betaFeedback

WorksPhysics and Astronomy

Geometric Ginzburg-Landau theory of charge ordering and commensurability

Abstract

The concept of quantum geometry has recently led to reinvigorated insights in a wide range of fields including physical responses, superconductivity, and optical transitions, with effects most pronounced in systems with nearly flat dispersion. Here, we show that it plays an essential role in charge density wave formation (CDW) – an important physical phenomenon that is responsible for driving various sharp changes in material transport properties including metal-insulator transitions. We derive an effective Ginzburg-Landau theory including uncharted contributions and, as a highlight, discover a general criterion for both CDW formation and commensurability transitions where underlying electron-phonon interactions manifest purely as electronic quantum geometric enhancements/suppressions. We benchmark our framework in a class of transition-metal dichalcogenides and resolve a longstanding puzzle where well-established purely kinetic CDW criteria fail in describing the correct ordering wavevector. Besides rendering robust criteria and fundamental insights that are immediately relevant to several experimental charge ordering systems, our theory can also be applied directly to other phonon-mediated phases such as superconductivity, and can be used as an important tool to explore the interplay between various such states. More generally, our framework provides a recipe for investigating the role of quantum geometry in phase transitions.

Introduction

Topology has provided a revolutionizing view on the characterization of quantum matter in the past decades. While topology, especially in nearly free fermion systems [?, ?, ?, ?, ?, ?], has increasingly been understood, quantum geometry [1, 2, 3, 4] has recently been galvanizing novel insights. The surge of interest in flatband systems that naturally provide for a playground in which interaction effects can enjoy synergies with geometrical and topological effects have led to several developments. In the absence of dispersive effects, the physics becomes inherently geometrically motivated, and effects on superconductivity and superfluid stiffness [5, 6, ?], wavefunction spreading [7, 8], fractionalization [?, ?, ?], as well as optical responses [?, ?, ?] are actively being explored through this geometrical lens.

These developments motivate a reinvigoration of well established phenomena as well as longer standing questions in such contexts. A key example in this regard concerns charge density wave formation in systems with a small bandwidth as compared to the band gap. Charge density waves (CDWs) [?, ?, ?] represent a main field of research as one of the driving mechanisms behind metal-insulator transitions and other sharp changes in transport properties [?, ?, ?], and due to the evolution of scanning tunneling microscopy (amongst others) [?, ?, ?], are observed on a daily basis. Although the foundations for such phenomena can be traced to Fermi surface nesting that underlies the Peierls instability [?] and the Kohn anomaly [?], experiments have shown that charge ordering can also occur in systems where such criteria are not met [?, ?, ?, ?, ?, ?, ?, ?]. What is more surprising perhaps is that there seem to be no hints in the electronic spectra regarding either the ordering itself or a preference of the system to a specific soft phonon wavevector [?, ?]. This naturally begs the question that geometry might be the driving factor, especially since these inconsistencies are most famously reported in relatively flat systems such as 2H polymorphs transition metal dichalcogenides (TMDs). While some recent work [?, ?] addressed that nesting conditions can indeed have a geometric component, their direct relation to classes of electron-phonon coupled Hamiltonians, and consequently real material modeling, has remained more elusive. TMDs in particular stand out because the CDW formation is known to not uniformly match these conditions in certain cases.

Here, we develop a definitive geometric theory for phases driven by electron-phonon interactions. Beginning with a generic electron-phonon Hamiltonian [?, ?], where the electronic quantum geometry manifests through the interaction form factors [?], we derive a geometric Ginzburg-Landau (GL) theory for CDW formation. Particularly, we find both dominant higher-order terms and geometric modifications to terms at all orders in the GL functional that drive CDW formation in materials long-thought to have no conclusive indicators of ordering in their electronic Hamiltonians. Since these materials often exhibit largely flat Lindhard susceptibilities (due to low dispersion), ordering finds its origins in large changes of the electronic flavor character (and hence, a large interband velocity matrix element) near the chemical potential. Specifically, we note that a system is most likely to condense into an ordered state at a wavevector q\mathbf{q} where Bloch states at q\mathbf{q}-separated momenta on the Fermi surface exchange their flavor character or equivalently, rotate into each other (see Figure 1). Further, we find that ordering can be enhanced by the presence of a third state on the Fermi surface connected via a symmetrically-equivalent q\mathbf{q}-vector, something that is highly likely in hexagonal systems which often have roughly triangular Fermi pockets (see Figure 1). As a highlight, we also discover that at commensurate ordering wavevectors =1/3 G= 1/3\,\mathbf{G}, umklapp scattering can interfere with the aforementioned triangular scattering process (see Figure 1), and the relative phase determines whether a commensurability condition can be expected. Notably, the phases of such scattering processes are purely geometric, thus suggesting that a commensurability transition is not a feature of the energetics at all. We benchmark our framework through ab initio calculations in several TMDs where the origin of the ordering wavevector has long been debated and find geometry-driven ordering peaks at experimentally-observed q\mathbf{q} along with associated commensurability transitions (where present). Further, where ordering peaks have similar magnitude, we predict multiple co-existing CDW phases in line with STM observations [?].

Geometric mechanism for charge-density-wave formation in flat-band systems

Figure 1. Geometric mechanism for charge-density-wave formation in flat-band systems. a Schematic illustration of the geometric origin of charge ordering. In flat and quasi-flat band systems, charge-density-wave instabilities are enhanced when Bloch states at momenta separated by an ordering wavevector QiQ_i undergo a strong change in flavor character, as illustrated by the color exchange in the band-structure inset. In hexagonal systems, three symmetry-related wavevectors, Q0Q_0, Q1Q_1, and Q2Q_2, further enable a higher-order triangular scattering process (purple) that contributes to wavevector selection. At commensurate wavevectors Qi=Γ/3Q_i = \Gamma/3, interference between triangular and Umklapp (magenta) scattering processes governs the commensurability transition. b Corresponding real-space charge-density-wave modulation resulting from the selected ordering wavevector.

Theory setting.—The effects of electronic band geometry on CDW formation can be traced to the form of the electron-phonon interaction. Generically, the interaction term in the electron-phonon Hamiltonian Hel−phH_{\mathrm{el-ph}} can be written in second-quantized form as [?]

Hel−ph=∑k,q,m,n,νγqν eqν⋅λk,k−qmn(b^qν+b^−qν†)c^m,k†c^n,k−q,(1)H_{\mathrm{el-ph}} = \sum_{\mathbf{k},\mathbf{q},m,n,\nu} \gamma_{\mathbf{q}\nu}\,\mathbf{e}_{\mathbf{q}\nu}\cdot\boldsymbol{\lambda}^{mn}_{\mathbf{k},\mathbf{k-q}}\left(\hat{b}_{\mathbf{q}\nu}+\hat{b}^{\dagger}_{-\mathbf{q}\nu}\right)\hat{c}^{\dagger}_{m,\mathbf{k}}\hat{c}_{n,\mathbf{k-q}}, \tag*{(1)}

where eqν\mathbf{e}_{\mathbf{q}\nu} is the phonon polarization at wavevector q\mathbf{q} and branch ν\nu, γqν\gamma_{\mathbf{q}\nu} is some term dependent only on the phonon spectrum (usually described as scaling with the phonon frequency Ωqν\Omega_{\mathbf{q}\nu} as Ωqν−1/2\Omega_{\mathbf{q}\nu}^{-1/2}), and b^qν\hat{b}_{\mathbf{q}\nu} and c^m,k\hat{c}_{m,\mathbf{k}} are the phonon and electron annihilation operators at their respective wavevectors and branch/band quantum numbers. Ref. [?] found the electron band geometry factor to be

λk,k′mn∝−i[U†(k)(∇kH0(k)−∇k′H0(k′))U(k′)]mn,\lambda^{mn}_{\mathbf{k},\mathbf{k}'} \propto-i\left[U^{\dagger}(\mathbf{k})\left(\nabla_{\mathbf{k}}H_0(\mathbf{k})-\nabla_{\mathbf{k}'}H_0(\mathbf{k}')\right)U(\mathbf{k}')\right]_{mn},

where H0(k)H_0(\mathbf{k}) is the bare electronic Hamiltonian at wavevector k\mathbf{k} which is diagonalized in the Bloch basis by unitary matrices Uαn(k)U_{\alpha n}(\mathbf{k}).

Integrating out the phononic degrees of freedom, the effective electronic Hamiltonian is found to contain an interaction term given by

Hint=−∑qρ^q(∑νD0,qν∣γqν∣2eqν⊗eqν∗)ρ^−q,(2)H_{\mathrm{int}}=-\sum_{\mathbf{q}}\hat{\rho}_{\mathbf{q}}\left(\sum_{\nu}D_{0,\mathbf{q}\nu}\left|\gamma_{\mathbf{q}\nu}\right|^2\mathbf{e}_{\mathbf{q}\nu}\otimes\mathbf{e}^{*}_{\mathbf{q}\nu}\right)\hat{\rho}_{-\mathbf{q}}, \tag*{(2)}

where ρ^q=∑m,n,kλk,k−qmnc^m,k†c^n,k−q\hat{\rho}_{\mathbf{q}}=\sum_{m,n,\mathbf{k}}\lambda^{mn}_{\mathbf{k},\mathbf{k-q}}\hat{c}^{\dagger}_{m,\mathbf{k}}\hat{c}_{n,\mathbf{k-q}} is an appropriately weighted density operator and D0,qνD_{0,\mathbf{q}\nu} is the bare phonon propagator. The interaction vertex 14gq=∑νD0,qν∣γqν∣2eqν⊗eqν∗\frac{1}{4}g_{\mathbf{q}}=\sum_{\nu}D_{0,\mathbf{q}\nu}|\gamma_{\mathbf{q}\nu}|^2\mathbf{e}_{\mathbf{q}\nu}\otimes\mathbf{e}^{*}_{\mathbf{q}\nu} consists of the summation of the weighted phonon propagator across all phonon branches and is therefore, expected to be a full rank tensor with a well-defined inverse. For the purposes of deriving a CDW nesting-like criterion dependent on only the electronic Hamiltonian, gqg_{\mathbf{q}} will be assumed to be fairly flat over the q\mathbf{q} states of interest. While this is not expected to be the case generally, such an assumption allows for an isolation of the cause of any experimentally-observed instability to be either the bare electronic Hamiltonian (as we study here) or some strong phononic driving factors.

Decoupling in the density channel via a Hubbard-Stratonovich field ϕq\phi_{\mathbf{q}} (equal to ∑m,n,k⟨gqλk,k−qmnc^m,k†c^n,k−q⟩\sum_{m,n,\mathbf{k}}\langle g_{\mathbf{q}}\lambda^{mn}_{\mathbf{k},\mathbf{k-q}}\hat{c}^{\dagger}_{m,\mathbf{k}}\hat{c}_{n,\mathbf{k-q}}\rangle at mean field level) and expanding near ϕq=0\phi_{\mathbf{q}}=0, a Ginzburg-Landau functional can be obtained as (see Methods)

F=∑qϕqgq−1ϕ−q+∑n,{qi}fi1…in,{qi}(n)ϕq1i1ϕq2i2⋯ϕqnin,(3)\mathcal{F}=\sum_{\mathbf{q}}\phi_{\mathbf{q}}g_{\mathbf{q}}^{-1}\phi_{-\mathbf{q}}+\sum_{n,\{\mathbf{q}_i\}}f^{(n)}_{i_1\ldots i_n,\{\mathbf{q}_i\}}\phi^{i_1}_{\mathbf{q}_1}\phi^{i_2}_{\mathbf{q}_2}\cdots\phi^{i_n}_{\mathbf{q}_n}, \tag*{(3)}

where the {qi}\{\mathbf{q}_i\} satisfy ∑inqi=G\sum_{i}^{n}\mathbf{q}_i=\mathbf{G} (for reciprocal lattice vectors G\mathbf{G}), and each coefficient f(n)f^{(n)} in the expansion is an $nindex contravariant tensor derived in Methods. Particularly, Equation (3) implies that the standard Lindhard susceptibility (ordinarily appearing in the coefficient of the quadratic term) is now modified to a generically rank-2 tensor given by

χ(q)=12∑m,n,knF(ξn,k)−nF(ξm,k+q)ξm,k+q−ξn,kλk,k+qmn⊗λk,k+qmn∗(4)\chi(\mathbf{q}) = \frac{1}{2} \sum_{m,n,\mathbf{k}} \frac{n_F(\xi_{n,\mathbf{k}})-n_F(\xi_{m,\mathbf{k}+\mathbf{q}})}{\xi_{m,\mathbf{k}+\mathbf{q}}-\xi_{n,\mathbf{k}}} \lambda^{mn}_{\mathbf{k},\mathbf{k}+\mathbf{q}} \otimes\lambda^{mn*}_{\mathbf{k},\mathbf{k}+\mathbf{q}} \tag*{(4)}

where nF(ξn,k=En(k)−μ)n_F(\xi_{n,\mathbf{k}} = E_n(\mathbf{k})-\mu) is the Fermi-Dirac distribution function representing the occupancy of the band nn with energy En(k)E_n(\mathbf{k}) at chemical potential μ\mu.

For a CDW at a single wavevector, mean-field theory would then suggest that the system is susceptible to charge ordering at wavevector q∗\mathbf{q}^*, where the largest eigenvalue of χ\chi is maximized. While systems with perfect nesting would still be dominated by the divergence of the energetic factor, suppression due to the geometric λ\lambda factors assumes a significant role in imperfectly nested and flatband systems.

Writing the Hamiltonian in terms of single-band projectors Pαβn(k)=Uαn(k)Unβ†(k)P^n_{\alpha\beta}(\mathbf{k}) = U_{\alpha n}(\mathbf{k})U^\dagger_{n\beta}(\mathbf{k}) as H0(k)=∑nEn(k)Pn(k)H_0(\mathbf{k}) = \sum_n E_n(\mathbf{k})P^n(\mathbf{k}), the velocity operator in λkk′mn\lambda^{mn}_{\mathbf{k}\mathbf{k}'} can be separated into a band velocity term v1(k)v_1(\mathbf{k}) and a purely geometric term v2(k)v_2(\mathbf{k}) since ∂kH0(k)=∑n(∂kEn(k)Pn(k)+En(k)∂kPn(k))\partial_{\mathbf{k}}H_0(\mathbf{k}) = \sum_n(\partial_{\mathbf{k}}E_n(\mathbf{k})P^n(\mathbf{k}) + E_n(\mathbf{k})\partial_{\mathbf{k}}P^n(\mathbf{k})). In perfectly flatband systems where ∂kEn(k)\partial_{\mathbf{k}}E_n(\mathbf{k}) and hence, v1(k)v_1(\mathbf{k}) is vanishing everywhere in the BZ, λkk′mn\lambda^{mn}_{\mathbf{k}\mathbf{k}'} for the flat bands can be rewritten approximately as λkk′mn≈−iEg[U†(k)(∂kPˉk−∂k′Pˉk′)U(k′)]mn\lambda^{mn}_{\mathbf{k}\mathbf{k}'} \approx-iE_g[U^\dagger(\mathbf{k})(\partial_{\mathbf{k}}\bar{P}_{\mathbf{k}}-\partial_{\mathbf{k}'}\bar{P}_{\mathbf{k}'})U(\mathbf{k}')]_{mn}, where EgE_g is the bandgap between the flat bands and the higher bands (see Methods). PkP_{\mathbf{k}} is the projector onto the flatband subspace SkS_{\mathbf{k}} at wavevector k\mathbf{k}, while Pˉk=1−Pk\bar{P}_{\mathbf{k}}=1-P_{\mathbf{k}} projects onto its complement Sˉk\bar{S}_{\mathbf{k}}. The quadratic susceptibility in Equation (4) is then (see Methods)

χ(q)∝Eg2∑kTr⁡[PˉkQk+q+(k↔k+q)†](5)\chi(\mathbf{q}) \propto E_g^2 \sum_{\mathbf{k}} \operatorname{Tr}\left[\bar{P}_{\mathbf{k}}Q_{\mathbf{k}+\mathbf{q}}+\left(\mathbf{k}\leftrightarrow\mathbf{k}+\mathbf{q}\right)^\dagger\right] \tag*{(5)}

where Qˉk,μν=Pˉk(∂kμPk∂kνPk)Pˉk\bar{Q}_{\mathbf{k},\mu\nu}=\bar{P}_{\mathbf{k}}(\partial_{k_\mu}P_{\mathbf{k}}\partial_{k_\nu}P_{\mathbf{k}})\bar{P}_{\mathbf{k}} is the non-Abelian quantum geometric tensor (QGT) in the subspace Sˉ\bar{S} [1, 2], and the trace is over band indices only. Importantly, the susceptibility of the system to charge ordering at wavevector q\mathbf{q} is determined solely by the overlap between the QGT and the flatband projector at q\mathbf{q}-separated quasimomenta.

In hexagonal lattices, if the C3C_3 symmetry remains unbroken, ordering must occur at three wavevectors related by 2π/32\pi/3 rotations, henceforth called triple-Q solutions. Further, if these solutions are parameterized as ϕQ=ΔQn^QeiαQ\phi_Q=\Delta_Q\hat{n}_Qe^{i\alpha_Q} with ΔQ∈R\Delta_Q\in\mathbb{R} and n^Q∈S1\hat{n}_Q\in S^1, C3C_3 enforces the order parameter at Q0Q_0 and rotated counterparts Qn=R2πn/3Q0Q_n=R_{2\pi n/3}Q_0 to be related by ΔQn=ΔQ0\Delta_{Q_n}=\Delta_{Q_0} and n^Qn=R2πn/3n^Q0\hat{n}_{Q_n}=R_{2\pi n/3}\hat{n}_{Q_0} (see Methods for an irreducible representation analysis where we find this to be the most general symmetry-allowed form). Explicitly looking for such solutions, the free energy at an ordering wavevector Q∈{Q0,Q1,Q2}\mathbf{Q}\in\{Q_0,Q_1,Q_2\} can be written to fourth order as

F=(3Tr⁡g+Π)Δ2+12(γΔ3ei∑n=02αn+c.c.)+βΔ4(6)\mathcal{F}=\left(\frac{3}{\operatorname{Tr}g}+\Pi\right)\Delta^2+\frac{1}{2}\left(\gamma\Delta^3e^{i\sum_{n=0}^{2}\alpha_n}+\mathrm{c.c.}\right)+\beta\Delta^4 \tag*{(6)}

where the coefficients Π\Pi, γ\gamma, and β\beta involve contractions of the coefficients fij…(n)({On})f^{(n)}_{ij\ldots}(\{O_n\}) in Equation (3) with n^Qn\hat{n}_{Q_n} and can be derived from the loop diagrams shown in Figure 4 (see Methods). Particularly, ΠQ\Pi_Q is proportional to the smallest eigenvalue of the modified Lindhard susceptibility in Equation (4). The dependence of all quantities on Q\mathbf{Q} has been suppressed for clarity.

Diagrammatic representations of GL coefficients

Figure 4. Diagrammatic representations of GL coefficients. The effective action and hence the GL free energy functional affords diagrammatic representations as shown of the electron-phonon interaction at a quadratic, b cubic, and c quartic order. d At commensurate wavevectors (for example, 2/3ΓM2/3\Gamma M in the TMDs considered here), the cubic term involves an extra contribution from umklapp scattering, whose phase interference with the normal cubic term in (b) determines a commensurability transition. e All terms involve interaction vertices of this form, where a bare electronic state in band nn and wavevector k\mathbf{k} is scattered by a phonon of wavevector q\mathbf{q} into a state in band mm and wavevector k+q\mathbf{k}+\mathbf{q}. The dependence on electronic geometry is encoded in the interaction vertex λk,k+qnm\lambda_{\mathbf{k},\mathbf{k}+\mathbf{q}}^{nm}. Note that the diagonal nature of the bare Green’s function has been used to concisely represent G0,(mk),(nk′)\mathcal{G}_{0,(\mathbf{m}\mathbf{k}),(\mathbf{n}\mathbf{k}^{\prime})} as G0,mk\mathcal{G}_{0,m\mathbf{k}}.

Importantly, we note that, owing to the three Q\mathbf{Q} solutions satisfying (due to the presence of C3C_3 symmetry) ∑n=02Qn=0\sum_{n=0}^{2}\mathbf{Q}_n=0, the cubic term provides a nonzero contribution and drives a (weakly) first-order transition into the CDW state. Additionally, the phase αQn\alpha_{Q_n} only enters the functional through the cubic term and only as ∑n=02αQn\sum_{n=0}^{2}\alpha_{Q_n}. Minimizing F\mathcal{F} implies ∑n=02αQn=−arg⁡(γQ)+π\sum_{n=0}^{2}\alpha_{Q_n}=-\arg(\gamma_Q)+\pi, leaving two phases undetermined revealing the two gapless sliding phason modes for an incommensurate CDW in 2 dimensions.

Setting the phase as above, the transition into the ordered CDW phase occurs when

3/Tr⁡gQ≡ΘQ:=−ΠQ+∣γQ∣24βQ,(7)3/\operatorname{Tr}g_Q\equiv\Theta_Q:=-\Pi_Q+\frac{|\gamma_Q|^2}{4\beta_Q}, \tag*{(7)}

where the order parameter ΔQ\Delta_Q discontinuously changes from 00 to ∣γ∣/2βQ|\gamma|/2\beta_Q. The wavevector of the CDW corresponds to the Q\mathbf{Q} for which Equation (7) is satisfied first as temperature is reduced. For fairly flat gQg_Q, the system is thus unstable to ordering at the wavevector Q∗\mathbf{Q}^* that maximizes ΘQ\Theta_Q, henceforth referred to as the “CDW driver”. Note that in the absence of any cubic term, the driver reduces to −ΠQ-\Pi_Q (or χQ\chi_Q), and the standard CDW instability condition is recovered where a peak in the Lindhard susceptibility corresponds to charge ordering. We emphasize however that in the absence of strong Q\mathbf{Q}-dependence/singular behavior of the Lindhard susceptibility, the inclusion of the cubic term in the driver is essential in selecting Q∗\mathbf{Q}^*. Further, in systems with small band velocities, the Q\mathbf{Q}-dependence of all coefficients in ΘQ\Theta_Q is strongly modified by the quantum geometry. Considering the full CDW driver then becomes essential for analyzing nearly flat and quasi-flat band systems (as observed, for example, in the TMDs considered in this work).

Universal commensurability condition.— As a highlight, we find that commensurate Q\mathbf{Q} vectors contribute an additional diagram to the free energy summation in Equation (6). Particularly, Q=1/3G\mathbf{Q}=1/3\mathbf{G}, as commonly seen in experiment, contributes an additional diagram to the cubic term where equal-Q\mathbf{Q} ϕ\phi's multiply together since G0,k=G0,k+G\mathcal{G}_{0,\mathbf{k}}=\mathcal{G}_{0,\mathbf{k}+\mathbf{G}}. This extra diagram depends on the individual phases of ϕQn\phi_{Q_n} through e3iαQne^{3i\alpha_{Q_n}} (in contrast to just their sum) and thus, gaps the two phason modes expected for a commensurate CDW. Additionally, residual C3C_3 symmetry forces the three phases to be equal (αQn=αQ0\alpha_{Q_n}=\alpha_{Q_0}) leading to γ→γ+γ′\gamma\to\gamma+\gamma' and $3\alpha_Q=-\arg(\gamma+\gamma')+\pi. In such a case, the free energy for the commensurate CDW must be compared to the free energy of the optimal \mathbf{Q} from Equation (6), and a commensurability-incommensurability transition can be predicted. Explicitly, after fixing the phase as above, the cubic coefficient of \Delta_Q is given by (up to a constant factor) ∣γQ∣|\gamma_Q| for incommensurate CDW and by ∣γQ+γQ′∣|\gamma_Q + \gamma'_Q| for commensurate CDW. The commensurability transition is thus fixed by the relative phase between γQ\gamma_Q and γQ′\gamma'_Q.

Application of theory to TMDs.— We apply our framework to three TMD systems where the origin of CDW formation (in the absence of nesting) has been debated and largely attributed to phonon spectra. Particularly, the bare Lindhard susceptibility exhibits a very broad peak near 2/3ΓM2/3\Gamma M in these systems, thereby leading to conclusions that inhomogeneities in the phononic spectra would be responsible for selecting the CDW wavevector. Here, we show that the CDW transition, as well as a sharp dependence on QQ to select the ordering wavevector, can actually be derived purely from the electronic Hamiltonian.

In Figure 2, we plot ΘQ\Theta_Q in the region where the bare Lindhard susceptibility is flat and inconclusive, and show that the geometry-enhanced higher-order terms lead to sharp changes in the driver and are responsible for the CDW transition in these materials (see SI for plots of individual coefficients and variation with temperature). While the coefficients themselves and hence, ΘQ\Theta_Q may get renormalized due to other interactions in the material, a sharp increase in the driver near 2/3ΓM2/3\Gamma M is strongly suggestive of a CDW transition at a wavevector near here. Interestingly, setting λ\lambda factors to be identity, the driver in just its energetic (bare) form (orange dashed line in Figure 2) is still insufficient (see SI for a detailed discussion), and consideration of the quantum geometry hiding in λ\lambda becomes essential for an accurate description of charge order in these materials. Further, an analysis of the temperature variation of ΘQ\Theta_Q (see SI) reveals that up to the experimentally measured transition temperature [?, ?, ?, ?], the quadratic term ΠQ\Pi_Q in ΘQ\Theta_Q remains largely constant. On the contrary, the geometry-enhanced higher-order terms vary significantly with temperature and are thus, primarily responsible for driving the CDW transition.

Figure 2. CDW transition driver ΘQ\Theta_Q as a function of QQ on the ΓM\Gamma M line for different materials. ΘQ\Theta_Q is plotted as a function of QQ between 1/2ΓM1/2\Gamma M and 8/9ΓM8/9\Gamma M for a NbSe2_2 (T≈23T \approx23 K), b TaSe2_2 (T≈35T \approx35 K), and c TaS2_2 (T≈49T \approx49 K), and peaks near/at 2/3ΓM2/3\Gamma M are observed for all three materials. The QQ-domain is so chosen since the bare Lindhard susceptibility (dotted) exhibits a broad peak here and is inconclusive in determining the CDW wavevector. Inclusion of higher-order terms in just their energetic form (orange dashed line) is still inconclusive. The geometric enhancements studied here (purple solid line) then become the dominant mechanism behind CDW formation. Importantly, note that an incommensurate CDW is predicted for both NbSe2_2 (QQ just below 2/3ΓM2/3\Gamma M as well as an additional coexisting QQ phase near 0.59ΓM0.59\Gamma M) and TaS2_2 (Q=0.71ΓMQ = 0.71\Gamma M), whereas a commensurate CDW is predicted for TaSe2_2, similar to experiment. The peaks of the bare curves are scaled to the geometric curve for faithful comparison and displaced vertically for clarity.

Concretely, writing the ordering wavevector as Q=(1+δ)⋅2/3ΓM\mathbf{Q} = (1+\delta)\cdot2/3\Gamma M, we discover a peak just below 2/3ΓM2/3\Gamma M for NbSe2_2 (δ=−0.01\delta= -0.01), and a peak above 2/3ΓM2/3\Gamma M for TaS2_2 (δ=0.06\delta= 0.06); for TaSe2_2, we find a commensurate peak at exactly 2/3ΓM2/3\Gamma M, all of which closely match experimental results [?, ?, ?]. For NbSe2_2, two peaks of similar magnitude are found near 2/3ΓM2/3\Gamma M and 4/7ΓM4/7\Gamma M (δ=−0.01\delta= -0.01 and −0.11-0.11 respectively), suggesting that different CDW orders may coexist, a prediction that is indeed corroborated in STM experiments [?].

The interference between the normal triangle scattering terms for the cubic coefficient and the additional umklapp scattering diagram for commensurate QQ can be explicitly seen in the kk-resolved integrand for the cubic coefficient (see Methods) in Figure 3. We emphasize the phase difference between the triangle scattering terms (top row) and umklapp scattering terms (bottom row) by using different colors for positive and negative contributions to Re⁡(γQei∑n=02αQn)\operatorname{Re}\left(\gamma_Q e^{i\sum_{n=0}^{2}\alpha_{Q_n}}\right). It is then easy to see that while NbSe2_2 exhibits destructive interference (the triangle and umklapp scattering at the same kk add with opposite signs), the two terms in TaSe2_2 add constructively. Since the triangle scattering can be expected to be smooth in QQ, destructive (constructive) interference with the umklapp scattering process in NbSe2_2 (TaSe2_2) suggests an incommensurate (commensurate) CDW, as seen in experiment. In TaS2_2, the two terms add with differing relative signs at different kk points, and the extra diagram turns out to not affect the cubic coefficient significantly (as in the smooth behavior of the driver near 2/3ΓM2/3\Gamma M in Figure 2). Such a weak dependence of the system on the commensurate umklapp diagram is likely to result in an incommensurate CDW as well (as corroborated by experiment).

Integrand for cubic coefficient at commensurate $Q = 2/3\Gamma M$

Figure 3. Integrand for cubic coefficient at commensurate Q=2/3ΓMQ = 2/3\Gamma M. Normalized contributions to the coefficient of Δ3\Delta^3 in the Ginzburg-Landau free energy of different materials has been shown as a function of k\mathbf{k} in the Brillouin Zone. The integrand consists of two terms: a-c triangle scattering processes where ∑Qi=0\sum\mathbf{Q}_i = 0, and d-f umklapp scattering processes where ∑Qi=G\sum\mathbf{Q}_i = \mathbf{G}, a reciprocal lattice vector. Corresponding diagrams for the g triangle and h umklapp scattering are also sketched. The phase of the respective integrands is emphasized using different colors for positive and negative contributions. Interference between the triangle and umklapp scattering suggests incommensurate CDW in NbSe2_2 and commensurate CDW in TaSe2_2. For TaS2_2, both constructive and destructive interference is present at different k\mathbf{k}, and thus, the extra diagram does not strongly affect the functional form for the CDW driver.

Discussions.— We discover two additional factors that contribute to charge ordering instabilities in imperfectly nested systems. In systems where the standard bare Lindhard susceptibility exhibits no sharp behavior as a function of ordering wavevector Q\mathbf{Q}, higher-order terms and geometric effects may combine to drive an instability. While the former modifies the transition condition by contributing an additive term on top of the quadratic susceptibility (see (7)), the latter modifies all terms in a significant Q\mathbf{Q}-dependent manner to drive the transition (see Figure 2).

We emphasize that the dominant higher-order contribution to the transition comes from the cubic term which is only expected to be non-zero for triple-Q\mathbf{Q} CDW states as in C3_3-symmetric materials like the TMDs considered here (in contrast to single-Q\mathbf{Q} or double-Q\mathbf{Q} ordering in 1D or 2D orthorhombic/tetragonal lattices). Additionally, owing to the form of the cubic term (see Methods), we do not expect it to show any divergences (unless the system is perfectly nested ξk=−ξk+q\xi_{\mathbf{k}} = -\xi_{\mathbf{k}+\mathbf{q}} and the quadratic term already diverges). Therefore, this contribution can only be expected to play a dominant role in imperfectly-nested hexagonal materials. In such materials, since the Fermi-surfaces themselves are C3_3 symmetric, the cubic term connects three points on the Fermi surface only for a narrow band of Q\mathbf{Q}-vectors and is hence, the driving factor in selecting the ordering wavevector (see Figure 2 and Figure 3).

Quantum geometric factors on the contrary may be dominant in enhancing or suppressing the susceptibility in all kinds of imperfectly nested materials. We reiterate that dependence on the electronic quantum geometry lives in the λkk′mn\lambda_{\mathbf{k}\mathbf{k}'}^{mn} factors, and a scaling analysis reveals when geometry can be expected to dictate the system’s instability. v1(k)v_1(\mathbf{k}), which is dependent on the band velocity, can generally be expected to scale as ∼Wa\sim Wa, where WW is the bandwidth and aa is the lattice constant. Since ∂kH0(k)\partial_{\mathbf{k}}H_0(\mathbf{k}) scales as ∼ta\sim ta, where tt is a measure of inter-atomic coupling, and t∼Wt \sim W usually, v2(k)v_2(\mathbf{k}) has a similar order of magnitude as v1(k)v_1(\mathbf{k}). However, in flatband systems where W→0W \to0, nontrivial geometry can cause v2(k)≫v1(k)v_2(\mathbf{k}) \gg v_1(\mathbf{k}), and geometric effects will dominate. Additionally, the W→0W \to0 limit is also likely to result in flat generalized Lindhard susceptibilities appearing in the fij...(n){qi}f_{ij...}^{(n)}\{\mathbf{q}_i\} coefficients in the GL functional (see (3) and Methods), therefore allowing the coefficients to be re-expressed purely in terms of geometric quantities.

Specifically for the quadratic susceptibility, we find that the geometric factor assumes a particularly simple form and is equal to the QGT in the non-occupied subspace Sˉk\bar{S}_{\mathbf{k}} projected onto the flatband subspace at a quasi-momentum q\mathbf{q} away (summed over all k\mathbf{k}). It is also straightforward to see from Equation (5) that this geometric enhancement is maximized when PkSˉk+q=Sˉk+q ∀kP_{\mathbf{k}}\bar{S}_{\mathbf{k}+\mathbf{q}} = \bar{S}_{\mathbf{k}+\mathbf{q}}\ \forall\mathbf{k}; in other words, we find that the system is most unstable to ordering if the Bloch bundle at k\mathbf{k} rotates into the Bloch bundle at k+q\mathbf{k}+\mathbf{q}. Formally, we require that the subspace VV in Sˉk\bar{S}_{\mathbf{k}} that the projector PkP_{\mathbf{k}} rotates into (that is, VV spanned by the eigenvectors of Qˉk\bar{Q}_{\mathbf{k}}) is strictly a subset of Sk+qS_{\mathbf{k}+\mathbf{q}}. A more careful analysis involving sub-leading order terms (see Methods) additionally requires ∂kPk=−∂k+qPk+q\partial_k P_k = -\partial_{k+q} P_{k+q} for χ(q)\chi(\mathbf{q}) to truly saturate its upper bound. Importantly, we note that these, being local criteria in k\mathbf{k}, are not well-described by the global (integrated over k\mathbf{k}) geometric nesting criteria defined in Refs. [?, ?]. Particularly, in real materials where the Hamiltonians and hence, the quantum geometry can deviate significantly from ideal scenarios, the conditions defined here, if satisfied for an extended subspace of the Brillouin Zone, will drive a CDW transition even if they are not globally satisfied (akin to Fermi surface nesting). Additionally, since they emerge naturally as modifications to the Ginzburg-Landau free energy coefficients, quantitative conclusions can be made both for systems where the aforementioned upper bound is not saturated and where inclusion of higher order terms (beyond the quadratic susceptibility in Ref. [?]) becomes necessary, thereby allowing for a more complete theory replete with an exact description of the order parameter as well.

Local criteria in k\mathbf{k} (as derived here) become even more important in quasi-flatband systems (such as the TMDs considered here) where kBT≪W≪Egk_B T \ll W \ll E_g, that is, when the system has a well-defined Fermi surface. Here, only states at or near the Fermi surface are relevant (quantitatively chosen by the Lindhard susceptibility), and ordering criteria necessarily need to be local in k\mathbf{k}. We find that strong variations of orbital character at the Fermi energy can result in large v2(k)v_2(\mathbf{k}). Particularly, it can be shown (see Methods) that to leading order in such systems, v2(k)∼−Eg∂kPkv_2(\mathbf{k}) \sim-E_g \partial_k P_k, and thus the strength of ∂kPk\partial_k P_k at the Fermi energy (where the bare Lindhard susceptibility is largest) determines the system’s instability to charge ordering. Explicitly, the same conditions as in the perfectly flat scenario must be satisfied here, but only within a region ∼±kBT\sim\pm k_B T of the chemical potential (see Figure 1). Intuitively, these translate to suggesting that geometric enhancements are most pronounced when the occupied states at k\mathbf{k} acquire the flavor character of unoccupied states at k+q\mathbf{k}+\mathbf{q} near the Fermi energy. Additionally, this effect is strongest when the states at k\mathbf{k} and k+q\mathbf{k}+\mathbf{q} completely exchange flavor character near the Fermi energy (for example, at band inversion points). Note that this condition is distinct from the two points already having similar flavor character since Pk∼Pk+q⇒χ(q)→0P_k \sim P_{k+q} \Rightarrow\chi(\mathbf{q}) \to0.

We also derive universal commensurability criteria grounded in the microscopics of a material. Specifically, we note that at commensurate Q\mathbf{Q}-vectors, the free energy can receive additional contributions from terms where the Q\mathbf{Q}-vectors add to a reciprocal lattice vector. Depending on whether such additional contributions are in-phase or out-of-phase with the incommensurate contributions, a commensurability-incommensurability transition can be predicted (see Figure 3). These extra terms also lock the phase of the order parameter, leading to the gapping of the phason Goldstone bosons.

While this work is focused on long-standing questions on CDW formation in imperfectly nested systems, the ideas behind the analysis here are easily generalizable to other phenomena too. For example, we are working on decoupling the interaction term in (2) in the pairing channel to determine geometric contributions to phonon-mediated superconductivity and pair-density wave formation. Since the TMDs considered here also exhibit superconductivity at low temperatures, an analysis of the competition between the pairing and density channel decouplings can reveal novel insights into the interplay between CDW formation and superconductivity hypothesized from experiment [?, ?, ?]. Additionally, we expect the superconducting gap to inherit the strong k\mathbf{k}-dependence of the λkk′mn\lambda_{kk'}^{mn}, thus providing insights into topological superconductivity as well. It would also be interesting to investigate how the addition of a Coulomb potential changes the form of the λkk′mn\lambda_{kk'}^{mn} factors and whether the system can be driven into a strongly correlated phase such as a Wigner crystal. Magnetic interaction terms also pose intriguing questions; here, the spin-spin order parameter (being dependent on Cartesian indices) acquires even more structure, and exotic magnetic textures may be investigated with associated geometric enhancements. Additionally, if these order parameters transform in different irreducible representations, the resulting Ginzburg-Landau theory could have important topological effects too.

Our work is qualified by the applicability of the electron-phonon matrix element assumed in ref. [?]. Generally, λkk′mn=∫ψmk∗∇Uψnk′ dr\lambda_{kk'}^{mn} = \int\psi_{mk}^{*} \nabla U \psi_{nk'} \, dr is the electronic matrix element of the gradient of the crystal potential U(r)U(r) [?]. It is related to the standard electron-phonon coupling constant λ\lambda in Eliashberg theory [?] by an average over the Fermi surface λ∝⟨λ2⟩\lambda\propto\langle\lambda^2 \rangle [?]. It can be argued that its equivalence with the difference of velocity operators is at leading order and is a good approximation for orbitals that are well-localized on the atomic sites (such as the dd orbitals in transition metal compounds) [?]. Alternatively, it can be derived exactly by employing a Gaussian approximation for the coupling strength as a function of distance, an assumption that is well-justified for ss orbitals as well as orbitals with ss-character in the dimensionality of the material (such pzp_z orbitals in 2D) [?]. In general, it can be expected to hold wherever angular contributions to the gradient of the crystal potential are either negligible or have the same dependence as the radial contributions to leading order.

Conclusions.—To summarize, we find that charge ordering in imperfectly nested systems can be driven by higher order terms and geometric enhancements at specific Q\mathbf{Q}-vectors. We apply our theory to TMDs where standard bare Lindhard susceptibility calculations exhibit no strong dependence on ordering wavevectors and are thus found to be insufficient to explain experimentally-found charge order near 2/3GM2/3\mathrm{GM}. We also predict commensurability transitions in such materials derived from the microscopic Hamiltonians. More generally, our theory can be applied to all systems where phononic effects play dominant roles and where existing calculations based on band energetics alone are found to be insufficient. We discover that systems where the Bloch orbital character rotates rapidly near the Fermi energy (thus imbuing non-trivial quantum geometry) are more susceptible to ordering. Finally, we note that analyses in a similar vein can be used to derive corresponding geometric conditions for other phenomena such as superconductivity and magnetic ordering as well.

Methods

Ginzburg-Landau functional

We construct the imaginary-time partition function for electronic fields ψm,k\psi_{m,k} in the presence of phonon-mediated interactions (as described by HintH_{\mathrm{int}} in (2))

Z=∫D[ψˉ,ψ]exp⁡(−Sel+T34∑qρqgqρ−q),(8)\mathcal{Z}=\int\mathcal{D}[\bar{\psi},\psi]\exp\left(-S_{\mathrm{el}}+\frac{T^{3}}{4}\sum_{q}\rho_{q}g_{q}\rho_{-q}\right), \tag*{(8)}

where Sel=T∑nkψˉn,k(−iωp+ξn,k)ψn,kS_{\mathrm{el}}=T\sum_{nk}\bar{\psi}_{n,k}(-i\omega_{p}+\xi_{n,k})\psi_{n,k}, and the geometric λkk′mn\lambda_{kk'}^{mn} factors are hidden in ρq=∑kλk+q,kmnψˉm,k+qψn,k\rho_{q}=\sum_{k}\lambda_{k+q,k}^{mn}\bar{\psi}_{m,k+q}\psi_{n,k}. The summations are carried out over k=(ωp,k)k=(\omega_{p},\mathbf{k}) and q=(ωp′,q)q=(\omega'_{p},\mathbf{q}), where ωp=π(2p+1)T\omega_{p}=\pi(2p+1)T are the fermionic Matsubara frequencies and ωp′=2πpT\omega'_{p}=2\pi pT are the bosonic Matsubara frequencies.

Introducing a Hubbard-Stratonovich decoupling field ϕq\phi_{q} in the density-density channel, the partition function can be rewritten as

Z=∫D[ψˉ,ψ,ϕ]exp⁡(−Sel−Sint),(9)\mathcal{Z}=\int\mathcal{D}[\bar{\psi},\psi,\phi]\exp(-S_{\mathrm{el}}-S_{\mathrm{int}}), \tag*{(9)}

where Sint=1T∑qϕˉqgq−1ϕq−T∑qρ−q⋅ϕqS_{\mathrm{int}}=\frac{1}{T}\sum_{q}\bar{\phi}_{q}g_{q}^{-1}\phi_{q}-T\sum_{q}\rho_{-q}\cdot\phi_{q}. Note that ϕq\phi_{q} inherits the structure of ρq\rho_{q} (to which it is coupled) such that ϕˉq=ϕ−q\bar{\phi}_{q}=\phi_{-q}.

Integrating out the fermionic degrees of freedom now, the effective action can be rewritten as

Seff[ϕ]=1T∑qϕˉqgq−1ϕq−Tr⁡ln⁡G−1,(10)S_{\mathrm{eff}}[\phi]=\frac{1}{T}\sum_{q}\bar{\phi}_{q}g_{q}^{-1}\phi_{q}-\operatorname{Tr}\ln\mathcal{G}^{-1}, \tag*{(10)}

where the electronic Green’s function G\mathcal{G} can be expressed using the non-interacting Green’s function G0,(mk),(nk′)=[T(−iωp+ξn,k)]−1δmnδkk′\mathcal{G}_{0,(mk),(nk')}=[T(-i\omega_{p}+\xi_{n,k})]^{-1}\delta_{mn}\delta_{kk'} as

G(mk),(nk′)−1:=(G0−1+Φ^)(mk),(nk′)(11)\mathcal{G}^{-1}_{(mk),(nk')}:=(\mathcal{G}_{0}^{-1}+\hat{\Phi})_{(mk),(nk')} \tag*{(11)}
=G0,(mk),(nk′)−1+T∑qλk′,kmn⋅ϕk′−k.(12)=\mathcal{G}^{-1}_{0,(mk),(nk')}+T\sum_{q}\lambda_{k',k}^{mn}\cdot\phi_{k'-k}. \tag*{(12)}

Near the CDW transition, we expect ϕq\phi_{q} to be small and can expand the action in powers of ϕq\phi_{q} to obtain the Ginzburg-Landau functional. Writing the second term in (10) as −Tr⁡ln⁡(G0−1(I+G0Φ^))=−Tr⁡ln⁡G0−1+∑j(−1)jTr⁡(G0Φ^)j/j-\operatorname{Tr}\ln(\mathcal{G}_{0}^{-1}(\mathbb{I}+\mathcal{G}_{0}\hat{\Phi}))=-\operatorname{Tr}\ln\mathcal{G}_{0}^{-1}+\sum_{j}(-1)^{j}\operatorname{Tr}(\mathcal{G}_{0}\hat{\Phi})^{j}/j, and ignoring the term independent of ϕ\phi, the effective action can be written as

Seff[ϕ]=1T∑q1g(q)∣ϕq∣2+∑j(−1)jjTr⁡(G0Φ^)j.(13)S_{\mathrm{eff}}[\phi]=\frac{1}{T}\sum_{q}\frac{1}{g(q)}|\phi_{q}|^{2}+\sum_{j}\frac{(-1)^{j}}{j}\operatorname{Tr}(\mathcal{G}_{0}\hat{\Phi})^{j}. \tag*{(13)}

Diagrammatically, Φ^\hat{\Phi} connects bare electronic propagators G0\mathcal{G}_{0} at different wavevectors, and the associated diagrams for the quadratic, cubic and quartic terms are shown in Figure 4.

Within a mean-field description and in the static limit (setting ωp′=0\omega'_{p}=0), (3) for the free energy is obtained with the coefficients f(n){qi}f^{(n)}\{\mathbf{q}_{i}\} given by

f(n){qi}=1n∑{mi},kL(n){ξmi,ki}⨂inλki,ki+1mimi+1,(14)f^{(n)}\{\mathbf{q}_{i}\}=\frac{1}{n}\sum_{\{m_{i}\},\mathbf{k}}L^{(n)}\{\xi_{m_{i},k_{i}}\}\bigotimes_{i}^{n}\lambda_{k_{i},k_{i+1}}^{m_{i}m_{i+1}}, \tag*{(14)}

where the ki=k+∑j=1i−1qj\mathbf{k}_{i}=\mathbf{k}+\sum_{j=1}^{i-1}\mathbf{q}_{j} form an nn-sided polygon in the BZ (as earlier, ∑inqi=G\sum_{i}^{n}\mathbf{q}_{i}=\mathbf{G}), and mn+1=m1m_{n+1}=m_{1}. A Matsubara summation has been carried out over all fermionic frequencies ωp\omega_{p} to obtain generalized bare susceptibilities L(n){ξmi,ki}=∑innF(ξmi,ki)/∏j≠i(ξmi,ki−ξmj,kj)L^{(n)}\{\xi_{m_{i},k_{i}}\}=\sum_{i}^{n}n_{F}(\xi_{m_{i},k_{i}})/\prod_{j\ne i}(\xi_{m_{i},k_{i}}-\xi_{m_{j},k_{j}}). Note that for n=2n=2, −f(2){qi}-f^{(2)}\{\mathbf{q}_{i}\} reduces to the χ(q)\chi(\mathbf{q}) in (4), and L(2)L^{(2)} reproduces the standard Lindhard susceptibility. While (14) may in general give complex f(n)f^{(n)} for a particular set {qi}\{\mathbf{q}_{i}\}, the GL expansion contains a term formed from the set {−qi}\{-\mathbf{q}_{i}\} too, which is the complex conjugate of the former and thus always results in real coefficients as expected.

Flatband limit

While (14) is general, we elaborate on the specific flatband limit where it considerably simplifies. Particularly, we investigate the scenario when the chemical potential lies within a set of NflatN_{\mathrm{flat}} bands (called flatbands) which are well-separated in energy from all other bands. Explicitly, their bandwidth WW is much smaller than the band gap EgE_{g}. For simplicity, we also assume that the thermal scale in the system is such that these NflatN_{\mathrm{flat}} bands are roughly equally occupied at all kk (this assumption can be relaxed later), and excitations across the band gap are negligible. Effectively, we work within the limit such that W≪kBT≪EgW\ll k_{B}T\ll E_{g}. As a result, the generalized bare susceptibilities L(n){ξmi,ki}L^{(n)}\{\xi_{m_{i},k_{i}}\} become approximately constant over kk, and at leading order, only contain contributions from bands mi∈[1,Nflat]m_{i}\in[1,N_{\mathrm{flat}}] (owing to the energy differences in the denominator, any other band contributions is suppressed by a factor of W/EgW/E_{g}. The Ginzburg-Landau coefficients thus simplify to

f(n){qi}≈1NflatTr⁡k⨂in(PkiΛki,ki+1),(15)f^{(n)}\{\mathbf{q}_{i}\}\approx\frac{1}{N_{\mathrm{flat}}}\operatorname{Tr}_{\mathbf{k}}\bigotimes_{i}^{n}\left(P_{k_{i}}\Lambda_{k_{i},k_{i+1}}\right), \tag*{(15)}

where Λk,k′,αβ=Uαm(k)λk,k′mnUβn†(k′)\Lambda_{k,k',\alpha\beta}=U_{\alpha m}(\mathbf{k})\lambda_{k,k'}^{mn}U_{\beta n}^{\dagger}(\mathbf{k}') and ⨂\bigotimes denotes normal matrix multiplication over flavor indices but a tensor product over Cartesian indices. The trace is over flavor indices as well, and as before, ki=k+∑j=1i−1qj\mathbf{k}_{i}=\mathbf{k}+\sum_{j=1}^{i-1}\mathbf{q}_{j}. In the small bandwidth limit, contributions to λk,k′mn\lambda_{k,k'}^{mn} are dominated entirely by the geometric part of the velocity operator v2(k)=∑nEn(k)∂kPn(k)\mathbf{v}_{2}(\mathbf{k})=\sum_{n}E_{n}(\mathbf{k})\partial_{\mathbf{k}}\mathcal{P}^{n}(\mathbf{k}). If we additionally assume that bands which share flavor character with the flat bands have bandwidth much smaller than EgE_g (the generalization to multiple well-separated subspaces of bands is just additive in the respective energy gaps), ∂kH0(k)≈Eg∂kPˉk\partial_{\mathbf{k}}H_0(\mathbf{k}) \approx E_g\partial_{\mathbf{k}}\bar{P}_{\mathbf{k}}, and λk,k′mn\lambda_{\mathbf{k},\mathbf{k}^{\prime}}^{mn} reduces to the expression given earlier involving differences of projector derivatives. Importantly, this allows the GL coefficients to be rewritten purely in terms of geometric quantities as

f(n){qi}=(iEg)n∑kTr⁡^⨂i=1n[Pki(∇kiPki−∇ki+1Pki+1)Pki+1].(16)f^{(n)}\{\mathbf{q}_i\}=(iE_g)^n\sum_{\mathbf{k}}\widehat{\operatorname{Tr}}\bigotimes_{i=1}^{n}\left[P_{\mathbf{k}_i}\left(\nabla_{\mathbf{k}_i}P_{\mathbf{k}_i}-\nabla_{\mathbf{k}_{i+1}}P_{\mathbf{k}_{i+1}}\right)P_{\mathbf{k}_{i+1}}\right]. \tag*{(16)}

For n=2n=2, Equation (16) results in two types of geometric terms, one of which is given in Equation (5). The other term is proportional to −1/2∑k(Pk∇kPˉk+Pk+q∇k+qPˉk+q)⊗b2+h.c.-1/2\sum_{\mathbf{k}}(P_{\mathbf{k}}\nabla_{\mathbf{k}}\bar{P}_{\mathbf{k}}+P_{\mathbf{k}+\mathbf{q}}\nabla_{\mathbf{k}+\mathbf{q}}\bar{P}_{\mathbf{k}+\mathbf{q}})\otimes b_2+\mathrm{h.c.}, which in general requires both terms inside the brackets to be significant (i.e., it requires the off-diagonal Sk→SˉkS_{\mathbf{k}}\to\bar{S}_{\mathbf{k}} block of the non-Abelian Berry connection to be non-negligible at both k\mathbf{k} and k+q\mathbf{k}+\mathbf{q}). Based on a degree-of-freedom counting analysis then, we argue that the QGT term in Equation (5) is the dominant contribution to χ(q)\chi(\mathbf{q}). Note that an additive contribution from these sub-leading order terms can be ensured if sgn⁡(∇kPˉk)=−sgn⁡(∇k+qPˉk+q)\operatorname{sgn}(\nabla_{\mathbf{k}}\bar{P}_{\mathbf{k}})=-\operatorname{sgn}(\nabla_{\mathbf{k}+\mathbf{q}}\bar{P}_{\mathbf{k}+\mathbf{q}}), which reproduces the conditions noted earlier for χ(q)\chi(\mathbf{q}) to saturate its upper bound. The

higher order terms can similarly be expanded in terms of projectors and their derivatives at different k\mathbf{k}-points; they do not, however, afford a closed form expression in terms of the QGT as in (5).

In quasi-flatband systems, W≪kBTW\ll k_BT is no longer a good approximation. However, the results from this section, albeit with some modifications, still apply. Working instead in the kBT≪W≪Egk_BT\ll W\ll E_g limit, the generalized bare susceptibilities L(n){ξmi,ki}L^{(n)}\{\xi_{m_i,k_i}\} are no longer constant over k\mathbf{k}, but are instead sharply peaked at the Fermi surface (with width in k-space given by ∼kBT/Wa≪1/a\sim k_BT/Wa\ll1/a). The summation over k\mathbf{k} in (14) then effectively reduces to a sum over the Fermi surface multiplied by some constant. (16) is similarly modified with the summation now running over k\mathbf{k} along the Fermi surface, along with an overall multiplicative scaling arising from the strength of L(n)L^{(n)} at the relevant q\mathbf{q}-vectors. Importantly, if L(n)L^{(n)} is relatively flat as a function of q\mathbf{q}, CDW formation is driven entirely by the same geometric factors derived in this section - this is the case for the TMDs considered in this work, where Figure 2 highlight the importance of the geometric factors in selecting the CDW Q-vectors.

Hexagonal lattices

In this section, we derive the coefficients f(n)f^{(n)} for hexagonal lattices where we get triple-Q charge order and show how the tensorial GL expansion simplifies to the scalar version in (6). Explicitly looking for such solutions effectively requires expressing FQj\mathcal{F}_{Q_j} as dependent only on ϕQj\phi_{Q_j} (for Qj=R2πj/3Q0Q_j = R_{2\pi j/3}Q_0) and minimizing it to obtain the ordering vectors QpQ_p. The allowed terms F(n)\mathcal{F}^{(n)} at any order nn in the GL expansion then become heavily constrained by the condition ∑iqi=G\sum_i q_i = G, and the free energy up to quartic order for incommensurate Q0Q_0 can be written as

FQ0=∑jF±Qj(2)+(FQ0,Q1,Q2(3)+c.c.)+∑i,jF±Qi,±Qj(4)(17)\mathcal{F}_{Q_0} = \sum_j \mathcal{F}^{(2)}_{\pm Q_j} + \left(\mathcal{F}^{(3)}_{Q_0,Q_1,Q_2}+\mathrm{c.c.}\right) + \sum_{i,j}\mathcal{F}^{(4)}_{\pm Q_i,\pm Q_j} \tag*{(17)}

where the dependence of each F(n)\mathcal{F}^{(n)} on the nn QQ-dependent order parameters is made explicit. For commensurate Q0Q_0 such that 3Q0=G3Q_0 = G, there are additional cubic ∑jFQj,Qj,Qj(3)+c.c.\sum_j \mathcal{F}^{(3)}_{Q_j,Q_j,Q_j}+\mathrm{c.c.} and quartic ∑jFQj,Qj,−Qj+1,−Qj+2(4)+c.c.\sum_j\mathcal{F}^{(4)}_{Q_j,Q_j,-Q_{j+1},-Q_{j+2}}+\mathrm{c.c.} terms.

The C3C_3 symmetry of the lattice forces FQ0\mathcal{F}_{Q_0} to transform under the trivial irrep (irreducible representation) and thus imposes additional structure on the order parameter. Specifically, owing to the C3C_3 symmetry of the H0(k)H_0(\mathbf{k}), each coefficient f(n)f^{(n)} transforms in the trivial irrep (f{qi}(n)→f{R2πj/3−1qi}(n)=f{qi}(n))(f^{(n)}_{\{q_i\}} \to f^{(n)}_{\{R_{2\pi j/3}^{-1}q_i\}} = f^{(n)}_{\{q_i\}}); consequently, the objects depending on the order parameter in F(n)\mathcal{F}^{(n)} must also transform in the trivial irrep. For example, for the quadratic term, this object is the covariant tensor ϕQj⊗ϕˉQj\phi_{Q_j}\otimes\bar{\phi}_{Q_j}, and transformation in the trivial irrep implies ϕR2πj/3Qj⊗ϕˉR2πj/3Qj=(R2πj/3ϕQj)⊗(R2πj/3ϕˉQj)\phi_{R_{2\pi j/3}Q_j}\otimes\bar{\phi}_{R_{2\pi j/3}Q_j} = (R_{2\pi j/3}\phi_{Q_j})\otimes(R_{2\pi j/3}\bar{\phi}_{Q_j}). Importantly, this leaves the overall phase of the order parameter across different QjQ_j unconstrained and allows it to be written as ϕQj=eiαQjΔQ0R2πj/3n^Q0\phi_{Q_j}=e^{i\alpha_{Q_j}}\Delta_{Q_0}R_{2\pi j/3}\hat{n}_{Q_0} (the corresponding cubic and quartic objects do not reduce the degrees of freedom of ϕ\phi any further).

For commensurate Q0Q_0, ϕQ03\phi_{Q_0}^3, and hence, ϕQj\phi_{Q_j} also transforms in the trivial irrep under C3C_3. The phases thus become constrained (αQj=αQ0 ∀j)(\alpha_{Q_j}=\alpha_{Q_0}\ \forall j), and two degrees of freedom in the order parameter are lost corresponding to the gapping of the two phason modes during the commensurability transition.

Premultiplying the tensorial coefficients f{Qj}(n)f^{(n)}_{\{Q_j\}} with tensor products of the relevant rotation matrices R2πj/3−1R^{-1}_{2\pi j/3}, the free energy can be expressed in terms of contractions with just the vectors n^Q0\hat{n}_{Q_0}. Suppressing the Q0Q_0 dependence henceforth for clarity, the free energy up to quartic order is

F=an^⊗2Δ2+Re⁡(cn^⊗3Δ3ei∑j=02αj)+bn^⊗4Δ4(18)\mathcal{F}=a\hat{n}^{\otimes2}\Delta^2+\operatorname{Re}\left(c\hat{n}^{\otimes3}\Delta^3e^{i\sum_{j=0}^{2}\alpha_j}\right)+b\hat{n}^{\otimes4}\Delta^4 \tag*{(18)}

where aa, cc and bb are totally symmetric rank-2, 3, and 4 tensors dependent on the f(n)f^{(n)} coefficients and are contracted with n^\hat{n} above. Identifying αj=α0 ∀j\alpha_j=\alpha_0\ \forall j reproduces the form for the commensurate case. Diagonalizing aa to get eigenvalues λi\lambda_i and eigenvectors ei,0e_{i,0}, the tensors cc and dd can be re-expressed in the eigenbasis of aa. The order parameter then in this basis becomes ϕj=(φe1,0+κe2,0)eiαj\phi_j=(\varphi e_{1,0}+\kappa e_{2,0})e^{i\alpha_j}.

Since the system is likely to exhibit a preference for condensing along a particular phonon polarization, the free energy is expected to be highly anisotropic in the φ\varphi-κ\kappa space (see SI for evidence regarding the same in the TMDs considered here). Concretely, we assume that near the transition temperature, one of the eigenvalues of aa is much smaller than the other (without loss of generality, λ1≪λ2\lambda_1\ll\lambda_2). Along the Γ\Gamma line in the TMDs studied here, f{±q}(2)f^{(2)}_{\{\pm q\}} was explicitly found to be highly anisotropic with a preference for the longitudinal polarization. Then, it is expected that the order parameter (φ∗,κ∗)(\varphi_*,\kappa_*) that minimizes F\mathcal{F} must be along the φ\varphi-axis to leading order, that is, κ∗\kappa_* is at least O(φ∗2)O(\varphi_*^2) (this claim will be self-consistently verified later). Up to quartic order in φ\varphi, the free energy reduces to

F=λ1φ2−c~111φ3+(b1111φ4−3c~112φ2κ+λ2κ2)(19)\mathcal{F}=\lambda_1\varphi^2-\tilde{c}_{111}\varphi^3+\left(b_{1111}\varphi^4-3\tilde{c}_{112}\varphi^2\kappa+\lambda_2\kappa^2\right) \tag*{(19)}

where the phase dependence has been absorbed into the cubic coefficients c~=−Re⁡cei∑j=02αj\tilde{c}=-\operatorname{Re}ce^{i\sum_{j=0}^{2}\alpha_j}. Minimizing with respect to κ\kappa reveals κ∗=3c~112φ2/2λ2\kappa_* = 3\tilde{c}_{112}\varphi^2/2\lambda_2. Substituting back into the free energy, the quartic coefficient becomes b1111′=b1111−9c~1122/4λ2b'_{1111}=b_{1111}-9\tilde{c}_{112}^2/4\lambda_2, and the global minimum shifts from φ∗=0\varphi_*=0 (disordered) to φ∗=c~111/2b1111′\varphi_*=\tilde{c}_{111}/2b'_{1111} (ordered) when λ1=c~1112/4b1111′\lambda_1=\tilde{c}_{111}^2/4b'_{1111}. The off-axis component of the minima is κ∗=3(λ1/λ2)c~112φ∗/c~111≪φ∗\kappa_*=3(\lambda_1/\lambda_2)\tilde{c}_{112}\varphi_*/\tilde{c}_{111}\ll\varphi_* since λ1/λ2≪1\lambda_1/\lambda_2\ll1 near the transition, thus verifying the earlier claim.

Writing Δ=φ2+κ2≈φ(1+9c~1122φ2/8λ22)\Delta=\sqrt{\varphi^2+\kappa^2}\approx\varphi\left(1+9\tilde{c}_{112}^2\varphi^2/8\lambda_2^2\right), and comparing the free energy coefficients in (19) with (6), one can identify (3/Tr⁡−Π)=λ1(3/\operatorname{Tr}-\Pi)=\lambda_1, γ=c111\gamma=c_{111}, and β=b1111′\beta=b'_{1111} (where for the quartic term, we use λ1≪λ2\lambda_1\ll\lambda_2). Importantly, the transition condition in (7) is recovered from λ1=c~1112/4b1111′\lambda_1=\tilde{c}_{111}^2/4b'_{1111}, and the explicit dependence of the coefficients on the microscopic f(n)f^{(n)} terms can be written for incommensurate Q0Q_0 as

ΠQ0=6λmin⁡[f{±Q0}(2)](20)\Pi_{Q_0}=6\lambda_{\min}\left[f^{(2)}_{\{\pm Q_0\}}\right] \tag*{(20)}
γQ0=2∑a≠b≠cfijk(3){Qa,Qb,Qc}e1,aie1,bje1,ck(21)\gamma_{Q_0}=2\sum_{a\ne b\ne c}f^{(3)}_{ijk}\{Q_a,Q_b,Q_c\}e^i_{1,a}e^j_{1,b}e^k_{1,c} \tag*{(21)}
βQ0=∑perm.a≤bfijkl(4){±Qa,±Qb}e1,aie1,aje1,bke1,bl(22)\beta_{Q_0}=\sum_{\substack{\mathrm{perm.}\\a\le b}}f^{(4)}_{ijkl}\{\pm Q_a,\pm Q_b\}e^i_{1,a}e^j_{1,a}e^k_{1,b}e^l_{1,b} \tag*{(22)}

where e1,0e_{1,0} is the eigenvector corresponding to the minimal eigenvalue of f{±Q0}(2)f^{(2)}_{\{\pm Q_0\}} in the expression for ΠQ0\Pi_{Q_0}, and e1,a=R2πa/3e1,0e_{1,a}=R_{2\pi a/3}e_{1,0}. The sums run over permutations of {qi}\{q_i\} in f(n)f^{(n)}. For commensurate Q0Q_0, γQ0→γQ0+γQ0′\gamma_{Q_0}\to\gamma_{Q_0}+\gamma'_{Q_0} and βQ0→βQ0+βQ0′\beta_{Q_0}\to\beta_{Q_0}+\beta'_{Q_0}, where the additional terms are given by

γQ0=6fijk(3){Q0,Q0,Q0}e1,0ie1,0je1,0k,(23)\gamma_{\mathbf{Q}_0}=6f^{(3)}_{ijk}\{\mathbf{Q}_0,\mathbf{Q}_0,\mathbf{Q}_0\}e^{i}_{1,0}e^{j}_{1,0}e^{k}_{1,0}, \tag*{(23)}
βQ0′=Re⁡∑perm.a≠b≠cfijkl(4){Qa,Qa,−Qb,−Qc}e1,aie1,aje1,bke1,cl.(24)\beta'_{\mathbf{Q}_0}=\operatorname{Re}\sum_{\substack{\mathrm{perm.}\\a\ne b\ne c}}f^{(4)}_{ijkl}\{\mathbf{Q}_a,\mathbf{Q}_a,-\mathbf{Q}_b,-\mathbf{Q}_c\}e^{i}_{1,a}e^{j}_{1,a}e^{k}_{1,b}e^{l}_{1,c}. \tag*{(24)}

In the above, it has been additionally assumed that the interaction vertex gqg_{\mathbf{q}} is itself strongly anisotropic, such that e2,0TgQ0−1e2,0≫c~1122/b1111e^{T}_{2,0}g^{-1}_{\mathbf{Q}_0}e_{2,0}\gg\tilde{c}_{112}^{2}/b_{1111} so that the hard mode (from κ⋆\kappa_{\star}) corrections to the quartic coefficient derived earlier are negligible. If this is not the case, βQ0\beta_{\mathbf{Q}_0} also contains contributions from λ2\lambda_2 and contractions of f(3)f^{(3)} with e2,0e_{2,0}.

First-principles Calculations

First-principles electronic-structure calculations for monolayer NbSe2_2, TaSe2_2, and TaS2_2 were performed within Density Functional Theory using the Perdew–Burke–Ernzerhof (PBE) exchange-correlation functional [?] and the projector augmented-wave (PAW) formalism [?] as implemented in VASP [?, ?]. Relativistic effects, including spin-orbit coupling, were fully taken into account. For all three materials, the plane-wave kinetic-energy cutoff was set to 400 eV and the electronic self-consistency threshold was set to 10−710^{-7} eV. The Brillouin zone was sampled using a 12×12×112 \times12 \times1 kk-mesh.

To calculate the generalized susceptibility coefficients, a six-band (10-band) tight-binding Hamiltonian was constructed for NbSe2_2 (TaS2_2 and TaSe2_2) using Wannier90 [?] with Nb 4dxy4d_{xy}, 4dx2−y24d_{x^2-y^2}, and 4dz24d_{z^2} (all Ta-5d5d) orbitals as the projection centres.

Acknowledgments

A. A. acknowledges funding from the Cambridge International Scholarship awarded by the Cambridge Trust. R.-J.S. acknowledges funding from an EPSRC ERC underwrite Grant No. EP/X025829/1. R.G. and M.S.B. gratefully acknowledge the Center for Computational Materials Science at Institute for Materials Research (IMR) for allocations on the MASAMUNE-IMR supercomputer system (Project No. 202112-SCKXX-0510). M.S.B. acknowledges support from Leverhulme Trust (Grant No. RPG-2023-253).

Author Contributions

R.-J. S. and A. A. conceptualized the idea and theory, where A. A. developed the framework under the supervision of R.-J. S. A. A. and R.-J. S. were informed on material candidates by M. S. B. R. G. carried out the first-principles calculations under the guidance of M. S. B., and A. A. performed the Ginzburg-Landau calculations on all these models with inputs from R. G. All authors contributed to the analysis and interpretation of the results. A. A. and R.-J. S. wrote the manuscript with input from all authors.

References

[1] Provost, J. & Vallee, G. Riemannian structure on manifolds of quantum states. Commun. Math. Phys. 76, 289–301 (1980). URL https://doi.org/10.1007/BF02193559.

[2] Ma, Y.-Q., Chen, S., Fan, H. & Liu, W.-M. Abelian and non-abelian quantum geometric tensor. Phys. Rev. B 81, 245129 (2010). URL https://link.aps.org/doi/10.1103/PhysRevB.81.245129.

[3] Bouhon, A., Timmel, A. & Slager, R.-J. Quantum geometry beyond projective single bands. arXiv:2303.02180 (2023). URL https://arxiv.org/abs/2303.02180.

[4] Yu, J. et al. Quantum geometry in quantum materials. npj Quantum Materials 10, 101 (2025). URL https://doi.org/10.1038/s41535-025-00801-3.

[5] Peotta, S. & Törmä, P. Superfluidity in topologically nontrivial flat bands. Nature Communications 6, 8944 (2015). URL https://www.nature.com/articles/ncomms9944.

[6] Herzog-Arbeitman, J., Peri, V., Schindler, F., Huber, S. D. & Bernevig, B. A. Superfluid weight bounds from symmetry and quantum geometry in flat bands. Phys. Rev. Lett. 128, 087002 (2022). URL https://link.aps.org/doi/10.1103/PhysRevLett.128.087002.

[7] Marzari, N. & Vanderbilt, D. Maximally localized generalized Wannier functions for composite energy bands. Phys. Rev. B 56, 12847–12865 (1997). URL https://link.aps.org/doi/10.1103/PhysRevB.56.12847.

[8] Resta, R. The insulating state of matter: a geometrical theory. The European Physical Journal B 79, 121–137 (2011).

[?] Parameswaran, S. A., Roy, R. & Sondhi, S. L. Fractional quantum hall physics in topological flat bands. Comptes Rendus Physique 14, 816–839 (2013).

[?] Roy, R. Band geometry of fractional topological insulators. Phys. Rev. B 90, 165139 (2014). URL https://link.aps.org/doi/10.1103/PhysRevB.90.165139.

[?] Fujimoto, M. et al. Higher vortexability: Zero-field realization of higher landau levels. Phys. Rev. Lett. 134, 106502 (2025). URL https://link.aps.org/doi/10.1103/PhysRevLett.134.106502.

[?] Ahn, J., Guo, G.-Y., Nagaosa, N. & Vishwanath, A. Riemannian geometry of resonant optical responses. Nat. Phys. 18, 290–295 (2021). URL https://doi.org/10.1038/s41567-021-01465-z.

[?] Grüner, G. The dynamics of charge-density waves. Rev. Mod. Phys. 60, 1129–1181 (1988). URL https://link.aps.org/doi/10.1103/RevModPhys.60.1129.

[?] Di Salvo, F. J. & Rice, T. M. Charge-density waves in transition-metal compounds. Phys. Today 32, 32–38 (1979).

[?] Imada, M., Fujimori, A. & Tokura, Y. Metal-insulator transitions. Rev. Mod. Phys. 70, 1039–1263 (1998). URL https://link.aps.org/doi/10.1103/RevModPhys.70.1039.

[?] Friend, R. H. & Jerome, D. Periodic lattice distortions and charge density waves in one- and two-dimensional metals. Journal of Physics C: Solid State Physics 12, 1441–1477 (1979).

[?] Peierls, R. E. Quantum Theory of Solids (Oxford University Press, Oxford, 1955).

[?]* Zhu, X., Cao, Y., Zhang, J., Plummer, E. W. & Guo, J. Classification of charge density waves based on their na

Supplementary Note 1: Summary of Ginzburg-Landau coefficient calculations

We present our calculations of the individual Ginzburg-Landau (GL) coefficients highlighting how the triangular scattering process in the cubic term is essential in selecting the ordering wavevector. The integrals were calculated on a 1080×10801080 \times1080 kk-grid with an adaptive mesh refining the coarse grid by a factor of 8 near the Brillouin Zone (BZ) peaks (convergence was tested using a refinement factor of 4, and the coefficients agreed to within a couple of percent).

As expected from literature [1, 2, 3, 4], the bare quadratic coefficient in Figure S1 exhibits broad peaks over the plotting domain for all three materials. Geometric enhancements also do not show any sharp selective behavior in the wavevector (as expected from the shape of the Fermi surface). The cubic coefficient (see Figure S1) however demonstrates sharp behavior near 2/3ΓM2/3\Gamma M owing to the triangular scattering processes connecting three points on the KK pocket Fermi surface (FS). Since such processes occur for the FSes of both of the (two) low-lying bands, the peak in ∣γQ∣|\gamma_{\mathbf{Q}}| is bimodal in nature. The shorter peak near 1/2ΓM1/2\Gamma M corresponds to scattering between states on the KK pocket FS and those on the Γ\Gamma pocket FS. However, owing to both the quadratic coefficient being smaller here, and suppression by the quartic in the expression for the CDW driver ΘQ\Theta_{\mathbf{Q}}, this peak does not result in any physical consequences. The quartic coefficient (see Figure S1) also shows strong variation with Q\mathbf{Q}, and through its minima, further refines the position of the peaks in the driver. Importantly, it is found to be positive throughout almost the entirety of the plotted domain and is therefore sufficient in bounding the GL functional from below without requiring higher order terms (see Supplementary Note 3 for a consideration of the scenario when it does become negative).

Figure S1. a-c Quadratic ΠQ\Pi_Q, d-f cubic ∣γQ∣|\gamma_Q| and g-i quartic βQ\beta_Q coefficients as a function of QQ for NbSe2_2 (T≈23T \approx23K), TaSe2_2 (T≈35T \approx35K), and TaS2_2 (T≈49T \approx49K). The bare quadratic susceptibility curve exhibits the broad peak that is commonly seen in literature [1, 2, 3, 4], and the geometric enhancements only marginally alter it and cannot explain the wavevector selection seen in experiment. The cubic coefficient, however, shows a sharp increase near 23ΓM\frac{2}{3}\Gamma\mathrm{M} for all three materials corresponding to ordering wavevectors connecting three points on the Fermi surface. The bimodal nature of the peak reflects the aforementioned being satisfied for both bands at different wavevectors. All three materials exhibit some sort of minima in their quartic coefficient near where the cubic is maximised, resulting in the CDW driver peaks reported in the main text. Importantly, βQ\beta_Q is found to be positive for almost the entire domain considered here, a property that is essential for the GL free energy to be bounded from below.

Supplementary Note 2: Comments on the insufficiency of bare higher-order terms

It is interesting to note that the bare higher-order terms in the CDW driver ΘQ\Theta_{\mathbf{Q}} contribute negligibly even where the corresponding geometrically modified terms are of similar order of magnitude to ΠQ\Pi_{\mathbf{Q}}. The origin for this lies in the functional form of L(3){ξmi,ki}L^{(3)}\{\xi_{m_i,\mathbf{k}_i}\}

L(3){ξmi,ki}=nF(ξm1,k1)(ξm1,k1−ξm2,k2)(ξm1,k1−ξm3,k3)+nF(ξm2,k2)(ξm2,k2−ξm1,k1)(ξm2,k2−ξm3,k3)+nF(ξm3,k3)(ξm3,k3−ξm1,k1)(ξm3,k3−ξm2,k2).(25)L^{(3)}\{\xi_{m_i,\mathbf{k}_i}\} = \frac{n_F(\xi_{m_1,\mathbf{k}_1})}{(\xi_{m_1,\mathbf{k}_1}-\xi_{m_2,\mathbf{k}_2})(\xi_{m_1,\mathbf{k}_1}-\xi_{m_3,\mathbf{k}_3})} + \frac{n_F(\xi_{m_2,\mathbf{k}_2})}{(\xi_{m_2,\mathbf{k}_2}-\xi_{m_1,\mathbf{k}_1})(\xi_{m_2,\mathbf{k}_2}-\xi_{m_3,\mathbf{k}_3})} + \frac{n_F(\xi_{m_3,\mathbf{k}_3})}{(\xi_{m_3,\mathbf{k}_3}-\xi_{m_1,\mathbf{k}_1})(\xi_{m_3,\mathbf{k}_3}-\xi_{m_2,\mathbf{k}_2})}. \tag*{(25)}

The transformation ξmi,ki→−ξmi,ki ∀{mi,ki}\xi_{m_i,\mathbf{k}_i} \to-\xi_{m_i,\mathbf{k}_i} \ \forall\{m_i,\mathbf{k}_i\} for T=0T = 0 (step function distribution) implies L(3){ξmi,ki}→−L(3){−ξmi,ki}L^{(3)}\{\xi_{m_i,\mathbf{k}_i}\} \to-L^{(3)}\{-\xi_{m_i,\mathbf{k}_i}\}. Since the largest contributions to the cubic coefficient arise from triangular scattering processes where all three states lie on the Fermi surface (i.e., where locally for small perturbations dki\mathrm{d}\mathbf{k}_i near the FS, ξmi,ki=vmi⋅dki\xi_{m_i,\mathbf{k}_i} = \mathbf{v}_{m_i} \cdot\mathrm{d}\mathbf{k}_i), the integral for the bare γQ\gamma_{\mathbf{Q}} necessarily involves summations of the form L(3){ξmi,ki}+L(3){−ξmi,ki}L^{(3)}\{\xi_{m_i,\mathbf{k}_i}\} + L^{(3)}\{-\xi_{m_i,\mathbf{k}_i}\}. Since these same contributions vanish, the bare γQ\gamma_{\mathbf{Q}} is entirely made of sub-leading order contributions from points further away from the FS and therefore contributes negligibly to the CDW driver. In other words, the integrand for the bare γQ\gamma_{\mathbf{Q}} is locally odd for all three k\mathbf{k} points near the FS and therefore vanishes there.

The same reasoning works for all L(n){ξmi,ki}L^{(n)}\{\xi_{m_i,\mathbf{k}_i}\}, where nn is odd. At T=0T = 0, nF(ξmi,ki)n_F(\xi_{m_i,\mathbf{k}_i}) is the step function θ(−ξmi,ki)\theta(-\xi_{m_i,\mathbf{k}_i}), and the transformation ξmi,ki→−ξmi,ki\xi_{m_i,\mathbf{k}_i} \to-\xi_{m_i,\mathbf{k}_i} sends L(n){ξmi,ki}=∑inθ(−ξmi,ki)/∏j≠i(ξmi,ki−ξmj,kj)→∑in(1−θ(−ξmi,ki))/∏j≠i(ξmi,ki−ξmj,kj)=−L(n){ξmi,ki}L^{(n)}\{\xi_{m_i,\mathbf{k}_i}\} = \sum_i^n \theta(-\xi_{m_i,\mathbf{k}_i})/\prod_{j \ne i}(\xi_{m_i,\mathbf{k}_i}-\xi_{m_j,\mathbf{k}_j}) \to\sum_i^n(1-\theta(-\xi_{m_i,\mathbf{k}_i}))/\prod_{j \ne i}(\xi_{m_i,\mathbf{k}_i}-\xi_{m_j,\mathbf{k}_j}) = -L^{(n)}\{\xi_{m_i,\mathbf{k}_i}\}.

Supplementary Note 3: Non-positive βQ\beta_Q

If βQ\beta_Q is found to be negative or close to zero (as it is at 2/3ΓM2/3\Gamma\mathrm{M} for NbSe2_2 and near 1/2ΓM1/2\Gamma\mathrm{M} for TaSe2_2), higher-order terms are needed to bound the GL free energy from below. These then also modify the transition condition (and hence the CDW driver) in a meaningful manner, and their evaluation is useful for a complete description of the transition. However, it is possible to make qualitative conclusions about the behaviour near these points from existing calculations. Particularly, the C3C_3-symmetry of the TMDs, along with the large cubic coefficients they exhibit, suggests that near these problematic points, a sextic term may be large enough to bound the GL functional. Assuming that βQ\beta_Q is negligible, the GL free energy with the inclusion of a sextic term of the form uQΔQ6u_Q\Delta_Q^6 with uQ>0u_Q > 0 affords a closed form solution for the transition given by

3TrgQ:=ΘQ′=−ΠQ+3∣γQ∣4(∣γQ∣4uQ)1/3.(26)\frac{3}{T_{\mathrm{rg}Q}} := \Theta'_Q = -\Pi_Q + \frac{3|\gamma_Q|}{4}\left(\frac{|\gamma_Q|}{4u_Q}\right)^{1/3}. \tag*{(26)}

ΘQ′\Theta'_Q here is the analog of the CDW driver studied in the main text.

While we do not explicitly calculate uQu_Q here, we do obtain an approximate lower bound estimate for it using the cubic coefficient. Based on the roughly triangular shape of the Fermi surfaces, we infer that the strongest contributions to uQu_Q must arise from the same triangular scattering processes (now at second order) that contributed to γQ\gamma_Q. In particular, where ∣γQ∣|\gamma_Q| involved an integral over k\mathbf{k} of some integrand IQ(k)/3I_Q(\mathbf{k})/3, one of the strongest contributors to uQu_Q is ∫BZIQ2(k)/6 d2k\int_{\mathrm{BZ}} I_Q^2(\mathbf{k})/6\,\mathrm{d}^2\mathbf{k} (the other significant contributor is a hexagonal scattering process particularly relevant for the Γ\Gamma pocket FS, but this is not considered here). Then, by the Cauchy-Schwarz inequality, uQ≳3∣γQ∣2/2u_Q \gtrsim3|\gamma_Q|^2/2, and ΘQ′≲−ΠQ+0.41∣γQ∣2/3\Theta'_Q \lesssim-\Pi_Q + 0.41|\gamma_Q|^{2/3}. Since we expect the transition condition to increase monotonically from the CDW driver ΘQ\Theta_Q in the main text to the modified version ΘQ′\Theta'_Q considered here as βQ\beta_Q decreases to zero, the approximate upper bound derived here for ΘQ′\Theta'_Q also applies to ΘQ\Theta_Q. Further, it can actually be used to signal where an inclusion of sextic terms may be useful even for positive (but small) βQ\beta_Q.

CDW drivers and approximate upper bounds for NbSe2, TaSe2, and TaS2

Figure S2. CDW driver ΘQ\Theta_Q for a NbSe2_2 (T≈23T \approx23 K), b TaSe2_2 (T≈35T \approx35 K), and c TaS2_2 (T≈49T \approx49 K), overlaid with approximate upper bound derived upon inclusion of a sextic term in the GL functional. Major corrections are only needed at 2/3ΓM2/3\Gamma\mathrm{M} for NbSe2_2 and near 1/2ΓM1/2\Gamma\mathrm{M} for TaSe2_2, where the original driver exhibits strongly negative peaks from a negative βQ\beta_Q. The ordering wavevector estimates from the unmodified driver as specified in the main text are still unlikely to change. Additionally, the secondary peak above 2/3ΓM2/3\Gamma\mathrm{M} for TaSe2_2 overshoots the upper bound due to small βQ\beta_Q and is expected to become shorter.

Supplementary Note 4: Anisotropy of the quadratic susceptibility χ\chi

We explicitly find that the quadratic susceptibility is indeed highly anisotropic, that is, one of the eigenvalues of χ\chi is much smaller than the other. It is then sufficient to work within the space of the larger eigenvalue (where the quadratic term in the GL free energy becomes close to zero near the transition temperature). In Figure S3, we plot the eigenvalues of χ\chi, the symmetric rank-2 tensor that forms part of the coefficient of n^⊗2Δ2\hat{\mathbf{n}}^{\otimes2}\Delta^2 in the GL free energy (along with the interaction vertex gQg_Q), and find that they indeed differ by at least one order of magnitude, thereby allowing the reduction of the free energy into a scalar order parameter Δ\Delta at leading order.

Figure S3. Both eigenvalues of the modified tensorial Lindhard susceptibility χ\chi as a function of QQ for a NbSe2_2 (T≈23 KT \approx23\,\mathrm{K}), b TaSe2_2 (T≈35 KT \approx35\,\mathrm{K}), and c TaS2_2 (T≈49 KT \approx49\,\mathrm{K}). Note that the eigenvalues have been plotted on a log scale, and the anisotropic nature of χ\chi is evident for all three TMDs considered in this work.

Supplementary Note 5: Variation of the CDW driver ΘQ\Theta_Q with temperature

Here, we present the variation of the CDW driver ΘQ\Theta_Q with temperature (see Figure S4). Recalling that the transition occurs when 3/TrgQ=ΘQ3/\mathrm{Tr}g_Q = \Theta_Q and that gQg_Q is independent of temperature, the variation of the driver with temperature signals when the charge ordering takes place. While the exact transition temperature cannot be estimated within the scope of this work since it depends on the phononic spectrum through gQg_Q (and is likely to get renormalized anyway due to higher order terms and other interactions), the trend does allow for qualitative conclusions to be made.

Figure S4. CDW transition driver ΘQ\Theta_Q plotted as a function of QQ for a NbSe2_2, b TaSe2_2, and c TaS2_2 at different temperatures TT. For all three TMDs, increasing TT smoothens out the peaks. However, the quadratic background (from −ΠQ-\Pi_Q) remains largely constant with TT, further proving that these geometry-enhanced higher-order-terms drive the CDW transition in these quasi-flatband hexagonal materials.

Firstly, as expected, higher temperatures smoothen out the peaks. Interestingly however, only the higher-order-terms in the driver seem to show any significant variation with temperature while the quadratic background largely stays the same. This is notable since it suggests that the transition is driven precisely by these same higher-order-terms; ΠQ\Pi_Q being largely constant with temperature implies that it cannot possibly equal 3/TrgQ3/\mathrm{Tr}g_Q by lowering the temperature, as would be required by a second-order transition driven solely by the quadratic susceptibility. Secondly, it is commonly reported in experiment [5, 6, 7, 8] that the transition is either second-order or weakly first order. Since the driver approaches the flat quadratic form at high temperatures, our theory is consistent with a weakly first-order transition; then, it is probable that 3/TrgQ3/\mathrm{Tr}g_Q rests slightly above this quadratic background, and geometry-enhanced higher order terms raise the driver to the ordering threshold upon lowering temperature.

References

References

  1. [1]Zhu, X., Cao, Y., Zhang, J., Plummer, E. W. & Guo, J. Classification of charge density waves based on their nature. Proceedings of the National Academy of Sciences 112, 2367–2371 (2015). URL https://doi.org/10.1073/pnas.1424791112.
  2. [2]Johannes, M. D., Mazin, I. I. & Howells, C. A. Fermi-surface nesting and the origin of the charge-density wave in NbSe₂. Phys. Rev. B 73, 205102 (2006). URL https://link.aps.org/doi/10.1103/PhysRevB.73.205102.DOI
  3. [3]Johannes, M. D. & Mazin, I. I. Fermi surface nesting and the origin of charge density waves in metals. Phys. Rev. B 77, 165135 (2008). URL https://link.aps.org/doi/10.1103/PhysRevB.77.165135.DOI
  4. [4]Flicker, F. & van Wezel, J. Charge order in NbSe₂. Phys. Rev. B 94, 235135 (2016). URL https://link.aps.org/doi/10.1103/PhysRevB.94.235135.
  5. [5]Moncton, D. E., Axe, J. D. & DiSalvo, F. J. Study of superlattice formation in 2H-NbSe₂ and 2H-TaSe₂ by neutron scattering. Phys. Rev. Lett. 34, 734–737 (1975). URL https://link.aps.org/doi/10.1103/PhysRevLett.34.734.
  6. [6]Moncton, D. E., Axe, J. D. & DiSalvo, F. J. Neutron scattering study of the charge-density wave transitions in 2H-TaSe₂ and 2H-NbSe₂. Phys. Rev. B 16, 801–819 (1977). URL https://link.aps.org/doi/10.1103/PhysRevB.16.801.
  7. [7]Friend, R. H. & Jérôme, D. Periodic lattice distortions and charge density waves in one- and two-dimensional metals. Journal of Physics C: Solid State Physics 12, 1441–1477 (1979).DOI
  8. [8]Liu, W. et al. Experimental observations and density functional simulations on the structural transition behavior of a two-dimensional transition-metal dichalcogenide. Scientific Reports 10, 18255 (2020).DOI

Paper details

vlnc:W_mucfuqvz
Version
v1 · latest
Published
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