight-binding model defined on the non- symmorphic lattice such as honeycomb, kagome, etc. A generic Hamiltonian, in such a case, can be written as

X † † X † † Figure 1. Comparison of the band-structure for non-interacting H = −t ci cj(bi bj) + h.c. ∓ εF (εB) ci ci(bi bi), Dirac bosons (left panel) vs fermions (right panel). (a) The hiji i energy spectrum for a single excitation induced by a ferromagnetic Heisenberg model defined on a honeycomb lattice (1) † † (see Eq. 15 and Refs. [18, 22]). The can occupy any where ci (bi ) is the creation operator for the fermionic state belonging to the two bands which cross each other at (bosonic) quasiparticles, like (magnons, the Dirac points (marked in red and black points). The near , , etc.), at site i within the unit cell vicinity area of the spectrum around these crossing points can be effectively described by a Dirac-like Hamiltonian. (b) In of the lattice. t- is the amplitude for the quasiparticle contrast, the Dirac fermionic excitations are described by the hopping between the nearest-neighbor sites hiji. The low-lying states from the Fermi energy ε at the Dirac node F lattice constant is denoted by a. εF (εB) denotes within a bandwidth ±W . the Fermi (an onsite) energy for fermionic (bosonic) quasiparticles. relevant results for the interacting Dirac magnons (see For example, in a honeycomb lattice, the Ref. [22]) as a prototype for Dirac bosons coupled Hamiltonian in Eq. 1 gives rise to band-structures as via short-range interaction. In a similar spirit, any shown in Fig. 1. In particular, for magnons (bosonic explicit analysis for the renormalization of the Dirac excitations, t = JS), one obtains a band-structure fermions subject to long-range interactions is dropped. as in Fig. 1a (see Ref. [22] for the details). The Instead, we rely on the well-known results obtained onsite energy εB(= 3JS) in this case, accounts for the in Refs. [3, 23]. We do not consider the role of positivity of the magnonic dispersions. In contrast, -orbit coupling or similar interactions that might for fermionic excitations, εF is typically the chemical lead to a gap opening at the Dirac point, and potential at the nodal point. The low-lying fermionic consequently, trigger a topological phase transition. states, within the bandwidth ±W around the nodal In this paper, we focus in the systems where the points (the red/black points in Fig. 1b), are typically Dirac crossing point is preserved. Only, the linear described by an effective Dirac Hamiltonian spectrum around the crossing point is modified due X † to various interactions. Hence, interaction induced Heff ' ΨkHkΨk; Hk = vDσ · k, (2) renormalization of Dirac spectrum does not change the k topology of the underlying system. where vD, the slope of the cone, is the Dirac velocity. † The rest of the paper is structured as follows. The spinor Ψk is composed of the states belonging We briefly revisit the free Dirac theory of bosons to the two inequivalent sub-lattices in the honeycomb and fermions in Sec. 2. Sec. 3 is divided into unit cell. The σx,y correspond to multiple subsections, where we systematically explore the pseudospin structure arising out of the sub-lattice the individual scenarios as discussed in the previous degrees of freedom. See Appendix A for the details paragraph. Although, the main focus of this paper of the derivation of fermionic and bosonic Dirac is on two-dimensional materials, we provide a short Hamiltonian. discussion on the interacting Dirac bosons in one and We notice that the bosonic Dirac excitations three dimensions in Sec. 4. We finally provide a are high-energy states compared to the band-bottom detailed discussion for our results on the interacting edge in Fig. 1a. As a result, it is sufficient to Dirac matter and conclude in Sec. 5. focus on the conical portion (see Fig. 1b) around the Dirac nodes for fermions, whereas for bosons 2. Dirac Materials: Noninteracting the entire band-structure is considered as illustrated quasiparticles in Fig. 1a. Despite these mutual differences, Dirac bosons and fermions display various common We begin by revisiting the Dirac theory of non- experimental features – for example, the Klein- interacting quasiparticles. Numerous material-specific tunneling in photons possessing Dirac-like dispersions realizations of the Dirac excitations of both fermionic of topological photonic insulators [25], the square- and bosonic origin have been observed in the last two root magnetic-field-dependent in decades. Ref. [4] provides an interesting review of Dirac phonons [26]. Motivated by these common Interacting Dirac Materials 4

Table 1. This table summarizes the primary results for various self-energy corrections in the case of interacting Dirac materials with both type of quasiparticles: fermions and bosons. The self-energy Σ corresponds to the upward region of the band-structure around the Dirac cone (see Fig. 1). Here, x corresponds to either a single or a combination of the variables – momentum k, frequency ω or temperature T (in the unit of Boltzmann constant, kB = 1; ~ = 1). Note: The respective referrences for the known results are provided in the entries where-ever applicable. The pseudo-relativistic invariance of the Dirac fermions has been used to deduce the momentum dependence of Σ in case of the Hubbard repulsion. At zero temperature, the bosonic quasiparticle excitations are absent and hence do not lead to any interactions. Hence, the self-energy correction is absent in these cases. However, this will not hold for a generic bosonic system.

Quantity Quasiparticle Long-range interaction Short-range interaction

Λ 2 3 3 T = 0 k log k , ω log ω [24] ∼-U max(k , ω ) h Λ i 3 3 T 6= 0 k log max(k,T ) -max(k , ω ) ReΣ(x) T = 0 Zero Zero [22] T 6= 0 T log T -T 2 (see Fig. 6) [22]

T = 0 -max(|k|,|ω|) [24] -max(k4, ω4) Fermion T 6= 0 -max(|k|,T, |ω|) -max(k4, ω4,T 4) ImΣ(x) T = 0 Zero Zero [22] Boson T 6= 0 -T log T -T 2 (see Fig. 7) [22]

experimental observations, we set out to investigate description of the Hartree-Fock calculation leading to whether such similarities exist in the interacting the logarithmic divergence) regime.  Λ H ' v 1 + α log σ · k, (3) k D k 3. Interacting Dirac materials where the second term in the bracket is the first- So far, we focused on the non-interacting Dirac order self-energy correction to the non-interacting 2 materials. Despite, their mutual differences for the Dirac Hamiltonian. α = e /0vD is the dimensionless quasiparticle type, both fermionic and bosonic Dirac coupling constant and Λ (∼ 2π/a) is an ultraviolet materials display numerous identical experimental cutoff. We notice that the renormalized Dirac velocity signatures. However, whether these similarities diverges as k → 0 implying an increase in the Dirac indicate an underlying universal description, especially velocity. in the interacting regime, remains unexplored. In This divergence of the renormalized Dirac velocity the subsequent sections, we focus on a subset of at k = 0, however, disappears at a finite temperature interacting Dirac systems and try to extract some T as 2   e vDΛ common features. δvD(k) = log , (4) 20 2kBT 3.1. Coulomb repulsion: Fermions where kB is the Boltzmann constant. The details of this result are provided in Appendix B. Such We begin with the most popularly studied system: a logarithmic temperature dependence has been Dirac fermions interacting via a Coulomb repulsion previously mentioned in Ref. [29] based on the scaling 2 Vq = e /0q. We witnessed an abundance of both argument. Here, we provide an alternative derivation theoretical and experimental investigations in this case, based on the perturbation theory. It is natural to over the past twenty years (see e.g. Refs. [3, 23, expect that at low temperature, the Dirac velocity 27, 28]). Therefore, without including any explicit should renormalize as derivations, we recollect the results in this case as (see   vDΛ Ref. [3] and at the end of Appendix B for a brief δvD(k) ∝ log . (5) max{vDk, 2kBT } Interacting Dirac Materials 5

(a) (b) (c)

Figure 2. (a) The spectral function A(k, ω) (in units of eV−1) for the non-interacting fermions on a honeycomb lattice is shown along the high-symmetry points in the Brillouin zone. The linear Dirac spectrum (with U = 0 in Eq. 9) is highlighted by the boxed area. (b) The renormalized spectral function A(k, ω) (in units eV−1) at zero temperature for three different strengths of the Hubbard interaction U = 0.2, 0.6, & 0.95 (marked by filled circles) as mentioned in the panel. (c) The spectral function, at fixed momentum kP (marked by the dotted white line in panel (b)), as function of the energy (positive valued) for various U values as shown in the panel. Note: the tight-binding hopping t is assumed to be 1 eV.

3.2. Screened Coulomb interaction: Fermions renormalized Dirac Hamiltonian as ! In the previous section, we discussed the renor- e2 Λ HY ' v 1 + log σ · k, (8) malization of Dirac fermions coupled via the un- k D p 2 0vD k2 + q screened Coulomb interaction. In real samples the TF bare Coulomb repulsion is screened. Such screening of where, once again we notice that the renormalized Coulomb interaction in two-dimension has been previ- Dirac velocity at k = 0 does not diverge due to the ously derived in Refs. [30, 31, 32] as presence of the cut-off qTF , which is related to the finite 2 range of the Yukawa potential. e  πqTF r   VSC (r) = 1 − H0(qTF r) − Y0(qTF r) , 0r 2 (6) 3.3. Onsite Hubbard interaction: Fermions 2 2 2 where qTF = 4e |µ|/0~ vD is the Thomas-Fermi Following the previous two sections, we observed that wave-vector, µ is the , and H0 and a moderate/long-range interaction qualitatively leads Y0 are the Struve and Neumann functions. Following to an upward renormalization of the Dirac velocity. To a similar analysis (see Appendix C) leading to the understand whether this generic feature extends over renormalized Dirac structure in Eq. 3, in this case we the entire range of the underlying interactions, here obtain we focus on the opposite limit – an onsite Hubbard  2  interaction. SC e Λ H ' vD 1 + log σ · k. (7) We perform the relevant calculations on a k  v k + q 0 D TF honeycomb lattice. The Hamiltonian can be written The Dirac velocity, of course, acquires a similar as logarithmic correction; however, the k = 0 divergence X † X X † disappears even at zero temperature. At finite H = −t ciσcjσ + U ni↑ni↓ − F ciσciσ, (9) temperature, we expect a similar modification to the hiji,σ i i,σ renormalized Dirac velocity as in Eq. 5 with δv (k) ∝ D where σ is the spin, and U denotes the strength of log[v Λ/max{v (k + q ), 2k T }]. It becomes D D TF B the onsite repulsion. Other parameters are defined in evident that the divergence of the Dirac velocity as Eq. 1. We assume t = 1 eV for further analysis. As in Eq. 3, arises out of the long-range nature of the discussed in Sec. 2, the non-interacting quasiparticles unscreened Coulomb interaction. When this true long- are described by a Dirac-like Hamiltonian at half- range character is modified to an intermediate/short- filling. Hubbard model on honeycomb lattice has range type, the divergence disappears as in Eq. 7. been extensively studied in the past. Especially, the In order to better understand this feature, we semi-metal to anti-ferromagnetic insulator transition analyze the renormalization of the non-interacting has been the primary focus of the state-of-the-art Dirac Hamiltonian subject to a Yukawa potential 2 numerical calculations [33, 34, 35]. While much of e −qTF r VY (r) = e . Indeed, performing a calculation 0r investigations [36, 37, 38] focus on understanding such similar to the one in Appendix C, we obtain the phase-transition, the semi-metallic phase for U  Uc has remained relatively less explored. Here, Interacting Dirac Materials 6

Figure 5. The variation of the imaginary part of the off- diagonal self-energy Σ12(K, ω = 0) as a function of temperature. The Hubbard coupling is taken to be U = 0.6t. Figure 3. The imaginary part (open circles) of the off-diagonal self-energy Σ12 (in the units of t) for the positive part of the frequency ω evaluated at the Dirac point at zero temperature the renormalized Green’s function G(k, ω) as for U = 0.6t. The numerical fit (solid red line) of our result with A(k, ω) = −Im Tr G(k, ω). B the ansatz function ImΣ12(ω) ∼ A · ω . The corresponding renormalized bands are shown in Fig. 2b for various interaction strengths U at zero temperature (only the highlighted portion of the Uc represents the critical strength of the Hubbard interaction. spectrum in panel Fig. 2a is considered here). On a Our goal, here, is to understand the modification closer look, we observe three separate lines as marked of this Dirac structure as a function of U in the weak by filled circles. However, to clearly distinguish these coupling regime. We obtain the following second-order three separated lines, we plot the spectral function self-energy correction (see Appendix D) as A(kP , ω) at a fixed momentum kP (marked by the dotted line in panel (b)) for different frequencies ω ! Σ(k, ω)Σ12(k, ω) within an energy window from 0.8 eV to 1.2 eV (see Σ(k, ω) = , (10) Fig. 2c). Indeed we observe three distinct peaks at Σ21(k, ω) Σ(k, ω) three different ω’s. Consequently, we infer that the where Σ(k, ω) denotes the 2 × 2- self-energy matrix bandwidth gradually decreases as a function of the arising out of the spinor structure of the Dirac coupling U, while keeping the Dirac point preserved. Hamiltonian. Individual components in the matrix Hence, the Dirac velocity also decreases as a function entries are provided in Appendix D. We begin with of U. analyzing the renormalized spectral function A(k, ω) from Eq. 10. The bare spectral function for the non- 3.3.1. Analytical structure of Σ(k, ω): We now fo- interacting case with U = 0 is shown in Fig. 2a. cus on analyzing the explicit frequency, momentum The linear band crossing at the Dirac point K is and temperature dependence of the self-energy com- shown within the highlighted area in Fig. 2a. In ponents. We first analyze the self-energy components order to track the consequent band renormalization at zero temperature. We observe that the diago- due to Hubbard repulsion, we analyze the spectral nal component Σ(ω) at the Dirac point exhibits very function A(k, ω). The latter is computed from weak frequency dependence; whereas the off-diagonal component Σ12(ω) varies strongly with frequency (see Fig. 4a,b). To extract the frequency dependence of ImΣ12, we depend on numerical fitting. We obtain the best fit to a numerical ansatz A · ωB with an exponent B ∼ 4. We further notice that the other remaining pa- rameter A is estimated to be extremely small in mag- nitude. The origin of such a small magnitude is possi- (a) (b) bly connected to the vanishing quasiparticle density-of- states at the Dirac point. Using the Kramers-Kronig Figure 4. The real (a) and imaginary (b) part of the off- 3 relation, we find that ReΣ12 exhibits a ω frequency diagonal self-energy Σ12 (in the units of t) as a function of the frequency ω evaluated at the Dirac point at zero temperature, dependence. for the strength of the Hubbard repulsion U = 0.6 (in unit of t). The imaginary part of the self-energy for a Interacting Dirac Materials 7

2 † 2D Fermi liquid typically follows ω log(ω) frequency where bi is a Cooper pair creation operator within dependence [24, 38]. Hence, our results dictate a a superconducting island at site i for the honeycomb deviation from the Fermi liquid picture. Considering lattice. The former is defined as the decrease of the band-width in Fig. 2b along with † X † † the relativistic invariance of the quasiparticles around bi = c↑rc↓r, (13) r∈G the Dirac point (ω = vDk), we predict the Dirac i velocity renormalizes as where the summation in Eq. 13 is over an individual † † 2 2 grain Gi at site i with c (c ) being the fermionic δvD ∝ U k . (11) ↑r ↓r creation operator with upward (downward) spin orientation. 3.3.2. Finite temperature: So far, we discussed The other parameters in Eq. 12 are defined as the renormalization of the Dirac spectrum at zero follows: (i) t is the hopping amplitude of the pairs temperature. Now, we focus on the temperature between the sites, (ii) U is an onsite potential, and evolution of the off-diagonal component Σ at 12 (iii) n denotes the mean-number of Cooper pairs in the Dirac point. The temperature evolution of 0 each grain. The collective excitations of this Cooper ImΣ (K, ω = 0) is shown in Fig. 5. Of course, it 12 pair network produce an energy spectrum similar to is very difficult to determine the form of temperature the one shown in Fig. 1a. The two inequivalent modes dependence just by examining our numerical results participating in this excitation spectrum around the at a small temperature. However, at a larger crossing point are effectively described by a Dirac- temperature, our result closely resembles a quartic √ like Hamiltonian with a velocity v = a 3JU with power-law dependence as in Fig. 3. Using the best D 4 fit, we predict a similar T 4 temperature dependence J (=n0t) representing the strength of the kinetic for the imaginary part of the self-energy. energy for these modes. However, when a nearest- 0 P neighbour interaction U hiji ninj is turned on, the Dirac velocity further modifies. A simple mean-field 3.4. Short-range interactions: Dirac bosons analysis predicts that the renormalized Dirac velocity In the previous sections, we observed an interesting decreases as 3U 0 similarity between the interacting Dirac fermions. δvD/vD ≈ − , (14) First, the low-energy fermionic excitations are still 2U 0 described by a modified Dirac-like Hamiltonian under for U . U  J (see Appendix A for the details of the the influence of the set of incorporated interactions. derivation of Eq. 14). Hence, again, in this case, the Secondly, long-range interactions typically lead to a algebraic structure of the Hamiltonian describing the logarithmic renormalization of the self-energy, whereas states around the crossing point remains isomorphic only power-law corrections are obtained for sufficiently to the non-interacting Dirac Hamiltonian, whereas localized interactions. We are, therefore, naturally the Dirac velocity acquires a quantitative downward tempted to analyze a few interacting bosonic systems modification. to find if there are also such similarities in comparison to the fermionic case. 3.4.2. Dirac magnons: Heisenberg interaction We begin with focusing on the two typical bosonic Previously, we described that the short-range inter- Dirac systems: (a) Cooper pairs that are described by actions lead to a decrease in the renormalized Dirac an effective Bose-Hubbard model and (b) interacting velocity of the interacting Dirac boson. However, magnons induced by a ferromagnetic Heisenberg model one might naturally question the robustness of such a defined on a honeycomb lattice. mean-field analysis. To answer this, we revisit our pre- vious work [22] on the ferromagnetic insulator CrBr3, 3.4.1. Cooper pairs: Extended Hubbard interac- where Cr3+ spins interact via Heisenberg exchange as tion Earlier theoretical work (see Ref. [17]), by some X of the present authors, analyzed a physical realization H = −J Si · Sj, (15) for the granular superconducting network for Dirac- hiji like bosons that arise from the collective excitations of the Cooper pairs. The corresponding Hamiltonian is where J denotes the strength of the Heisenberg described by a Bose-Hubbard model on a honeycomb exchange interaction between the nearest-neighbor lattice as sites hiji of a honeycomb lattice. Utilizing Holstein- Primakoff (HP) transformation, the spin operators X † X 2 S , in the above Hamiltonian, can be written H = −t bi bj + h.c. + U (ni − n0) , (12) i hiji i in terms of the magnon creation and annihilation operators corresponding to the induced spin-wave Interacting Dirac Materials 8

(a) (b) that the magnonic excitations around the Dirac point acquire momentum-dependent decay rates as shown in Fig. 7. Both the real and imaginary parts of the self- energy contain a T 2 temperature dependence.

3.5. Long-range interactions: Dirac bosons So far, we discussed the renormalization behavior of various interacting Dirac systems. The results derived or collected from the existing literature, throughout Figure 6. Comparison of the renormalized band-structure of Sec. 3.2 – Sec. 3.4 already point to an interesting the interacting Dirac bosons (left panel) vs fermions (right panel) similarity between the fermionic and bosonic Dirac around the crossing point. (a) The conical Dirac spectrum (solid matter: short-range interactions qualitatively leads to line) of the non-interacting spin-wave excitations in a honeycomb a decreased Dirac velocity with power-law self-energy ferromagnet, arising from the Holstein-Primakoff transformation (HPT) in Eq. 15. The leading higher order terms of the HPT corrections. In contrast, long-range interactions yield leads to a short-range interaction resulting in a renormalized an increased Dirac velocity for fermions with non- spectrum (dashed line). Along with a finite shift in the Dirac polynomial (logarithmic) self-energy renormalization. point energy the Dirac velocity decreases. Reproduced from Hence, one may ask whether a similar behavior is Ref. [22]. (b) The bare conical spectrum (solid line) of the Dirac fermions renormalizes due to the long-range interaction leading replicated in case for Dirac bosons coupled via a long- to an increased Dirac velocity as illustrated in the modified range interaction. spectrum (dashed line). In this section, we consider a model for charged bosonic quasiparticles interacting via the long-range Coulomb repulsion on a honeycomb lattice. The excitations. The leading order terms in the Holstein- corresponding Hamiltonian is written as Primakoff expansion leads to the bosonic version of the Hamiltonian (see Eq. 1) already discussed in Sec. 2. ∗2 X † X † e X ninj The subleading higher-order terms, essentially H = −t bi bj + 3t bi bi + , (16) 0 |ri − rj| yield an effective model of the Dirac magnons hiji i ij interacting via short-range interactions, almost similar † to nearest-neighbor Hubbard interaction discussed in where bi denotes the creation operator for the the previous section. The resulting renormalization of bosonic excitation with charge e∗, t is hopping the magnons has been studied in detail within a second- amplitude between nearest-neighbor sites on the order perturbation theory in our previous work [22]. honeycomb lattice and the last term denotes the For the purpose of the recent paper, here we skip Coulomb interaction between the two different sites the corresponding explicit derivations and recollect the i, j. Motivated by the free spin-wave theory on a primary results: (a) the renormalized spectrum around honeycomb lattice (see Heisenberg model discussion in the Dirac crossing point (see Fig. 6a), and (b) induced Sec. 3.4), we introduced the onsite energy 3t in the decay rate of the spin-wave excitations (see Fig. 7). model in Eq. 16 to constraint positive definite spectrum We observe that the renormalized Dirac velocity, for the bosons. in this case, decreases (the slope of the modified cone To analyze the effect of the Coulomb interaction (dashed line) is smaller than the slope of bare Dirac on the Dirac spectrum, the Hamiltonian (Eq. 16) is cone (solid line) in Fig. 6a). We further observe rewritten in the diagonal basis. We focus on a subset of the scattering processes which are relevant at low temperature (see Appendix E for the details). We (a) (b) notice that there are eight different scattering processes that take part in the Coulomb interaction in the diagonal basis. We assume that one of the interacting bosons (dk0 ) is always excited by the thermal energy, and due to low temperature will be occupied near the bottom of the band-structure as illustrated in Fig. 1a. Figure 7. Scattering rates of the up (a) and down (b) The other bosonic excitations can, in principle, be bands of the spin-wave excitation spectrum (see Fig. 1a), due created anywhere in the band-structure using some to the leading higher order terms of the Holstein-Primakoff external source (viz. in case of magnons, this external transformation of the Heisenberg model in Eq. 15. The rates source could be inelastic neutron scattering). Under (in the units of JS) are shown along the high-symmetry points in the Brillouin zone of a honeycomb lattice. Reproduced from this approximation, we retain only three of all the Ref. [22]. Interacting Dirac Materials 9 scattering channels (see Eq.E.4) as 4. Dirac materials in other dimensions

X h † † In this section, we briefly discuss a few Dirac Hint = Vq 2 cos Φuk−qdk0+qukdk0 + (17) q,k,k0 systems in one and three dimensions and outline the role of interactions. We start from the most † † 2 cos Φ−cos Ψ  † † i i sin Φdk−qdk0+qukdk0 + 4 dk−qdk0+qdkdk0 , common example of one dimensional (1D) Dirac excitations – the solutions of the Su-Schrieffer-Heeger where the angles Φ, &Ψ are defined in Appendix E. (SSH) model [39] which describes spinless fermions Assuming the strength of the Coulomb interaction is on a dimerized 1D lattice with staggered hopping ∗2 small compared to kinetic energy (e /0as < t, where amplitudes. In this case, the effective Hamiltonian as is the Born scattering length), we set up a second- around the edge of the Brillouin zone describes massive order perturbation theory. The self-energy for the Dirac fermions. The mass term disappears when respective channels are derived in Appendix E as the two in-equivalent hopping amplitudes become

2 d identical, and consequently leads to nodal Dirac X |Vq,k| b(ε 0 ) Σ(k) =∼ k . (18) fermions. It is natural to expect that a bosonic εk − εk−q − εq + iδ k0,q analog of SSH model would lead to similar Dirac bosons at the edge of the one-dimensional Brillouin We notice that the temperature dependence of the zone. Indeed, there have been recent theoretical self-energy in all the three channels is governed by predictions about such collective bosonic excitations P d the factor k0 b(εk0 ); where the dispersion for the on plasmons [40], magnons [41], [42], etc. d thermally excited bosons is approximated as εk0 ≈ While the primary focus on these one dimensional 3 0 2 4 t|k | . To evaluate this integral, we introduce an Dirac systems has been on the topological aspects of −1 infrared cut-off k0 ∼ L , where L is the linear system the corresponding system, far less investigation is done size. The integral is consequently evaluated as on the electronic interaction effects. Previous studies of short-range interactions on fermionic SSH model based Z ∞ k0dk0 4π log(3βt/4N 2) 2π ≈ − (19) on the renormalization group analysis [43] indicates 3βt k02 3βt k0 e 4 − 1 that the low-energy Dirac-like behavior belongs to a two-component Gross-Neveu class, in a similar way where L = Na and β = 1/T in the unit of to the two-dimensional Dirac fermions. Based on our Boltzmann constant k = 1. At this point, we B analysis of two dimensional systems in this paper, we would like to mention that the possibility of the predict that the Dirac velocity will also renormalize formation of a Bose-Einstein condensate is ignored by downward in the 1D case for short-range interactions. assuming that the relevant temperature is always larger In this spirit, the renormalization should be upward for than the corresponding condensation temperature (the any long-range interactions. condensation temperature for parabolic bosons is Renormalization effects of three-dimensional (3D) absolute zero in the two-dimensions). Consequently, Weyl-Dirac fermions were studied in detail in Ref. [44], we obtain which found that the Dirac velocity is renormalized 2 2 4T π h4TN i X |Vq,k| upwards in the presence of Coulomb interaction. This Σ(k) ≈ log . result is consistent within our predictions that long- 3t 3t εk − εk−q − εq + iδ q range interaction generically leads to an increase in (20) the Dirac velocity. Consistently, the short-range The diverging logarithmic pre-factor in the bosonic interactions remain marginally irrelevant [45] and lead case is in stark contrast to the logarithmic divergence to a decrease in the Dirac velocity. To investigate for the Dirac fermions case (see Eq. 4). In the latter whether this feature persists in the 3D bosonic case, it is associated with an ultraviolet energy scale Dirac systems, we need to incorporate explicit model (Λ), whereas for the bosonic system, we have an calculation for various bosonic systems, which is left infrared energy scale (N) which dictates the overall for future work. size of the underlying system. The momentum- dependent renormalization is consequently governed by the remaining integral in Eq. 20. The relevant analysis 5. Discussion and conclusion of the momentum-dependence of the self-energy is In this paper, we present a systematic study of the provided in Appendix F. The renormalized spectrum renormalization effects in Dirac materials for various is shown in Fig. F2. We notice that the Coulomb interactions in the weak coupling regime. The results repulsion does not lead to any gap opening at the recorded in Table 1, give a unified summary of the Dirac point. Instead, the Dirac velocity is renormalized effect of the interactions on the Dirac quasiparticles upward. for the models studied/discussed in this paper. The Interacting Dirac Materials 10 form of the quasiparticle interactions qualitatively strong enhancement of this feature and removes the affects the velocity renormalization – contact or short- divergences [29]. Hence, experimental observation of range interactions generically lead to a decrease in the diamagnetic susceptibility can, in principle, display the Dirac velocity, whereas long-range interactions the features of the non-linear renormalization for long- (both screened and unscreened) typically leads to an range interactions in the two-dimensional fermionic increase. The quasiparticle statistics – i.e. whether systems. We notice that the emergence of non- we are dealing with bosons or fermions – does affect Fermi liquid behavior for the Dirac fermions under the power laws fixing the temperature-dependence of the influence of long-range interaction is responsible the self-energy corrections for short-range interactions, for the dramatic changes in the corresponding physical but does not influence the logarithmic behavior for the observables. long-range Coulomb repulsion. However, in the case of the short-range interac- In addition to the discussion on the overall tions in Dirac fermions, such non-Fermi liquid charac- renormalization of the Dirac velocity, we found non- teristics arise only to the sub-leading correction in the Fermi liquid behavior for interacting Dirac fermions. self-energy. Hence, it becomes a challenging task to Our analysis in Sec. 3.3 on the Hubbard model observe the fingerprints of such corrections (Sec. 3.3) on honeycomb lattice leads to a quartic frequency- in a physical quantity. Precise measurements of vari- dependent ImΣ(ω), in comparison to the conventional ous quantities viz. magnetoresistance, quantum Hall- ω2 log ω structure for the two-dimensional Fermi liquid. effect, etc may show deviations from the Fermi liquid We furthermore mention that the primary purpose of predictions which can be associated with the renormal- the bosonic analysis (in Sec. 3.5 and Appendix F) is ization corrections described earlier. to capture the overall trend for the renormalized band For the bosonic systems, the experimental obser- structure around the Dirac point without aiming for a vation of the features of the spectral renormalization precise quantitative estimate. around the high-energy crossing point is complicated, A closer look at Table 1 implies that the power- as the low-energy bosons near the ground state would law exponents in the self-energy for the short-range naturally couple to any experimental probe. Addition- interactions crucially depend on the specific model. ally, the quasiparticle excitations are neutral in most of However, the common feature in all these situations the conventional bosonic systems. We expect that the is the overall decrease of the Dirac velocity governed measurement of various thermal quantities may show by a polynomial type correction for both the real and the effects of the renormalization of the conical Dirac imaginary part of the self-energy. On the other hand, spectrum. the long-range interactions in both the fermionic and To conclude, our results show how the quasiparti- bosonic cases, yield logarithmic self-energy corrections. cles in Dirac materials are modified by various interac- In both these cases, the renormalized Dirac velocity tions. While our results are based on a specific subset consists of divergences. Yet, the nature of such of the diverse realizations of Dirac materials, a uni- divergences differs between the two statistical species versal description of the same for both fermionic and – fermions have an ultraviolet divergence compared to bosonic quasiparticles remains an interesting possibil- an infrared singularity in the bosonic case. We admit ity. that there are definitely certain intrinsic differences between fermionic and bosonic Dirac materials in the 6. Acknowledgments interacting regime. At the same time, the qualitative similarity of the results based on Table 1 is naturally S.B. acknowledges special thanks to Arno P. Kampf. reflected based on (a) slowing down or speeding up We are grateful to D. Vollhardt, M. Kollar, W. A. the Dirac quasiparticles and (b) the appearance of Atkinson, F.F. Assaad, D.N. Basov, L. Chioncel, P. polynomial or non-polynomial correction in the self- Hofmann and S.S. Pershoguba for discussions. S. B. energy. acknowledges support by the Deutsche Forschungs- It is important to note that many of the fermionic gemeinschaft through TRR 80. H. Å. acknowledges renormalization features analyzed in this paper are the Knut and Alice Wallenberg Foundation for finan- also verified in various experimental observations, cial support (Grant No. KAW-2013.0020). Work at viz. Quantum Oscillations experiments in suspended KTH was supported by ERC DM 321031, the KAW- graphene [28] can be used to extract the non- 2013.0096 and the VILLUM FONDEN via the Centre linear momentum dependence of the Dirac velocity of Excellence for Dirac Materials (Grant No. 11744). (Sec. 3.1,3.2). On the other hand, the diamagnetic susceptibility in two dimensional Dirac fermionic systems typically diverges at zero temperature [46, 47]. However, the unscreened Coulomb repulsion leads to a Interacting Dirac Materials 11

Appendix A. Derivations of the effective Dirac bosonic Hamiltonian in the momentum-space as Hamiltonian X  3 −γ  H = Ψ† H Ψ , H = JS k , (A.7) In this section, we provide the details of the derivation k k k k −γ∗ 3 k k of the effective Dirac Hamiltonian starting from the tight-binding description on the honeycomb lattice. where the spinor Ψk is defined in terms of the two The fermionic Hamiltonian in Eq. 1 can be rewritten inequivalent boson operators on the two sublattice sites in momentum space as for an honeycomb lattice. γk is the same as defined h i earlier in Eq. A.2. The diagonal terms 3JS arises X † ∗ † z z H = −t γkcakcbk + γkcbkcak , (A.1) from the Si Sj coupling in the Heisenberg interaction k of Eq. 15. In contrast to the fermionic case, we keep the entire structure in the full Brillouin zone. The band where the chemical potential µ is assumed to be zero, structure from the Hamiltonian in Eq. A.7 is shown in c† creates an on the α sublattice site and γ αk k Fig. 1a. is defined as 3 Finally, we provide the details of the mean-field X ik·δi γk = e , (A.2) result in Eq. 14. We start from an extended Bose- i=1 Hubbard model defined on a honeycomb lattice as where δ are the three nearest-neighbor vectors as i X † X 2 0 X shown in Fig. D1a H = −t bi bj +U (ni −n0) +U ninj, (A.8) hiji i hiji 1 √ 1 √ 1 δ1 = (1, 3), δ2 = (1, − 3), δ3 = (−1, 0), (A.3) † 2 2 2 where bi is the bosonic quasiparticle creation operator and U, U 0 are the strengths of the onsite and nearest- in the unit of the lattice constant a. Please note neighbor Hubbard interactions, respectively. We that the irrelevant spin degrees of freedom has been further assume that there are n mean number of neglected. The Hamiltonian in Eq. A.1 can now be 0 bosons per grains. Now we translate the bosonic rewritten in the matrix form as operator to their charge-density representation as   †A p A iθA †B p B iθB X † 0 γk b = n e i and b = n e i , where A/B H = Ψ H Ψ , H = −t , (A.4) i i i i k k k k γ∗ 0 corresponds to the two sublattice sites. Assuming k k the bosons are now Cooper pairs on a granular where Ψk is a spinor composed of the two sublattice superconductor as described, in Sec. 3.4.1, we focus † † † operators as Ψk = (cak, cbk). Expanding the 2 × 2- on the low-energy effective theory of the fluctuations 0 Hamiltonian in Eq. A.4 near the Dirac point K = of the phase θi’s. For small U and U , we focus on the ( 2π , 2√π ) we obtain 3 3 3 small phase fluctuations around mean boson density n0 and consequently obtain the following non-interacting   3t 0 kx − iky Hamiltonian as HK+k = − . (A.5) 2 kx + iky 0 X h A A B B A B H = 3Jδθk δθ−k + 3Jδθk δθ−k − γkδθk δθ−k Rewriting the effective Hamiltonian with Pauli matri- k ces we obtain the effective Hamiltonian in Eq. 2 with − γ∗δθBδθA + UδnAδnA + UδnBδnB the Dirac velocity v = 3t . k k −k k −k k −k D 2 U 0 U 0 i A bosonic analogue of the effective Hamiltonian A B ∗ A A + γkδnk δn−k + γkδnk δn−k , (A.9) resulting from the Heisenberg interaction in Eq. 15, is 2 2 obtained by rewriting the spin operators Si in terms of where the differential phases and densities are small magnon operators. The latter is obtained by Holstein- fluctuations around the ground state (n0, θ0) and J = Primakoff transformation (HPT) [22] as n0t. We now solve for the normal mode frequencies from the Hamiltonian in Eq. A.9 and obtain the √ † ! x y ai aiai 3 S + iS = 2S a − + O(1/S 2 ), frequencies for the two modes as [17] i i i 4S  0 0  ! 3U U 2 √ † † ω1,2(k) = J 3U ∓ |γk| ± U|γk| − |γk| . x y † ai ai ai 3 S − iS = 2S a − + O(1/S 2 ), 2 2 i i i 4S (A.10) The modification of Dirac velocity can now be easily Sz = S − a†a , (A.6) i i i extracted as in Eq. 14 by taking the limit U 0 to zero. where S is the total spin. We substitute the leading order terms of HPT in Eq. 15 to obtain the effective Interacting Dirac Materials 12

1 Appendix B. Coulomb repulsion at finite where β = k T and kB is the Boltzmann constant. 0 B temperature Relabeling ωn = m + ωn and q → q − k, we simplify the integrand as In Sec. 3.1, we discussed the disappearance of the 0 2 logarithmic divergence of the Dirac velocity at a finite 0 0 iωn + vDσ · q e G(q, iωn)D(q−k, iωn−iωn) = 0 2 2 2 , temperature as in Eq. 4. Here, we outline the details (iωn) − vDq 0|k − q| of the derivation of this result. A generic two-body 0 interaction V (r, r0) can be formulated as where V (r, r ) is taken to be the Coulomb interaction. Now, we compute the frequency summation in Eq. B.5 Z 0 † † 0 as H = drdr ψrψr0 V (r, r )ψr0 ψr, (B.1) 0 1 X iωn + vDσ · q where ψr is a fermionic field. Eq. B.1, which contains 0 2 2 2 β 0 (iωn) − vDq four fermionic fields, can be recasted in terms of two- ωn fermionic fields by introducing an auxiliary bosonic 1 X 1h 1 1 i vDσ · q = + + field φ as 0 0 0 2 2 2 r β 0 2 iωn − vDq iωn + vDq (iωn) − vDq iωn Z Z † † 0 φrφr0 βvD q − βvD q 1h i 1he 2 − e 2 iσ · q H = drφrψrψr + drdr 0 . (B.2) V (r, r ) = − f(vDq) + f(−vDq) − βvD q − βvD q 2 2 e 2 + e 2 q Eq. B.2 is obtained from Eq. B.1 using the Hubbard- 1 1 βv q  σ · q Stratonovich transformation. It is important to = − − tanh D , (B.6) 2 2 2 q mention that the original Hamiltonian in Eq. B.1 can be recovered by integrating out the auxiliary bosonic where f(ε) is the Fermi-Dirac distribution function. field φ , which can be thought of as an instantaneous r Hence, the self-energy Σ(k, iωn) in Eq. B.5 becomes mediator of the exchange interaction V (r, r0) between "   # 2 the two fermionic fields. The dynamics of φr is X βvDq σ · q e Σ(k) = 1 + tanh , governed by its Green’s function 2 q 2 |k − q| q 0 D † E 0 φr,tφr0,t0 = V (r, r ),D(q, im) = V (q), (B.3) (B.7) where D(q, im) is the Green’s function in momentum where the irrelevant frequency dependence in the self- space, with V (q) being the Fourier transform of energy has been dropped. The first term in the bracket 0 V (r, r ) with m being the bosonic Matsubara in Eq. B.7 corresponds to a constant shift in energy. frequency. We now consider the non-interacting Hence, we focus on the second term in the ultraviolet fermionic Green’s function. Following the Hamiltonian (large momentum behavior of the integrand in Eq. B.7) 2 in Eq. 2, we obtain the latter as the regime as (in the unit of e /20)   1 βvD kλ G(k, iωn) = , (B.4) kσ Z (cos θ, sin θ) tanh 2 iωn − vDσ · k Σ(k) ∝ · λdλdθ , 2 p1 + λ2 − 2λ cos(γ − θ) where ωn is the fermionic Matsubara frequency. Next, we consider the lowest order (in the (B.8) dimensionless coupling α) self-energy correction to the where the variable λ is defined as λ = q/k. The Green’s function in Eq. B.4 as (see Fig.B1) exchange wavevector is written in polar coordinate 1 X as q = q(cos θ, sin θ) and the angle between k and Σ(k, iω ) = − G(k + q, iω + i )D(q, i ), n β n m m q is γ − θ. For large q → Λ (where Λ ∼ 2π/a is ωn,q (B.5) the ultraviolet cutoff for the momentum in honeycomb lattice), the integrand in Eq. B.8 reduces to

Λ kσ Z k h Σ(k) ∝ · dλdθ (cos θ, sin θ) cos(γ − θ)· 2 βv kλ i tanh D 2 Λ σ Z k dλ βv kλ = · tanh D (k cos γ, k sin γ) Figure B1. Hartree-Fock self-energy diagram that leads 2 λ 2 to the self-energy expression in Eq. B.5 and the consequent σ · k  v Λ  temperature-dependent renormalization of the Dirac velocity in ===⇒ log D . (B.9) Eq. 4. Λ→∞ 2 2kBT Interacting Dirac Materials 13

Hence, we obtain the self-energy correction (see Eq. 4) (a) (b) 2   as Σ(k) = e log vD Λ σ · k. 20 2kB T To provide a pedagogical description, we finally provide a brief outline of the derivation leading to the well-known logarithmic renormalization of the Dirac velocity at zero temperature, as in Eq. 3. In this case, we compute the Hartree-Fock correction to the quasiparticle energy as Figure D1. (a) The honeycomb lattice for the Hubbard model. Red (blue) symbols signify the inequivalent lattice Z e2 1 k · k0  δE ≈ ± dk0 ± , (B.10) sites A (B), with δi’s being its three nearest-neighbour vectors. k 0 0 (b) The second-order Feynman diagram for the local Hubbard 0|k − k | 2 |k||k | interaction, with α(β) representing the sub-lattice indices. where ± sign corresponds to the electrons and holes, respectively and the second term originates from the integrand in Eq. C.4 reduces to overlap of the eigenfunctions from the Hamiltonian Λ in Eq. A.5. Performing a similar large momentum Z k0 λdλ k Σ(k) ∝ k0σ · dθ(cos θ, sin θ) cos(γ − θ) expansion we find the divergent part of the integral λ k0 Λ in Eq. B.10 is proportional to ± log . Λ k Z k dλ = σ · (k cos γ, k sin γ) λ Appendix C. Hartree-Fock correction:  Λ  Screened Coulomb interaction ===⇒ σ · k log . (C.5) Λ→∞ k + qTF In this appendix, we outline the derivation of the Hence, we obtain the self-energy correction (see Eq. 7) renormalization that led to Eq. 7 and Eq.8 in Sec. 3.2. 2   as Σ(k) = e log Λ σ·k. Clearly, the finite range We first begin with the Fourier transformation of the 0 k+qTF of the screened Coulomb interaction (signaled by non- screened Coulomb interaction VSC (r) (see Eq. 6) as zero qTF ) leads to the disappearance of the divergence e2 1 as in Eq. 3. VSC (q) = . (C.1) 0 q + qTF A similar analysis for the Yukawa potential, would then generate an analogous cutoff in the Following the similar Hubbard-Stratonovich transfor- 2   renormalization as Σ(k) = e log √ Λ σ · k. mation as in Appendix Appendix B, we obtain the zero 0 2 2 k +qTF temperature Green’s function for auxiliary field φr as Appendix D. Fermionic second-order e2 1 D(q, ) = . (C.2) perturbation theory: Hubbard interaction 0 q + qTF In this part, we outline the detailed analysis of the Performing a Hartree-Fock analysis (replace the φ- second-order perturbation theory for the half-filled propagator in Fig. B1 by Eq. C.2, we obtain the lowest- Hubbard model on a honeycomb lattice. The tight- order (in the dimensionless coupling α) self-energy binding Hamiltonian of Eq. 9 can be written in correction (in the unit of e2/ ) as 0 momentum-space as Z (cos θ, sin θ) X   Σ(k) ∝ σ · qdqdθ , (C.3) H = −t γ a† b + γ∗b† a , (D.1) |q − k| + q 0 k k k k k k TF k where we wrote the exchange wavevector in polar † † where ak(bk) represents the fermionic creations coordinate as q = q(cos θ, sin θ). Choosing γ − θ as P ik·δi operator on sublattice A(B), with γk = δ e . the angle between the two wavevectors k and q, we i Here δi are the three nearest-neighbour vectors on the rewrite Eq. C.3 as iφk lattice as shown in Fig. D1a. Rewriting γk = |γk|e , we obtain the Green’s function in sublattice basis as 0 R λdλ(cos θ,sin θ)dθ Σ(k) ∝ k σ · q 2 , (C.4) λ2−2 kλ cos(γ−θ)+ k + qTF " # k0 k02 k0 1 1 1 G (k, iω ) = G (k, iω ) = + 11 n 22 n 2 iω + ε iω − ε where λ = q/k0 as in the previous case with k0 = n k n k " φk φk # k + qTF . Similarly, for large q → Λ (where Λ ∼ 2π/a φ −i k cos 2 i sin 2 is the ultraviolet cutoff for the lattice-momentum), the G12(k, iωn) = e 2 + iωn + εk iωn − εk (D.2) Interacting Dirac Materials 14 where Gαβ (α, β = 1, 2 label the two sublattices Eq. D.3 has been rewritten as A, B) corresponds to the Green’s function between ν0ν00βν sublattices α and β with the dispersion written as Vk0,k00,k,k000 = Uδν0ν00 δν0βδβν . (D.4) εk = |γk| (in unit of the tight-binding parameter t). ∗ G21(k, iωn) = G12(k, iωn → −iωn) . The self-energy Using Eq. D.3 and Eq. D.4, the second order self-energy (upto second order in U) in sublattice basis then follows Σαβ is obtained (see Fig. D1b for the corresponding from the Hubbard interaction as Feynman diagram) as

2 X 2 Σαβ = U T Gαβ(k + q)χαβ(q) (D.5) UT X 0 UT X Σ (k, iω ) = δ G (k ) − q ↑,αβ n αβ N ↓,αα N 2 k0 k0,k00 X χαβ(q) = −T Gαβ(p)Gβα(p + q), (D.6) ν0ν00βν 0 0 00 000 p G↑,αν0 (k )G↑,αν0 (k )G↓αν00 (k ) · G↓,να(k )Vk0,k00,k,k000 , (D.3) where the spin-index has been suppressed. The po- larization function χαβ(k) corresponds to the bubble where the momentum conservation at the vertices shown in Fig. D1b. Using the sublattice representa- implies k000 = k0 +k00 −k. Here, we used the short-hand tions of the Green’s function in Eq. D.2, we obtain the notation as k = (k, iωn). The Hubbard interaction in four components of the polarization function χαβ as

" # 1 X f(−εp) − f(−εp0 ) f(−εp) − f(εp0 ) f(εp) − f(−εp0 ) f(εp) − f(εp0 ) χ11(q) = χ22(q) = + + + , (D.7) 4 i + ε 0 − ε i − ε 0 − ε i + ε 0 + ε i + ε 0 + ε p m p p m p p m p p m p p " X φp φp0 f(−εp) − f(−εp0 ) φp φp0 f(εp) − f(εp0 ) χ12(q) = cos cos + sin sin − 2 2 i + ε 0 − ε 2 2 i − ε 0 + ε p m p p m p p # φ φ 0 f(−ε ) − f(ε 0 ) φ φ 0 f(ε ) − f(−ε 0 ) i cos p sin p p p + i sin p cos p p p , (D.8) 2 2 im − εp0 − εp 2 2 im + εp0 + εp " X φp φp0 f(−εp) − f(−εp0 ) φp φp0 f(εp) − f(εp0 ) χ21(q) = cos cos + sin sin − 2 2 i + ε 0 − ε 2 2 i − ε 0 + ε p m p p m p p # φ φ 0 f(ε ) − f(−ε 0 ) φ φ 0 f(−ε ) − f(ε 0 ) i sin p cos p p p + i cos p sin p p p . (D.9) 2 2 im + εp0 + εp 2 2 im − εp0 − εp

Now, we compute the self-energy, defined in forming the summation over the relevant Matsub- Eq. D.5, in the sublattice basis with the 2 × ara frequencies we obtain the second order self- 2- matrix-valued polarization function. After per- energy Σαβ(k, iωn) in the sublattice basis as

2 " U X f(εp0 )f(−εp)f(−εk0 ) + f(−εp0 )f(εp)f(εk0 ) Σ11(k) = Σ22(k) = + 8 iω + ε 0 − ε 0 + ε p,q n k p p

f(−ε 0 )f(−ε )f(−ε 0 ) + f(ε 0 )f(ε )f(ε 0 ) p p k p p k + iωn + εk0 + εp0 + εp

f(ε 0 )f(ε )f(−ε 0 ) + f(−ε 0 )f(−ε )f(ε 0 ) f(−ε 0 )f(ε )f(−ε 0 ) + f(ε 0 )f(−ε )f(ε 0 ) p p k p p k + p p k p p k + iωn + εk0 − εp0 − εp iωn + εk0 + εp0 − εp

f(ε 0 )f(ε )f(ε 0 ) + f(−ε 0 )f(ε )f(−ε 0 ) f(−ε 0 )f(−ε )f(ε 0 ) + f(ε 0 )f(ε )f(−ε 0 ) p p k p p k + p p k p p k + iωn − εk0 − εp0 + εp iωn − εk0 + εp0 + εp # f(ε 0 )f(ε )f(ε 0 ) + f(−ε 0 )f(−ε )f(−ε 0 ) f(−ε 0 )f(ε )f(ε 0 ) + f(ε 0 )f(−ε )f(−ε 0 ) p p k p p k + p p k p p k , (D.10) iωn − εk0 − εp0 − εp iωn − εk0 + εp0 − εp Interacting Dirac Materials 15

where Σ11(k) and Σ22(k) are the identical diagonal entries are even longer and are provided as follows entries of the 2×2-self-energy matrix. The off-diagonal

" i φ 0 φ φ 0 f(ε 0 )f(−ε )f(−ε 0 ) + f(−ε 0 )f(ε )f(ε 0 ) 2 X (φ 0 −φp−φ 0 ) k p p p p k p p k Σ12(k) = U e 2 p k cos cos cos + 2 2 2 iω + ε 0 − ε 0 + ε p,q n k p p

φ 0 φ φ 0 f(−ε 0 )f(ε )f(−ε 0 ) + f(ε 0 )f(−ε )f(ε 0 ) φ 0 φ φ 0 cos k sin p sin p p p k p p k + sin k cos p sin p × 2 2 2 iωn + εk0 + εp0 − εp 2 2 2

f(−ε 0 )f(−ε )f(ε 0 ) + f(ε 0 )f(ε )f(−ε 0 ) φ 0 φ φ 0 p p k p p k − sin k sin p cos p × iωn − εk0 + εp0 + εp 2 2 2

f(ε 0 )f(ε )f(ε 0 ) + f(−ε 0 )f(−ε )f(−ε 0 ) φ 0 φ φ 0 p p k p p k − i cos k cos p sin p × iωn − εk0 − εp0 − εp 2 2 2

f(−ε 0 )f(−ε )f(−ε 0 ) + f(ε 0 )f(ε )f(ε 0 ) φ 0 φ φ 0 p p k p p k + i cos k sin p cos p × iωn + εk0 + εp0 + εp 2 2 2

f(ε 0 )f(ε )f(−ε 0 ) + f(−ε 0 )f(−ε )f(ε 0 ) φ 0 φ φ 0 p p k p p k + i sin k cos p cos p × iωn + εk0 − εp0 − εp 2 2 2

f(ε 0 )f(−ε )f(ε 0 ) + f(−ε 0 )f(ε )f(−ε 0 ) p p k p p k + iωn − εk0 − εp0 + εp # φ 0 φ φ 0 f(−ε 0 )f(ε )f(ε 0 ) + f(ε 0 )f(−ε )f(−ε 0 ) i sin k sin p sin p p p k p p k , (D.11) 2 2 2 iωn − εk0 + εp0 − εp " i φ 0 φ φ 0 f(ε 0 )f(−ε )f(−ε 0 ) + f(−ε 0 )f(ε )f(ε 0 ) 2 X (φp+φ 0 −φ 0 ) k p p p p k p p k Σ21(k) = U e 2 k p cos cos cos + 2 2 2 iω + ε 0 − ε 0 + ε p,q n k p p

φ 0 φ φ 0 f(−ε 0 )f(ε )f(−ε 0 ) + f(ε 0 )f(−ε )f(ε 0 ) φ 0 φ φ 0 cos k sin p sin p p p k p p k − sin k sin p cos p × 2 2 2 iωn + εk0 + εp0 − εp 2 2 2

f(ε 0 )f(ε )f(ε 0 ) + f(−ε 0 )f(−ε )f(−ε 0 ) φ 0 φ φ 0 p p k p p k + sin k cos p sin p × iωn − εk0 − εp0 − εp 2 2 2

f(−ε 0 )f(−ε )f(ε 0 ) + f(ε 0 )f(ε )f(−ε 0 ) φ 0 φ φ 0 p p k p p k − i cos k sin p cos p × iωn − εk0 + εp0 + εp 2 2 2

f(ε 0 )f(ε )f(−ε 0 ) + f(−ε 0 )f(−ε )f(ε 0 ) φ 0 φ φ 0 p p k p p k + i cos k cos p sin p × iωn + εk0 − εp0 − εp 2 2 2

f(−ε 0 )f(−ε )f(−ε 0 ) + f(ε 0 )f(ε )f(ε 0 ) φ 0 φ φ 0 p p k p p k − i sin k cos p cos p × iωn + εk0 + εp0 + εp 2 2 2

f(ε 0 )f(−ε )f(ε 0 ) + f(−ε 0 )f(ε )f(−ε 0 ) p p k p p k − iωn − εk0 − εp0 + εp # φ 0 φ φ 0 f(−ε 0 )f(ε )f(ε 0 ) + f(ε 0 )f(−ε )f(−ε 0 ) i sin k sin p sin p p p k p p k . (D.12) 2 2 2 iωn − εk0 + εp0 − εp

We have utilized the following identity where Einstein) distributions, f(x)(b(x)) corresponds to the Fermi-Dirac (Bose-

h ih i f(y) − f(x) b(x − y) + f(z) = f(−x)f(y)f(z) + f(x)f(−y)f(−z). (D.13)

These long expressions for the self-energy in Eqs. D.10 diagonal entries as simplify at zero temperature. We therefore obtain the

" # U 2 X 1 1 Σ(k, ω) = + (D.14) 8 ω − ε 0 − ε 0 − ε + iδ ω + ε 0 + ε 0 + ε + iδ p,q k p p k p p Interacting Dirac Materials 16

where we relabeled the identical diagonal entries ical continuation to real frequency as iωn → ω + iδ. Σ11(k) and Σ22(k) as Σ(k) and performed the analyt- The off-diagonal entries are similarly simplied as

" i 2 (φp0 −φp−φk0 ) 2 X φk0 φp φp0 e Σ12(k, ω) = −U sin sin cos + 2 2 2 ω − ε 0 − ε 0 − ε + iδ p,q k p p

i (φ 0 −φ −φ 0 ) # φ 0 φ φ 0 e 2 p p k i cos k cos p sin p 2 2 2 ω + εk0 + εp0 + εp + iδ

" i 2 (φp+φk0 −φp0 ) 2 X φk0 φp φp0 e Σ21(k, ω) = −U sin sin cos − 2 2 2 ω − ε 0 − ε 0 − ε + iδ p,q k p p

i (φ +φ 0 −φ 0 ) # φ 0 φ φ 0 e 2 p k p i cos k cos p sin p , (D.15) 2 2 2 ω + εk0 + εp0 + εp + iδ

Hamiltonian in the diagonal basis we obtain

Now, we focus on the spectral function for this X h u † d † i effective two-band system as H = εkukuk + εkdkdk + Hint, (E.1) k X  † † † † H = V a a a 0 a + b b b 0 b A(k, ω) = −Im Tr G(k, ω), (D.16) int q k−q k0+q k k k−q k0+q k k q,k,k0 † † i + a b 0 bk0 ak , where the matrix Green’s function is defined as k−q k +q (E.2)   ω − Σ(k, ω) γ − Σ (k, ω) † † G−1(k, ω) = k 12 . where a (b ) is the creation operator for bosons on γ∗ − Σ (k, ω) ω − Σ(k, ω) k k k 21 sublattice A(B) and u† (d† ) corresponds to bosonic (D.17) k k operators in the diagonal basis with the dispersion (u/d) εk = t(3 ± |γk|). Here, γk is identical to the one e∗2 1 discussed in the previous Appendix and Vq = . 0 |q| Appendix E. Bosonic second-order The diagonal and sub-lattice operators are related to perturbation theory: Coulomb interaction each other via

i φk (uk + dk) −i φk (−uk + dk) ak = e 2 √ , bk = e 2 √ , (E.3) In this part, we examine the Hamiltonian (see 2 2 iφ Eq. 16) for the Dirac bosons interacting via long- where γk = |γk|e k . Next, we rewrite the in- range Coulomb interaction. Rewriting the bosonic teracting Hamiltonian in the diagonal basis as " X † † † † † † Hint = Vq A(Φ, Ψ)uk−quk0+quk0 uk + i sin Φuk−quk0+quk0 dk + A(Φ, Ψ)uk−quk0+qdk0 dk q,k,k0 # † † † † † † † † + i sin Φuk−qdk0+quk0 uk + 2 cos Φuk−qdk0+quk0 dk + i sin Φuk−qdk0+qdk0 dk + A(Φ, Ψ)dk−qdk0+qdk0 dk + h.c.

(E.4)

where the two angles are defined as Φ = (φk + φk0 − consideration. We compute the correction to the bare φk−q − φk0+q)/2 and Ψ = (φk0+q + φk − φk−q − Dirac spectrum within a second-order perturbation 1 1  φk0 )/2 and A(Φ, Ψ) = 2 cos Φ − 4 cos Ψ . However, theory in the strength of the Coulomb interaction not all of these scattering channels are relevant in following the Feynman diagram (see Fig.E1) and the low temperature window due to the kinematic Interacting Dirac Materials 17

Appendix F. Interacting Dirac bosons: Analysis

In this appendix, we analyze the momentum- dependence of the self-energy correction (see Eq. 20) in detail. First of all, the bare Coulomb interaction for a charged particle always is screened. Hence, we incorporate a similar modification (see Sec. 3.2) to the S e∗2 1 potential Vq as V = . Of course, the phase q 0 q+qTF factors that emerge as a result of the change of basis Figure E1. The self-energy diagram in the second-order perturbation theory for the Dirac bosons interacting via the long- from the sublattice to the diagonal basis as in Eq. E.4, range Coulomb interaction. remain the same. The scattering rates for the relevant three channels are shown in Fig. F1a,b (solid lines). Note that there are two channels for the scattering of obtain the former as “up” bosons compared to only one channel for “down” 2 0 X |Vq,k,k0 | F (k; k , q) bosons. The scattering rates for the corresponding two Σ(ω, k) = ; (E.5) channels of the “up” bosons are added and illustrated ω + εk0 − εk−q − εk0+q + iδ k0,q in Fig. F1a. However, the same prescription for the self-energy where V 0 is the relevant scattering strength for the q,k,k (Eq. 20) leads to a negative correction for the down- particular channel following Eq.E.4 and F (k; k0, q) is band. This feature dictates that the renormalized the corresponding thermodynamic factor. The later energy for the Goldstone modes (down bosons near is obtained after performing the summation over the the Γ point) becomes negative, which is unphysical. Matsubara frequency as [22] At this point, we notice that our starting bosonic 0 h ih i tight-binding model (see Eq. 16) corresponds to the F (k; k , q) = 1 + b(ε ) 1 + b(ε 0 ) b(ε 0 )− k−q k +q k collective excitations of some underlying particles such h i as spins that give rise to a similar magnon model. b(εk−q)b(εk0+q) 1 + b(εk0 ) , (E.6) Hence, the matrix elements in Eq. E.4 are an effective description and are not the full theory. We believe that where b(x) corresponds to Bose-Einstein distribution the emergence of this negative bosonic self-energy is an function. On a closer inspection, we observe that artifact of such an incomplete model. the first term in the above expression corresponds to To cure such an unphysical result, we focus on the direct scattering process, i.e. where the original understanding the effect of the phase-dependent matrix boson with momentum k and the thermal boson with elements. The relevant scattering rates for the up and momentum k0 scatter into bosons with momenta k − q down band (both with and without the corresponding and k0 + q, whereas the second term to the reverse phase factors in the matrix elements in Eq.‘E.4 are scattering process. For very low temperature T  shown in Fig. F1. The results are plotted only along t, the direct scattering process dominates over the the Γ → K, instead of the full Brillouin zone to reverse one, and F (k; k0, q) can be approximated as 0 d illustrate the renormalization around the Dirac point. F (k; k , q) ≈ b(ε 0 ). Following this approximation, the k The solid lines are the results of the exact matrix on- shell self-energy further simplifies as elements derived in Eq. E.4, whereas the dashed lines 2 d X |Vq,k,k0 | b(ε 0 ) Σ(k) =∼ k . (E.7) εk − εk−q − εq + iδ k0,q

At the same time, only three scattering processes among the eight different scattering channels (see Eq.E.4) are relevant at low temperature due to kinematic considerations. These processes are labeled (a) (b) as u → u + d, u → d + d and d → d + d, where an “up” (“down”) boson effectively scatters with a “ghost-down” Figure F1. The scattering rates (from Eq. 20 with the screened boson (thermally excited bosons around the bottom of Coulomb interaction) for the “up” (a) and “down” (b) bosons in ∗4 2 2 the bosonic band-structure as illustrated in Fig.1a) into the unit of e /0t as a function of the momenta along the Γ → an “up” (“down”) and “down” bosons. K point in the Brillouin zone at temperature T = 0.1t. Dashed lines – without the phase-dependent matrix elements; Solid lines – with the phase-dependent matrix elements. Inset – Magnified illustration of the solid line in panel (b). Interacting Dirac Materials 18

A A 2005 Nature 438 197–200 ISSN 1476-4687 [2] Geim A K and Novoselov K S 2007 Nat. Mater. 6 183–191 ISSN 1476-1122 [3] Castro Neto A H, Guinea F, Peres N M R, Novoselov K S and Geim A K 2009 Rev. Mod. Phys. 81(1) 109–162 [4] Wehling T, Black-Schaffer A and Balatsky A 2014 Adv. Phys. 63 1–76 [5] Vafek O and Vishwanath A 2014 Annu. Rev. Condens. Matter Phys. 5 83–112 [6] Wang J, Deng S, Liu Z and Liu Z 2015 Natl. Sci. Rev. 2 22–39 [7] Volovik G E 1992 Exotic Properties of Superfluid 3He (WORLD SCIENTIFIC) [8] Hasan M Z and Kane C L 2010 Rev. Mod. Phys. 82(4) 3045–3067 Figure F2. The real part of the self-energy (from Eq. 20 [9] Qi X L and Zhang S C 2011 Rev. Mod. Phys. 83(4) 1057– with the screened Coulomb interaction as in Eq. F.1) for the 1110 “up” (solid line) and “down” (dashed line) bosons in the unit of [10] Chen Y L, Liu Z K, Analytis J G, Chu J H, Zhang H J, ∗4 2 2 e /0t as a function of the momenta along the Γ → K point Yan B H, Mo S K, Moore R G, Lu D H, Fisher I R, in the Brillouin zone at temperature T = 0.1t. Zhang S C, Hussain Z and Shen Z X 2010 Phys. Rev. Lett. 105(26) 266401 [11] Yang S A, Pan H and Zhang F 2014 Phys. Rev. Lett. 113(4) are the results without the phase factors. Please note 046401 that the screened Coulomb interaction is not modified [12] Abergel D, Apalkov V, Berashevich J, Ziegler K and Chakraborty T 2010 Adv. Phys. 59 261–482 in both the cases, only the phase factors are dropped [13] Gomes K K, Mar W, Ko W, Guinea F and Manoharan H C in the case for Fig. F1b. Apart from the overall 2012 Nature 483 306–310 ISSN 0028-0836 magnitude, the scattering rates are almost identical [14] Polini M, Guinea F, Lewenstein M, Manoharan H C and (Note: There is an additional kink in Fig. F1a for the Pellegrini V 2013 Nat. Nano 8 625–633 ISSN 1748-3387 [15] Young S M and Kane C L 2015 Phys. Rev. Lett. 115(12) up-band). This feature dictates that the phase factors 126803 do not play that crucial role in the specific momentum- [16] Katsnelson M I, Novoselov K S and Geim A K 2006 Nat. dependence of the scattering rates. Motivated by this Phys. 2 620–625 ISSN 1745-2473 fact, we modify the phase-dependent part of all the [17] Banerjee S, Fransson J, Black-Schaffer A M, Ågren H and Balatsky A V 2016 Phys. Rev. B 93(13) 134502 matrix elements in Eq. E.4 as [18] Fransson J, Black-Schaffer A M and Balatsky A V 2016 Phys. Rev. B 94(7) 075401 ∗2     e 1 h |k| ih |q| i [19] Haldane F D M and Raghu S 2008 Phys. Rev. Lett. 100(1) Vq,k = 1 − exp 1 − exp · 013904 0 q + qTF λ λ [20] Dubois M, Shi C, Zhu X, Wang Y and Zhang X 2017 Nat. h |k − q| i 1 − exp , (F.1) Commun. 8 14871 λ [21] Fei Z, Andreev G O, Bao W, Zhang L M, S McLeod A, Wang C, Stewart M K, Zhao Z, Dominguez G, Thiemens where λ is a constant. Now, we substitute this in Eq. 20 M, Fogler M M, Tauber M J, Castro-Neto A H, Lau C N, to compute the renormalization of the band structure. Keilmann F and Basov D N 2011 Nano Lett. 11 4701– 4705 Without aiming for a numerical estimate, we focus here [22] Pershoguba S S, Banerjee S, Lashley J C, Park J, Ågren on the overall trend around the Dirac crossing point. H, Aeppli G and Balatsky A V 2018 Phys. Rev. X 8(1) The real part of the self-energy for the two 011010 [23] González J, Guinea F and Vozmediano M 1994 Nucl. Phys. bands are consequently computed and illustrated in B 424 595 – 618 ISSN 0550-3213 Fig. F2. Once again we stress that we do not [24] Das Sarma S, Hwang E H and Tse W K 2007 Phys. Rev. B aim for any numerical estimate from our model but 75(12) 121406 only focus on evaluating the overall trend of the [25] Ni X, Purtseladze D, Smirnova D A, Slobozhanyuk A, Alù A and Khanikaev A B 2018 Sci. Adv. 4 energy renormalization around the Dirac crossing [26] Wen X, Qiu C, Qi Y, Ye L, Ke M, Zhang F and Liu Z 2019 point. Fig. F2 clearly dictates that the Dirac velocity Nat. Phys. 15 352–356 ISSN 1745-2481 increases for the Coulomb interaction. At the same [27] Barnes E, Hwang E H, Throckmorton R E and Das Sarma time, we observe that the weak Coulomb interaction S 2014 Phys. Rev. B 89(23) 235431 [28] Elias D C, Gorbachev R V, Mayorov A S, Morozov S V, does not open any gap at the Dirac point. This result is Zhukov A A, Blake P, Ponomarenko L A, Grigorieva I V, directly comparable to its fermionic counterpart, where Novoselov K S, Guinea F and Geim A K 2011 Nat. Phys. we also observed an increased Dirac velocity and no gap 7 701–704 ISSN 1745-2473 opening. [29] Sheehy D E and Schmalian J 2007 Phys. Rev. Lett. 99(22) 226803 [30] Giuliani G and Vignale G 2005 Quantum Theory of the References Electron Liquid (Cambridge University Press) [31] Katsnelson M I 2006 Phys. Rev. B 74(20) 201401 [1] Novoselov K S, Geim A K, Morozov S V, Jiang D, [32] Kotov V N, Uchoa B, Pereira V M, Guinea F and Katsnelson M I, Grigorieva I V, Dubonos S V and Firsov Castro Neto A H 2012 Rev. Mod. Phys. 84(3) 1067–1125 Interacting Dirac Materials 19

[33] Parisen Toldin F, Hohenadler M, Assaad F F and Herbut I F 2015 Phys. Rev. B 91(16) 165108 [34] Otsuka Y, Yunoki S and Sorella S 2016 Phys. Rev. X 6(1) 011029 [35] Sorella S, Otsuka Y and Yunoki S 2012 Sci. Rep. 2 992 [36] Tang H K, Leaw J N, Rodrigues J N B, Herbut I F, Sengupta P, Assaad F F and Adam S 2018 Science 361 570–574 ISSN 0036-8075 [37] Herbut I F 2006 Phys. Rev. Lett. 97(14) 146401 [38] Wu W and Tremblay A M S 2014 Phys. Rev. B 89(20) 205128 [39] Heeger A J, Kivelson S, Schrieffer J R and Su W P 1988 Rev. Mod. Phys. 60(3) 781–850 [40] Downing C A and Weick G 2017 Phys. Rev. B 95(12) 125426 [41] Pirmoradian F, Zare Rameshti B, Miri M and Saeidian S 2018 Phys. Rev. B 98(22) 224409 [42] Downing C A, Sturges T J, Weick G, Stobińska M and Martín-Moreno L 2019 Phys. Rev. Lett. 123(21) 217401 [43] Fradkin E and Hirsch J E 1983 Phys. Rev. B 27(3) 1680– 1697 [44] Throckmorton R E, Hofmann J, Barnes E and Das Sarma S 2015 Phys. Rev. B 92(11) 115101 [45] Jian S K and Yao H 2017 Phys. Rev. B 96(15) 155112 [46] Ghosal A, Goswami P and Chakravarty S 2007 Phys. Rev. B 75(11) 115123 [47] Ando T 2007 Physica E Low Dimens. Syst. Nanostruct. 40 213 – 227 ISSN 1386-9477