betaFeedback

WorksPhysics and Astronomy

Geometric Stiffness in Interlayer Exciton Condensates

Abstract

Recent experiments have confirmed the presence of interlayer excitons in the ground state of transition metal dichalcogenide (TMD) bilayers. The interlayer excitons are expected to show remarkable transport properties when they undergo Bose condensation. In this work, we demonstrate that quantum geometry of Bloch wavefunctions plays an important role in the phase stiffness of the Interlayer Exciton Condensate (IEC). Notably, we identify a geometric contribution that amplifies the stiffness, leading to the formation of a robust condensate with an increased BKT temperature. Our results have direct implications for the ongoing experimental efforts on interlayer excitons in materials that have non-trivial quantum geometry. We provide quantitative estimates for the geometric contribution in TMD bilayers through a realistic continuum model with gated Coulomb interaction, and find that the substantially increased stiffness allows for an IEC to be realized at amenable experimental conditions.

Introduction

Advancements in topological quantum matter have drawn attention to the crucial role of Bloch wavefunctions in diverse condensed matter systems. While the influence of Berry curvature on non-interacting electrons is well-understood as an anomalous velocity [1], a closely related quantity, the quantum metric, has recently gained significant attention particularly in the context of flat-band superconductivity and related experiments in moiré heterostructures [2]. Derived from the geometric properties of Bloch wavefunctions, the quantum metric has profound effects on various facets of superconductivity. Notably, it modifies the mass of Cooper pairs [3, 4, 5], phase stiffness [6, 7, 8, 9, 10, 11, 12], spectral weight [13, 14, 15, 16], and potentially the critical temperature [17, 18]. Other than superconductivity, quantum geometry is known to appear in current noise spectrum [19], dielectric response [20], electron-phonon coupling [21], plasmons [22] and nonlinear response [23, 24, 25, 26, 27]. In this work, we target Interlayer Exciton Condensates (IECs) and reveal a significant geometric contribution to the phase stiffness that results in a more robust condensate characterized by a higher Berezinskii-Kosterlitz-Thouless (BKT) transition temperature.

Excitons, bound states of electrons and holes in semiconductors, are bosons that have long been proposed to form a Bose-Einstein condensate (BEC) at low temperatures [28, 29]. Unlike conventional BECs, exciton condensates conserve total particle number and instead break a U(1)e×U(1)hU(1)_e \times U(1)_h symmetry that corresponds to separate conservation of electrons and holes [30, 31]. This symmetry is experimentally realizable in a bilayer system with a spacer that suppresses single-particle tunneling between the layers. If the electrons reside on the top (tt) and holes in bottom (bb) layers, IEC is formed by the spontaneous breaking of U(1)t×U(1)bU(1)_t \times U(1)_b symmetry [32, 33, 34, 35]. Its superfluid properties have been observed in quantum Hall systems [36] [37].

Although an exciton condensate arising intrinsically in a real material has been a challenge, there has been progress in three-dimensional semimetal 1T-TiSe2_2 [38], monolayer WTe2_2 [39, 40], bilayer WSe2_2 [41], and TMD bilayer WSe2_2/MoSe2_2 [42]. In particular, ref. [43] has established the existence of interlayer excitons in the ground state by capacitance measurements and characterized the exciton Mott transition [44–48] as a function of the density of electron-hole pairs. These interlayer excitons have finite dipole moment and interact via dipole-dipole interaction. Thus, as a virtue of interacting bosons in 2D, it is possible that there is condensation at low enough temperatures [49–53]. Confirming the existence of the condensate requires transport experiments [31] [54], which have remained elusive until two recent reports [55, 56].

IECs display fascinating transport properties, most notably dissipationless counterflow transport. When equal and opposite fields are applied to the two layers, excitons flow without resistance in the condensate [36]. The longitudinal counterflow conductivity, σCF(ω)\sigma_{\mathrm{CF}}(\omega), diverges in the dc limit with σCF(ω)=Dsδ(ω)+⋯\sigma_{\mathrm{CF}}(\omega) = D_s\delta(\omega) + \cdots, where the weight of the delta function, DsD_s, represents the phase stiffness that governs the free energy cost of phase fluctuations in the condensate [57]. In this study, we demonstrate that DsD_s, in addition to a conventional contribution from band dispersion, has a significant geometric contribution that arises from the wavefunctions of the non-interacting electron and hole bands.

The critical role of wavefunctions in exciton condensation is evident from quantum Hall bilayers where excitons exhibit macroscopic coherence and condense [58, 59] despite the non-interacting bands having flat dispersion and infinite mass. The non-trivial wavefunctions of the Landau level endow mobility to excitons even when the constituent electrons and holes are immobile. Similar phenomenology is anticipated in materials without a magnetic field, such as WSe2_2/MoSe2_2, as a virtue of their non-trivial wavefunctions [60–62]. However, there are key distinctions due to finite dispersion and asymmetric electron-hole bands, and it remains to be seen if such effects preserve or diminish the geometric contribution.

We address these fundamental questions in this paper. We begin by deriving the phase stiffness of a general exciton Hamiltonian, separating it into conventional and geometric components. The latter arises primarily from the non-interacting electron/hole wavefunctions. To demonstrate this effect, we consider a realistic continuum model endowed with screened Coulomb interactions applicable to bilayer TMD devices and provide estimates for the geometric contribution to stiffness in MoTe2_2 homobilayers [63] as well as WSe2_2/MoSe2_2 heterobilayers [43]. Finally, we propose experimental setups that can serve as validation platforms for our theory.

Phase Stiffness.— Thermodynamic stability of a condensate depends on the free energy cost associated with spatial fluctuations in the phase, θ→θ(r)\theta\to\theta(\mathbf{r}), which is quantified as F=(1/2)Ds∫dr∣∇θ(r)∣2\mathcal{F} = (1/2)D_s \int d\mathbf{r}|\nabla\theta(\mathbf{r})|^2, where DsD_s represents the phase stiffness. For IECs, the phase stiffness is calculated as a linear response coefficient when equal and opposite vector potentials are applied to the two layers [37]. The antisymmetric vector potential couples symmetrically to the exciton, since electron and holes have opposite charges, inducing infinitesimal phase fluctuations in the condensate.

A general model Hamiltonian for the IEC is given by Hex=H0+Hint\mathcal{H}_{\mathrm{ex}}=\mathcal{H}_0+\mathcal{H}_{\mathrm{int}}, where H0=Ht+Hb\mathcal{H}_0=\mathcal{H}_t+\mathcal{H}_b describes the non-interacting properties with Bloch Hamiltonians {Hν}\{\mathcal{H}_\nu\} for the two layers ν={t,b}\nu=\{t,b\}, and Hint\mathcal{H}_{\mathrm{int}} is a density-density interaction that gives rise to excitons. We write

H0=∑kΨk†H0(k)Ψk\mathcal{H}_0=\sum_{\mathbf{k}}\Psi_{\mathbf{k}}^\dagger\mathcal{H}_0(\mathbf{k})\Psi_{\mathbf{k}}

where H0(k)=[Ht(k)⊕Hb(k)]+Vbτz\mathcal{H}_0(\mathbf{k})=[\mathcal{H}_t(\mathbf{k})\oplus\mathcal{H}_b(\mathbf{k})]+V_b\tau_z and Ψk\Psi_{\mathbf{k}} is a spinor that has internal labels (α,β)(\alpha,\beta) that are omitted for brevity along with layer label ν\nu: Ψk=(ck,α,t,⋯ ,ck,β,b,⋯ )T\Psi_{\mathbf{k}}=(c_{\mathbf{k},\alpha,t},\cdots,c_{\mathbf{k},\beta,b},\cdots)^{T}, and k∈BZ\mathbf{k}\in\mathrm{BZ}. The last term describes a bias voltage VbV_b that tunes the gap between the conduction and valence bands (see Figure 1) with τz\tau_z being the Pauli matrix in the layer subspace. With vanishing single-particle tunneling between the layers, the non-interacting model has a U(1)t×U(1)bU(1)_t\times U(1)_b symmetry that is spontaneously broken by the IEC.

Image not extracted

Figure 1. Device setup with the two layers, orange and blue, with electrons in one and holes in the other. The hBN spacer (green) suppresses single-particle tunneling between the layers. Bias Voltage, VbV_b, tunes the gap between lowest unoccupied band and highest occupied band. It is important that these are direct excitons. The bands are shifted horizontally to highlight the layers.

Density-density interactions remain unaffected by the external vector potential. Consequently, when equal and opposite A=Ax^\mathbf{A}=A\hat{x} are applied to the two layers, the current operators depend solely on the Bloch Hamiltonians H0(k,A)=Ht(k−eA/ℏ)⊕Hb(k+eA/ℏ)\mathcal{H}_0(\mathbf{k},\mathbf{A})=\mathcal{H}_t(\mathbf{k}-e\mathbf{A}/\hbar)\oplus\mathcal{H}_b(\mathbf{k}+e\mathbf{A}/\hbar). We assume spatially isotropic systems to suppress the tensor nature of currents and stiffness and leave the extension to anisotropic systems to App. A6. The current operator can be expressed as j=−δH0/δA=jP+jDj=-\delta\mathcal{H}_0/\delta A=j_P+j_D, where jP=(e/ℏ)∑kΨk†[τz2∂kH0(k)]Ψkj_P=(e/\hbar)\sum_{\mathbf{k}}\Psi_{\mathbf{k}}^\dagger[\tau_z^2\partial_k\mathcal{H}_0(\mathbf{k})]\Psi_{\mathbf{k}} and jD=−(e/ℏ)2∑kΨk†∂k2H0(k)Ψkj_D=-(e/\hbar)^2\sum_{\mathbf{k}}\Psi_{\mathbf{k}}^\dagger\partial_k^2\mathcal{H}_0(\mathbf{k})\Psi_{\mathbf{k}} are the paramagnetic and diamagnetic currents respectively. The stiffness is determined by the Kubo formula, given by Ds=−[⟨jD⟩−χjPjP(q⊥→0,ω=0)]/4D_s=-[\langle j_D\rangle-\chi_{j_Pj_P}(\mathbf{q}_\perp\to0,\omega=0)]/4, where χjPjP\chi_{j_Pj_P} is the longitudinal current-current correlator [57, 64, 65]. Calculating the stiffness requires a complete enumeration of the eigenstates of the full interacting Hamiltonian Hex\mathcal{H}_{\mathrm{ex}}. To make progress beyond the formal definition, we focus on mean-field theories where the interaction term breaks down into fermionic bilinears. We can then utilize the eigenspectrum Em,kE_{m,\mathbf{k}} and eigenstates ∣um,k⟩|u_{m,\mathbf{k}}\rangle of the mean-field Hamiltonian to express the stiffness as:

Ds=ℏ24e2A[∑k,mfm,k⟨um,k|∂k2H0(k)|um,k⟩−∑k,m,nfm,k−fn,kEn,k−Em,k∣⟨um,k|τz2∂kH0(k)|un,k⟩∣2](1)D_s=\frac{\hbar^2}{4e^2A}\left[\sum_{\mathbf{k},m}f_{m,\mathbf{k}}\left\langle u_{m,\mathbf{k}}\middle|\partial_k^2\mathcal{H}_0(\mathbf{k})\middle|u_{m,\mathbf{k}}\right\rangle-\sum_{\mathbf{k},m,n}\frac{f_{m,\mathbf{k}}-f_{n,\mathbf{k}}}{E_{n,\mathbf{k}}-E_{m,\mathbf{k}}}\left|\left\langle u_{m,\mathbf{k}}\middle|\tau_z^2\partial_k\mathcal{H}_0(\mathbf{k})\middle|u_{n,\mathbf{k}}\right\rangle\right|^2\right] \tag*{(1)}

where fm,k=f[Em,k]f_{m,\mathbf{k}}=f[E_{m,\mathbf{k}}] is the Fermi occupation factor and AA is the volume normalization factor which is the area of the sample in 2D. The stiffness measures coherence between transport in the two layers. In the absence of inter-layer tunneling, the only way to get finite stiffness is for the interaction to admit an off-diagonal exciton term Ψk†[τiφ^(k)]Ψk\Psi_{\mathbf{k}}^\dagger[\tau^i\hat{\varphi}(\mathbf{k})]\Psi_{\mathbf{k}} with i=x,yi=x,y within mean-field. The matrix function φ^(k)\hat{\varphi}(\mathbf{k}) is a mean-field ansatz that needs to be evaluated self-consistently.

Although the dependence of the phase stiffness on wavefunctions is evident in Eq. (1) from the matrix structure of the current operators, the geometric and energetic terms are intertwined with no clear route to separation. To tackle this challenge, we introduce a projected low-energy model [9], which is applicable when excitons predominantly arise from the lowest lying electron band (ee) and the highest hole band (hh)

Hex=∑kψk†(ϵe(k)−Σe(k)φ(k)φ(k)∗ϵh(k)+Σh(k))ψk(2)\mathcal{H}_{\mathrm{ex}}=\sum_{\mathbf{k}}\psi_{\mathbf{k}}^\dagger \begin{pmatrix} \epsilon_e(\mathbf{k})-\Sigma_e(\mathbf{k}) & \varphi(\mathbf{k})\\ \varphi(\mathbf{k})^* & \epsilon_h(\mathbf{k})+\Sigma_h(\mathbf{k}) \end{pmatrix} \psi_{\mathbf{k}} \tag*{(2)}

where ϵe/h\epsilon_{e/h} are the bare electron/hole dispersions, ψk=(ck,e,ck,h)T\psi_{\mathbf{k}}=(c_{\mathbf{k},e},c_{\mathbf{k},h})^T is the low-energy basis state, Σe/h(k)\Sigma_{e/h}(\mathbf{k}) are the self-energies (including VbV_b) and φ(k)\varphi(\mathbf{k}) is a shorthand for

φ(k)=∑α∈t,β∈b[Ut(k)]α,e∗φαβ(k)[Ub(k)]β,h.(3)\varphi(\mathbf{k})=\sum_{\substack{\alpha\in t,\beta\in b}}[U_t(\mathbf{k})]_{\alpha,e}^*\varphi_{\alpha\beta}(\mathbf{k})[U_b(\mathbf{k})]_{\beta,h}. \tag*{(3)}

A crucial aspect to highlight is that Ut(k)U_t(\mathbf{k}) and Ub(k)U_b(\mathbf{k}) are independent unitary matrices that diagonalize Ht(k)\mathcal{H}_t(\mathbf{k}) and Hb(k)\mathcal{H}_b(\mathbf{k}) respectively. The subscripts ee and hh indicate the electron and hole bands involved in the exciton pairing. It is important to note that, unlike in BCS theory where the redundancy in Nambu basis introduces a particle-hole symmetry in the BdG Hamiltonian, Ut(k)U_t(\mathbf{k}) and Ub(k)U_b(\mathbf{k}) do not have such constraints. Hence our framework is a generalization of phase stiffness, that is designed for WSe2_2/MoSe2_2 bilayers where the electron and hole bands originate from distinct materials.

We next assume ϵe(k)=−ϵh(k)=ϵ(k)\epsilon_e(\mathbf{k})=-\epsilon_h(\mathbf{k})=\epsilon(\mathbf{k}), while allowing Ut(k)U_t(\mathbf{k}) to differ from Ub(k)U_b(\mathbf{k}). This assumption is justified as the effect of asymmetric dispersion on exciton is a well understood textbook problem [66] that only complicates our analysis without offering additional insight (see also App. § 3.2). On the other hand, the presence of Ut(k)U_t(\mathbf{k}) and Ub(k)U_b(\mathbf{k}) has a non-trivial consequence for φ(k)\varphi(\mathbf{k}) in the presence of vector potential:

φ(k,A)≈φ(k)−eAℏP(k)−e2A22ℏ2D(k),(4)\varphi(\mathbf{k},\mathbf{A})\approx\varphi(\mathbf{k})-\frac{eA}{\hbar}\mathcal{P}(\mathbf{k})-\frac{e^2A^2}{2\hbar^2}\mathcal{D}(\mathbf{k}), \tag*{(4)}

where P(k)\mathcal{P}(\mathbf{k}) and D(k)\mathcal{D}(\mathbf{k}) are the paramagnetic and diamagnetic terms that involve derivatives of Uν(k)U_\nu(\mathbf{k}) (see App. A2 for the full expression). These terms are in addition to the energetic terms arising from ∂kϵ(k)\partial_k\epsilon(\mathbf{k}) and ∂k2ϵ(k)\partial_k^2\epsilon(\mathbf{k}) in the current operator and are key to our analysis. After performing a lengthy but straightforward calculation that is detailed in App. A2, we find two main contributions to the stiffness Ds=Dsc+DsgD_s=D_s^c+D_s^g with

Dsc=12A∑k∂ki2ϵ(k)vk2(5)D_s^c=\frac{1}{2A}\sum_{\mathbf{k}}\partial_{k_i}^2\epsilon(\mathbf{k})v_{\mathbf{k}}^2 \tag*{(5)}
Dsg=14A∑kG(k)E(k)+14A∑kRe⁡[P(k)φ(k)∗]2E(k)3(6)D_s^g=\frac{1}{4A}\sum_{\mathbf{k}}\frac{\mathcal{G}(\mathbf{k})}{E(\mathbf{k})}+\frac{1}{4A}\sum_{\mathbf{k}}\frac{\operatorname{Re}[\mathcal{P}(\mathbf{k})\varphi(\mathbf{k})^*]^2}{E(\mathbf{k})^3} \tag*{(6)}

where vk2=(1−ξ(k)/E(k))/2v_{\mathbf{k}}^2=(1-\xi(\mathbf{k})/E(\mathbf{k}))/2 is the momentum occupation factor with ξ(k)=ϵ(k)−Σ(k)\xi(\mathbf{k})=\epsilon(\mathbf{k})-\Sigma(\mathbf{k}), and E(k)=ξ(k)2+∣φ(k)∣2E(\mathbf{k})=\sqrt{\xi(\mathbf{k})^2+|\varphi(\mathbf{k})|^2} is the mean-field quasiparticle dispersion. The second part of the equation includes the geometric quantity G(k)=Re⁡[D(k)φ(k)∗]−∣P(k)∣2\mathcal{G}(\mathbf{k})=\operatorname{Re}[\mathcal{D}(\mathbf{k})\varphi(\mathbf{k})^*]-|\mathcal{P}(\mathbf{k})|^2. We note that all terms in eq. (6) are individually gauge invariant (see App. A4) and constitute the main result of our paper.

A few comments are in order. The geometric contribution in Eq. (6) arises explicitly from the Taylor expansion of the off-diagonal term φ(k,A)\varphi(\mathbf{k},A). If the wavefunctions were trivial, say independent of k\mathbf{k}, this contribution would vanish since P(k)=D(k)=0\mathcal{P}(\mathbf{k})=\mathcal{D}(\mathbf{k})=0. As one might expect, there are some connections with superconductivity as well. If the bands were particle-hole symmetric, eq. (2) would reduce to a BdG matrix and the function G(k)\mathcal{G}(\mathbf{k}) will become precisely the quantum metric, g(k)g(\mathbf{k}). This is not surprising since superfluid stiffness is known to be enhanced by quantum metric in superconductors within mean-field [2]. Since the particle-hole symmetry is not enforced in an exciton mean-field theory, our results present a generalization of quantum geometric phase stiffness beyond pairing between particle-hole symmetric states. We further note that orbital embedding modifies our calculation in a similar fashion as in superconductivity [11]. It enforces the minimal condition where the solution to gap equation is kept real in the presence of external vector potential [67]. Our result demonstrate a generalization of excitonic phase stiffness beyond particle-hole symmetry discussed in the context of quantum Hall bilayers [58].

Hubbard Model with tunable quantum metric.— To illustrate the phenomenology with a simple model, we consider an interlayer on-site orbital-diagonal Hubbard interaction Hint=V∑i,αn^i,α,αn^i,α,β\mathcal{H}_{\mathrm{int}}=V\sum_{i,\alpha}\hat{n}_{i,\alpha,\alpha}\hat{n}_{i,\alpha,\beta}. The labels ii, α\alpha pertaining to unit cell index and orbitals, are not crucial for our discussion as long as the interaction is interlayer in character. Details about the mean-field theory can be found in App. B1.

For the Bloch hamiltonian, we consider a tunable quantum metric model that has two orbitals on a square lattice with the Bloch Hamiltonian

H(ζ,k)=2t(2−pk)+tFcos⁡(ζpk)σx+tFsin⁡(ζpk)σy(7)H(\zeta,\mathbf{k})=2t(2-p_{\mathbf{k}})+t_F\cos(\zeta p_{\mathbf{k}})\sigma^x+t_F\sin(\zeta p_{\mathbf{k}})\sigma^y \tag*{(7)}

where the Pauli matrices σi\sigma^i act in orbital space, pkp_{\mathbf{k}} is a periodic function given by pk=cos⁡kx+cos⁡kyp_{\mathbf{k}}=\cos k_x+\cos k_y and the parameter ζ\zeta controls the quantum geometry by introducing long range hoppings [68]. More precisely, ζ\zeta is an overall scaling factor for the quantum geometric tensor Q(k,ζ)μν=(ζ2/4)sin⁡kμsin⁡kνQ(\mathbf{k},\zeta)_{\mu\nu}=(\zeta^2/4)\sin k_\mu\sin k_\nu, which is real because of inversion and time-reversal symmetries. The Berry curvature F(k)=−Im⁡[Q(k)xy/2F(\mathbf{k})=-\operatorname{Im}[Q(\mathbf{k})_{xy}/2 vanishes identically while the quantum metric gμν(k)=Re⁡[Qk]μνg_{\mu\nu}(\mathbf{k})=\operatorname{Re}[Q_{\mathbf{k}}]_{\mu\nu} is finite. Another convenient aspect of this model is that the band dispersions ϵ±,k=2t(2−pk)±tF\epsilon_{\pm,\mathbf{k}}=2t(2-p_{\mathbf{k}})\pm t_F are independent of ζ\zeta. This permits the use of ζ\zeta to tune quantum geometry without affecting the band dispersion, capturing the discussion around Eq. (6).

Symmetric bilayers, where both layers are described by the same parameter ζ\zeta, give rise to excitons (see Figure 2a) where the stiffness has a significant enhancement coming from the geometric term DsgD_s^g. The extra contribution is indeed proportional to the trace of the quantum metric in this particular case, as Eq. (6) simplifies to Dsg=(φ2/4A)∑kg(k)/E(k)D_s^g=(\varphi^2/4A)\sum_{\mathbf{k}}g(\mathbf{k})/E(\mathbf{k}) [69]. While the geometric contribution dominates at higher bias voltage ∣Vb∣|V_b|, it is important to note that a high ∣Vb∣|V_b| results in higher densities of electron-hole pairs and may ultimately drive the system to the exciton Mott transition [45–48]. The case of asymmetric wavefunctions follows along similar lines and is outlined in App. § 3.2 and Figure S3.

Image not extracted

Figure 2. a. Mean-field gap solution for the toy model with tunable quantum metric at fixed interaction U=6tU=6t. b. Mean-field phase stiffness. All energy scales, φ\varphi, DsD_s, VbV_b are plotted in units of single-particle hopping tt (see Eq. (58)). Subscripts cc and gg denote conventional and geometric contributions respectively.

Image not extracted

Figure S3. a) Non-interacting band structure (dashed) and interacting excitonic energy (solid). The color denotes the layer polarization showing the mixing between the different bands in the excitonic state. The inset shows the intralayer and interlayer Coulomb interaction as a function of q/qΛq/q_{\Lambda} in unit of Vq=0V_{q=0}. Notice that VqtbV^{tb}_{q} decays exponentially with qq. b) Gap function Tr⁡Δ†(k)Δ(k)\sqrt{\operatorname{Tr}\Delta^{\dagger}(k)\Delta(k)} denoting interlayer phase coherence. Calculations are performed for ΔE=(Δt+Δb−2Vb)/2=55meV\Delta E=(\Delta_{t}+\Delta_{b}-2V_{b})/2=55\mathrm{meV} where the single particle spectrum is gapped, vt=vb=2.19eVA˚v_{t}=v_{b}=2.19\mathrm{eV}\mathring{\mathrm{A}}, Δt=Δb=0.9eV\Delta_{t}=\Delta_{b}=0.9\mathrm{eV}, we set the screening length ξ=12nm\xi=12\mathrm{nm} and d=1nmd=1\mathrm{nm}. We fix the UV cutoff qΛa=2π/3q_{\Lambda}a=2\pi/3 corresponding to half-distance between the two corners KK and K′K^{\prime} of the TMD Brillouin zone. These parameters are the same ad those employed for the results on homobilayer reported in the maintext.

Continuum Model for TMDs.— With the formalism in place, we now aim to quantify the role of quantum geometric effects in TMD exciton bilayers. TMDs are semi-conductors with valley optical selection rules [70] and hence their minimal model is that of a gapped Dirac cone. We consider the continuum Hamiltonian introduced in ref. [71] with spin-valley locking. The continuum k.pk.p Hamiltonian for a given spin and valley, to first order in kk, reads

Hν(k)=Δνσz+vνk⋅σ,(8)H_{\nu}(\mathbf{k}) = \Delta_{\nu}\sigma_z + v_{\nu}\mathbf{k} \cdot\boldsymbol{\sigma}, \tag*{(8)}

where σi\sigma^i is the Pauli matrix in the internal sublattice degrees of freedom for the conduction and valence bands, and dispersion ϵkν±=±Δν2+vν2∣k∣2\epsilon_{\mathbf{k}\nu\pm} = \pm\sqrt{\Delta_{\nu}^{2} + v_{\nu}^{2}|\mathbf{k}|^{2}}. The two-band aspect of the model is justified for the valence band where the Ising spin-orbit coupling (SOC) gap is of the order of 180 meV in MoX2_2 and 440 meV in WX2_2 [72]. We also neglect the quadratic trigonal warping term and the quadratic particle-hole mass imbalance since it does not play an important role in our phenomenological discussion.

The geometric properties of the model are encoded in the form factors Fk,q=Uν†(k)Uν(k+q)F_{\mathbf{k},\mathbf{q}} = U_{\nu}^{\dagger}(\mathbf{k})U_{\nu}(\mathbf{k}+\mathbf{q}) which has non-trivial momentum dependence. In order to isolate the geometric contribution, we will compare our results with the gapped Dirac cone in eq. (8) to that of a parabolic band Hν(k)=±ϵkνσzH_{\nu}(\mathbf{k}) = \pm\epsilon_{\mathbf{k}\nu}\sigma_z with energies ϵkν≈Δν+k2/2mν\epsilon_{\mathbf{k}\nu} \approx\Delta_{\nu} + k^{2}/2m_{\nu} with mν=vν2/mνm_{\nu} = v_{\nu}^{2}/m_{\nu} and trivial form factor Fk,q=δk,qF_{\mathbf{k},\mathbf{q}} = \delta_{\mathbf{k},\mathbf{q}}.

We overlay this model with a gated Coulomb interaction

Hint=12A∑qΛ∑ν,ν′=t,bVqνν′n^q,νn^−q,ν′,(9)\mathcal{H}_{\mathrm{int}} = \frac{1}{2A}\sum_{\mathbf{q}}^{\Lambda}\sum_{\nu,\nu'=t,b} V_{\mathbf{q}}^{\nu\nu'}\hat{n}_{\mathbf{q},\nu}\hat{n}_{-\mathbf{q},\nu'}, \tag*{(9)}

where n^q,ν\hat{n}_{\mathbf{q},\nu} is the layer resolved density operator n^q,ν=∑kΛ∑a=K,K′ψk,aν†ψk+q,aν\hat{n}_{\mathbf{q},\nu} = \sum_{\mathbf{k}}^{\Lambda}\sum_{a=K,K'}\psi_{\mathbf{k},a\nu}^{\dagger}\psi_{\mathbf{k}+\mathbf{q},a\nu} and

Vqtt=Vqbb=e2ϵϵ0tanh⁡qξ2q,Vqtb=Vqbt=e−dqVqtt.(10)V_{\mathbf{q}}^{tt} = V_{\mathbf{q}}^{bb} = \frac{e^{2}}{\epsilon\epsilon_{0}}\frac{\tanh q\xi}{2q}, \qquad V_{\mathbf{q}}^{tb} = V_{\mathbf{q}}^{bt} = e^{-dq}V_{\mathbf{q}}^{tt}. \tag*{(10)}

Here dd is the inter-layer distance, Λ\Lambda is the UV cutoff, and ξ\xi is the screening length of the bilayer which is defined as the distance between the bilayer and the metallic gates. As representative values [41, 43], we set ξ=12\xi= 12 nm, ϵ≈6\epsilon\approx6 and d≈1d \approx1 nm. We focus on the regime of low-density of electron-hole pairs ne−h≤0.075n_{\mathrm{e-h}} \le0.075 to reduce screening effects leading to the exciton Mott transition [44–48]. We now proceed with a Hartree-Fock calculation. We find that while the Hartree term vanishes as a consequence of charge neutrality, the Fock term gives rise to a self energy correction to the single-particle Hamiltonian. The self-consistent equations for the Fock self-energy are solved employing an iterative scheme, described in App. § 3.3.

The Coulomb interaction induces intravalley excitons as well as charge transfer between the two layers ϕz=⟨τzσ0⟩=2ne−h\phi_z = \langle\tau^z\sigma^0\rangle= 2n_{\mathrm{e-h}}. The operators τ±σa\tau^{\pm}\sigma^a probe spontaneous interlayer coherence where τ±=(τx±iτy)/2\tau^{\pm} = (\tau^x \pm i\tau^y)/2 are raising/lowering operators in the layer degree of freedom and σa\sigma^a are Pauli matrices in the band space. A finite expectation value ⟨τ±σ0⟩≠0\langle\tau^{\pm}\sigma^0\rangle\ne0 corresponds to an interlayer intravalley exciton condensate that breaks the U(1)e/hU(1)_{\mathrm{e/h}} symmetry. The first panel of Figure 3 shows the evolution of ϕ=∑a=x,y,z⟨τxσa⟩2\phi= \sqrt{\sum_{a=x,y,z}\langle\tau^x\sigma^a\rangle^2} in the gauge where interlayer symmetry is broken along τx\tau^x as a function of VbV_b for the homobilayer (Figure 3a) and heterobilayer (Figure 3b) case. We notice that for the continuum model the energy VbV_b is given by Vb=(Δt+Δb−2Ez)/2V_b = (\Delta_t+\Delta_b-2E_z)/2 with EzE_z electric displacement field applied to the bilayer. Above a critical value of the electrostatic potential Vb∗≈70/75V_b^* \approx70/75 meV the system turns into a semiconductor where the energy gap EgapE_{\mathrm{gap}} grows linearly with the applied bias. Finally, the phase stiffness is calculated using (1) where the expectation value is taken with the Hartree-Fock wavefunctions. The results are shown in the right panel of Figure 3a–Figure 3b: blue data for parabolic bands and green ones for a gapped Dirac cone. We emphasize that the non-trivial structure of the wavefunctions appears in the geometric contribution of the superfluid stiffness, while it does not change the exciton binding energy (energy gap) and the size of the order parameter.

Image not extracted

Figure 3. We show the interlayer exciton coherence ϕ\phi and the phase stiffness DsD_s as a function of VbV_b. Green and Blue data corresponds to a gapped Dirac cone and parabolic bands with mass m=v2/Δm = v^2/\Delta, respectively. a: Homobilayer with velocity v=2.16v = 2.16 eVÅ, energy gap Δ=0.9\Delta= 0.9 eV relevant for MoTe2_2. b: Heterobilayer with velocity vt=2.2v_t = 2.2 eVÅ and vb=2.6v_b = 2.6 eVÅ, energy gap Δt=1.21\Delta_t = 1.21 eV and Δb=1.16\Delta_b = 1.16 eV representative values for WSe2_2/MoSe2_2. In both cases, we find spontaneous formation of excitons around Vb∗≈70/75V_b^* \approx70/75 meV. The calculations were done with gate distance ξ=12\xi= 12 nm and spacing between the two layers d=1d = 1 nm.

Remarkably, the geometric component increases DsD_s and, correspondingly, the BKT transition temperature (TBKTT_{\mathrm{BKT}}) related to the superfluid stiffness by the Nelson-Kosterlitz relation kBTBKT=π2(lim⁡T→TBKTDs(T))k_{\mathrm{B}}T_{\mathrm{BKT}} = \frac{\pi}{2}\left(\lim_{T \to T_{\mathrm{BKT}}}D_s(T)\right) [73]. Approximating Ds(T)D_s(T) with its T=0T = 0 value (from last panel in Figure 3b), we infer that geometric contribution increases TBKTT_{\mathrm{BKT}} roughly from 12 to 36 Kelvin at Vb≈40 meVV_b \approx40\,\mathrm{meV}. This threefold increase should be observable in experiments.

Conclusion.— Phase stiffness is significantly modified by the constituent electron and hole wavefunctions of the exciton. We investigated two limiting scenarios, one with local Hubbard interaction and the other with gated Coulomb interaction projected onto the low-energy bands, and remarkably, we find geometric contributions that play a vital role in either case. Our findings complement existing research on the effects of quantum geometry in exciton spectrum [74–76], exciton wavefunction [77] and its possible realizations in moiré heterostructures [69, 78, 79]. Importantly, we find a novel gauge-invariant quantum geometric quantity G(k)\mathcal{G}(\mathbf{k}) whose properties require further investigation.

Bilayer TMD devices offer an ideal platform to validate our predictions. Quite recently, there have been two reports on exciton superfluidity in WSe2_2/MoSe2_2 bilayer [55, 56]. The BKT temperature of 20 K appears to be consistent with the quantum geometric contribution, although a more sophisticated calculation is needed for a more quantitative prediction. Critical counterflow current from non-linear transport, similar to TBG superconductor [80], may also provide clues for the said geometric contribution.

Acknowledgment.— Research on geometric properties of exciton condensates is supported as part of Programmable Quantum Materials, an Energy Frontier Research Center funded by the U.S. Department of Energy (DOE), Office of Science, Basic Energy Sciences (BES), under award DE-SC0019443. We benefited from the 2023 Quantum Geometry Working Group meeting that took place at the Flatiron institute. We thank Päivi Törmä and Jonah Herzog-Arbeitman for their comments on orbital embeddings and minimal quantum metric. We acknowledge discussions with Kin Fai Mak, Pavel Volkov, Yongxin Zeng and Ilia Komissarov. D.G. and R.Q. acknowledge support from the Flatiron Institute, a division of the Simons Foundation.

References

Supplementary material for “Geometric Stiffness in Interlayer Exciton Condensates”

Nishchhal Verma; Daniele Guerci; Raquel Queiroz

Mean-field Phase Stiffness

Kubo formula

Exciton condensates exhibit the remarkable phenomenon of dissipationless counterflow transport when equal and opposite fields are applied to the two layers. The asymmetric field drives excitons in a coherent fashion. The response kernel to layer anti-symmetric vector potential is defined as [37]

jCF=−4e2ℏ2DsA(11)j_{\mathrm{CF}}=-\frac{4e^2}{\hbar^2}D_s A \tag*{(11)}

where DsD_s is the phase stiffness and A=Ax^\mathbf{A}=A\hat{x} is the external vector potential that is applied along +x^+\hat{x} in top layer and −x^-\hat{x} in bottom layer. As a matter of fact, this can be an in-plane magnetic field as A=zAx^\mathbf{A}=zA\hat{x} gives By=∂zAxB_y=\partial_z A_x where zz is the direction normal to the interface [81]. The stiffness is the weight of the delta function in counterflow conductivity Re⁡[σCF(ω)]=Dsδ(ω)+⋯\operatorname{Re}[\sigma_{\mathrm{CF}}(\omega)]=D_s\delta(\omega)+\cdots. By utilizing the Kramers-Kronig relation, DsD_s can be extracted from the 1/ω1/\omega tail of Im⁡[σCF(ω)]\operatorname{Im}[\sigma_{\mathrm{CF}}(\omega)], which follows Im⁡[σCF(ω)]∝Ds/ω\operatorname{Im}[\sigma_{\mathrm{CF}}(\omega)]\propto D_s/\omega. Stiffness controls the cost of phase fluctuations of the condensate and is formally defined by the Kubo formula [57]

Ds=−14(⟨jD⟩−χjPjP(q⊥=0,ω=0)).(12)D_s=-\frac{1}{4}\left(\langle j_D\rangle-\chi_{j_Pj_P}(\mathbf{q}_{\perp}=0,\omega=0)\right). \tag*{(12)}

The diamagnetic (jDj_D) and paramagnetic (jPj_P) currents are given by a Taylor expansion of the Hamiltonian

Hex(A)=∑kΨk†[(Ht(k−eA/ℏ)00Hb(k+eA/ℏ))+Vbτz]Ψk+Hint(13)\mathcal{H}_{\mathrm{ex}}(A)=\sum_{\mathbf{k}}\Psi_{\mathbf{k}}^\dagger\left[\begin{pmatrix}\mathcal{H}_t(\mathbf{k}-e\mathbf{A}/\hbar)&0\\0&\mathcal{H}_b(\mathbf{k}+e\mathbf{A}/\hbar)\end{pmatrix}+V_b\tau^z\right]\Psi_{\mathbf{k}}+\mathcal{H}_{\mathrm{int}} \tag*{(13)}
=Hex(0)−eAℏ∑kΨk†(∂kHt(k)00−∂kHb(k))Ψk+e2A22ℏ2∑kΨk†(∂k2Ht(k)00∂k2Hb(k))Ψk(14)=\mathcal{H}_{\mathrm{ex}}(0)-\frac{eA}{\hbar}\sum_{\mathbf{k}}\Psi_{\mathbf{k}}^\dagger\begin{pmatrix}\partial_k\mathcal{H}_t(\mathbf{k})&0\\0&-\partial_k\mathcal{H}_b(\mathbf{k})\end{pmatrix}\Psi_{\mathbf{k}}+\frac{e^2A^2}{2\hbar^2}\sum_{\mathbf{k}}\Psi_{\mathbf{k}}^\dagger\begin{pmatrix}\partial_k^2\mathcal{H}_t(\mathbf{k})&0\\0&\partial_k^2\mathcal{H}_b(\mathbf{k})\end{pmatrix}\Psi_{\mathbf{k}} \tag*{(14)}

which gives

jP=eℏ∑kΨk†(∂kHt(k)00−∂kHb(k))Ψk,jD=−e2ℏ2∑kΨk†(∂k2Ht(k)00∂k2Hb(k))Ψk.(15)j_P=\frac{e}{\hbar}\sum_{\mathbf{k}}\Psi_{\mathbf{k}}^\dagger\begin{pmatrix}\partial_k\mathcal{H}_t(\mathbf{k})&0\\0&-\partial_k\mathcal{H}_b(\mathbf{k})\end{pmatrix}\Psi_{\mathbf{k}},\qquad j_D=-\frac{e^2}{\hbar^2}\sum_{\mathbf{k}}\Psi_{\mathbf{k}}^\dagger\begin{pmatrix}\partial_k^2\mathcal{H}_t(\mathbf{k})&0\\0&\partial_k^2\mathcal{H}_b(\mathbf{k})\end{pmatrix}\Psi_{\mathbf{k}}. \tag*{(15)}

Next, in order to calculate the expectation values in Eq. (12), we assume a mean-field Hamiltonian with eigenspectrum Em,kE_{m,\mathbf{k}} and eigenstates ∣um,k⟩\lvert u_{m,\mathbf{k}}\rangle to find

⟨jD⟩=1A∑k,mf[Em,k]⟨um,k∣∂k2H0(k)∣um,k⟩,χjPjP=1A∑k,m≠nf[Em,k]−f[En,k]En,k−Em,k∣⟨um,k∣τz∂kH0(k)∣un,k⟩∣2(16)\langle j_D\rangle=\frac{1}{A}\sum_{\mathbf{k},m}f[E_{m,\mathbf{k}}]\langle u_{m,\mathbf{k}}\lvert\partial_k^2\mathcal{H}_0(\mathbf{k})\rvert u_{m,\mathbf{k}}\rangle,\qquad\chi_{j_Pj_P}=\frac{1}{A}\sum_{\mathbf{k},m\ne n}\frac{f[E_{m,\mathbf{k}}]-f[E_{n,\mathbf{k}}]}{E_{n,\mathbf{k}}-E_{m,\mathbf{k}}}\left\lvert\langle u_{m,\mathbf{k}}\lvert\tau^z\partial_k\mathcal{H}_0(\mathbf{k})\rvert u_{n,\mathbf{k}}\rangle\right\rvert^2 \tag*{(16)}

which gives Eq. (1) of the main text. The expression not only captures the conventional contribution coming from energies but also includes all geometric contributions that are hidden inside the matrix elements. These geometric factors arise from the underlying electron and hole wavefunctions that are not directly visible in this equation. We will now describe a projected mean-field model which systematically uncovers these geometric contributions.

Projected Phase Stiffness

If the electron and hole bands are sufficiently isolated from other bands, we can focus on the low-energy Hamiltonian and couple to the external vector potential

Hex(A)=∑kψk†(ϵ(k−eA/ℏ)−Σ(k)φ(k,A)ϕ φ(k,A)∗−(ϵ(k+eA/ℏ)−Σ(k)))ψk=∑kψk†Hex(A)ψk.(17)\mathcal{H}_{\mathrm{ex}}(A)=\sum_{\mathbf{k}}\psi_{\mathbf{k}}^\dagger\begin{pmatrix}\epsilon(\mathbf{k}-eA/\hbar)-\Sigma(\mathbf{k})&\varphi(\mathbf{k},A)\\ \phi\,\varphi(\mathbf{k},A)^*&-(\epsilon(\mathbf{k}+eA/\hbar)-\Sigma(\mathbf{k}))\end{pmatrix}\psi_{\mathbf{k}}=\sum_{\mathbf{k}}\psi_{\mathbf{k}}^\dagger H_{\mathrm{ex}}(A)\psi_{\mathbf{k}}. \tag*{(17)}

The vector potential dependence on the off-diagonal component is key to the geometric contribution

φ(k,A)=∑αβ[Ut(k−eA/ℏ)]e,α∗φαβ(k)[Ub(k+eA/ℏ)]β,h=φ(k)−eAℏP(k)−e2A22ℏ2D(k)(18)\varphi(\mathbf{k}, A) = \sum_{\alpha\beta} [U_t(\mathbf{k}-eA/\hbar)]^{*}_{e,\alpha}\varphi_{\alpha\beta}(\mathbf{k})[U_b(\mathbf{k}+eA/\hbar)]_{\beta,h} = \varphi(\mathbf{k})-\frac{eA}{\hbar}\mathcal{P}(\mathbf{k})-\frac{e^2A^2}{2\hbar^2}\mathcal{D}(\mathbf{k}) \tag*{(18)}

where P(k)\mathcal{P}(\mathbf{k}) and D(k)\mathcal{D}(\mathbf{k}) are defined as

P(k)=∑αβ∂k[Ut(k)]e,α∗φαβ(k)[Ub(k)]β,h−[Ut(k)]e,α∗φαβ(k)∂k[Ub(k)]β,h(19)\mathcal{P}(\mathbf{k}) = \sum_{\alpha\beta} \partial_k[U_t(\mathbf{k})]^{*}_{e,\alpha}\varphi_{\alpha\beta}(\mathbf{k})[U_b(\mathbf{k})]_{\beta,h} - [U_t(\mathbf{k})]^{*}_{e,\alpha}\varphi_{\alpha\beta}(\mathbf{k})\partial_k[U_b(\mathbf{k})]_{\beta,h} \tag*{(19)}
D(k)=∑αβ2∂k[Ut(k)]e,α∗φαβ(k)∂k[Ub(k)]β,h−∂k2[Ut(k)]e,α∗φαβ(k)[Ub(k)]β,h−[Ut(k)]e,α∗φαβ(k)∂k2[Ub(k)]β,h.(20)\mathcal{D}(\mathbf{k}) = \sum_{\alpha\beta} 2\partial_k[U_t(\mathbf{k})]^{*}_{e,\alpha}\varphi_{\alpha\beta}(\mathbf{k})\partial_k[U_b(\mathbf{k})]_{\beta,h} - \partial_k^2[U_t(\mathbf{k})]^{*}_{e,\alpha}\varphi_{\alpha\beta}(\mathbf{k})[U_b(\mathbf{k})]_{\beta,h} - [U_t(\mathbf{k})]^{*}_{e,\alpha}\varphi_{\alpha\beta}(\mathbf{k})\partial_k^2[U_b(\mathbf{k})]_{\beta,h}. \tag*{(20)}

Moving onto the stiffness calculation, we will use the thermodynamic definition instead of using the Kubo formula. Focusing on T=0T=0 and assuming that the mean-field band is completely gapped (even for Hex(k,A)H_{\mathrm{ex}}(\mathbf{k}, A)), the stiffness is given by

Ds=ℏ24e2(∂2∂A2F(A))=ℏ24e2∑k∂2∂A2E−(k,A)(21)D_s = \frac{\hbar^2}{4e^2}\left(\frac{\partial^2}{\partial A^2}\mathcal{F}(A)\right) = \frac{\hbar^2}{4e^2}\sum_{\mathbf{k}}\frac{\partial^2}{\partial A^2}\mathcal{E}_{-}(\mathbf{k}, A) \tag*{(21)}

where E−(k,A)\mathcal{E}_{-}(\mathbf{k}, A) is the lower band corresponding to Hex(k,A)H_{\mathrm{ex}}(\mathbf{k}, A), given by

E±(k,A)=ξ−(k,A)±ξ+(k,A)2+∣φ(k,A)∣2(22)\mathcal{E}_{\pm}(\mathbf{k}, A) = \xi_{-}(\mathbf{k}, A) \pm\sqrt{\xi_{+}(\mathbf{k}, A)^2 + |\varphi(\mathbf{k}, A)|^2} \tag*{(22)}

where

ξ±(k,A)=12[(ϵ(k−eA/ℏ)−Σ(k))±(ϵ(k+eA/ℏ)−Σ(k))](23)\xi_{\pm}(\mathbf{k}, A) = \frac{1}{2}\left[(\epsilon(\mathbf{k}-eA/\hbar)-\Sigma(\mathbf{k})) \pm(\epsilon(\mathbf{k}+eA/\hbar)-\Sigma(\mathbf{k}))\right] \tag*{(23)}

that has convenient properties like

ξ−(k,0)=0,ξ+(k,0)=ϵ(k)−Σ(k)≡ξ(k)(24)\xi_{-}(\mathbf{k},0)=0,\qquad\xi_{+}(\mathbf{k},0)=\epsilon(\mathbf{k})-\Sigma(\mathbf{k})\equiv\xi(\mathbf{k}) \tag*{(24)}
∂Aξ−(k,A)∣A=0=−eℏ∂kϵ(k),∂Aξ+(k,A)∣A=0=0(25)\partial_A\xi_{-}(\mathbf{k},A)|_{A=0}=-\frac{e}{\hbar}\partial_k\epsilon(\mathbf{k}),\qquad\partial_A\xi_{+}(\mathbf{k},A)|_{A=0}=0 \tag*{(25)}
∂A2ξ−(k,A)∣A=0=0,∂A2ξ+(k,A)∣A=0=e2ℏ2∂k2ϵ(k).(26)\partial_A^2\xi_{-}(\mathbf{k},A)|_{A=0}=0,\qquad\partial_A^2\xi_{+}(\mathbf{k},A)|_{A=0}=\frac{e^2}{\hbar^2}\partial_k^2\epsilon(\mathbf{k}). \tag*{(26)}

We are now ready to evaluate DsD_s:

Ds=ℏ24e2∂2∂A2(∑kE−(k,A))∣A=0=ℏ24e2∂2∂A2(∑k(ξ−(k,A)−ξ+(k,A)2+∣φ(k,A)∣2))∣A=0(27)D_s = \frac{\hbar^2}{4e^2}\frac{\partial^2}{\partial A^2}\left(\sum_{\mathbf{k}}\mathcal{E}_{-}(\mathbf{k},A)\right)\bigg|_{A=0} = \frac{\hbar^2}{4e^2}\frac{\partial^2}{\partial A^2}\left(\sum_{\mathbf{k}}\left(\xi_{-}(\mathbf{k},A)-\sqrt{\xi_{+}(\mathbf{k},A)^2+|\varphi(\mathbf{k},A)|^2}\right)\right)\bigg|_{A=0} \tag*{(27)}
=ℏ24e2∑k∂A2ξ−(k,A)∣A=0−∂2∂A2(ξ+(k,A)2+∣φ(k,A)∣2)∣A=0(28)= \frac{\hbar^2}{4e^2}\sum_{\mathbf{k}}\partial_A^2\xi_{-}(\mathbf{k},A)\big|_{A=0}-\frac{\partial^2}{\partial A^2}\left(\sqrt{\xi_{+}(\mathbf{k},A)^2+|\varphi(\mathbf{k},A)|^2}\right)\bigg|_{A=0} \tag*{(28)}
=−14∑k∂k2ϵ(k)ξ(k)E(k)+ℏ28e2∑k−∂A2∣φ(k,A)∣2E(k)+ℏ24e2∑k(∂A∣φ(k,A)∣2/2)2E(k)3.(29)= -\frac{1}{4}\sum_{\mathbf{k}}\partial_k^2\epsilon(\mathbf{k})\frac{\xi(\mathbf{k})}{E(\mathbf{k})}+\frac{\hbar^2}{8e^2}\sum_{\mathbf{k}}\frac{-\partial_A^2|\varphi(\mathbf{k},A)|^2}{E(\mathbf{k})}+\frac{\hbar^2}{4e^2}\sum_{\mathbf{k}}\frac{\left(\partial_A|\varphi(\mathbf{k},A)|^2/2\right)^2}{E(\mathbf{k})^3}. \tag*{(29)}

Using the fact that ∑k∂k2ϵ(k)=0\sum_{\mathbf{k}}\partial_k^2\epsilon(\mathbf{k})=0, we add and subtract this term to write

Ds=14∑k∂k2ϵ(k)[1−ξ(k)E(k)]+ℏ28e2∑k−∂A2∣φ(k,A)∣2E(k)+ℏ24e2∑k(∂A∣φ(k,A)∣2/2)2E(k)3(30)D_s = \frac{1}{4}\sum_{\mathbf{k}}\partial_k^2\epsilon(\mathbf{k})\left[1-\frac{\xi(\mathbf{k})}{E(\mathbf{k})}\right]+\frac{\hbar^2}{8e^2}\sum_{\mathbf{k}}\frac{-\partial_A^2|\varphi(\mathbf{k},A)|^2}{E(\mathbf{k})}+\frac{\hbar^2}{4e^2}\sum_{\mathbf{k}}\frac{\left(\partial_A|\varphi(\mathbf{k},A)|^2/2\right)^2}{E(\mathbf{k})^3} \tag*{(30)}

where we can further use

∣φ(k,A)∣2=∣φ(k)∣2−2eAℏRe⁡[[P(k)φ(k)∗]−e2A2ℏ2(Re⁡[[D(k)φ(k)∗]−∣P(k)∣2)(31)|\varphi(\mathbf{k},A)|^2=|\varphi(\mathbf{k})|^2-2\frac{eA}{\hbar}\operatorname{Re}[[\mathcal{P}(\mathbf{k})\varphi(\mathbf{k})^*]-\frac{e^2A^2}{\hbar^2}\left(\operatorname{Re}[[\mathcal{D}(\mathbf{k})\varphi(\mathbf{k})^*]-|\mathcal{P}(\mathbf{k})|^2\right) \tag*{(31)}

to arrive at the result quoted in the main text

Ds=12∑k∂k2ϵ(k)vk2+14∑kRe⁡[D(k)φ(k)∗]−∣P(k)∣2E(k)+14∑kRe⁡[P(k)φ(k)∗]2E(k)3.(32)D_s = \frac{1}{2}\sum_{\mathbf{k}}\partial_k^2\epsilon(\mathbf{k})v_{\mathbf{k}}^2+\frac{1}{4}\sum_{\mathbf{k}}\frac{\operatorname{Re}[\mathcal{D}(\mathbf{k})\varphi(\mathbf{k})^*]-|\mathcal{P}(\mathbf{k})|^2}{E(\mathbf{k})}+\frac{1}{4}\sum_{\mathbf{k}}\frac{\operatorname{Re}[\mathcal{P}(\mathbf{k})\varphi(\mathbf{k})^*]^2}{E(\mathbf{k})^3}. \tag*{(32)}

Reduction to Quantum Metric

We can validate our calculations by making two assumptions: first, ϕ^(k)=ϕ1\hat{\phi}(\mathbf{k})=\phi\mathbf{1}, ϕ∈R\phi\in\mathbb{R} and second, the two layers are identical, giving [Ut(k)]e,α=[Ub(k)]h,α ∀ α,k[U_t(\mathbf{k})]_{e,\alpha}=[U_b(\mathbf{k})]_{h,\alpha}\ \forall\ \alpha,\mathbf{k}. We can express the matrix elements in a more concise form (dropping e,he,h labels):

φ(k)=ϕ,P(k)=2ϕ∑α∂kUα∗(k)Uα(k)≡2ϕ⟨∂kuk∣uk⟩,D(k)=4ϕ∑α∂kUα∗(k)∂kUα(k)≡4ϕ⟨∂kuk∣∂kuk⟩(33)\varphi(\mathbf{k})=\phi,\quad\mathcal{P}(\mathbf{k})=2\phi\sum_{\alpha}\partial_k U_\alpha^*(\mathbf{k})U_\alpha(\mathbf{k})\equiv2\phi\langle\partial_k u_k|u_k\rangle,\quad\mathcal{D}(\mathbf{k})=4\phi\sum_{\alpha}\partial_k U_\alpha^*(\mathbf{k})\partial_k U_\alpha(\mathbf{k})\equiv4\phi\langle\partial_k u_k|\partial_k u_k\rangle \tag*{(33)}

and the resulting geometric contribution to stiffness greatly simplifies

Dsg=14A∑k1E(k)(Re⁡[D(k)φ(k)∗]−∣Im⁡[P(k)]∣2)=1A∑kϕ2E(k)(Re⁡[⟨∂kuk∣∂kuk⟩]−∣⟨∂kuk∣uk⟩∣2)(34)D_s^g=\frac{1}{4A}\sum_{\mathbf{k}}\frac{1}{E(\mathbf{k})}\left(\operatorname{Re}[\mathcal{D}(\mathbf{k})\varphi(\mathbf{k})^*]-|\operatorname{Im}[\mathcal{P}(\mathbf{k})]|^2\right)=\frac{1}{A}\sum_{\mathbf{k}}\frac{\phi^2}{E(\mathbf{k})}\left(\operatorname{Re}[\langle\partial_k u_k|\partial_k u_k\rangle]-|\langle\partial_k u_k|u_k\rangle|^2\right) \tag*{(34)}
=1A∑kϕ2E(k)Re⁡[Q(k)].(35)=\frac{1}{A}\sum_{\mathbf{k}}\frac{\phi^2}{E(\mathbf{k})}\operatorname{Re}[\mathcal{Q}(\mathbf{k})]. \tag*{(35)}

Here Re⁡[Q(k)]=g(k)\operatorname{Re}[\mathcal{Q}(\mathbf{k})]=g(\mathbf{k}) is the quantum metric of the electron (or hole) band.

Gauge Invariance

There is a gauge redundancy in non-interacting electron/hole wavefunctions since Bloch wavefunctions are only defined upto a phase. Physical quantities must be invariant under such gauge transformations. Specifically, introducing phases in the wavefunctions should not modify any observable

Ut(k)α,e→eiϕt(k)Ut(k)α,e,Ub(k)β,h→eiϕb(k)Ub(k)β,h(36)U_t(\mathbf{k})_{\alpha,e}\to e^{i\phi_t(\mathbf{k})}U_t(\mathbf{k})_{\alpha,e},\quad U_b(\mathbf{k})_{\beta,h}\to e^{i\phi_b(\mathbf{k})}U_b(\mathbf{k})_{\beta,h} \tag*{(36)}

where ϕν(k)\phi_\nu(\mathbf{k}) are real but arbitrary functions in the BZ. It turns out that φ(k)\varphi(\mathbf{k}), P(k)\mathcal{P}(\mathbf{k}) and D(k)\mathcal{D}(\mathbf{k}) are sensitive to the gauge choice

φ(k)→e−i(ϕt(k)−ϕb(k))φ(k)(37)\varphi(\mathbf{k})\to e^{-i(\phi_t(\mathbf{k})-\phi_b(\mathbf{k}))}\varphi(\mathbf{k}) \tag*{(37)}
P(k)→e−i(ϕt(k)−ϕb(k))[P(k)−iφ(k)[∂kϕt(k)+∂kϕb(k)]](38)\mathcal{P}(\mathbf{k})\to e^{-i(\phi_t(\mathbf{k})-\phi_b(\mathbf{k}))}\left[\mathcal{P}(\mathbf{k})-i\varphi(\mathbf{k})[\partial_k\phi_t(\mathbf{k})+\partial_k\phi_b(\mathbf{k})]\right] \tag*{(38)}
D(k)→e−i(ϕt(k)−ϕb(k))[D(k)−φ(k)[∂kϕt(k)+∂kϕb(k)]2+2i[∂kϕt(k)+∂kϕb(k)]P(k)].(39)\mathcal{D}(\mathbf{k})\to e^{-i(\phi_t(\mathbf{k})-\phi_b(\mathbf{k}))}\left[\mathcal{D}(\mathbf{k})-\varphi(\mathbf{k})[\partial_k\phi_t(\mathbf{k})+\partial_k\phi_b(\mathbf{k})]^2+2i[\partial_k\phi_t(\mathbf{k})+\partial_k\phi_b(\mathbf{k})]\mathcal{P}(\mathbf{k})\right]. \tag*{(39)}

However, the combinations that enter geometric contributions to phase stiffness are gauge invariant. Eq. (31) offers a simple proof. The left-hand side is gauge invariant and hence the right-hand side has to be gauge invariant for arbitrary expansions in AA.

It can also be checked by an explicit calculation. The first term, Re⁡[P(k)φ(k)∗]\operatorname{Re}[\mathcal{P}(\mathbf{k})\varphi(\mathbf{k})^*], gets an additional term which is purely imaginary

P(k)φ(k)∗→P(k)φ(k)∗−i∣φ(k)∣2[∂kϕt(k)−∂kϕb(k)](40)\mathcal{P}(\mathbf{k})\varphi(\mathbf{k})^*\to\mathcal{P}(\mathbf{k})\varphi(\mathbf{k})^*-i|\varphi(\mathbf{k})|^2[\partial_k\phi_t(\mathbf{k})-\partial_k\phi_b(\mathbf{k})] \tag*{(40)}

and hence Re⁡[P(k)φ(k)∗]\operatorname{Re}[\mathcal{P}(\mathbf{k})\varphi(\mathbf{k})^*] is gauge invariant. The second term follows along similar lines

D(k)φ(k)∗→D(k)φ(k)∗−∣φ(k)∣2[∂kϕt(k)+∂kϕb(k)]2+2iP(k)φ(k)∗[∂kϕt(k)+∂kϕb(k)](41)\mathcal{D}(\mathbf{k})\varphi(\mathbf{k})^*\to\mathcal{D}(\mathbf{k})\varphi(\mathbf{k})^*-|\varphi(\mathbf{k})|^2[\partial_k\phi_t(\mathbf{k})+\partial_k\phi_b(\mathbf{k})]^2+2i\mathcal{P}(\mathbf{k})\varphi(\mathbf{k})^*[\partial_k\phi_t(\mathbf{k})+\partial_k\phi_b(\mathbf{k})] \tag*{(41)}
∣P(k)∣2→∣P(k)∣2+∣φ(k)∣2[∂kϕt(k)+∂kϕb(k)]2−2Im⁡[P(k)φ(k)∗][∂kϕt(k)+∂kϕb(k)](42)|\mathcal{P}(\mathbf{k})|^2\to|\mathcal{P}(\mathbf{k})|^2+|\varphi(\mathbf{k})|^2[\partial_k\phi_t(\mathbf{k})+\partial_k\phi_b(\mathbf{k})]^2-2\operatorname{Im}[\mathcal{P}(\mathbf{k})\varphi(\mathbf{k})^*][\partial_k\phi_t(\mathbf{k})+\partial_k\phi_b(\mathbf{k})] \tag*{(42)}

where the last two terms exactly cancel and Re⁡[D(k)φ(k)∗]−∣P(k)∣2\operatorname{Re}[\mathcal{D}(\mathbf{k})\varphi(\mathbf{k})^*]-|\mathcal{P}(\mathbf{k})|^2 is gauge invariant.

Orbital Embedding

In addition to the gauge freedom mentioned above, there is an another transformation that changes the wavefunctions, but does not affect the energies

Ut(k)α,e→eiϕt,α(k)Ut(k)α,e,Ub(k)β,h→eiϕb,β(k)Ub(k)β,h(43)U_t(\mathbf{k})_{\alpha,e}\to e^{i\phi_{t,\alpha}(\mathbf{k})}U_t(\mathbf{k})_{\alpha,e},\quad U_b(\mathbf{k})_{\beta,h}\to e^{i\phi_{b,\beta}(\mathbf{k})}U_b(\mathbf{k})_{\beta,h} \tag*{(43)}

where the phases depend on orbital index α\alpha (and β\beta). The transformation corresponds to physically moving the orbitals within the unit cell while preserving all hopping amplitudes. It changes Berry curvature [82], quantum metric [11] and the geometric contribution to stiffness. It has been argued that orbital embedding should not change the phase stiffness [67] and the correct prescription is discussed in detail in ref. [11]. The gauge physically corresponds to orbital locations that keep order parameter solutions real in the presence of vector potential.

Spatially Anisotropic Systems

Underneath all our discussion so far, we have assumed isotropic systems where the vector potential is applied along x^\hat{x} with A=Ax^\mathbf{A}=A\hat{x} and the longitudinal response is calculated along x^\hat{x} while suppressing all spatial indices. In this section, we will outline the extension to spatially anisotropic systems. Introducing indices i,j={x,y}i,j=\{x,y\}, we find that the phase stiffness in Eq. (16) becomes a tensor

[Ds]ij=−14A(⟨[jD]ij⟩−χ[jP]i[jP]j(q⊥=0,ω=0))(44)[D_s]_{ij}=-\frac{1}{4A}\left(\langle[j_D]_{ij}\rangle-\chi_{[j_P]_i[j_P]_j}(\mathbf{q}_{\perp}=0,\omega=0)\right) \tag*{(44)}
=ℏ24e2A[∑k,mf[Em,k]⟨um,k∣[j^D(k)]ij∣um,k⟩−∑k,m≠nf[Em,k]−f[En,k]En,k−Em,k⟨um,k∣[j^P(k)]i∣un,k⟩⟨un,k∣[j^P(k)]j∣um,k⟩](45)=\frac{\hbar^2}{4e^2A}\left[\sum_{\mathbf{k},m}f[E_{m,\mathbf{k}}]\left\langle u_{m,\mathbf{k}}\left|[\hat{j}_D(\mathbf{k})]_{ij}\right|u_{m,\mathbf{k}}\right\rangle -\sum_{\mathbf{k},m\ne n}\frac{f[E_{m,\mathbf{k}}]-f[E_{n,\mathbf{k}}]}{E_{n,\mathbf{k}}-E_{m,\mathbf{k}}} \left\langle u_{m,\mathbf{k}}\left|[\hat{j}_P(\mathbf{k})]_i\right|u_{n,\mathbf{k}}\right\rangle \left\langle u_{n,\mathbf{k}}\left|[\hat{j}_P(\mathbf{k})]_j\right|u_{m,\mathbf{k}}\right\rangle\right] \tag*{(45)}

where the current operators are

[j^P(k)]i=eℏ(∂kiHt(k)00−∂kiHb(k)),[j^D(k)]ij=−e2ℏ2(∂ki∂kjHt(k)00∂ki∂kjHb(k)).(46)[\hat{j}_P(\mathbf{k})]_i=\frac{e}{\hbar}\begin{pmatrix}\partial_{k_i}H_t(\mathbf{k})&0\\0&-\partial_{k_i}H_b(\mathbf{k})\end{pmatrix},\qquad [\hat{j}_D(\mathbf{k})]_{ij}=-\frac{e^2}{\hbar^2}\begin{pmatrix}\partial_{k_i}\partial_{k_j}H_t(\mathbf{k})&0\\0&\partial_{k_i}\partial_{k_j}H_b(\mathbf{k})\end{pmatrix}. \tag*{(46)}

The projected stiffness in Eq. (32) follows along similar lines

[Ds]ij=14A∑k[∂ki∂kjϵ(k)(1−ξ(k)E(k))+∑k[G(k)]ijE(k)+∑kRe⁡[[P(k)]iφ(k)∗]Re⁡[[P(k)]jφ(k)∗]E(k)2](47)[D_s]_{ij}=\frac{1}{4A}\sum_{\mathbf{k}}\left[\partial_{k_i}\partial_{k_j}\epsilon(\mathbf{k})\left(1-\frac{\xi(\mathbf{k})}{E(\mathbf{k})}\right)+\sum_{\mathbf{k}}\frac{[G(\mathbf{k})]_{ij}}{E(\mathbf{k})}+\sum_{\mathbf{k}}\frac{\operatorname{Re}[[P(\mathbf{k})]_i\varphi(\mathbf{k})^*]\operatorname{Re}[[P(\mathbf{k})]_j\varphi(\mathbf{k})^*]}{E(\mathbf{k})^2}\right] \tag*{(47)}

where [G(k)]ij=Re⁡[D(k)]ijφ(k)∗−[P(k)]i[P(k)]j∗[G(\mathbf{k})]_{ij}=\operatorname{Re}[D(\mathbf{k})]_{ij}\varphi(\mathbf{k})^*-[P(\mathbf{k})]_i[P(\mathbf{k})]_j^* and the functions [P(k)]i[P(\mathbf{k})]_i, [D(k)]ij[D(\mathbf{k})]_{ij} are

[P(k)]i=∑αβ[∂ki[Ut(k)]e,α∗φαβ(k)[Ub(k)]β,h−[Ut(k)]e,α∗φαβ(k)∂ki[Ub(k)]β,h](48)[P(\mathbf{k})]_i=\sum_{\alpha\beta}\left[\partial_{k_i}[U_t(\mathbf{k})]_{e,\alpha}^*\varphi_{\alpha\beta}(\mathbf{k})[U_b(\mathbf{k})]_{\beta,h}-[U_t(\mathbf{k})]_{e,\alpha}^*\varphi_{\alpha\beta}(\mathbf{k})\partial_{k_i}[U_b(\mathbf{k})]_{\beta,h}\right] \tag*{(48)}
[D(k)]ij=∑αβ[∂ki[Ut(k)]e,α∗φαβ(k)∂kj[Ub(k)]β,h+∂kj[Ut(k)]e,α∗φαβ(k)∂ki[Ub(k)]β,h−∂ki∂kj[Ut(k)]e,α∗φαβ(k)[Ub(k)]β,h−[Ut(k)]e,α∗φαβ(k)∂ki∂kj[Ub(k)]β,h].(49)[D(\mathbf{k})]_{ij}=\sum_{\alpha\beta}\left[\partial_{k_i}[U_t(\mathbf{k})]_{e,\alpha}^*\varphi_{\alpha\beta}(\mathbf{k})\partial_{k_j}[U_b(\mathbf{k})]_{\beta,h}+\partial_{k_j}[U_t(\mathbf{k})]_{e,\alpha}^*\varphi_{\alpha\beta}(\mathbf{k})\partial_{k_i}[U_b(\mathbf{k})]_{\beta,h} -\partial_{k_i}\partial_{k_j}[U_t(\mathbf{k})]_{e,\alpha}^*\varphi_{\alpha\beta}(\mathbf{k})[U_b(\mathbf{k})]_{\beta,h}-[U_t(\mathbf{k})]_{e,\alpha}^*\varphi_{\alpha\beta}(\mathbf{k})\partial_{k_i}\partial_{k_j}[U_b(\mathbf{k})]_{\beta,h}\right]. \tag*{(49)}

Lastly, the BKT temperature is determined by the determinant of the stiffness tensor kBTc=(π/2)det⁡[Ds(Tc)]ijk_BT_c=(\pi/2)\sqrt{\det[D_s(T_c)]_{ij}}.

Mean-Field theory for the Hubbard interaction

We consider an interlayer on-site orbital-diagonal Hubbard interaction Hint=V∑i,αn^i,α,tn^i,α,b\mathcal{H}_{\mathrm{int}}=V\sum_{i,\alpha}\hat{n}_{i,\alpha,t}\hat{n}_{i,\alpha,b}. The labels i,αi,\alpha pertaining to unit cell index and orbitals, are not crucial for our discussion as long as the interaction is interlayer in character. We decouple the four-fermion interaction with a spatially uniform ansatz

Vαβci,α,t†ci,α,tci,β,b†ci,β,b≈Vαβnα,tci,β,b†ci,β,b+Vαβci,α,t†ci,α,tnβ,b+ϕαβci,β,b†ci,α,t+ϕαβ†ci,α,t†ci,β,b(50)V_{\alpha\beta}c^\dagger_{i,\alpha,t}c_{i,\alpha,t}c^\dagger_{i,\beta,b}c_{i,\beta,b}\approx V_{\alpha\beta}n_{\alpha,t}c^\dagger_{i,\beta,b}c_{i,\beta,b}+V_{\alpha\beta}c^\dagger_{i,\alpha,t}c_{i,\alpha,t}n_{\beta,b}+\phi_{\alpha\beta}c^\dagger_{i,\beta,b}c_{i,\alpha,t}+\phi_{\alpha\beta}^\dagger c^\dagger_{i,\alpha,t}c_{i,\beta,b} \tag*{(50)}

where {nα,ν}\{n_{\alpha,\nu}\} and {ϕαβ}\{\phi_{\alpha\beta}\} are given by

nα,ν=1A∑i⟨ci,α,ν†ci,α,ν⟩=1N∑k⟨ck,α,ν†ck,α,ν⟩,ϕαβ=−VαβA∑i⟨ci,α,t†ci,β,b⟩=−VαβN∑k⟨ck,α,t†ck,β,b⟩.(51)n_{\alpha,\nu}=\frac{1}{A}\sum_i\langle c^\dagger_{i,\alpha,\nu}c_{i,\alpha,\nu}\rangle=\frac{1}{N}\sum_{\mathbf{k}}\langle c^\dagger_{\mathbf{k},\alpha,\nu}c_{\mathbf{k},\alpha,\nu}\rangle,\qquad \phi_{\alpha\beta}=-\frac{V_{\alpha\beta}}{A}\sum_i\langle c^\dagger_{i,\alpha,t}c_{i,\beta,b}\rangle=-\frac{V_{\alpha\beta}}{N}\sum_{\mathbf{k}}\langle c^\dagger_{\mathbf{k},\alpha,t}c_{\mathbf{k},\beta,b}\rangle. \tag*{(51)}

Here A=NΩA=N\Omega is the area of the system, NN is the number of unit cells and Ω\Omega is the area of unit cell. This pre-factor serves as a normalization factor. When we substitute this result back into the full Hamiltonian, we obtain the Hartree shifts

[H~(ζt,k)]αα′=[H(ζt,k)]αα′+δα,α′Vαβnβ,b,[H~(ζb,k)]ββ′=[H(ζb,k)]ββ′+δβ,β′Vαβnα,t(52)[\widetilde{H}(\zeta_t,\mathbf{k})]_{\alpha\alpha'}=[H(\zeta_t,\mathbf{k})]_{\alpha\alpha'}+\delta_{\alpha,\alpha'}V_{\alpha\beta}n_{\beta,b},\qquad [\widetilde{H}(\zeta_b,\mathbf{k})]_{\beta\beta'}=[H(\zeta_b,\mathbf{k})]_{\beta\beta'}+\delta_{\beta,\beta'}V_{\alpha\beta}n_{\alpha,t} \tag*{(52)}

in addition to the interlayer exciton term ∝ϕαβci,β,b†ci,α,t\propto\phi_{\alpha\beta}c^{\dagger}_{i,\beta,b}c_{i,\alpha,t}. Combining the two, we get the mean-field Hamiltonian

HMF=∑kΨk†HMF(k)Ψk\mathcal{H}_{\mathrm{MF}}=\sum_{\mathbf{k}}\Psi_{\mathbf{k}}^{\dagger}H_{\mathrm{MF}}(\mathbf{k})\Psi_{\mathbf{k}} where

HMF(k)=[H~(ζt,k)⊕(−H~(ζb,k))]+τzVb+τxϕ.(53)H_{\mathrm{MF}}(\mathbf{k})=\left[\tilde{H}(\zeta_t,\mathbf{k})\oplus\left(-\tilde{H}(\zeta_b,\mathbf{k})\right)\right]+\tau^{z}V_b+\tau^{x}\phi. \tag*{(53)}

The last term breaks the U(1)t×U(1)bU(1)_t\times U(1)_b symmetry as discussed before.

To establish the notation, let us review the simplest model consisting of two identical bands with mass mm: one particle-like and one hole-like, that belong to two different layers. We consider the system at charge neutrality, which means that the sum of the particle density ntn_t and the hole density nbn_b equals 11, and introduce an inter-layer bias VbV_b. The interaction term is Un^t(n^b−1)U\hat{n}_t(\hat{n}_b-1), specifically chosen to vanish when the lower band is fully occupied (nb=1n_b=1). We then decouple the interaction into two components: a real exciton field denoted by ϕ\phi, which can be expressed as ϕ=−U(⟨cbct⟩+⟨ct†cb†⟩)/2\phi=-U(\langle c_b c_t\rangle+\langle c_t^\dagger c_b^\dagger\rangle)/2, and an inter-layer polarization field denoted by ϕz\phi_z which is ϕz=U⟨ct†ct⟩\phi_z=U\langle c_t^\dagger c_t\rangle. Combining these, we obtain the mean-field Hamiltonian

HMF=∑kψk†[(k22m+Vb−ϕz)τz+ϕτx]ψk.(54)\mathcal{H}_{\mathrm{MF}}=\sum_{\mathbf{k}}\psi_{\mathbf{k}}^\dagger\left[\left(\frac{\mathbf{k}^2}{2m}+V_b-\phi_z\right)\tau_z+\phi\tau_x\right]\psi_{\mathbf{k}}. \tag*{(54)}

where ψk=(ce,k,ch,k)T\psi_{\mathbf{k}}=(c_{e,\mathbf{k}},c_{h,\mathbf{k}})^T (since electron/hole labels are identical to the layer labels). Next, we define ξk=k2/2m+ϵh−ϕz\xi_{\mathbf{k}}=\mathbf{k}^2/2m+\epsilon_h-\phi_z and Ek=ξk2+ϕ2E_{\mathbf{k}}=\sqrt{\xi_{\mathbf{k}}^2+\phi^2} to write the self-consistent equations (at T=0T=0) for ϕ\phi and ϕz\phi_z

1V=1N∑k12Ek,ϕz=V2N∑k(1−ξkEk).(55)\frac{1}{V}=\frac{1}{N}\sum_{\mathbf{k}}\frac{1}{2E_{\mathbf{k}}},\qquad\phi_z=\frac{V}{2N}\sum_{\mathbf{k}}\left(1-\frac{\xi_{\mathbf{k}}}{E_{\mathbf{k}}}\right). \tag*{(55)}

The solutions can be obtained numerically with a lattice regularization k2/2m→ϵk=2t(2−cos⁡kx−cos⁡ky)\mathbf{k}^2/2m\to\epsilon_{\mathbf{k}}=2t(2-\cos k_x-\cos k_y) where tt can be thought of as a nearest-neighbor hopping amplitude on a square lattice.

The multi-band generalization follows along similar lines. First we compute the eigenvalues and eigenvectors {Em(k),∣um,k⟩}\{E_m(\mathbf{k}),\lvert u_{m,\mathbf{k}}\rangle\} of the mean-field Hamiltonian. The mean-field states follow the thermal occupation

⟨γk,m†γk,m′⟩=f[Em(k)]δm,m′.(56)\langle\gamma^\dagger_{\mathbf{k},m}\gamma_{\mathbf{k},m'}\rangle=f[E_m(\mathbf{k})]\delta_{m,m'}. \tag*{(56)}

where ff is the Fermi function and the operators {γk}\{\gamma_{\mathbf{k}}\} are related to {ck}\{c_{\mathbf{k}}\} by a unitary transformation. We use the new operators {γk}\{\gamma_{\mathbf{k}}\} to rewrite Eq. (51) which turns out to have a concise matrix structure

ϕαβ=−VαβN∑kf[Em(k)]⟨um,k∣∂HMF∂ϕαβ∗∣um,k⟩,nα,t=VαβN∑kf[Em(k)]⟨um,k∣∂HMF∂nβ,b∣um,k⟩.(57)\phi_{\alpha\beta}=-\frac{V_{\alpha\beta}}{N}\sum_{\mathbf{k}}f[E_m(\mathbf{k})]\left\langle u_{m,\mathbf{k}}\left|\frac{\partial H_{\mathrm{MF}}}{\partial\phi_{\alpha\beta}^{*}}\right|u_{m,\mathbf{k}}\right\rangle,\qquad n_{\alpha,t}=\frac{V_{\alpha\beta}}{N}\sum_{\mathbf{k}}f[E_m(\mathbf{k})]\left\langle u_{m,\mathbf{k}}\left|\frac{\partial H_{\mathrm{MF}}}{\partial n_{\beta,b}}\right|u_{m,\mathbf{k}}\right\rangle. \tag*{(57)}

The equation is dependent on {nα,t,ϕαβ}\{n_{\alpha,t},\phi_{\alpha\beta}\} on both sides, thus requiring an iterative approach. We start with a guess and use FindRoot function in Mathematica to find the solution. The solutions are shown in Fig. S1.

Tunable Quantum Metric Model

We reprint the Bloch hamiltonian for the tunable quantum metric model for convenience

H(ζ,k)=2t(2−pk)+tFcos⁡(ζpk)σx+tFsin⁡(ζpk)σy.(58)H(\zeta,\mathbf{k})=2t(2-p_{\mathbf{k}})+t_F\cos(\zeta p_{\mathbf{k}})\sigma^x+t_F\sin(\zeta p_{\mathbf{k}})\sigma^y. \tag*{(58)}

Here the Pauli matrices σi\sigma^i act in orbital space, pkp_{\mathbf{k}} is a periodic function given by pk=cos⁡kx+cos⁡kyp_{\mathbf{k}}=\cos k_x+\cos k_y and ζ\zeta is an overall scaling factor for the quantum geometric tensor Q(k,ζ)μν=(ζ2/4)sin⁡kμsin⁡kν\mathcal{Q}(\mathbf{k},\zeta)_{\mu\nu}=(\zeta^2/4)\sin k_\mu\sin k_\nu.

We find a significant simplification with ϕαβ(k)=ϕδαβ\phi_{\alpha\beta}(\mathbf{k})=\phi\delta_{\alpha\beta}. This relatively simple form can be attributed to three key factors. The on-site nature makes momentum independent ϕ^\hat{\phi} a natural ansatz. The orbital-diagonal aspect n^α,tn^α,b\hat{n}_{\alpha,t}\hat{n}_{\alpha,b} removes the possibility of off-diagonal elements in ϕ^\hat{\phi}. Lastly, the two orbitals in the toy model are related by an anti-plane inversion which ultimately reduces the exciton field to be proportional to identity with a single scalar ϕ^=ϕ1\hat{\phi}=\phi\mathbf{1}.

When both layers are described by the same parameter ζ\zeta (see Figure 2a), the stiffness has a significant enhancement coming from the geometric term DsgD_s^g. The extra contribution is indeed proportional to the trace of the quantum metric in this particular case Dsg=(ϕ2/4A)∑kg(k)/E(k)D_s^g=(\phi^2/4A)\sum_{\mathbf{k}}g(\mathbf{k})/E(\mathbf{k}) [69]. While the geometric contribution dominates at higher bias voltage ∣Vb∣\lvert V_b\rvert, it is important to note that a high ∣Vb∣\lvert V_b\rvert results in higher densities of electron-hole pairs and may ultimately drive the system to the exciton Mott transition [45–48].

Image not extracted

Figure S1. Mean-field theory for symmetric excitons in the minimal model. a Both the particle and hole bands have the same dispersion and wavefunction. The non-interacting bands (dashed line) have a gap VbV_b that increases further with the formation of excitons (solid line). b Self-consistent solution of ϕ\phi together with DsD_s. Both require a finite interaction strength U/t>0U/t > 0 to activate. c Geometric contribution to stiffness scales with ζ2\zeta^2 as expected from trace of the quantum metric. d and e show the temperature dependence of ϕ\phi and DsD_s respectively. The mean-field temperature (at which ϕ\phi vanishes) is clearly larger than the BKT temperature (intersection of Ds(T)D_s(T) and 2T/π2T/\pi). The condensate is limited by phase fluctuations.

Excitons with Asymmetric Bands

In this section, we discuss various aspects of particle-hole asymmetry in the geometric contribution to stiffness. Let us begin by ignoring the effects of wavefunctions and recall the theory of excitons with asymmetric electron and hole masses. This is a common scenario in most semiconductors where the conduction and valence bands exhibit different dispersions, characterized by me≠mhm_e \ne m_h. It leads to a slightly generalized form of Eq. (54) with the mean-field Hamiltonian:

HMF=∑kψk†[k22M1+(k22m+Vb−ϕz)τz+ϕτx]ψk.(59)\mathcal{H}_{\mathrm{MF}} = \sum_{\mathbf{k}} \psi_{\mathbf{k}}^\dagger\left[ \frac{k^2}{2M}\mathbb{1} + \left(\frac{k^2}{2m} + V_b - \phi_z\right)\tau_z + \phi\tau_x \right]\psi_{\mathbf{k}} . \tag*{(59)}

Here, M−1=(me−mh)/2memhM^{-1} = (m_e-m_h)/2m_em_h and m−1=(me+mh)/2memhm^{-1} = (m_e+m_h)/2m_em_h are effective masses. The inclusion of the identity term modifies the mean-field dispersion relation E±(k)=ξM(k)±E(k)E_{\pm}(\mathbf{k}) = \xi_M(\mathbf{k}) \pm E(\mathbf{k}), where E(k)=ξ(k)2+ϕ2E(\mathbf{k}) = \sqrt{\xi(\mathbf{k})^2+\phi^2} represents the previous quasiparticle dispersion and ξM(k)=k2/2M\xi_M(\mathbf{k}) = k^2/2M. The self-consistent order parameter equation changes as well, mainly via the thermal factors, which were suppressed in Eq. (B6),

1V=1N∑kf[E−(k)]−f[E+(k)]2E(k).(60)\frac{1}{V} = \frac{1}{N}\sum_{\mathbf{k}} \frac{f[E_-(\mathbf{k})]-f[E_+(\mathbf{k})]}{2E(\mathbf{k})}. \tag*{(60)}

It is easy to check that the equation reduces to Eq. (55) if E+(k)=−E−(k)=E(k)E_+(\mathbf{k})=-E_-(\mathbf{k})=E(\mathbf{k}). The effects stemming from the asymmetry can be visualized as a momentum-dependent Zeeman term in analogy to the one encountered in BCS theory. It is well known that large anisotropy can act against exciton binding [83, 84] and destroy the condensate. Phase stiffness is affected as well since it is a convolution of ϕ\phi with the counterflow current operator j^CF\hat{j}_{\mathrm{CF}}. If ϕ\phi decreases, DsD_s will decrease as well.

Next, we consider the more subtle scenario of symmetric energies but asymmetric wavefunctions. We again utilize the minimal model presented in the previous section, eq. (B9), with ζt=1\zeta_t=1 for the top layer and ζb=ζ\zeta_b=\zeta in the bottom layer. This configuration results in a particle-hole symmetric spectrum but with distinct electron and hole wavefunctions. The overlap ϑ(k)=cos⁡[(ζ−1)pk/2]\vartheta(\mathbf{k})=\cos[(\zeta-1)p_k/2] exhibits significant variation across the Brillouin zone (see Figure S2). This variation plays a crucial role in the quasiparticle energy E(k)=ξ(k)2+ϕ2∣ϑ(k)∣2E(\mathbf{k})=\sqrt{\xi(\mathbf{k})^2+\phi^2|\vartheta(\mathbf{k})|^2}, which, in turn, affects the gap equation 1/V=(1/N)∑k1/E(k)1/V=(1/N)\sum_{\mathbf{k}}1/E(\mathbf{k}). The reduction in ∣ϑ(k)∣|\vartheta(\mathbf{k})| from unity leads to a decrease in ϕ\phi compared to the particle-hole symmetric case where ϑ(k)=1\vartheta(\mathbf{k})=1 (see Figure S2 and Figure S2). The stiffness is also affected, as shown in Figure S2. Two factors contribute to this alteration: pre-factor ϕ\phi and G(k)G(\mathbf{k}) which explicitly depends on ζ\zeta through ϑ(k)\vartheta(\mathbf{k}), P(k)P(\mathbf{k}), and D(k)D(\mathbf{k}). Although it may be model specific, we find that the first contribution is symmetric around ζ=1\zeta=1 while the geometric contribution is skewed towards higher ζ\zeta (see Figure S2). Additionally, it is important to highlight that the geometric contribution is not necessarily positive definite, see Figure S2. This stands in contrast to the geometric contributions reported in superfluid stiffness [2] and electron-phonon coupling [21].

Image not extracted

Figure S2. Asymmetric excitons in the minimal model with tunable quantum metric. a The overlap function between two bands defined by ζt=1\zeta_t=1, ζb=ζ=2\zeta_b=\zeta=2. The gap equation (b) and stiffness (c) are both modified as the asymmetry in wavefunction is tuned by ζ\zeta (with Vb=0.1tV_b=0.1t). Notice that the conventional contribution is symmetric around ζt=ζb=1\zeta_t=\zeta_b=1. d Overlap function averaged over the BZ. Departure from unity results in a smaller mean-field solution for ϕ\phi. The isotropic case ζt=ζb\zeta_t=\zeta_b results in the strongest exciton. e Sign of the geometric contribution in the minimal model. There are parameter regions where the contribution is negative. Intuitively, quantum geometry can be detrimental to stiffness if the bands are very different.

Motivated by these two simple scenarios, we can deduce what happens in a realistic situation with both asymmetric dispersions and wavefunctions. The dispersion asymmetry results in redefinition of effected masses mm and MM. On top of that, the wavefunction asymmetry modifies both ϕ\phi and G(k)G(\mathbf{k}). Both these effects are at play simultaneously, the details of which depend on the model and we are left limited to numerical results.

Continuum model for exciton condensation in AA-stacked TMDs

To start with we consider spinless electrons described by the effective Hamiltonian:

H=∑k,aΛψk,a†[Ht,a(k)−μ+Ez00Hb,a(k)−μ−Ez]ψk,a+12A∑qΛ∑ν,ν′=t,bVqνν′nq,νn−q,ν′(61)H=\sum_{\mathbf{k},a}^{\Lambda}\psi^\dagger_{\mathbf{k},a}\begin{bmatrix}H_{t,a}(\mathbf{k})-\mu+E_z&0\\0&H_{b,a}(\mathbf{k})-\mu-E_z\end{bmatrix}\psi_{\mathbf{k},a}+\frac{1}{2A}\sum_{\mathbf{q}}^{\Lambda}\sum_{\nu,\nu'=t,b}V_{\mathbf{q}}^{\nu\nu'}n_{\mathbf{q},\nu}n_{-\mathbf{q},\nu'} \tag*{(61)}

where ψk,a\psi_{\mathbf{k},a} is a 4-dim spinor in the sublattice and layer indeces, ψk,a=[ψk,a+t,ψk,a−t,ψk,a+b,ψk,a−b]\psi_{\mathbf{k},a}=[\psi_{\mathbf{k},a+t},\psi_{\mathbf{k},a-t},\psi_{\mathbf{k},a+b},\psi_{\mathbf{k},a-b}], KK and K′K' refers to the two different valleys, Λ\Lambda is the UV cutoff and EzE_z the electric displacement field controlling the energy offset between the two layers. The interaction is obtained by taking the Fourier transform of Vrtt=Vrbb=e2/(4πϵ0ϵr)V_{\mathbf{r}}^{tt}=V_{\mathbf{r}}^{bb}=e^2/(4\pi\epsilon_0\epsilon r) and Vrtb=Vrbt=e2/(4πϵ0ϵr2+d2)V_{\mathbf{r}}^{tb}=V_{\mathbf{r}}^{bt}=e^2/(4\pi\epsilon_0\epsilon\sqrt{r^2+d^2}) with dd interlayer distance. From

∫d2q2πe−dq+iq⋅rq=∮∣z∣=12dz2πrz2+1−2dz/(ir)=1r2+d2,(62)\int\frac{d^2\mathbf{q}}{2\pi}\frac{e^{-d q+i\mathbf{q}\cdot\mathbf{r}}}{q}=\oint_{\lvert z\rvert=1}\frac{2dz}{2\pi r z^2+1-2dz/(ir)}=\frac{1}{\sqrt{r^2+d^2}}, \tag*{(62)}

we readily find:

Vqtt=Vqbb=Vq=e22ϵ0ϵq,Vqtb=e−qdVq.(63)V_{\mathbf{q}}^{tt}=V_{\mathbf{q}}^{bb}=V_{\mathbf{q}}=\frac{e^2}{2\epsilon_0\epsilon q},\qquad V_{\mathbf{q}}^{tb}=e^{-qd}V_{\mathbf{q}}. \tag*{(63)}

Accounting for the presence of the metallic gates at distance ξ\xi from the sample we replace 1/q→tanh⁡(qξ/2)/q1/q\to\tanh(q\xi/2)/q, which regularizes the singularity at q=0q=0. We use σ\sigma for dz2/dx2−y2±idxyd_{z^2}/d_{x^2-y^2}\pm i d_{xy} at valley K/K′K/K' while τ\tau for layer. The chemical potential μ\mu enforces the charge neutrality condition:

1N∑kΛ⟨ψk,a†ψk,a⟩−nCNP=0.(64)\frac{1}{N}\sum_{\mathbf{k}}^{\Lambda}\langle\psi^\dagger_{\mathbf{k},a}\psi_{\mathbf{k},a}\rangle-n_{\mathrm{CNP}}=0. \tag*{(64)}

Model for TMDs

We assume that due to the strong Ising spin orbit coupling the conduction (CB) and valence band (VB) of each layer around KK and K′K' points are spin polarized. This assumption is realized under the assumption of large Ising spin-orbit coupling τΔvbszτ\tau\Delta_{vb}s_z^\tau and τΔcbszτ\tau\Delta_{cb}s_z^\tau with τ=±\tau=\pm for the two valleys. For MX2_2 the SOC gap is 2Δvb=148,186,2192\Delta_{vb}=148,186,219meV with X={S2,Se2,Te2}\{\mathrm{S}_2,\mathrm{Se}_2,\mathrm{Te}_2\}. For WX2_2 the gap is larger due to the larger SOC 2Δvb=429,466,4842\Delta_{vb}=429,466,484meV for VB and 2Δcb=−32,−37,−532\Delta_{cb}=-32,-37,-53meV for CB with X={S2,Se2,Te2}\{\mathrm{S}_2,\mathrm{Se}_2,\mathrm{Te}_2\}. In the conduction band the SOC is small and corrections from the nearby conduction band might play an important role. Keeping terms up to the second order in momentum the k⋅pk\cdot p Hamiltonian reads:

Hν,K(k)=[Δν+ανk2vνk−vνk+−mν−γνk2],Ht/b,K′(k)=Ht/b,K∗(−k).(65)H_{\nu,\mathrm{K}}(\mathbf{k})=\begin{bmatrix}\Delta_{\nu}+\alpha_{\nu}k^2&v_{\nu}k_-\\v_{\nu}k_+&-m_{\nu}-\gamma_{\nu}k^2\end{bmatrix},\qquad H_{t/b,\mathrm{K}'}(\mathbf{k})=H^*_{t/b,\mathrm{K}}(-\mathbf{k}). \tag*{(65)}
MoS2_2MoSe2_2MoTe2_2WS2_2WSe2_2WTe2_2
2m2m [eV]2.762.331.822.882.421.82
vv [eV Å]2.222.202.162.592.602.79
α\alpha [eV Å2^2]0.52-0.54-1.662.031.080.17
γ\gamma [eV Å2^2]-6.65-6.20-5.78-7.96-7.58-7.31

caption: Relevant parameters of the k.pk.p Hamiltonian for different TMDs [72] computed performing GW calculations [85, 86].

:::

where the latter relation enforces the time reversal symmetry. The typical values of the parameters of the model are given in tbl:1. Different configurations are characterized by a typical value of the ratio v/m≈2v/m \approx2 Å. Finally, in the main text we compare results obtained with gapped Dirac cones and parabolic bands. The latter is obtained expanding the Dirac dispersion around KK and K′K', ±Δ2+v2k2≃±Δ±k2/2m\pm\sqrt{\Delta^2+v^2k^2}\simeq\pm\Delta\pm k^2/2m with m=Δ/v2m=\Delta/v^2 and gap 2Δ2\Delta.

Symmetries

The model is invariant under the transformation:

ψk,at→eiϕtψk,at,ψk,at→eiϕbψk,at,(66)\psi_{k,\mathrm{at}}\to e^{i\phi_t}\psi_{k,\mathrm{at}},\qquad\psi_{k,\mathrm{at}}\to e^{i\phi_b}\psi_{k,\mathrm{at}}, \tag*{(66)}

where ϕb+ϕt\phi_b+\phi_t is the total phase associate to U(1)charge while the relative phase ϕt−ϕb\phi_t-\phi_b is related to the U(1)layer symmetry. The latter is spontaneously broken in the excitonic state which is characterized by interlayer coherence. In addition, we observe that the model in Eq. (65) breaks the particle-hole symmetry for α≠γ\alpha\ne\gamma. The low-energy model is also invariant under the continuous space rotational rotational symmetry k→Rθk\mathbf{k}\to R_\theta\mathbf{k} and Uθ=diag⁡(e−iθ/2,eiθ/2)U_\theta=\operatorname{diag}(e^{-i\theta/2},e^{i\theta/2}) which acts on the sublattice degree of freedom:

Uθ†Hν,K/K′(Rθk)Uθ=Hν,K/K′(k).(67)U_\theta^\dagger H_{\nu,K/K'}(R_\theta\mathbf{k})U_\theta=H_{\nu,K/K'}(\mathbf{k}). \tag*{(67)}

Including higher order terms in the k.pk.p expansion lower the aforementioned symmetry to C3zC_{3z}. Finally, M\mathcal{M} acting as (x,y)→(−x,y)(x,y)\to(-x,y) and ↑↔↓\uparrow\leftrightarrow\downarrow is a symmetry of the model.

Diagonalizing the single-particle Hamiltonian

The eigenvalues of the non-interacting Hamiltonian are given by E±,t/b(k)=±ϵt/b(k)E_{\pm,t/b}(k)=\pm\epsilon_{t/b}(k), ϵt/b(k)=vt/b2k2+Δt/b2\epsilon_{t/b}(k)=\sqrt{v_{t/b}^{2}k^2+\Delta_{t/b}^{2}} while the matrix of eigenstates is:

U(k)=12[eiϕk(cos⁡φk2+sin⁡φk2)eiϕk(cos⁡φk2−sin⁡φk2)(cos⁡φk2−sin⁡φk2)−(cos⁡φk2+sin⁡φk2)],(68)U(\mathbf{k})=\frac{1}{\sqrt{2}}\begin{bmatrix} e^{i\phi_{\mathbf{k}}}\left(\cos\frac{\varphi_k}{2}+\sin\frac{\varphi_k}{2}\right) & e^{i\phi_{\mathbf{k}}}\left(\cos\frac{\varphi_k}{2}-\sin\frac{\varphi_k}{2}\right)\\ \left(\cos\frac{\varphi_k}{2}-\sin\frac{\varphi_k}{2}\right) & -\left(\cos\frac{\varphi_k}{2}+\sin\frac{\varphi_k}{2}\right) \end{bmatrix}, \tag*{(68)}

where tan⁡φk=Δ/(vk)\tan\varphi_k=\Delta/(vk) and we dropped the layer and valley labels. Applying the unitary transformation:

U(k)=[Ut(k)00Ub(k)],(69)U(\mathbf{k})=\begin{bmatrix} U_t(\mathbf{k}) & 0\\ 0 & U_b(\mathbf{k}) \end{bmatrix}, \tag*{(69)}

the kinetic part becomes:

Hkin=∑a=K,K′∑kΨk,a†[ϵt(k)σz+(Vb−μ)σ000ϵb(k)σz+(−Vb−μ)σ0]Ψk,a,(70)H_{\mathrm{kin}}=\sum_{a=K,K'}\sum_{\mathbf{k}}\Psi^\dagger_{\mathbf{k},a}\begin{bmatrix} \epsilon_t(k)\sigma^z+(V_b-\mu)\sigma^0 & 0\\ 0 & \epsilon_b(k)\sigma^z+(-V_b-\mu)\sigma^0 \end{bmatrix}\Psi_{\mathbf{k},a}, \tag*{(70)}

where Ψk,aν=Uaν†(k)ψk,aν\Psi_{\mathbf{k},a\nu}=U^\dagger_{a\nu}(\mathbf{k})\psi_{\mathbf{k},a\nu} with a=K/K′a=K/K' and ν=t,b\nu=t,b. We now look at the interaction projected in the basis of the eigenstates of the kinetic term. The density operator for the layer ν=t/b\nu=t/b reads:

ρqν=∑kψkν†ψk+qν=∑k∑αα′[Fk,qν]αα′Ψkνα†Ψk+qνα′,(71)\rho^\nu_{\mathbf{q}}=\sum_{\mathbf{k}}\psi^\dagger_{\mathbf{k}\nu}\psi_{\mathbf{k}+\mathbf{q}\nu}=\sum_{\mathbf{k}}\sum_{\alpha\alpha'}[F^\nu_{\mathbf{k},\mathbf{q}}]_{\alpha\alpha'}\Psi^\dagger_{\mathbf{k}\nu\alpha}\Psi_{\mathbf{k}+\mathbf{q}\nu\alpha'}, \tag*{(71)}

where we drop for the moment the valley index, and the form factor is obtained projecting the density operator in the Bloch basis

Fk,kν=Uν†(k)Uν(k+q),[Fk,qν]αα′=⟨uk,αν∣uk+q,α′ν⟩.(72)F^\nu_{\mathbf{k},\mathbf{k}}=U^\dagger_\nu(\mathbf{k})U_\nu(\mathbf{k}+\mathbf{q}),\qquad[F^\nu_{\mathbf{k},\mathbf{q}}]_{\alpha\alpha'}=\langle u^\nu_{\mathbf{k},\alpha}|u^\nu_{\mathbf{k}+\mathbf{q},\alpha'}\rangle. \tag*{(72)}

The orthogonality condition implies that [Fk,0ν]αα′=δαα′\left[F^{\nu}_{\mathbf{k},0}\right]_{\alpha\alpha'}=\delta_{\alpha\alpha'}. As a result the interaction takes the form:

Hint=12A∑qkk′∑νν′∑aa′[Fk,qν,a]αα′Vqνν′[Fk′,qν′,a′]γγ′Ψk,aνα†Ψk+q,aνα′Ψk′+q,a′ν′γ†Ψk′,a′ν′γ′(73)H_{\mathrm{int}}=\frac{1}{2A}\sum_{\mathbf{q}kk'}\sum_{\nu\nu'}\sum_{aa'}\left[F^{\nu,a}_{\mathbf{k},\mathbf{q}}\right]_{\alpha\alpha'}V^{\nu\nu'}_{\mathbf{q}}\left[F^{\nu',a'}_{\mathbf{k}',\mathbf{q}}\right]_{\gamma\gamma'}\Psi^\dagger_{\mathbf{k},a\nu\alpha}\Psi_{\mathbf{k}+\mathbf{q},a\nu\alpha'}\Psi^\dagger_{\mathbf{k}'+\mathbf{q},a'\nu'\gamma}\Psi_{\mathbf{k}',a'\nu'\gamma'} \tag*{(73)}

At mean-field level we decouple the interaction in the Hartree and Fock (exchange) terms that are derived in the following sections.

Hartree term

The Hartree term corresponds to:

HH=∑k∑a=K,K′∑ν=t,bδμνΨk,aν†Ψk,aν−12A∑νν′V0νν′∑kk′∑aa′=K,K′δρaν(k)δρa′ν′(k′)(74)H_H=\sum_k\sum_{a=K,K'}\sum_{\nu=t,b}\delta\mu_\nu\Psi^\dagger_{k,a\nu}\Psi_{k,a\nu}-\frac{1}{2A}\sum_{\nu\nu'}V^{\nu\nu'}_0\sum_{kk'}\sum_{aa'=K,K'}\delta\rho^{a\nu}(k)\delta\rho^{a'\nu'}(k') \tag*{(74)}

where we have introduced the charge density measured with respect to charge neutrality:

δρaν(k)=∑α=±⟨Ψk,aνα†Ψk,aνα⟩−Nδk,0.(75)\delta\rho^{a\nu}(k)=\sum_{\alpha=\pm}\langle\Psi^\dagger_{k,a\nu\alpha}\Psi_{k,a\nu\alpha}\rangle-N\delta_{k,0}. \tag*{(75)}

the contribution δμν\delta\mu_\nu is given by:

δμν=1N∑k∑a=K,K′∑α=±∑ν′δρααaν′(k)V0νν′.(76)\delta\mu_\nu=\frac{1}{N}\sum_k\sum_{a=K,K'}\sum_{\alpha=\pm}\sum_{\nu'}\delta\rho^{a\nu'}_{\alpha\alpha}(k)V^{\nu\nu'}_0. \tag*{(76)}

Before moving on we observe that

1N∑k∑a=K,K′∑α=±δρααaα(k)=nν−1=sign⁡(ν)ϕz,(77)\frac{1}{N}\sum_k\sum_{a=K,K'}\sum_{\alpha=\pm}\delta\rho^{a\alpha}_{\alpha\alpha}(k)=n_\nu-1=\operatorname{sign}(\nu)\phi_z, \tag*{(77)}

where ϕz\phi_z measure the charge density transferred between the layers ϕz=(nt−nb)/2=nt−1=1−nb\phi_z=(n_t-n_b)/2=n_t-1=1-n_b since at charge neutrality nt+nb=2n_t+n_b=2 with this definition sign⁡(t)=1\operatorname{sign}(t)=1 and sign⁡(b)=−1\operatorname{sign}(b)=-1. Thus, we have:

δμt=ϕz(V0tt−V0bt),δμb=ϕz(V0tb−V0bb).(78)\delta\mu_t=\phi_z\left(V^{tt}_0-V^{bt}_0\right),\qquad\delta\mu_b=\phi_z\left(V^{tb}_0-V^{bb}_0\right). \tag*{(78)}

From Eq. (63) we find:

δμt=δμ,δμb=−δμ,δμ=ϕz(V0tt−V0bt),ϕz=nt−nb2,(79)\delta\mu_t=\delta\mu,\qquad\delta\mu_b=-\delta\mu,\qquad\delta\mu=\phi_z\left(V^{tt}_0-V^{bt}_0\right),\qquad\phi_z=\frac{n_t-n_b}{2}, \tag*{(79)}

which corresponds to electrostatic repulsion. We notice that V0tt=V0tbV^{tt}_0=V^{tb}_0 and δμ=0\delta\mu=0.

Fock term

The exchange term is given by:

HX=−∑k∑a=K,K′∑νν′Ψk,aν†ΣF,αβνν′,a(k)Ψk,aν′β.(80)H_X=-\sum_k\sum_{a=K,K'}\sum_{\nu\nu'}\Psi^\dagger_{k,a\nu}\Sigma^{\nu\nu',a}_{F,\alpha\beta}(k)\Psi_{k,a\nu'\beta}. \tag*{(80)}

where we have introduced the quantity:

ΣX,αβνν′,a(k)=1A∑k′Vk′−kνν′∑α′β′[Fk,k′−kν′,a]β′β[Fk,k′−kν,a]αα′δρβ′α′ν′,a(k′).(81)\Sigma^{\nu\nu',a}_{X,\alpha\beta}(k)=\frac{1}{A}\sum_{k'}V^{\nu\nu'}_{k'-k}\sum_{\alpha'\beta'}\left[F^{\nu',a}_{k,k'-k}\right]_{\beta'\beta}\left[F^{\nu,a}_{k,k'-k}\right]_{\alpha\alpha'}\delta\rho^{\nu',a}_{\beta'\alpha'}(k'). \tag*{(81)}

where we have introduced the quantity:

δρβ′α′ν′,a(k)=⟨Ψk,aν′β′†Ψk,aνα′⟩−δν,ν′δα′β′δα′−.(82)\delta\rho^{\nu',a}_{\beta'\alpha'}(k)=\left\langle\Psi^\dagger_{k,a\nu'\beta'}\Psi_{k,a\nu\alpha'}\right\rangle-\delta_{\nu,\nu'}\delta_{\alpha'\beta'}\delta_{\alpha'-}. \tag*{(82)}

Σ^(k)\hat{\Sigma}(k) is a 4-dim matrix in the layer and band basis. It includes an intralayer terms which renormalize the electronic band structure and an interlayer term giving rise to interlayer phase coherence and exciton condensation. The Hartree-Fock Hamiltonian takes the form:

HHF=∑kΛΨk†[H^kin(k)−Σ^(k)]Ψk.(83)H_{\mathrm{HF}}=\sum_{\mathbf{k}}^{\Lambda}\Psi_{\mathbf{k}}^{\dagger}\left[\hat{H}_{\mathrm{kin}}(k)-\hat{\Sigma}(k)\right]\Psi_{\mathbf{k}}. \tag*{(83)}

The self-consistency equation D21 for the self-energy is solved at every kk-point in momentum space employing an iterative approach. Figure S3 shows the typical band structure and the interlayer gap function, defined as Tr⁡Δ†(k)Δ(k)\sqrt{\operatorname{Tr}\Delta^{\dagger}(k)\Delta(k)} with Δ^(k)\hat{\Delta}(k) interlayer component of the self-energy Δ^(k)=Tr⁡[τ+Σ^(k)]\hat{\Delta}(k)=\operatorname{Tr}\left[\tau^{+}\hat{\Sigma}(k)\right]. Finally, we remark that the excitonic order parameter ϕ\phi has been computed as:

ϕx=TA∑k∑iωTr⁡[τx⊗σaG^(k,iω)],(84)\phi_{x}=\frac{T}{A}\sum_{\mathbf{k}}\sum_{i\omega}\operatorname{Tr}\left[\tau^{x}\otimes\sigma^{a}\hat{G}(k,i\omega)\right], \tag*{(84)}

in the gauge where the interlayer coherence is broken along τx\tau^{x} and the Green’s function G^\hat{G} is obtained including all the bands of the model from Eq. (83).

References

  1. [1]D. Xiao, M.-C. Chang, and Q. Niu, Rev. Mod. Phys. 82, 1959 (2010).
  2. [2]P. Törmä, S. Peotta, and B. A. Bernevig, Nat. Rev. Phys. 4, 528 (2022).
  3. [3]P. Törmä, L. Liang, and S. Peotta, Phys. Rev. B 98, 20511 (2018).
  4. [4]M. Iskin, Phys. Rev. A 97, 033625 (2018).
  5. [5]M. Iskin, Phys. Rev. A 103, 053311 (2021).
  6. [6]S. Peotta and P. Törmä, Nat. Commun. 6, 8944 (2015).
  7. [7]A. Julku, S. Peotta, T. I. Vanhala, D.-H. Kim, and P. Törmä, Phys. Rev. Lett. 117, 045303 (2016).
  8. [8]L. Liang, T. I. Vanhala, S. Peotta, T. Siro, A. Harju, and P. Törmä, Phys. Rev. B 95, 024515 (2017).
  9. [9]F. Xie, Z. Song, B. Lian, and B. A. Bernevig, Phys. Rev. Lett. 124, 167002 (2020).
  10. [10]J. Herzog-Arbeitman, V. Peri, F. Schindler, S. D. Huber, and B. A. Bernevig, Phys. Rev. Lett. 128, 087002 (2022).
  11. [11]K.-E. Huthinen, J. Herzog-Arbeitman, A. Chew, B. A. Bernevig, and P. Törmä, Phys. Rev. B 106, 014518 (2022).
  12. [12]D. Mao and D. Chowdhury, Proc. Natl. Acad. Sci. U.S.A. 120, e2217816120 (2023).
  13. [13]M. Tomvasyan, S. Peotta, P. Törmä, and S. D. Huber, Phys. Rev. B 94, 245149 (2016).
  14. [14]N. Verma, T. Hazra, and M. Randeria, Proc. Natl. Acad. Sci. U.S.A. 118 (2021).
  15. [15]J. Ahn and N. Nagaosa, Phys. Rev. B 104, L100501 (2021).
  16. [16]D. Mao and D. Chowdhury, arXiv, 2304.07318 (2023).
  17. [17]J. S. Hoffmann, E. Berg, and D. Chowdhury, Phys. Rev. B 102, 201112 (2020).
  18. [18]V. Peri, Z.-D. Song, B. A. Bernevig, and S. D. Huber, Phys. Rev. Lett. 126, 027002 (2021).
  19. [19]T. Neupert, C. Chamon, and C. Mudry, Phys. Rev. B 87, 245103 (2013).
  20. [20]I. Komissarov, T. Holder, and R. Queiroz, arXiv, 2306.08035 (2023).
  21. [21]J. Yu, C. J. Ciccariello, R. Bianco, I. Errea, P. Narang, and B. A. Bernevig, arXiv, 2305.02340 (2023).
  22. [22]A. Arora, M. S. Rudner, and J. C. W. Song, Nano Lett. 22, 9351 (2022).
  23. [23]M. F. Lapa and T. L. Hughes, Phys. Rev. B 99, 121111 (2019).
  24. [24]Y. Gao and D. Xiao, Phys. Rev. Lett. 122, 227402 (2019).
  25. [25]V. Kozii, A. Avdoshkin, S. Zhong, and J. E. Moore, Phys. Rev. Lett. 126, 156602 (2021).
  26. [26]J. Ahn, G.-Y. Guo, N. Nagaosa, and A. Vishwanath, Nat. Phys. 18, 290 (2022).
  27. [27]S. Chaudhary, C. Lewandowski, and G. Refael, Phys. Rev. Res. 4, 013164 (2022).
  28. [28]J. M. Blatt, K. W. Böer, and W. Brandt, Phys. Rev. 126, 1691 (1962).
  29. [29]B. I. Halperin and T. M. Rice, Rev. Mod. Phys. 40, 755 (1968).
  30. [30]Y. E. Lozovik and V. Yudson, JETP Lett.(USSR)(Engl. Transl.) 22 (1975).
  31. [31]Y. E. Lozovik and V. Yudson, Zh. Eksp. Teor. Fiz 71, 738 (1976).
  32. [32]D. Yoshioka and A. H. MacDonald, Journal of the Physical Society of Japan 59, 4211 (1990).
  33. [33]K. Yang, Phys. Rev. Lett. 87, 056802 (2001).
  34. [34]S. De Palo, F. Rapisarda, and G. Senatore, Phys. Rev. Lett. 88, 206401 (2002).
  35. [35]Y. N. Joglekar, A. V. Balatsky, and S. Das Sarma, Phys. Rev. B 74, 233302 (2006).
  36. [36]J. P. Eisenstein and A. H. MacDonald, Nature 432, 691 (2004).

Paper details

vlnc:W_yakajcru
Version
v1 · latest
Published
Fresh — added to the Hub within the last 30 days
Added
License
CC BY 4.0
Usage

views · 0

downloads · 0

A view counts once per browser per day, a download each time the PDF is requested.

  • Physics and Astronomy

Not recorded for this paper

View statement
Looking for related papers…
Contents