[name=Dibya, color=cyan]d [name=Adam, color=purple]a [name=Herb, color=orange]h

Impact of a Lifshitz Transition on the onset of spontaneous coherence

Abstract

Lifshitz transitions are topological transitions of a Fermi surface, whose signatures typically appear in the conduction properties of a host metal. Here, we demonstrate, using an extended Falicov-Kimball model of a two-flavor fermion system, that a Lifshitz transition which occurs in the non-interacting limit impacts interaction-induced insulating phases, even though they do not host Fermi surfaces. For strong interactions we find a first order transition between states of different polarization This transition line ends in a very unusual quantum critical endpoint, whose presence is stabilized by the onset of inter-flavor coherence. We demonstrate that the surfaces of maximum coherence in these states reflect the distinct Fermi surface topologies of the states separated by the non-interacting Lifshitz transition. The phenomenon is shown to be independent of the band topologies involved. Experimental realizations of our results are discussed for both electronic and optical lattice systems.

Introduction.

—In recent years, topology has become increasingly appreciated in condensed matter physics as a framework for understading diverse physical phenomena. These include the Thouless pump [1, 2], topological defects [3, 4, 5, 6, 7, 8, 9], quantized Hall effects [10, 11, 12, 13, 14, 15, 16, 17, 18], magnetic breakdown [19, 20, 21, 22], and topological insulators [23, 24, 25, 26, 27, 28, 29, 30, 31, 32, 33]. In addition to its utility for theoretical understanding, topology is physically significant because it leads to phenomena that are robust with respect to various perturbations [34, 35, 36, 37, 38].

Lifshitz transitions [39, 40, 41, 42, 43, 44] are an example of this. They occur when the topology of a Fermi surface changes with system parameters such as pressure, do** or external magnetic field [40, 41, 42, 44], and typically are observable as anomalies in magneto-oscillation periods as the system passes through such transitions. Because a Lifshitz transition is a Fermi surface phenomenon, its impact is normally only expected in metallic systems. In this work, we demonstrate that such transitions can also impact systems outside their metallic regime. In particular, we report on a system where a Lifshitz transition in the non-interacting limit leaves a clear signature when gap-opening interactions eliminate the Fermi surface.

      (a)

      (b)

       (c)

         (d)

           (e)

Figure 1: (1), (1), (1): the coherence as a function of crystal momentum. Regions of largest coherence form topologically distinct curves. (1) and (1): Fermi surfaces on either side of the Lifshitz transition in absence of spontaneous coherence.

Specifically, we examine a two-component system with interactions such that inter-component coherent phases can be supported. We demonstrate that the system supports different phases where the loops of maximum coherence in the Brillouin Zone are topologically distinct. An example of this is presented in Fig. 1. The topologies of the Fermi surfaces on either side of the Lifshitz transition mirror the topologies of these maximum coherence loops, demonstrating that the Lifshitz transition “seeds” the quantum phase transition of the interacting system. Importantly, the two different coherent states are separated by a first order transition for relatively strong interactions. We find that, within the models we examine, the transition line hosts a very unusual quantum critical endpoint (QCEP) [45], reminiscent of a thermodynamic $\mathcal{Z}_{2}$ critical point [9, 46]. As in the the latter, beyond the endpoint the evolution between coherent phases becomes continuous, with no sharp distinction between the states. Examples of this are illustrated in Fig. 2.

                            (a)

                            (b)

                                                                     (c)                                                                      (d)

              (e)

Figure 2: (2) and (2): Energy states solving Hartree-Fock equations, demonstrating hysteresis for large $U$ which is absent for small $U$. (c) and (d): Phase diagrams for four-band models for crossing bands of opposite and the same Chern numbers, respectively. (e): Phase diagram of a two-band model for crossing bands of opposite Chern numbers.

The phenomenon we describe in this work is robust with respect to band topology: we find it occurs whether the non-interacting bands that correlate through coherence have the same or different Chern numbers. In the latter case, the first order transition can also entail a transition in the occupied band topology [47, 48], but in such situations one does not find distinct coherent phases on either side of the transition. Our study demonstrates that for systems where a non-interacting Lifshitz transition resides in a setting where spontaneous breaking of a continuous symmetry can occur, a rich set of transitions results, separating phases that offer insight into the different Fermi surface topologies that the non-interacting system supports.

Model Hamiltonian.

—We adopt a minimal model capturing the physics in which we are interested, with Hamiltonian of the form

$$\hat{H}=\sum_{\bf k}\sum_{\ell=t,b}\sum_{p=\pm 1}\left[pE_{\bm{k}}+\Delta_{\ell}\right]\hat{c}_{{\bf k},\ell,p}^{{\dagger}}\hat{c}_{{\bf k},\ell,p}+U\sum_{\bf k}\hat{\rho}_{-\bm{k},t}\hat{\rho}_{\bm{k},b},$$ (1)

where $\hat{c}_{{\bf k},\ell,\pm 1}^{{\dagger}}$ creates a particle in an eigenstate of a Bernevig-Hughes-Zhang Hamiltonian [17] $h_{\ell,\bm{k}}$, with eigenvalues $\pm E_{\bf k}$ and eigenvectors $\chi_{{\bf q},\pm 1}$. (Details may be found in the Supplementary Material (SM) [49].) The second term is an interlayer contact interaction [30, 16] that involves density operators $\hat{\rho}_{\bm{k},\ell}=\sum_{{\bf q},p=\pm 1}\chi^{{\dagger}}_{{\bf k}+{\bf q},p}\cdot\chi_{{\bf q},p}\hat{c}_{{\bf q}+{\bf k},\ell,p}^{{\dagger}}\hat{c}_{{\bf q},\ell,p}$. Eq. 1 preserves the total number of fermions of $t$- and $b$-flavor separately, and supports a $U(1)$ flavor symmetry. Such models belong to the general class of extended Falicov-Kimball models (EFKM), which have been studied as candidates for describing electronic excitonic physics, ferroelectricity and bilayers phases [50, 51, 52, 53, 54, 55, 56, 57, 58, 59]. The impacts of Lifshitz transitions and band topology on such systems, the focus of this study, are to our knowledge unknown.

The model can also be interpreted as a bilayer system without tunneling between layers [60, 61, 62, 63]), in which intralayer interactions have been neglected. In what follows we adopt this realization as a paradigm for such systems, and refer to the flavors as layers. We expect the behavior of this model at half-filling to apply when bands of each flavor are relatively far apart energetically, while there is a crossing of bands of two different flavors: with short-range interactions, Fermi statistics suppresses short-range intra-flavor interactions, allowing inter-flavor interactions to dominate. Beyond this, the model can also be mapped onto a two-species fermionic atomic gas system in an optical lattice, as we discuss below.

Hartree-Fock Analysis.

—We consider ground states of $\hat{H}$ within the Hartree-Fock (HF) approximation, in situations where the system is half-filled. Details of the analysis may be found in the SM [49]. For $\Delta=0$, the spectrum consists of one filled and one empty band for each layer, with an intervening gap that is present even in the absence of interactions. With increasing $|\Delta|$, bands of opposite flavor approach one another, eventually crossing when spontaneous symmetry breaking is not considered. The Fermi surfaces consist of matching loops for each flavor surrounding the $\Gamma$ point of the square Brillouin zone (BZ). With growing $|\Delta|$ these loops eventually touch the $M$ points at the BZ edge, signaling a Lifshitz transition. For still larger $|\Delta|$ the loop topology changes, now surrounding the $X$ points (corners) of the BZ.

Multi-flavor systems with matching Fermi surfaces are known to be unstable to spontaneous inter-flavor coherence in the presence of interactions [64, 65]. Figs. 2 (c) and (d) illustrates what happens, within our HF analysis, when two bands pass fully through one another as a function of $\Delta$, with a Lifshitz transition separating different Fermi surface topologies (see SM [49] for an illustration.) We consider both crossing bands of the same Chern number $\mathcal{C}=1$ [panel (c)], and crossing bands of opposite Chern number $\mathcal{C}=\pm 1$ [panel (d)]. Continuous transitions are indicated as dashed lines while solid lines indicate first order transitions. The resulting phases include a non-interacting gap (NIG) phase, which is continuously connected to the $\Delta=0$ non-interacting state; inter-layer coherent phases (Coh I and II); and a layer-polarized (LP) phase.

To characterize the polarization of a phase, we define a polarization function $P(\bm{k})\equiv\sum_{i=1}^{4}\langle\psi_{i}(\bm{k})|\tau_{z}\otimes\mathbbm{1}|\psi_{i}(\bm{k})\rangle f(E_{i}(\bm{k}))\equiv\sum_{i=1}^{4}P_{i}(\bm{k})f(E_{i}(\bm{k})),$ where $|\psi_{i}({\bf k})\rangle$ are the four HF wavefunctions at wavevector ${\bf k}$ with energy $E_{i}({\bf k})$, $f$ is the Fermi function, and the $\tau_{z}$ is a Pauli matrix acting in layer-space. Spontaneous coherence in the system arises when order parameters of the form $\langle c^{{\dagger}}_{{\bf k},\ell,p}c_{{\bf k},\ell^{\prime},p^{\prime}}\rangle\neq 0$ for $\ell\neq\ell^{\prime}$. This always entails values of $P_{i}({\bf k})$ which are not equal to either $-1$ or $1$, so that non-vanishing values of a coherence function, $C(\bm{k})\equiv\sum_{\bm{k}}\sum_{i=1}^{4}(1-|P_{i}(\bm{k})|)f(E_{i}(\bm{k})),$ signal the presence of interlayer coherence.

Fig. 1 illustrates the behavior of $C({\bf k})$ as a function of ${\bf k}$ for two coherent states, and their correlation with the non-interacting Fermi surface. The loci of maximum $C({\bf k})$ have two different topologies reflecting the behavior of the Fermi surfaces on either side of the non-interacting Lifshitz transition. For small $U$ the evolution from one behavior to the other as a function of $\Delta$ is continuous, while for larger values there is a first order transition between them (see Figs. 2 (a) and (b)). The transition line is quite interesting. It hosts a very unusual QCEP, which at the mean-field level is highly analogous to the critical point of a thermal $\mathcal{Z}_{2}$ transition. Indeed, the behavior of the full polarization, $P=\sum_{{\bf k}}P({\bf k})$, when the bias $\Delta$ is varied, displays a jump that continuously vanishes at the QCEP, as illustrated in Fig. 3.

Figure 3: Plot of $P$ versus $1/\Delta$ for fixed $U$, illustrating a closing discontinuous polarization jump.

Beyond this, as $U$ drops from large values, the transition line gives birth to two continuous transitions, in which coherence sets in from either the NIG or the LP phase. Moreover, although the situations illustrated in Figs. 2 (c) and (d) look very similar, they have an important difference. In (2), the transition out of the NIG phase involves a change in Chern number. The transition occurs without a gap closing at large $U$ [47, 66, 67, 28], while at smaller $U$ the onset of coherence occurs continuously and simultaneously with the topological transition. The topological transition is in this way tied up with the symmetry-breaking transition. By contrast, in (2) there is no topological transition of the filled bands, and there is no closing of the band gap at any of the quantum phase transitions. Fig. 4 illustrates this difference in behavior.

Two Band Model.

—Several of the results described above can be qualitatively understood within a two band model, in which one retains only the two bands that cross one another in the model. These can have the same or opposite topologies; Fig. 2(e) illustrates the resulting phase diagram for the case of opposite Chern numbers. Although there is a change in the locations of the phase boundaries, particularly at large $U$, the system retains the same basic phases and features of the four band model. Most prominent is the first order transition line drop** from large $U$, which ends at a QCEP.

The phases of this model are characterized by two order parameters: (i) $B_{p}\equiv{U\over 2V}\sum_{\bf k}\langle\hat{c}^{{\dagger}}_{{\bf k},t}\hat{c}_{{\bf k},t}-\hat{c}^{{\dagger}}_{{\bf k},b}\hat{c}_{{\bf k},b}\rangle$, where $\hat{c}_{{\bf k},\ell}$ annihilates a particle in the retained band of layer $\ell$, which is a measure of the polarization of the system; and (ii) an interlayer coherence $B^{tb}({\bf k})\equiv{U\over V}\sum_{{\bf k}_{1}}|\chi_{{\bf k},t}^{{\dagger}}\cdot\chi_{{\bf k}_{1},b}|^{2}\langle c^{{\dagger}}_{{\bf k}_{1},t}c_{{\bf k}_{1},b}\rangle$. ($V$ is the system area). Within mean-field theory the self-consistent equations for these have the form [49]

$$\frac{B_{p}}{U}=\dfrac{1}{2V}\sum_{{\bf k}_{2}} \dfrac{\widetilde{\xi}({\bf k}_{2})}{\sqrt{|\widetilde{\xi}({\bf k}_{2})|^{2}+|B^{tb}({\bf k}_{2})|^{2}}}\equiv F(B_{p}+\Delta);$$ (2)

$$B^{tb}({\bf k}_{1})={U\over{2V}}\sum_{{\bf k}_{2}} \dfrac{f(\theta_{{\bf k}_{1}},\theta_{{\bf k}_{1}};\mathcal{C}_{rel})\,B^{tb}({\bf k}_{2})}{\sqrt{|\widetilde{\xi}({\bf k}_{2})|^{2}+|B^{tb}({\bf k}_{2})|^{2}}},$$ (3)

where $\tilde{\xi}({\bf k}_{2})=E_{{\bf k}_{2}}+\Delta+B_{p}$, $\mathcal{C}_{rel}=1(-1)$ when the product of the Chern numbers for the crossing bands is 1(-1), $f(\theta_{{\bf k}_{1}},\theta_{{\bf k}_{1}};1)=(1+\cos\theta_{{\bf k}_{1}}\cos\theta_{{\bf k}_{2}})/{2}$ and $f(\theta_{{\bf k}_{1}},\theta_{{\bf k}_{1}};-1)=\left(\cos^{2}\theta_{{\bf k}_{1}}/2\right)\left(\cos^{2}\theta_{{\bf k}_{2}}/2\right)$. In Eq. 3 we have assumed that $\langle c^{{\dagger}}_{{\bf k}_{1},t}c_{{\bf k}_{1},b}\rangle$ is real and has $C_{4}$ rotational symmetry in the HF ground state, which we indeed find in our more general numerical analysis.

       (a)

       (b)

Figure 4: Gap vs. bias $\Delta$ for $U=1.8$. (4): Crossing bands have the same Chern number $\mathcal{C}=1$. (4): Crossing bands have opposite Chern number $\mathcal{C}=\pm 1$. Insets: Gap remains open in first case but closes in second.

Eq. 2 provides particular insight into the connection between the first-order transition lines in Fig. 2 and a thermal $\mathcal{Z}_{2}$ transition. Direct plots of the left- and right-hand sides of the equation show that at large positive (negative) values of $\Delta$, one finds a single solution with maximal negative (positive) values of $B_{p}$, but in some transition region of $\Delta$ there are three solutions. (See SM for details [49].) The physical state of the system jumps between two of the three solutions when their energies cross, behavior which is highly reminiscent of what one finds in a mean-field treatment of, for example, the liquid-gas transition [68]. Such a jump always occurs provided $\max_{x}\frac{dF(x)}{dx}>\frac{1}{U}$. Interestingly, because of the Lifshitz transition, one finds a point for which the left-hand side of the inequality diverges when $B^{tb}=0$, so that there is a first order jump for any positive value of $U$, and no QCEP is manifested. However, for $|B^{tb}|>0$, this divergence is smoothed over, $\max_{x}\frac{dF(x)}{dx}<\infty$, and for sufficiently small (but non-vanishing) $U$ the first order jump gives way to a continuous cross-over: a QCEP is stabilized. This mean-field phenomenology is highly reminiscent of that of a classical liquid-gas critical point. Remarkably, in order to be realized in this setting, spontaneous coherence, a purely quantum phenomenon, must be manifested between layers.

Eq. 3 also allows an understanding of the behaviors of the coherence onset at the NIG-Coh I boundary for small $U$, which are different depending on whether $C_{rel}=\pm 1$. In this regime, the bias is such that the single-particle energies of the two bands are very close near the $\Gamma$ point. Writing $\Delta_{0}$ for the bias at which the two bands touch at the $\Gamma$ point, we define $\tilde{\Delta}\equiv\Delta-\Delta_{0}$, and simplify our model by assuming the bands to be quadratic, in which case they have a constant density of states $g_{0}$. Within this model and for small $k$ one may assume $\cos\theta_{{\bf k}}=\cos\theta(E_{\bf k})\approx-1+\alpha(E_{\bf k}-E_{0})$, with $\alpha$ a parameter of order the bandwidth $W$. As shown in the SM [49], one may then show $B^{tb}\approx b_{0}(1-\cos\theta_{\bf k})$ for $\mathcal{C}_{rel}=1$, and $B^{tb}\approx b_{0}\cos^{2}\theta_{\bf k}/2$ for $\mathcal{C}_{rel}=-1$, with

$$b_{0}\approx W \exp\left[-\frac{2}{Ug_{0}}\right]\quad\quad\quad\mathcal{C}_{rel}=1;$$

$$b_{0}\approx 4\sqrt{\frac{W}{\alpha^{2}|\tilde{\Delta}|}} \exp\left[\frac{-4}{\alpha^{2}\tilde{\Delta}^{2}Ug_{0}}\right]\Theta(-\tilde{\Delta})\quad\mathcal{C}_{rel}=-1;$$

for small $|\tilde{\Delta}|$. In these expressions, $\Theta$ is a Heaviside step function. These results show that for bands of opposite Chern number ($\mathcal{C}_{rel}=-1$) the band closing ($\tilde{\Delta}\rightarrow 0$) associated with a change of topology of the occupied band occurs simultaneously with the onset of coherence ($b_{0}\neq 0$). This contrasts with the situation for $\mathcal{C}_{rel}=1$, in which $b_{0}$ is non-zero for small $|\tilde{\Delta}|$ even if the corresponding non-interacting bands do not cross. Thus the single particle energy gap never closes in this case. The differing Berry’s curvatures of the two bands for $\mathcal{C}_{rel}=-1$ case frustrates the formation of coherence in the system. This is apparent in the insets of Fig. 4.

Discussion.

—These results suggest a number of interesting questions. One set of these addresses the universality classes of the critical point as well as those of the coherence onset regions when the bands first cross. In the former case, the presence of a broken $U(1)$ symmetry and its accompanying Goldstone mode suggests that its critical behavior will be different than that of a classical thermal $\mathcal{Z}_{2}$ transition. In the latter case, the presence of gapless fermions for $\mathcal{C}_{rel}=-1$ suggests the transition will be in a different universality class than for $\mathcal{C}_{rel}=1$. Effects of real thermal fluctuations on the system, and the form a quantum critical region [69, 70] takes, is import to understand in settings where temperature effects cannot be ignored. Another set of questions involve how this phase diagram might be manifested in different physical realizations. One involves an optical lattice [71] hosting two species of atoms with an inter-species Feshbach resonance [72, 73]. A particle-hole transformation maps this onto an EFKM; in this case coherence is realized in superconducting states of the system. For appropriate parameters, we expect two such states, separated by a first order transition. Finally, van der Waals materials [74] offer platforms for electron bilayer realizations of this system, which support layer polarized states [75, 76] and/or interlayer coherence [77, 78, 79, 80, 81], whose interaction and competition could lead to novel quantum phase boundaries and transitions such as those we have described in this study.