Intertwined geometries in collective modes of two dimensional Dirac fermions Ankan Biswas, Avraham Klein Abstract It is well known that the time-dependent response of a correlated system can be inferred from its spectral correlation functions. As a textbook example, the zero sound collective modes of a Fermi liquid appear as poles of its particle-hole susceptibilities. However, the Fermi liquid’s interactions endow these response functions with a complex analytic structure, so that this time/frequency relationship is no longer straightforward. We study how the geometry of this structure is modified by a nontrivial band geometry, via a calculation of the zero sound spectrum of a Dirac cone in two dimensions. We find that the chiral wavefunctions, that encode the band geometry, fundamentally change the analytic structure of the response functions, which encode its Riemannian geometry. As a result, isotropic interactions can give rise to a variety of unconventional zero sound modes, that, due to the geometry of the functions in frequency space, can only be identified via time-resolved probes. These modes are absent in a conventional Fermi liquid with similar interactions, so that these modes can be used as a sensitive probe for the existence of Dirac points in a band-structure. I Introduction Understanding how a strongly correlated quantum system responds to a strong pulse of electromagnetic radiation combines questions of fundamental physics with exciting prospects for future applications [ 1 ] . After being exposed to such a pulse, quantum systems can exhibit strong nonequilibrium dynamics, leading to emergent macroscopic behavior [ 2 ] and metastable states with no equilibrium counterpart, properties that could be crucial for developing quantum switches and devices [ 3 , 4 ] . Such so-called “ultrafast” laser pulses, lasting from attoseconds to picoseconds and spanning frequencies from terahertz (THz) to X-ray [ 5 , 6 , 7 , 8 ] , are particularly significant when applied in the low-frequency THz range for quantum materials and devices. By carefully controlling the wavepacket shape of an incoming pulse, it is possible to probe and manipulate the emergent, low energy collective modes of an interacting quantum system, and measure their time dependent response in a pump-probe configuration. Experiments have demonstrated the response of varied quantum phenomena to ultrafast pulses, including magnetism, superconductivity and many other phenomena [ 9 , 10 ] . Theoretical descriptions of a pumped system’s evolution in the time domain are challenging. One of the reasons of this is that even the simplest out-of-equilibrium observable is a function of two times (e.g. center-of-mass and relative time), or of simultaneous time and frequency [ 11 ] . In fact, even within a linear response regime, the idea of pump-probe experiments raises a fundamental question on the relationship of time and frequency in interacting systems: Even in equilibrium , does the time domain response fully encode the frequency domain response (and vice versa), or do these domains convey different information? In other words, are the frequency and time domain linear responses of an interacting system just naive Fourier transforms of one another, or is there some deeper relationship, potentially with non-local aspects tied to topology? Figure 1: Representation of the intertwined geometries of the nontrivial band structure and particle-hole susceptibility \chi(\bm{\mathrm{q}},\Omega) . Illustration of the Dirac cone, showing how a particle-hole excitation (red wiggly line) moves an electron by momentum \bm{\mathrm{q}} and frequency \Omega . Each such process is defined on a Riemann surface (a sphere) of complex \Omega . The zero sound collective modes appear as poles on this two-sheeted Riemann surface, of which the top one represents the “physical” sheet (whose real axis represents physical frequencies), and the bottom is “unphysical”. The nontrivial geometry is evident in the gluing of lips of the branch cuts, red to red and green to green. Different colored disks indicate poles representing a variety of collective modes on the two sheets : red disks - conventional underdamped (cu) modes; purple disks - conventional overdamped (co) modes; orange disks - hidden (h) modes, hidden from the physical real axis; and brown disks - mirage (m) modes on the unphysical sheet. The properties of the modes are reviewed in detail in the paper. The time-dependent response is given by a Fourier transform, whose integration contour (blue dashed line) traces over the entire surface, picking up the contributions from the nontrivial geometry. (Inset [in arbitrary units]) The time-dependent response to a sharp pulse, for different angles \alpha between the mode wavevector and its polarization, in a Dirac liquid with a repulsive isotropic interaction. The anisotropic frequency and decay rates are a result of the chiral bandstructure, absent in a conventional Fermi liquid. At long times the three responses cross over to a single universal behavior (see Fig. 7 ). The numerical value used in the inset is F_{0}=10 (see the following sections for details). An efficient way to study this issue is by analyzing the low energy collective excitations of an itinerant fermion system, i.e. a Fermi liquid. The fundamental excitations of a Fermi liquid (in its normal state), are zero sound (ZS) modes. ZS modes, which are collective excitations from coherent fluctuations of fermions near the Fermi surface, manifest as sharp propagating excitations for repulsive quasiparticle interactions, whereas for attractive interactions, they generally appear as damped resonances originating from quasi-bound states [ 12 , 13 , 14 , 15 , 16 , 17 ] . However, as is well-known, the particle-hole response of a Fermi-liquid is non-analytic on the complex frequency plane [ 18 ] . Elementary theorems of complex analysis imply that the response function is therefore defined a Riemann surface, suggesting that the interactions endow the system with a nontrivial geometry and topological structure [ 19 ] . This structure manifests itself through several unconventional zero-sound (ZS) modes [ 20 ] , which we will describe in detail. For example, even in the simplest model of an interacting Fermi liquid, an itinerant paramagnet, weak attraction in the spin channel gives rise to a so-called “hidden” mode. This is a sharp propagating mode which is generated by the attraction, rather than repulsion, but which is hidden in the sense that it cannot be detected via standard spectral measurements, due to the nontrivial geometry of the Riemann surface. The existence of a nontrivial geometric structure arising purely from a Fermi liquid’s interactions raises another question: in itinerant systems with a nontrivial bandstructure geometry and topology, what is the relationship between the two? This relationship is depicted pictorially in Fig. 1 . In this paper, we examine this interplay by studying what is possibly the simplest such case: the ZS spectrum of a Dirac liquid. As we shall show, in the low energy limit this relationship is encoded in the particle-hole polarization at finite doping. We computed the zero-temperature one-loop polarization response function, \displaystyle\chi^{b}_{\mu\nu}(\bm{\mathrm{q}},i\Omega)\approx-\int\frac{d\omega d^{2}k}{\left(2\pi\right)^{3}}\bra{u_{\bm{\mathrm{k}}_{-}}}\tau_{\mu}\ket{u_{\bm{\mathrm{k}}_{+}}}\bra{u_{\bm{\mathrm{k}}_{+}}}\tau_{\nu}\ket{u_{\bm{\mathrm{k}}_{-}}}\quad \displaystyle G(\bm{\mathrm{k}}_{-},i\omega_{-})G(\bm{\mathrm{k}}_{+},i\omega_{+}). (1) Here, \epsilon_{\bm{\mathrm{k}}} is the dispersion, G=(i\omega-\epsilon_{\bm{\mathrm{k}}})^{-1} is a fermionic Green’s function, \tau_{\mu} are pseudospin Pauli matrices, and \ket{u_{\bm{\mathrm{k}}}} is the wavefunction in the conduction band, and shifted frequencies and momenta are defined as \omega_{\pm}=\omega\pm\frac{\Omega}{2} and \bm{\mathrm{k}}_{\pm}=k\pm\frac{\mathbf{q}}{2} , where \Omega and \mathbf{q} denote the external frequency and momentum, respectively. The existence of the nontrivial form-factors \bra{u_{\bm{\mathrm{k}}}}\tau_{\mu}\ket{u_{\bm{\mathrm{k}}}} imprints the band geometry on the polarization. When the bare polarization bubble is renormalized by an RPA sum, this gives rise to collective modes, also “stamped” with the band geometry, once the frequency is rotated back to the real axis. To investigate the interplay of geometries, we identified the collective modes, analyzed their behavior as a function of doping, and compared their characteristics to that of a “conventional” Fermi liquid with a parabolic bandstructure and trivial topology. We found, that unconventional ZS modes only appear at finite chemical potential and vanish at the charge neutrality point, implying that the nontrivial geometry of the interaction is a result of Landau damping and the breaking of Lorentz invariance to Galilean invariance. At finite doping, we found that the chirality of the bandstructure introduces an angular, p-wave like behavior into the off-diagonal ( \mu\neq\nu ) terms in Eq. ( 1 ), which creates a host of unconventional ZS modes (see e.g. Fig. 1 ). The chiral terms arise from the projection of the wavefunctions onto the pseudospin vector, which is isomorphic to the Berry connection of the system, and indicates that the bandstructure geometry is directly imprinted onto the collective modes. The modes are qualitatively different from those appearing in a conventional Fermi liquid with a trivial bandstructure, even when accounting for the Riemann surface. Finally, we did not find any induced change of the genus of the Riemann surface due to the Dirac band structure, indicating that the topological aspects of the system are robust. The fact that the ZS modes in a doped Dirac liquid are qualitatively different those of a Fermi liquid, even in the limit of infinitesimal frequency and momentum transfer, when the Dirac point is “infinitely far away” from the excitations, means that the two types of geometries are inherently intertwined. In particular, as we will discuss later, the implication is that a knowledge of both structures is necessary for a correct interpretation of the time-dependent response of the ZS collective modes in an interacting system, and conversely, that these modes can be utilized to probe both structures. We demonstrate this further in the Appendix by studying several additional models with parabolic bands, both with and without nontrivial band geometry. The rest of this paper is organized as follows. In Sec. II we introduce our model and review how the Riemannian geometry arises and affects the relationship between time and frequency. In Sec. III we calculate the ZS collective modes of a Dirac liquid and compare them to that of a conventional Fermi liquid. We make some concluding remarks in Sec. IV . II Model and Theoretical Framework II.1 Electronic band structure and interactions Figure 2: The RPA sum for the polarization bubbles that determined \chi_{\mu\nu} . To study the collective modes in our work, we start with the bare Hamiltonian of a single Dirac cone in the sublattice representation. Neglecting spin degrees of freedom, the Hamiltonian near the Dirac point is expressed as: \displaystyle\hat{H}=\sum_{\bm{\mathrm{k}}}c^{\dagger}_{\bm{\mathrm{k}}}\left(-\mu\tau_{0}+v_{F}\bm{\mathrm{\tau}}\cdot\bm{\mathrm{k}}\right)c_{\bm{\mathrm{k}}}. (2) where we use hat notation to denote matrices in the sublattice space. Here \mu is the chemical potential, \bm{\mathrm{\tau}} Pauli matrices representing the pseudospin sublattice degree of freedom for the fermions c_{\bm{\mathrm{k}}} , and v_{F} is the Fermi velocity. This Hamiltonian captures the low-energy fermionic states near the Dirac point, giving rise to the well-known cone structure, as illustrated in Fig. 1 . In physical systems, there are typically several different cones in the Brillouin zone, e.g. six in the case of the tight binding model for graphene (with an extended Brillouin zone) [ 21 ] . For most of the calculations in this work, the additional cones just add a numerical prefactor, which we neglect for simplicity. When multiple cones are important, we note it explicitly (see e.g. Sec. III.3 ). Inverting the Hamiltonian in Eq. ( 2 ), the Green’s function for the system, on the Matsubara axis, is given by: \hat{G}(\bm{\mathrm{k}},i\omega)=-\frac{i(\omega-i\mu)\tau_{0}-v_{F}\bm{\mathrm{\tau}}\cdot\bm{\mathrm{k}}}{(\omega-i\mu)^{2}+v^{2}_{F}|\bm{\mathrm{k}}|^{2}}, (3) The bare particle-hole susceptibility is therefore, \chi_{\mu\nu}^{b}(\bm{\mathrm{q}},i\Omega)=-\int\frac{d\omega d^{2}k}{\left(2\pi\right)^{3}}\mbox{Tr}\left[\tau_{\mu}\hat{G}(\bm{\mathrm{k}}_{+},\omega_{+})\tau_{\nu}\hat{G}(\bm{\mathrm{k}}_{-},\omega_{-})\right] (4) where the trace is over the sublattice indices. We use Greek subscripts to denote 2+1 space dimensions, and Latin subscripts for spatial coordinates. The exact result obtained by diagonalizing the Green’s function can be found, along with most computational details, in the Appendix A . In what follows, we focus on the low energy excitations. We rotate to the real axis via i\Omega\rightarrow\Omega+i\delta , where \delta is an infinitesimal regularizer for the retarded susceptibility. (Later on we will assume \delta is a small static disorder.) Finally, we assume the limit v_{F}|\bm{\mathrm{q}}|,\Omega\to 0,s=\frac{\Omega}{v_{F}|\bm{\mathrm{q}}|}\to\mbox{finite}. (5) The implication is that except at the charge neutrality point, which we discuss later, we may always assume the response is dominated by the upper band intraband processes. In that case the full expression in Eq. ( 4 ) simplifies to Eq. ( 1 ), where the upper band wavefunctions are \ket{u_{\bm{\mathrm{k}}}}=\frac{1}{\sqrt{2}}\left(\begin{array}[]{c}e^{-i\theta_{\bm{\mathrm{k}}}}\\ 1\end{array}\right), (6) and \theta_{\bm{\mathrm{k}}} is the azimuthal angle on the FS. To study the collective modes, we follow the ideas of Fermi liquid (FL) theory. As usual, we assume that the fermionic properties are modified by renormalizing the free propagator in Eq. ( 3 ) with a quasiparticle residue Z , \hat{G}\to Z\hat{G} and an effective mass m^{*} , v_{F}\to v_{F}(m/m^{*}) , and interact via a static interaction on the FS. Typically, these interactions are decomposed into spin and charge channels. Here, instead, we consider the interactions decomposed into the sublattice channels [ 22 ] , and identify the fully renormalized, static, and antisymmetric interactions between low-energy fermions, represented by the vertex \Gamma_{ij,kl}(\bm{\mathrm{k}}-\bm{\mathrm{p}}) . The harmonic expansion of \Gamma_{ij,kl}(\bm{\mathrm{k}}-\bm{\mathrm{p}}) over orbital moments l is possible for rotationally and SU(2)-invariant systems, such as those we are considering here, and give the sublattice version of Landau parameters F^{c(s)}_{l} , namely \Gamma_{ij,kl}=\nu_{F}^{-1}F^{c}(\bm{\mathrm{k}}-\bm{\mathrm{p}})\tau_{0}^{ik}\tau_{0}^{jl}+\nu_{F}^{-1}F^{s}(\bm{\mathrm{k}}-\bm{\mathrm{p}})\bm{\mathrm{\tau}}^{ik}\cdot\bm{\mathrm{\tau}}^{jl}, (7) where F^{c(s)}(\bm{\mathrm{k}}-\bm{\mathrm{p}})=F^{c(s)}_{0}+2\sum_{l=1}^{\infty}F^{c(s)}_{l}\cos l(\theta_{\bm{\mathrm{k}}}-\theta_{\bm{\mathrm{p}}}) (8) and \nu_{F}=\frac{\mu}{v_{F}^{2}}Z^{2}\left(\frac{m^{*}}{m}\right) , where \theta_{\bm{\mathrm{k}}},\theta_{\bm{\mathrm{p}}} are the azimuthal angle of \bm{\mathrm{k}} , \bm{\mathrm{p}} . As a starting point for the analysis, let’s compute the (retarded) density-density reponse \chi^{R}_{00} . We sum over the sublattice degrees of freedom, go through the algebra and find [ 20 ] , for finite \mu , \chi^{b}_{00}(s)=\frac{\mu}{v^{2}_{F}}l_{0}(s), (9) where l_{0}(s)=1+\frac{is}{\sqrt{1-s^{2}}}. (10) Up to the global prefactor, this is precisely the same response as for a conventional FL [ 20 ] . At the charge neutrality point (CNP), we have [ 23 , 24 ] , \chi^{CNP}_{00}(\bm{\mathrm{q}},s)=\frac{|\bm{\mathrm{q}}|}{32v_{F}}\frac{1}{\sqrt{1-s^{2}}}. (11) Viewed as a function of complex s=\frac{\Omega}{v_{F}|\bm{\mathrm{q}}|} , \chi_{00}(s) has branch points at \pm s_{b}=\pm 1-i\delta, (12) where, as we recall, \delta is an infinitesimal regularizer that insures \hat{\chi} is analytic in the upper half plane. It is these branch points that introduce the nontrivial geometry and topology into the structure of \hat{\chi} , as we now discuss. II.2 The analytic structure of the collective modes Let us now turn to discuss the ZS collective mode spectrum and its associated algebraic structure. The ZS modes are the poles of the quasiparticle susceptibility, computed from the RPA sum of the polarization bubbles and the interaction vertex, in the limit of vanishing momentum and frequency, see Eq. ( 5 ). For a generic interaction the computation of the sum can be quite involved. Here, to clarify the algebraic structure we will consider the simplest limit of a single Landau parameter F_{0}^{(s)}=F_{0} . We choose to use the pseudospin channel rather than the charge channel to avoid issues of the Coloumb interaction and plasmons [ 25 ] . As we shall see, for the Dirac case the existence of chiral wavefunctions qualitatively changes the collective mode behavior from, e.g. a spin fluctuation channel. For the single Landau parameter, the sum is straightforward and given by Fig. 2 . The vertex F=F_{0}\tau_{m}\tau_{m},m=x,y , and the sum is, \chi^{qp}_{\mu\nu}=\chi^{b}_{\mu\nu}-F_{0}\chi^{b}_{\mu k}\chi^{b}_{k\nu}+\cdots=\left[(\hat{1}+F_{0}\nu_{F}^{-1}\hat{\chi}^{b})^{-1}\hat{\chi}^{b}\right]_{\mu\nu}, (13) where the superscript “qp” denotes the “quasiparticle” susceptibility [ 26 , 27 , 18 , 28 , 29 , 30 , 31 , 32 , 33 , 34 ] . Henceforth we supress the “qp” notation. To properly understand the pole structure, we must recall that the bare susceptibility [ 12 , 13 , 35 ] , and hence quasiparticle susceptibility as well, includes a square-root singularity and has two branch points in the lower half-plane, right below s=\pm 1 , see Eqs. ( 9 )-( 12 ). As a result, \hat{\chi} in Eq. ( 13 ) is not defined on a complex plane but rather on a two sheeted Riemann surface. One sheet which we denote the “physical” sheet includes the physical real s axis and is analytic in the upper half plane. The other sheet, the “unphysical” one, does not need to be analytic in the upper half plane. These sheets are depicted in the illustration in Fig. 1 , showing how poles of \hat{\chi} on the surface can exist in three of the four available half-planes. To construct the surface, we “cut” the sheets along branch cuts, chosen here to run from \pm s_{b}...\pm\infty-i\delta , where we recall that on the sphere all points at infinity converge. The genus of the surface, which in our simple case is just zero (a closed sphere), and the number of poles on it, are topological invariants. As discussed in the Introduction, the nontrivial Riemann structure has a profound impact on the time-dependent response. For an instantaeous pulse exciting the FL, the response will be given by the Fourier transfer of \hat{\chi} , \hat{\chi}(\bm{\mathrm{q}},t)=\int_{-\infty}^{\infty}\frac{ds}{2\pi}e^{-isv_{F}|\bm{\mathrm{q}}|t}\hat{\chi}(s). (14) As seen in Fig. 1 , the complex integration contour runs over the surface, and does not have to be chosen strictly on the physical sheet. This implies that the time-dependent response can be dominated by poles that are not “conventional”, in the sense of being situated below the physical s axis and which show up as the well known sharp Lorentzians or broad damped peaks in the spectral function. For a concrete example of this, let’s imagine the interaction of Eq. ( 8 ) is in fact that of a conventional parabolic FL, such that F_{0} from Eq. ( 13 ) is just an isotropic spin interaction (for an itinerant ferromagnet), and we are interested in the spin response \chi_{ij} . The sum is completely straightforward such that, \chi_{ij}\propto\delta_{ij} (15) and the poles are just the well-known poles of a FL [ 12 , 13 ] ZS mode , 1+F_{0}l_{0}(s)=0. (16) Since l_{0}(s=0)=1 , Eq. ( 16 ) is a manifestation of the Pomeranchuk criterion, requiring -10 ) one obtains the well-known ZS propagating modes, [ 13 ] , [ 18 ] , [ 27 ] . For a weak attractive interaction -1/2