Excitonic instabilities and spontaneous time-reversal symmetry breaking on the Honeycomb lattice

Wei Liu Physics Department, City College of the City University of New York, New York, NY 10031, USA

Alexander Punnoose∗ Instituto de F´ısica Te´orica - Universidade Estadual Paulista, R. Dr. Bento Teobaldo Ferraz 271, Barra Funda, S˜aoPaulo - SP, 01140-070, Brazil (Dated: October 12, 2018) We elucidate the close relationship between spontaneous time-reversal symmetry breaking and the physics of excitonic instabilities in strongly correlated multiband systems. The underlying mecha- nism responsible for the spontaneous breaking of time-reversal symmetry in a many-body system is closely related to the Cooper-like pairing instability of interband particle-hole pairs involving higher order symmetries. Studies of such pairing instabilites have, however, mainly focused on the mean-field aspects of the virtual exciton condensate, which ignores the presence of the underlying collective Fermi liquid excitations. We show that this relationship can be exploited to systematically derive the coupling of the condensate order-parameter to the intraband Fermi liquid particle-hole excitations. Surprisingly, we find that the static susceptibility is negative in the ordered phase when the coupling to the Fermi liquid collective excitations are included, suggesting that a uniform condensate of virtual excitons, with or without time reversal breaking, is an unstable phase at T = 0.

PACS numbers: 05.30.Fk, 11.30.Er, 71.27.+a

I. INTRODUCTION

In multiband systems, the normal Fermi liquid (FL) state of a partially filled band becomes unstable when the 1 exciton (electron-hole pair) binding energy, EB, is tuned EB to become larger than the direct excitation threshold en- μ ED = 2| | EG ergy ED < |EB| (see Fig.1). When this happens vir- tual excitons form spontaneously in the groundstate and destabilizes the FL state. As a cure for this instability, it E was proposed many decades ago2,3 that a new Hartree- Fock groundstate may be constructed by hybridizing the bands. Such a state is characterized by a non-zero ex- † † † pectation value hakbki 6= 0, where ak/ak and bk/bk are the electron creation/annihilation operators in the rele- FIG. 1. A simple direct bandgap (EG) model with quadratic vant bands. As a result, the relative phase between the bands of equal masses is shown. The location of the exciton level E = E − E is marked by the dashed line. E is bands is locked to form a uniform (q = 0) condensate of D B B the exciton binding energy and ED is the direct interband interband particle-hole pairs. excitation threshold from the chemical potential µ (measured To distinguish condensates of interband particle-hole from the center of the gap). When EB = ED, the cost of pairs from quantum liquids that arise from instabilities in creating virtual exciton pairs become negligible. the intraband particle-hole channel found in single band systems, like the Pomeranchuk instability4, we refer to the former here as an “excitonic-liquid” (XL). the gapped excitonic-insulator state, the XL state has The XL is to be contrasted with the much studied only one species of free carriers, hence away from half- arXiv:1401.4150v1 [cond-mat.str-el] 16 Jan 2014 excitonic-insulator state formed when the number of free filling the chemical potential lies in the band leading to conduction electrons and valence holes are finite. The a metallic state with a sharp Fermi surface (at least in latter can occur either naturally when the bands cross at the Hartree-Fock approximation). This raises the inter- the Fermi level (semimetal with bandgap EG < 0) or can esting question of the stability of the XL in the presence be created artificially by optically pumping a semicon- of gapless fermions at the Fermi surface. ductor (EG > 0). The two carrier types have their own The issue of the coupling of bosonic modes in the or- chemical potentials which are tied together by thermody- dered phase to gapless fermions has received considerable namic considerations. The excitonic-insulator problem attention recently, particularly concerning the applicabil- has been shown to be mathematically equivalent to the ity of the Hertz theory10 to continuous quantum phase BCS (Bardeen-Cooper-Schrieffer) model with electron- transitions at T = 0. The issue is that integrating the hole pairs instead of electron-electron pairs5–9. Unlike fermions out completely has been shown to give rise to 2 a non-local (in time or frequency) bosonic theory that serving condensate remains in one-to-one correspondence is dominated by Landau damping, which in most cases with the FL states with well-defined quasiparticle exci- lead to an infinite number of marginal terms in the ef- tations, at the Hartree-Fock level the condensates away fective action11. (For a review of this and other related from half-filling are always metallic with a continuum of issues, see Ref. 12.) In all these studies, involving a sin- gapless fermionic excitations at the Fermi surface. gle band, it is usually implicitly assumed, to ensure the We show that the hybridization of the bands necessar- stability of the condensate, that the opposite limit, i.e., ily induces a coupling of the XL order-parameter to the the static limit, of the fluctuation of the order-parameter gapless fermions at the Fermi surface. In the following, 2 limq→0h|δφ(q, ω = 0)| i > 0 is positive. We find that we analyze the static susceptibility originating from this this assumption is generally violated in a multi-band sys- coupling in two different two-band models, both of which tem when the coupling of the order-parameter of the XL stabilizes an XL phase at the mean-field level. The first state to the collective intraband particle-hole excitations model is a simplified continuum degenerate semiconduc- is considered. We note that the issues originating from tor model in which the T symmetry is left intact in the the dynamics of the gapless fermions leading to the Lan- condensed phase, and the second model is a lattice model dau damping of the bosonic mode is irrelevant for us in that supports a T symmetry broken phase. establishing the stability of the condensed phase. There is considerable interest to understand if a uni- form XL can in principle exist. In fact, excitonic sin- II. THEORY OF THE XL STATE. gularities in multiband systems is recognized to be one of the very few known mechanisms capable of introduc- Our objective in this section is to analyze a simple ing singularities in the irreducible interactions leading to 13 model that undergoes a quantum phase transition to a the breakdown of FL theory . At the mean-field level, uniform XL state. We consider the well studied, con- the Hamiltonian couples linearly to the bilinear opera- † tinuum, two-band, spinless fermion model with a direct tors a bk via a vertex function A(k). The symmetries k bandgap EG > 0 shown in Fig.1. Keeping only the di- of A(k) are dictated by the lattice symmetries. As a rect intraband scattering of the electrons and holes, the † result, a non-zero expectation value A(k)hakbki= 6 0 par- model is described by the Hamiltonian tially breaks the point group symmetry. Furthermore, depending on the transformation properties of A(k) un- X † 1 X † † K = ψ ( τ − µ) ψ − V 0 a b b a 0 der the corresponding magnetic group, time-reversal (T ) k k 3 k Ld kk k+q k k0 k +q symmetry may or may not be spontaneously broken in k k0,k;q (1) the groundstate. Special interest in understanding the The 2 × 2 Pauli matrices τ ’s act on the internal band properties of electronic systems with broken T symme- i space; τ is the z-component. The spinor ψ† = (b† , a† ) try stem from its proposed relevance to high temperature 3 k k k creates electrons in the conduction (b) and the valence superconductivity. An intriguing proposal, put forward (a) bands. The bands are assumed to be parabolic with originally by Varma as a candidate for a Marginal Fermi equal masses  = E /2+k2/2m. The chemical potential liquid, posited a non-magentic, translationally invariant k G P † groundstate with spontaneously broken T invariance 14. µ constraints the electron number N = k ψkψk. Hole More generally, such groundstates are novel in that doping is assumed throughout, i.e., partially filled valence they can exhibit anomalous Hall and Kerr effects with- band and empty conduction band, hence µ < 0. All out magnetic fields. Various non-magnetic, translation- energies are measured from the center of the gap. ally invariant states with broken T symmetry that are For the interaction, only the dominant scattering chan- ubiquitous in Hartree-Fock approximation schemes have nel, the direct intraband scattering of the electrons and been classified and analyzed in Ref. 15. They were first holes, are retained. The Hamiltonian K has been used demonstrated by Haldane in a simple two-dimensional extensively as a minimal model to analyze the excitonic properties of optically pumped degenerate semiconduc- model of non-interacting electrons on the honeycomb lat- 9,18,19 tice with a periodic magnetic field arranged in such a way tors . We note that because optical pumping fixes P † † that the field averages to zero in each unit cell thus pre- the number of electrons and holes Nph = k bkbk +akak, serving the translational symmetry of the lattice 16. It the chemical potential, µph, in these systems appear 9,20 P † was later shown that interacting electrons on the honey- as Kph = H−µphNph = k ψk[(k−µph)τ3]ψk+Hint. comb lattice with next-nearest neighbor interactions can † A particle-hole transformation, ak → c−k↓ and bk → ck↑, form exactly the same state as proposed by Haldane that maps Kph to the continuum fermion model with at- spontaneously breaks T invariance in the groundstate 17. tractive interactions21,22 whose mean-field solution cor- Our main purpose in this paper is to analyze the sta- responds exactly to the BCS Hamiltonian. In contrast, bility of the mean-field solutions of particle-hole con- the same transformation maps K in Eq. (1) to the BCS densates both with and without T symmetry breaking. Hamiltonian in a Zeeman field with field hZ = µ and We first show that the XL’s are metallic away from fixed BCS chemical potential µBCS = −EG/2. When half-filling. Indeed, since the single-particle states in EG < 0 (semi-metal) the mean-field solution of K maps any mean-field approximation of a particle number con- exactly to the BCS Hamiltonian in a Zeeman field23,24. 3

Since EG > 0 when the XL state is formed, the particle- We show below, using a more elementary quantum hole transformation does not provide much further in- mechanical calculation, that the above instability is re- sights. However, we show that the instability is closely lated to the formation of Cooper-like pairs of interband related to the Cooper-like pairing of interband particle- particle-hole pairs with zero energy. This is done by writ- hole pairs at the Fermi surface. ing the equation of motion for a single particle-hole pair Further analytical progress is possible by restricting in the background of a ‘rigid’ Fermi sea of particles7. our analysis to the s-wave pairing state. We make We assume a linear combination of pair wavefunctions, the standard approximation of a separable screened po- ϕk, of single particle-hole pairs created by annihilating 18 0 tential , Vkk0 = V A(k)A(k ), where A(k) = 1 for an electron from inside of the Fermi surface, |FSi, of the | − k − µ| < Λ and zero otherwise (around the hole partially filled valence band and recreating it at the same Fermi surface). The cutoff is of the order Λ ∼ O(EG). k value in the empty conduction band This simplification allows the interaction to be written X00 † as a product of bilinear operators |φ0i = ϕkb ak|FSi (7) k k V X H = − Φˆ †(q)Φ(ˆ q) (2) The sum extends from k to the upper cutoff. int Ld F q The pair energy is obtained by solving the stationary Schr¨odingerequation ˆ P0 † + P0 † + where, Φ(q) = k ψkτ ψk+q = k bkak+q, and τ is X00 † X00 h † i the band index raising operator. The prime on the k- E ϕk b ak = ϕk K, b ak (8) k k k k sum explicitly enforces the restriction on A(k). Thus we   arrive at our simplified model Hamiltonian X00 V X0 † † † = 2kϕk − ϕk0 (a 0 ak0 − b 0 bk0 ) b ak (9) k Ld k0 k k k X V X K = ψ† [ τ − µ] ψ − Φˆ †(q)Φ(ˆ q) (3) k k 3 k Ld The Cooper model assumes a rigid Fermi sea, which is k q equivalent to decoupling the interaction terms by averag- † † − + ing over the Fermi sea hak0 ak0 − bk0 bk0 i = f(k0 ) − f(k0 ). One then obtains the Bethe-Goldstone equation: A. Fermi surface instability: Mean-field theory X00 Eϕk = 2kϕk − V ϕk0 (10) We explore the possibility of a uniform XL groundstate k0 with a finite expectation value V hΦ(ˆ q)i/Ld = φ δ . 0 q,0 Substituting φ = V P00 ϕ and inverting the above Such a state is anticipated from the behavior of the ef- 0 k k fective interaction, U(q) = V Γ(q), which in the ladder equation gives the self-consistent equation diagram approximation takes the general form X00 1 1 = −V (11) 1 k E − 2k Γ(q) = (4) 1 − V χ0(q) Setting E = 0 gives Eq. (6). Since E is measured from the chemical potential, when V > Vc, it becomes favorable It follows from the form of K in Eq. (3) that χ0 is the non-interacting interband susceptibility to create particle-hole pairs out of the Fermi sea thus destabilizing the normal groundstate. + − A possible cure for this instability is to assume a new 1 X0 f(k+q) − f(k ) χ0(q) = − (5) Hartree-Fock solution built of virtual excitons 2,3. This Ld k + − − k+q k is equivalent to assigning a non-zero expectation value ± to hΦ(ˆ q)i ∼ φ0δq,0. We note that the order-parameter is The energies  = ±k − µ are the upper and lower band θ k i τ3 energies measured from µ. Since we assume hole dop- not invariant under a global U(1) rotation e 2 ψk of the + relative phase between the two bands. However, Hint in ing, at T = 0 the Fermi function f(k ) = 0 for all k, and − Eq. (2) conserves the particle number in each band sepa- f(k ) = 1 for |k| > kF and zero otherwise. kF is the non- interacting Fermi wavevector that fixes the filling. Sub- rately and is therefore invariant. Hence the condensation spontaneously breaks the U(1) symmetry. The associ- stituting for the Fermi functions in (5), we get χ0(0) > 0 in the limit q = 0. Hence the uniform Γ(0) diverges at ated Goldstone mode is easily identified with the phase of the XL order-parameter. We note, however, that the a critical Vc satisfying 1 − Vcχ0(0) = 0, indicating a FL enhanced U(1) symmetry of Hint exists because terms instability for the states at the chemical potential. The † † † † critical V that determines the QCP equals such as a a bb and a a ab that do not conserve the par- c ticle number in each band are omitted in our model. It is 1 X00 1 therefore only an approximate symmetry in general and = (6) Vc k 2k when spontaneously broken will give rise to a massive pseudo-Goldstone mode25,26. For the sake of generality, [The sum P00 = L−d P0 . (Remember that the sin- k |k|>kF we suppress the gapless phase fluctuations in our simpli- gle prime denotes the upper cutoff imposed by A(k).)] fied model by assuming φ0 to be real. 4

∗ We proceed to look for a mean-field solution by sub- tegrated out to give Z = R D[φ∗]D[φ]e−Seff[φ ,φ], where stituting φ0 in Eq. (3). The mean-field Hamiltonian d L X 2 S [φ∗, φ] = −Tr ln G−1 + |φ(q)| (14) eff βV 0 d q X † L 2 KMF = ψk [kτ3 − µ − φ0τ1] ψk + φ0 (12) −1 0 k V G (k, k ) = (−in − µ + kτ3) δk,k0 1 1 − φ∗(k0 − k)τ+ − φ(k − k0)τ− (15) Since φ0 is assumed real, the linear terms in KMF that β β originate on decoupling the interaction terms are com- † 0 ˆ † ˆ The shorthand notations k ≡ (k, ikn) and k − k = q ≡ bined as φ0(Φ (0) + Φ(0)) = φ0ψkτ1ψk. [From Eq. (3), the q = 0 component Φ(0)ˆ = P0 ψ† τ+ψ = P0 b† a .] (q, iqm), where kn/qm are the odd/even Matsubara fre- k k k k k k quencies, are used throughout. The Fourier transform Diagonalizing K , we obtain ξ± = ±E − µ, where MF k k is defined as φ(x) = β−1 P L−d P e−iqmτ+iq·rφ(q). the energy E = p2 + φ2. When E > 0, the lev- m q k k 0 G Since only real φ(x) is considered, φ(q) = φ∗(−q). els never cross and all states up to the non-interacting The mean-field solution corresponds to the saddle Fermi wavevector kF are filled. This point is worth em- point φ(q) = βδm,0δq,0φ0. The magnitude of φ0 is ob- phasizing again, which is that because KMF commutes † † tained by minimizing the action δSeff[φ0]/δφ0 = 0. The with the individual number operatorn ˆk = akak + bkbk, saddle point condition at T = 0 generalizes Eq. (6) to every k-state up to the non-interacting kF remains oc- cupied as V is tuned through the QCP. It implies the 1 X00 1 Fermi wavevector k is not renormalized and a sharp = (16) F V k 2Ek Fermi surface exists in the condensed phase. √ The most suggestive way to write the XL wavefunction, The mean-field scaling φ0 ∝ V − Vc is recovered close to the QCP. Next, we investigate the stability of the |XLi, so that kF and the exciton nature are both appar- Q † saddle point solution by analyzing the Gaussian fluctu- ent is |XLi = k[uk + vkbkak]|FSi. The ‘vacuum’ is Q0 † ations around φ0. This is done by expanding the action the non-interacting Fermi surface |FSi = |k|>k ak|0i F to quadratic order in the deviation δφ(x) = φ(x) − φ0. corresponding to the partially filled valence band. De- To this end, we first separate the φ0 and the δφ contri- spite the formal similarity with the BCS wavefunction, it −1 −1 butions in G in Eq. (15) as G = G (1 − ΣG0), where can be shown explicitly that |XLi does not possess off- 0 7 diagonal long-ranged order (ODLRO) due to the sharp 0 (in + µ) + kτ3 − φ0τ1 G0(k, k ) = − δk,k0 (17) cutoff at kF . In standard notation, the hybridized states + − (in − ξk )(in − ξk ) are written as ck+ = ukbk −vkak and ck− = ukak +vkbk, 0 1 0 corresponding to ±Ek, respectively. The coefficients are Σ(k, k ) = δφ(k − k )τ1 (18) found by minimizing the free energy. Setting all the β phases to zero, which is justified when φ0 is real, we get The trace in Eq. (14) is then expanded in the stan- −1 −1 dard way using the formula Tr ln G = Tr ln G0 −     P 1 Tr(G Σ)n. The order n = 2 terms are collected 2 1 k 2 1 k n n 0 uk = 1 + , vk = 1 − (13) to derive the |δφ|2 corrections. The first order correc- 2 Ek 2 Ek tion, n = 1, vanishes since the saddle-point is an ex- tremum. We write the expansion to quadratic order of (2) d P −1 2 Seff as Seff = Seff[φ0] + L q[V Γ(q)] |δφ(q)| , where B. Order-parameter fluctuations Γ−1(q) = 1 − V χ(q) is the generalization of Eq. (4). The susceptibility χ(q) in the ordered phase equals

The existence of a sharp Fermi surface in the XL 1 X χ(q) = − Tr [G (k + q)τ G (k)τ ] (19) phase implies that gapless particle-hole excitations ex- 2βLd 0 1 0 1 ists at the Fermi surface, which can couple to the fluc- k tuations of the order-parameter of the XL. The fluctua- Note that the Green’s function G as defined in Eq. (17) tions about the mean-field solution are most easily calcu- 0 27,28 is a 2 × 2 matrix written in the original FL basis (b, a) lated using the standard functional integral method . and is therefore not diagonal due to the hybridization of Briefly, the partition function is written as an imagi- the bands in the XL phase. Also note that the τ1 in R ¯ − R β dτL nary time integral Z = D[ψ, ψ]e 0 , where the the trace originates from Σ ∼ δφτ (see Eq. (18)). The P ¯ 1 Lagrangian L = k ψk(τ)∂τ ψk(τ) + K. Substitut- diagrams corresponding to the various terms from the P ¯ ing for K from Eq. (3), we get L = k ψk(τ)(∂τ − expansion of the trace are shown in Fig.2. Compared V P ¯ µ + kτ3)ψk(τ) − Ld q Φ(q, τ)Φ(q, τ). Next, Hubbard- to the interband FL susceptibility χ0 (see Eq. (5)), there Stratonovich fields, φ(q, τ), are introduced to decouple are two new contributions in the XL phase that originate the quartic Φ(¯ q, τ)Φ(q, τ) term. The resulting action is from the off-diagonal terms of G0. They are shown in 2 quadratic in the fermionic fields and can therefore be in- the last line of Fig.2, they vanish as ∼ φ0. 5

k + q From the self-consistent equation derived in Eq. (16), the first term equals 1/V at the saddle-point and cancels the −1 τ1 τ1 constant in Γ (q) = 1 − V χ(q). The remaining contri- Tr[G0τ1 G0 τ1 ] q 2 2 butions of O(φ0, q ) to the bosonic propagator equal k −1 2 2 2 [V Γ(q)] = γ⊥(4φ0 − ω + c|q| /2m) − χk(q) (24) b b a a = [G0 ]11[G0 ]22 G G + [ 0 ]22[ 0 ]11 The occurrence of the pole ω2 = (2φ )2 + c|q|2/2m, a a b b 0 starting at ω = 2φ0, is identified with the collective excitations of the transverse field Ising model29,30. To elucidate this further, we rewrite the fermionic bilinears b a a b in the Hamiltonian K in Eq. (3) in terms of pseudospin + [G0 ]12[G0 ]12 + [G0 ]21[G0 ]21 31 † operators σi(k) = ψkτiψk. They obey the usual SU(2) a b b a 0 commutation relations [σi(k), σj(k )] = 2iijkσk(k)δk,k0 . The dynamics of the uniform state is described by the FIG. 2. The diagrammatic expansion of the trace in the def- q = 0 term of K. The reduced Hamiltonian, K0, equals inition of χ(q) in Eq. (19) is shown. The new off-diagonal diagrams (last line) exist only in the XL phase, they van- X V X − + 0 2 K0 = kσ3(k) − σ (k)σ (k ) (25) ish as φ0 close to the QCP. The interaction of the bosonic Ld propagator, corresponding to the amplitude fluctuations of k k,k0 the order-parameter, with the gapless fermions at the chemi- 2 (The constant µ is suppressed here, it plays a crucial cal potential generates a bosonic mass ∼ φ0. The sign of the mass is derived in the main text and is shown to be negative. role only when the scattering at the Fermi surface in- volving q 6= 0 are included.) As V increases, a transition occurs from the paramagnetic state in which all states In the mean-field basis ck,± defined in Eq. (13) (the XL point down (occupied valence band) to a correlated Ising basis) χ(q) separates into inter and intraband contribu- ferromangnetic state with spins pointing in the x − y tions, which we write as χ(q) = χ⊥(q) + χk(q). After plane (hybridized states). Our choice of a real order- summing over the internal energy sums, we get † parameter, hΦ(0)ˆ i = hΦˆ (0)i ∝ φ0, breaks the symmetry 1 0 h i P X in the x direction, i.e., khσ1(k)i 6= 0 (see Eq. (12)). χ⊥(q) = F⊥(k, q) Π+−(k, q) + Π−+(k, q) (20) 2Ld k The key point is that this Ising symmetry is lost when 1 X0 h i the finite q 6= 0 terms involving particle-hole scattering χk(q) = Fk(k, q) Π++(k, q) + Π−−(k, q) (21) 2Ld k at the Fermi surface are included. The scattering is rep- 2 resented by χk(q), which appears as a self-energy correc- The form factors F⊥(k, q) = (ukuk+q − vkvk+q) and 2 tion to the bosonic propagator in Eq. (24). We show next Fk(k, q) = (ukvk+q +vkuk+q) , where uk/vk are defined 2 in Eq. (13). The polarization functions that the static limit of χk(q) ∼ φ0 is positive (implying a negative mass contribution to Γ(q)) and therefore has s s0 f(ξk+q) − f(ξk ) the potential to destabilize the condensate. Note that at Π 0 (k, q) = (22) ss s s0 half-filling χk = 0, hence the condensate is always stable. iqm − (ξk+q − ξk ) First we show that the condensate is stable when χk 0 where s, s = +/− are the upper/lower band indices. is suppressed. This requires that the constants γ⊥ and Note that the polarization operator, Π−−, corresponds c > 0 are positive in Eq. (24). The integral for γ⊥ is to the intraband particle-hole bubble in the XL phase, ultraviolet (Λ) convergent in dimensions d < 6 and hence ± i.e., the energies ξk correspond to the XL bands. (At we set Λ → ∞ and restrict ourselves to d < 6. In the limit T = 0, the upper band is empty, hence Π++ = 0.) φ0, q = 0, only two parameters remain, the dimension d The hybridization of the bands in the XL phase cou- and the hole doping factor x = (1 − EG/2F ). d/2−1 ples the order-parameter to the gapless fermions at the Separating the factor ν() = νd( − EG/2) where d/2 Fermi surface. The intraband polarization Π−− repre- νd = (m/2π) /Γ(d/2), corresponding to the density of 2 sents this coupling, it vanishes as Fk ∼ φ0 close to the states, the constant γ⊥ can be written as QCP. Physically it generates an additional “mass” for the bosonic propagator. Depending on the sign of the mass, X00 1 1 d/2−3 (d) γ⊥ = 3 = νdF R⊥ (x) (26) the condensate may or may not be stable. To determine k (2k) 8 the sign, we expand both χ⊥(q) and χk(q) for small q and take the static limit (mass term) of χ(q). Restricting ourselves to d = 2, 3 and 4, we get Since EG > 0, the interband terms Π+−/−+ in χ⊥ have   n (2) (3) (4)o 1 π  3x 1 a regular expansion in q. After analytical continuation, R⊥ ,R⊥ ,R⊥ = , 1 + , (1 + x) (27) + 2 2 2 8 2 2 iqm = ω + i0 , we get at T = 0 to O(q , φ0) 00 X 1 2 2 2 Only the leading O(x) correction is shown in d = 3. We χ⊥(q) = − γ⊥(4φ0 − ω + c|q| /2m) (23) k 2Ek find that the constant c = 2EG/(4 − d) + O(x), which is 6 ultraviolet convergent only for d < 4. (The derivation of c is not shown here as it is not crucial to the discussion.) We therefore restrict our analysis to below d = 4. s Finally, the effect of the self-energy correction χk in ˆ s ˆ Eq. (24) is derived. From Eq. (13) it follows that the 01 00 2 2 2 form-factor Fk ∼ (ukvk) ∼ (φ0/2Ek) . Hence to φ0 order, it is sufficient to set φ0 = 0 in the polariza- 2 P00 2 tion function: χk(q) = φ0 k Π0(q)/2k, where Π0(q) =

Π−−(q)|φ0=0 is the standard Lindhard-type function. It has the well known FL singularities FIG. 3. The ordered states can be pictured as two separate current carrying loops running along the sides of the two tri- ∂f(ξ−)  −q · ∇ ξ−  angular sub-lattices composed of A (white vertices) and B 0 k k k (black vertices) sites that form the honeycomb lattice. (The Π (q) ≈ − − × − (28) ∂ξk ω − q · ∇kξk atoms at A and B are assumed identical.) The current loops break the T symmetry16. Since the total moment in a unit cell with different limiting values32 when ω, |q| → 0: the dy- is zero, they generate translationally invariant patterns when ω namic limit Π0 = limω→0 Π0(q = 0, ω) = 0, and the extended over the whole lattice. The pattern generated by q − − ˆ s† ˆ s† static limit Π0 = lim|q|→0 Π0(q, ω = 0) = −∂f(ξk )/∂ξk . hΦ01(q = 0)i has C6 symmetry, while hΦ00(q = 0)i has only Importantly, this FL singularity induces a singularity in a reduced C3v symmetry that breaks inversion symmetry. χk(q) which in turn induces a singularity in Γ(q). Keep- ing the same notation for the static and dynamic limits, q 2 we get χ = −γk(2φ0) , where k quadratic dispersions is unstable in d = 2. We extend 1 this analysis to study the stability of the XL phase in γ = ν d/2−3R(d)(x) (29) k 8 d F k an interacting system in which T symmetry is sponta- neously broken in the groundstate. To this end, we con- (d) d/2−1 sider the extended Hubbard model of spinless electrons The function Rk (x) = −x derives its form from the density of states at the chemical potential. on a honeycomb lattice with next-nearest-neighbor (nnn) repulsion. The Hamiltonian is defined as We now combine the two contributions γ⊥ + γk to de- termine the sign of the mass term. To this end, we write q the static limit Γ in Eq. (24) as X † † X A A B B H = −t (Ai Bj +Bi Aj)+V (ni nj +ni nj ) (32) 1 nn nnn [V Γq]−1 ≡ γ(2φ )2 = ν d/2−3R(d)(x)(2φ )2 (30) 0 8 d F 0 † † From Eqs. (26) and (29) we obtain the following expres- The operators Ai /Ai and Bi /Bi are the on-site electron sions for the leading x dependence for R(d) = R(d) + R(d) creation/annihilation operators on the respective sub- ⊥ k A/B in dimensions d = 2, 3 and 4 lattices and ni are the number operators. The spin de- grees of freedom are suppressed to avoid any spin-related n o  1 π √  1  T symmetry breaking effects33. R(2),R(3),R(4) = − , − x , (1 − x) (31) 2 8 2 The non-interacting spectrum with nearest-neighbor Note that the filling factor x < 1 away from half-filling. (nn) tunneling is a semi-metal with two inequivalent de- Hence for all d < 4 the function R(d)(x) becomes negative generacy points, ±kD, called Dirac points, located at the (and correspondingly γ becomes negative) beyond some corners of the Brillouin Zone. At the mean-field level, beyond a critical interaction strength Vc the interaction critical doping xc. In particular, xc = 0 in d = 2, and 2 lifts the degeneracy at the Dirac points and opens a gap xc = (π/8) ≈ 0.15 in d = 3 (the exact value is slightly 17 higher at ≈ 0.2 corresponding to a hole Fermi energy to stabilize a Topological Mott Insulator (TMI) phase at half-filling. Unlike the conventional Mott phase, the ∼ EG/8). A negative γ in Eq. (30) implies an instability of the XL phase (negative mass). TMI breaks both chirality (C) and T symmetries but is This is our main result in this section, namely, the invariant under the combined CT transformation. It is uniform XL obtained self-consistently at the mean-field therefore a type-II state according to the classification level is mostly unstable in d < 4 (small doping) and is in Ref. 15. (See Fig.3 for a physical description of these always unstable in d = 2. states.) The properties of the TMI phase are insensitive to the nn repulsion U when U/Vc  1. We therefore neglect U in our model. (Further details about the inter- III. XL PHASE WITH BROKEN T SYMMETRY play of U and V can be found in Ref. 17.) The methods developed in the last section are applied In the last section, we demonstrated that the uniform here to study the stability of the broken T phase away XL phase with s-wave pairing in a two-band system with from half-filling. We first write the Hamiltonian (32) in 7 k-space (we set t = 1 and use N for the number of sites) The bilinear operators are obtained by combining the Ak and Bk operators as ψk (see Eq. (34)) X −iθk † X iθk † H = − ke A Bk − ke B Ak k k i θ q τ − i θ q τ ˆ s X † 2 k+ 1 2 k− 1 k k Φ0j(q) = Skψ q e 2 τje 2 ψk− q (37) k+ 2 2 V X † † k q q − g(k − p)Ak+ q Ak− Ap− q Ap+ N 2 2 2 2 s k,p;q For notational simplicity, we define Sk ≡ A0(k). Note that our choice of the basis in Eq. (34) introduces phases V X † † q q − g(k − p)Bk+ q Bk− Bp− q Bp+ (33) in the interaction. The indices j = 0, 1 label the Pauli N 2 2 2 2 k,p;q matrices τ0 (identity) and τ1. Physically they correspond to adding or subtracting the two loop currents on each iθk P 17 The real part of the kinetic term ke = l exp(ik · dl) sublattice as described in Fig.3. Mean-field analysis of p P ˆ s† gives the energy k = 3 + 2 l cos(k · tl), it vanishes at the two patterns favors the condensation of hΦ01(q = 0)i. ˆ the Dirac points, i.e., ±kD = 0. The dl vectors connect We therefore neglect Φ00 and arrive at our minimal model the nearest neighbor atoms and tl’s are the basis vectors of the hexagonal Bravais lattice. (l = 1, 2 and 3). In the X † V X s† K = ψ [ τ − µ]ψ − Φˆ (q)Φˆ s (q) (38) following, we transform to the basis in which the kinetic T k k 3 k 2N 01 01 q term is diagonal, namely, k

The Hamiltonian KT is a generalization of K derived    i θ − i θ    bk 1 −e 2 k e 2 k Bk in Eq. (3) and the mean-field analysis and the fluctuation ψk = = √ i θ − i θ (34) ak 2 e 2 k e 2 k Ak calculations follow exactly as detailed in the last section. To avoid repetition we present only the main steps below. The various interaction induced symmetry breaking possibilities consistent with the lattice symmetries can be P A. Mean-field theory gleaned by expressing g(k − p) = l cos[(k − p) · tl] in terms of the distinct irreducible representations (irreps) ˆ s† ˆ s of the underlying lattice. One possible decomposition We note that since Φ01(q) = Φ01(−q), the order- V ˆ s involving separable irreps of the planar C6v symmetry parameter φT = N hΦ01(0)i is real. Hence, the mean-field group of the honeycomb lattice is shown below Hamiltonian reads

X † N 2 2 KT,MF = ψk [kτ3 − µ − φT Skτ1] ψk + φT (39) X s∗ s c∗ c 2V g(k − p) = Aν (k)Aν (p) + Aν (k)Aν (p) (35) k ν=0 ± It follows that the energies ξk = ±Ek − µ, where s/c p 2 2 2 Ek = k + φT Sk. Thus, a gap ∼ φT |SkD | opens at Here, A0 belong to the one-dimensional representations s/c s/c ±kD. However, since Sk is an odd function, the sign B1/A1, while (A1 , A2 ) form the basis for the two- of the gaps are in opposite directions at the two Dirac dimensional representations E1/E2. (See, e.g., Ref. 34 points. This is the origin of the anomalous quantum Hall for the notation used here.) Explicitly, the functions 16 effect as described by Haldane . A non-zero φT breaks s 1 i 2νπ i 4νπ A (k) = √ [sin(k·t )+e 3 sin(k·t )+e 3 sin(k·t )], 35 ν 3 1 2 3 T symmetry but does not break inversion symmetry . c and the Aν (k) functions have sines replaced by cosines. Minimizing the action we obtain the self-consistent 1 P0 2 We first note that in order to have the closed loop equation at T = 0 as 1/V = N BZ Sk/Ek. As be- configurations shown in Fig3, it is necessary that the fore, we consider hole doping, i.e., µ < 0. At T = 0, all order-parameter has a minimum of C3 symmetry. The states |k| > kF measured from ±kD are filled, leaving two-dimensional irreps E1 and E2 have characters -1 un- small hole pockets at the Dirac points. The lower cutoff der C3 and therefore can not form closed loops. Further- kF is denoted by the prime on the summation. Since it is more, since under a time-reversal operation an arbitrary a convergent integral no upper-cutoff is required and the function A(k) transforms as TA(k)T −1 = A∗(−k), only summation extends over the whole Brillouin Zone (BZ). s the B1 representation, or A0, breaks the T symmetry. Finally, as before we write the eigenfunctions in the c s (A0 does not break T symmetry.) A0 has a reduced C6 form ck+ = ukbk − vkak and ck− = ukak + vkbk, corre- symmetry and hence breaks the C6v symmetry to C6. sponding to ±Ek. After minimizing the free energy, we 2 2 In the ladder approximation, the interactions in the get for uk and vk the same relations as in Eq. (13) with s an additional phase for v given as (φ > 0 is assumed) A0 channel diverge near the QCP and hence the other k T channels can be neglected. This simplifies the interaction q S q part of the Hamiltonian (33) to just two terms 2 k 2 vk = vk = vk sign(Sk) (40) |Sk| V X H = − Φˆ s† (q)Φˆ s (q) + Φˆ s† (q)Φˆ s (q) (36) This additional phase does not appear in the calculation int 2N 01 01 00 00 q of the static susceptibility as shown below. 8

B. Order-parameter fluctuations models, both of which stabilizes an uniform XL phase at the mean-field level. Our results therefore suggest that Since only the stability of the condensate is under ques- a uniform condensate of virtual excitons, with or with- tion, we only examine the static limit of the susceptibil- out spontaneous time reversal symmetry breaking, is an ities below. We follow the same steps as in the previous unstable phase at T = 0. We arrive at this conclusion section. First, the matrices in Eqs. (17) and (18) are by analyzing the static limit of the effective interaction modified for the honeycomb lattice as: in the particle-hole channel and showing it to be nega- tive. It follows that a Ginzburg-Landau type description 15 0 (in + µ)τ0 + kτ3 − φT Skτ1 of the ordered phase is in general not possible. G0(k, k ) = − + − δk,k0 (41) Finally, a few remarks about the relevance of the higher (in − ξk )(in − ξk ) order terms in the expansion of the action Seff in Eq. (14). i 0 1 0 (θ −θ 0 )τ Σ(k, k ) = δφ(k − k )S e 2 k k 1 τ (42) In general, the bosonic action can be expanded as β k+ 1 ∞ Z 0 (2) X 2 2n−1 2n The momentum k+ = (k + k )/2. Secondly, the form Seff = Seff + (dωd q) b2n(δφ(q)) (44) factors in Eqs. (20) and (21) are modified to include the n=2 2 2 2 structure factor Sk as F⊥(k, q = 0) = Sk(uk − vk) and 2 2 2 The coefficients b2n were calculated in Ref. 11 and shown Fk(k, q = 0) = 4S u v . As noted below Eq. (40), only k k k to contain universal singular contributions that makes the squares of uk and vk appear in these expressions. 2 q −1 2 the coefficients anomalously large in the dynamic limit Finally, to O(φT ) we get [V Γ ] = γφT for the static limit of the interaction amplitude [equivalent to Eq. (30)] ω → 0 and q = 0, leading to the general conclusion that the Hertz theory is incomplete. (The correspond- (2) 0 4  − −  ing dynamical corrections in S is the familiar Landau 1 X Sk f(ξk ) ∂f(ξk ) eff γ = + (43) damping term.) In the opposite limit, i.e., the static N k 2  ∂ξ− k k k limit, no such anomalous contributions exists and the It includes contributions from the inter and intraband Hertz assumption that the vertices are local is restored, susceptibility, i.e., γ = γ + γ , as explained in Eqs. (26) allowing for a controlled expansion in powers of δφ. To ⊥ k establish the stability of the mean-field groundstate, it and (29). Since S4 is a sharply peaked function around k is necessary that the static susceptibilites for all possible k , the integrals can be evaluated by linearizing the spec- D perturbations of the groundstate are positive. For the trum around k as  ∼ α|k − k |, where α = 3/2 (in D k D order parameter fluctuations, this translates to the sign units of lattice spacing). The lower cut-off is from k and F of the coefficient of the (δφ)2 term in a finite field φ . hence there is no divergence away from half-filling and an 0 2 17 Close to the transition, we calculate the contributions to expansion in φT is possible (unlike at half-filling ). 2 4 order φ0, which are shown in Fig.2. Note that there are After linearizing, we get γk = −2ν0S /F . The neg- kF no vertex corrections in this order. ative sign originates from the derivative of the Fermi Our results suggest that XL condensates with a sharp function.√ The density of states around each kD equals Fermi surface tends to be unstable in d = 2. Of course, ν0 = 1/ 3π and the factor 2 accounts for the contri- any condensate can be stabilized if a gap opens, how- butions from ±kD. The interband contribution is eas- ever, the mechanism to open a gap in the XL condensate ily shown to satisfy γ⊥ < |γk|, implying that γ < 0 in is unclear at the moment 36. Another possible cure for Eq. (43). Hence, we conclude that the uniform T bro- this instability might be to assume a non-uniform (q 6= 0) ken state on the honeycomb lattice, which is stable at 17 mean-field state similar to the spin-bag models proposed half-filling , is unstable to infinitesimal doping. in the case of doped anti-ferromagnetism37,38. This pos- sibility is not analyzed in this paper.

IV. DISCUSSION V. ACKNOWLEGEMENTS We have shown that the mean-field solution for the in- terband particle-hole condensate with a sharp Fermi sur- We acknowledge insightful conversations with J. Bir- face, which we call an excitonic liquid (XL) in this paper, man (CCNY), R. Matheus (IFT), A. R. Rocha (IFT) and is unstable in the presence of the gapless fermions at the C. M. Varma (UCR). We would like to thank the CMB Fermi surface. The origin of the instability is closely re- and the Physics Departments at Tulane University for lated to the singularity of the FL polarization function their kind hospitality where part of this work was done. in Eq. (28). We demonstrate this destabilization in two Partial support was provided by PSC-CUNY Award 41.

∗ Corresponding Author: [email protected] 1 N. F. Mott, Philos. Mag. 6, 287 (1961). 9

2 R. S. Knox, “Theory of excitons,” in Solid State Physics 21 C. A. R. S´ade Melo, M. Randeria, and J. R. Engelbrecht, Suppl., Vol. 5, edited by F. Seitz and D. Turnbull (Aca- Phys. Rev. Lett. 71, 3202 (1993). demic Press, Inc., New York, 1963) p. 100. 22 V. M. Loktev, R. M. Quick, and S. G. Sharapov, Phys. 3 J. des Cloizeaux, J. Phys. Chem. Solids 26, 259 (1965). Rep. 349, 1 (2001). 4 I. Pomeranchuk, Sov. Phys. JETP 8, 361 (1958). 23 L. Balents and C. M. Varma, Phys. Rev. Lett. 84, 1264 5 L. V. Keldysh and Y. V. Kopaev, Sov. Phys. Solid State (2000). 6, 2219 (1965). 24 V. Barzykin and L. P. Gorkov, Phys. Rev. Lett. 84, 2207 6 A. N. Kozlov and L. A. Maksimov, Sov. Phys. JETP 21, (2000). 790 (1965). 25 Y. Nambu and G. Jona-Lasinio, Phys. Rev. 122, 345 7 D. Jerome, T. M. Rice, and W. Kohn, Phys. Rev. 158, (1961). 462 (1967). 26 Y. Nambu and G. Jona-Lasinio, Phys. Rev. 124, 246 8 B. I. Halperin and T. M. Rice, Rev. Mod. Phys. 40, 755 (1961). (1968). 27 V. N. Popov, Functional Integrals and Collective Excita- 9 C. Comte and P. Nozi`eres,J. Phys. (Paris) 43, 1069 (1982). tions (Cambridge Univ. Press, Cambridge, 1987). 10 J. A. Hertz, Phys. Rev. B 14, 1165 (1976). 28 H. Kleinert, Fortsch. Phys. 26, 565 (1978). 11 A. Abanov and A. Chubukov, Phys. Rev. Lett. 93, 255702 29 P. G. de Gennes, Solid State Commun. 1, 132 (1963). (2004). 30 R. Brout, K. A. M¨uller,and H. Thomas, Solid State Com- 12 H. v. L¨ohneysen, A. Rosch, M. Vojta, and P. W¨olfle,Rev. mun. 4, 507 (1966). Mod. Phys. 79, 1015 (2007). 31 H. J. Lipkin, Lie Groups for Pedestrians (Dover Publica- 13 C. M. Varma, Z. Nussinov, and W. van Saarloos, Phys. tions, New York, 2002). Rep. 361, 267 (2002). 32 A. A. Abrikosov, L. P. Gorkov, and I. E. Dzyaloshinski, 14 C. M. Varma, Phys. Rev. B 55, 14554 (1997). Methods of Quantum Field Theory in Statistical Physics 15 K. Sun and E. Fradkin, Phys. Rev. B 78, 245122 (2008). (Dover Publications, New York, 1975). 16 F. D. M. Haldane, Phys. Rev. Lett. , 2015. 33 B. A. Volkov, Y. V. Kopaev, and A. I. Rusinov, Sov. Phys. 17 S. Raghu, X.-L. Qi, C. Honerkamp, and S.-C. Zhang, Phys. JETP 41, 952 (1976). Rev. Lett. 100, 156401 (2008). 34 M. Tinkham, Group Theory and Quantum Mechanics 18 J. Gavoret, P. Nozi`eres,B. Roulet, and M. Combescot, J. (Dover Publications, New York, 1992). Phys. France 30, 987 (1969). 35 J. L. Ma˜nes,F. Guinea, and M. A. H. Vozmediano, Phys. 19 P. Nozi`eresand C. Comte, J. Phys. (Paris) 43, 1083 (1982). Rev. B 75, 155424 (2007). 20 R. Cˆot´eand A. Griffin, Phys. Rev. B 37, 4539 (1988). 36 C. M. Varma, Phys. Rev. Lett. 83, 3538 (1999). 37 A. Singh and Z. Teˇsanovi´c,Phys. Rev. B 41, 614 (1990). 38 A. Singh, Z. Teˇsanovi´c, and J. H. Kim, Phys. Rev. B (R) 44, 7757 (1991).