betaFeedback

WorksPhysics and Astronomy

Universal relation between energy gap and dielectric constant

Abstract

We establish a universal relation between the energy gap and the static dielectric constant for all insulating states. This relation yields an upper bound on the energy gap, which only depends on the electron density and electronic dielectric constant. We identify two types of energy gaps associated with transverse and longitudinal excitations at long wavelength, which correspond to the optical gap and the plasmon energy respectively. Their upper bounds are set by the dielectric constant and its inverse respectively. The transverse gap bound is calculated for a wide range of materials and compared with the measured optical gap. A remarkable case is cubic boron nitride, in which the direct gap reaches \SI{72}{\percent} of the bound. Our results are derived from the Kramers-Kronig relation and the ff-sum rule, and therefore rest on general physical principles.

Insulating states of matter with a bulk energy gap are ubiquitous: examples include semiconductors, Mott insulators, and quantum Hall states. If the ground state is separated from excited states by a finite energy gap, zero-temperature conductivity necessarily vanishes at low frequency. In recent decades, significant progress has been made in understanding ground state properties of various types of insulating states [1, 2, 3, 4, 5], including quantum Hall states and topological insulators. On the other hand, the energy gap of insulating states has received little attention, despite its fundamental importance. The band gap of semiconductors determines their electrical and optical properties. The energy gap of topological insulators limits the temperature for observing quantized conductance.

Recently, we found a fundamental upper bound on the energy gap for (integer or fractional) Chern insulators, which is simply determined by the electron density and the ground state Chern number [6]. Interestingly, this bound is saturated for quantum Hall states in Landau level systems, and is fairly tight for Chern insulators in twisted semiconductor bilayers at zero magnetic field. Thus, it offers a guiding principle for searching large-gap topological materials, and reveals a deep connection between the ground-state topology and the excitation spectrum.

In this work, by exploring the connection between ground state property, thermodynamic response and energy gap, we derive an upper bound on the energy gap of insulating states based on the dielectric constant. We identify two energy gaps in electronic solids that are associated with excitations that couple to longitudinal and transverse electric field respectively. We find two distinct sum rules for the dielectric constant, from which the bounds for longitudinal and transverse energy gaps are derived. Our gap bound applies to all insulators—topological or trivial. We show that the bound is practically useful for a wide range of materials, especially wide-bandgap semiconductors. A remarkable example is cubic boron nitride (c-BN), where the direct gap reaches up to 72 % of our gap bound!

We start by considering generalized optical weights for insulating states [6], defined as the negative moments of the absorptive part of the optical conductivity σ(ω)\sigma(\omega),

Wαβi≡∫0∞dωσαβabs(ω)ωi,(1)W_{\alpha\beta}^{i} \equiv\int_{0}^{\infty} \mathrm{d}\omega\frac{\sigma_{\alpha\beta}^{\mathrm{abs}}(\omega)}{\omega^{i}}, \tag*{(1)}

where σαβabs≡(σαβ+σβα∗)/2\sigma_{\alpha\beta}^{\mathrm{abs}} \equiv(\sigma_{\alpha\beta}+\sigma_{\beta\alpha}^{*})/2 is a Hermitian tensor consisting of the real part of longitudinal optical conductivity and the imaginary part of optical Hall conductivity. For example, σαβabs=Re⁡(σxx)δαβ+iIm⁡(σxy)ϵαβ\sigma_{\alpha\beta}^{\mathrm{abs}}=\operatorname{Re}(\sigma_{xx})\delta_{\alpha\beta}+i\operatorname{Im}(\sigma_{xy})\epsilon_{\alpha\beta} in isotropic two-dimensional systems.

The zero-th moment Re⁡[Wαβ0]=∫0∞dωRe⁡σαβabs\operatorname{Re}[W_{\alpha\beta}^{0}]=\int_{0}^{\infty}\mathrm{d}\omega\operatorname{Re}\sigma_{\alpha\beta}^{\mathrm{abs}} is the standard optical spectral weight. It is well-known from the ff-sum rule that optical spectral weight is a ground state property given by [7]

Re⁡Wαβ0=π2V⟨∂2H^(A)∂Aα∂Aβ∣A=0⟩,(2)\operatorname{Re}W_{\alpha\beta}^{0}=\frac{\pi}{2V}\left\langle\left.\frac{\partial^{2}\hat{H}(\mathbf{A})}{\partial A_{\alpha}\partial A_{\beta}}\right|_{\mathbf{A}=0}\right\rangle, \tag*{(2)}

where H^(A)\hat{H}(\mathbf{A}) is the Hamiltonian of the system coupled to a uniform vector potential A\mathbf{A} and VV is the volume of the system. The fundamental microscopic Hamiltonian of electronic solids includes electron’s kinetic energy p2/(2m)p^{2}/(2m), but no other terms involving pnp^{n} with n≥2n\geq2 (where p\mathbf{p} is the momentum). Then, Eq. (2) reduces to the universal ff-sum rule

Wαα0=πne2/(2m),(3)W_{\alpha\alpha}^{0}=\pi ne^{2}/(2m), \tag*{(3)}

where nn is the electron density (including core electrons), mm is the bare mass, and e(<0)e(<0) is the charge of electron.

For systems with a finite energy gap in the bulk, the negative-ii-th moment WiW^{i} is well-defined for any i≥1i\geq1. Previously, we derived the topological gap bound by utilizing W1W^{1}, including both real and imaginary parts [6]. The real part of W1W^{1} (termed “quantum weight”) is related to the quantum metric of the ground states over twisted boundary conditions [8], while its imaginary part is related to the quantized dc Hall conductivity, or the many-body Chern invariant. We also note that the positive second moment was previously discussed in relation to optical properties and charge distribution [9]. In this work, we shift our focus to the negative-second moment of optical conductivity, W2W^2. Since the charge current is the time derivative of the polarization, the optical conductivity of insulating states is directly related to the polarizability as:

−iωχ(ω)=σ(ω).(4)-\mathrm{i}\omega\chi(\omega)=\sigma(\omega). \tag*{(4)}

Here σ\sigma and χ\chi describe respectively the current and polarization induced by an applied uniform electric field: j=σE0\mathbf{j}=\sigma E_0 and P=χE0\mathbf{P}=\chi E_0, where E0E_0 couples to the position of charges −eE0⋅∑iri-eE_0\cdot\sum_i\mathbf{r}_i. Substituting this relation into Eq. (1), we can rewrite Re⁡W2\operatorname{Re} W^2 as

Re⁡W2=12∫−∞∞dω Im⁡χ(ω)ω.(5)\operatorname{Re} W^2=\frac{1}{2}\int_{-\infty}^{\infty}\mathrm{d}\omega\,\frac{\operatorname{Im}\chi(\omega)}{\omega}. \tag*{(5)}

The Kramers-Kronig relation for χ\chi relates the right hand side of Eq. (5) to the thermodynamic response function χ0\chi^0 which measures the change of polarization in response to an externally applied static electric field:

χ(0)=−lim⁡ω→0Im⁡σω≡χ0.(6)\chi(0)=-\lim_{\omega\to0}\frac{\operatorname{Im}\sigma}{\omega}\equiv\chi^0. \tag*{(6)}

This leads to a sum rule relating optical conductivity and thermodynamic polarizability:

∫0∞dω Re⁡σ(ω)ω2=π2χ0.(7)\int_0^\infty\mathrm{d}\omega\,\frac{\operatorname{Re}\sigma(\omega)}{\omega^2}=\frac{\pi}{2}\chi^0. \tag*{(7)}

Using generalized optical weights, we can derive a series of upper bounds on the optical gap EgE_g of insulating states. Eg≡ℏωgE_g\equiv\hbar\omega_g is defined by the threshold frequency ωg\omega_g for optical absorption so that Re⁡σ(ω<ωg)=0\operatorname{Re}\sigma(\omega<\omega_g)=0. Since Re⁡σ(ω)\operatorname{Re}\sigma(\omega) is always semipositive, we can find an upper bound on the generalized optical weights as Wααi≤(ℏ/Eg)iWαα0W^i_{\alpha\alpha}\leq(\hbar/E_g)^iW^0_{\alpha\alpha}, where α\alpha can be any direction. This inequality immediately leads to a bound on the optical gap EgE_g given by W0W^0 and WiW^i:

Eg≤ℏ(Wαα0Wααi)1/i.(8)E_g\leq\hbar\left(\frac{W^0_{\alpha\alpha}}{W^i_{\alpha\alpha}}\right)^{1/i}. \tag*{(8)}

This inequality applies to any system at all dimensions, independent of any microscopic details.

We also note that the optical gap EgE_g is, by definition, greater than or equal to the spectral gap Δ\Delta, which is defined as the energy difference between the ground state and the first excited state. Therefore, our optical gap bound, Eq. (8), also gives an upper bound on the spectral gap. For example, in noninteracting band insulators, the optical gap is the minimum direct gap, while the spectral gap can be indirect and thus smaller. In the presence of the Coulomb interaction, the optical gap and the spectral gap can be influenced by the excitonic effects. Note that both the optical gap and the spectral gap are defined with charge-neutral states of the system and should not be confused with the electronic gap, which is the energy cost to add or remove one electron.

Setting i=2i=2 in Eq. (8) and using Eq. (7), we find a gap bound based on the polarizability χ0\chi^0:

Eg≤ℏ2Wαα0πχαα0.(9)E_g\leq\hbar\sqrt{\frac{2W^0_{\alpha\alpha}}{\pi\chi^0_{\alpha\alpha}}}. \tag*{(9)}

For systems where the universal ff-sum rule for W0W^0, Eq. (3), is applicable, the bound simplifies to

Eg≤ℏne2mχαα0.(10)E_g\leq\hbar\sqrt{\frac{ne^2}{m\chi^0_{\alpha\alpha}}}. \tag*{(10)}

This inequality applies to any gapped system with a conserved total particle number.

From now on, we shall focus on three-dimensional charged systems with long-range Coulomb interaction, such as electrons in real solids. Our goal is to relate the thermodynamic response χ0\chi^0 to experimentally observable quantities. In order to proceed properly, we need to consider wavevector- and frequency-dependent conductivity σ(q,ω)\sigma(q,\omega) and take the limit q→0q\to0 with great care. It is also important to note that throughout this work, σ(q,ω)\sigma(q,\omega) is defined by the current in response to the externally applied electric field E0E_0, which we shall refer to as external conductivity. On the other hand, the physical conductivity is experimentally measured, which we denote as σp\sigma_p, is defined by the current in response to the total electric field E\mathbf{E}. E\mathbf{E} includes the external field E0E_0 and possibly additional field from induced charge density, i.e., due to the screening effect.

To proceed, we decompose the external conductivity σ(q,ω)\sigma(q,\omega) into longitudinal and transverse parts:

σμν(q,ω)=qμqνq2σμνL(q,ω)+(δμν−qμqνq2)σμνT(q,ω).\sigma_{\mu\nu}(q,\omega)=\frac{q_\mu q_\nu}{q^2}\sigma^L_{\mu\nu}(q,\omega)+\left(\delta_{\mu\nu}-\frac{q_\mu q_\nu}{q^2}\right)\sigma^T_{\mu\nu}(q,\omega).

where σL,T(q,ω)\sigma^{L,T}(q,\omega) relates the current to an applied electric field E0E_0 parallel or perpendicular to the wavevector qq, respectively. In the limit q→0q\to0, although the same physical conductivity σp\sigma_p is obtained for longitudinal and transverse cases, one generally gets different results for σL\sigma^L and σT\sigma^T. Since the transverse electric field does not couple to charge density directly, the transverse conductivity σT(q→0,ω)\sigma^T(q\to0,\omega) is equal to the physical conductivity at q=0q=0: σT(q→0,ω)=σp(ω)\sigma^T(q\to0,\omega)=\sigma_p(\omega). However, the longitudinal conductivity σL(q→0,ω)\sigma^L(q\to0,\omega) is generally different from σp(ω)\sigma_p(\omega), because a longitudinal electric field induces charge density, which screens the applied electric field. Due to this screening effect, the total electric field is reduced from the external field by the complex dielectric constant ϵ\epsilon: E0=ϵEE_0=\epsilon E. Therefore we have σL(q→0,ω)=σp(ω)/ϵ(ω)\sigma^L(q\to0,\omega)=\sigma_p(\omega)/\epsilon(\omega) for isotropic systems. The same argument holds for anisotropic systems with well-defined principle axes, in which ϵ\epsilon takes a diagonal form for all frequencies.

Further noting the defining relation between physical conductivity and complex dielectric constant

ϵ(ω)=1+iσp(ω)ϵ0ω,(11)\epsilon(\omega)=1+\frac{i\sigma_p(\omega)}{\epsilon_0\omega}, \tag*{(11)}

we can relate transverse and longitudinal external conductivities to ϵ(ω)\epsilon(\omega): σT(q→0,ω)=−iωϵ0(ϵ(ω)−1)\sigma^T(\mathbf{q}\to0,\omega)=-i\omega\epsilon_0(\epsilon(\omega)-1), σL(q→0,ω)=−iωϵ0(1−ϵ−1(ω))\sigma^L(\mathbf{q}\to0,\omega)=-i\omega\epsilon_0(1-\epsilon^{-1}(\omega)). These general relations were originally established using diagrammatic methods by Ambegaokar and Kohn [10]; related discussions can be found in Refs. [11, 12]. For our purpose, it is useful to consider transverse and longitudinal polarizabilities of insulating states defined by the corresponding external conductivities as in Eq. (4):

χT(ω)≡iσT(q→0,ω)ω=ϵ0(ϵ(ω)−1),(12)\chi^T(\omega)\equiv\frac{i\sigma^T(\mathbf{q}\to0,\omega)}{\omega}=\epsilon_0(\epsilon(\omega)-1), \tag*{(12)}
χL(ω)≡iσL(q→0,ω)ω=ϵ0(1−ϵ−1(ω)),(13)\chi^L(\omega)\equiv\frac{i\sigma^L(\mathbf{q}\to0,\omega)}{\omega}=\epsilon_0(1-\epsilon^{-1}(\omega)), \tag*{(13)}

which are related to the complex dielectric constant ϵ(ω)\epsilon(\omega) and its inverse ϵ−1(ω)\epsilon^{-1}(\omega), respectively. As response functions of insulators, both χT\chi^T and χL\chi^L are analytic in the upper half plane of ω\omega including at ω=0\omega=0 (where they are finite), and tend to zero as ∣ω∣→∞|\omega|\to\infty (note ϵ→1\epsilon\to1 at high frequency). Therefore, we can apply the Kramers-Kronig relation to them:

∫−∞∞dωIm⁡χL,T(ω)ω=πRe⁡χL,T(0).(14)\int_{-\infty}^{\infty}d\omega\frac{\operatorname{Im}\chi^{L,T}(\omega)}{\omega}=\pi\operatorname{Re}\chi^{L,T}(0). \tag*{(14)}

Combining Eq. (14) with Eqs. (12) and (13), we obtain two sum rules:

1ϵ0∫−∞∞dωRe⁡σT(q→0,ω)ω2=∫−∞∞dωIm⁡ϵ(ω)ω=π(ϵ−1)(15)\frac{1}{\epsilon_0}\int_{-\infty}^{\infty}d\omega\frac{\operatorname{Re}\sigma^T(\mathbf{q}\to0,\omega)}{\omega^2} =\int_{-\infty}^{\infty}d\omega\frac{\operatorname{Im}\epsilon(\omega)}{\omega} =\pi(\epsilon-1) \tag*{(15)}

and

1ϵ0∫−∞∞dωRe⁡σL(q→0,ω)ω2=−∫−∞∞dωIm⁡ϵ−1(ω)ω=π(1−ϵ−1)(16)\frac{1}{\epsilon_0}\int_{-\infty}^{\infty}d\omega\frac{\operatorname{Re}\sigma^L(\mathbf{q}\to0,\omega)}{\omega^2} =-\int_{-\infty}^{\infty}d\omega\frac{\operatorname{Im}\epsilon^{-1}(\omega)}{\omega} =\pi(1-\epsilon^{-1}) \tag*{(16)}

where ϵ=ϵ(ω=0)\epsilon=\epsilon(\omega=0) is the static dielectric constant. It is remarkable that these sum rules associated with longitudinal and transverse external conductivity give us sum rules for the imaginary part of ϵ(ω)−1\epsilon(\omega)-1 and ϵ−1(ω)−1\epsilon^{-1}(\omega)-1, respectively. We emphasize this is the consequence of the relation between complex dielectric constant and the response function to external perturbation χL,T\chi^{L,T} shown in Eq. (12) and (13). The causality of the response function allows us to apply Kramers-Kronig relation. These sum rules for ϵ\epsilon and ϵ−1\epsilon^{-1} were previously derived from the mathematical property of the complex dielectric constant in Landau and Lifshitz [13] (see also Refs. [14, 15]). Our derivation above reveals their relations to the response to transverse and longitudinal electric fields.

Combining the above sum rules for W2W^2 and the standard ff sum rule for W0W^0 allows us to derive upper bounds on two distinct energy gaps of real solids. In gapped systems, the real part of conductivity is nonzero only above a threshold frequency, thus taking the form

Re⁡σL,T(q→0,ω)=0 for ℏω<EgL,T.(17)\operatorname{Re}\sigma^{L,T}(\mathbf{q}\to0,\omega)=0\ \text{for }\hbar\omega<E_g^{L,T}. \tag*{(17)}

This expression defines two energy gaps EgL,TE_g^{L,T} associated with the lowest energy excitations at q→0\mathbf{q}\to0 that couple to transverse and longitudinal electric fields, respectively. By studying the response of a charged system to a transverse or longitudinal electric field, we can probe transverse and longitudinal excitations respectively.

It is important to note that for three-dimensional charged systems, the transverse and longitudinal gaps are generally different because of the long-range Coulomb interaction. A well-known example is the Wigner crystal state of electrons in uniform positive charge background. This system has a longitudinal phonon at the plasma frequency but a vanishing transverse phonon frequency at a long wavelength [16] (in the presence of a periodic lattice potential, the transverse mode shifts to finite frequency, leading to an insulating state with an energy gap). A transverse collective mode is manifested as a resonance in the dielectric constant ϵ(ω)\epsilon(\omega), while the longitudinal plasma mode is manifested as a zero in ϵ(ω)\epsilon(\omega) or a resonance in ϵ−1(ω)\epsilon^{-1}(\omega).

Combining the sum rule Eq. (15) with the standard ff sum rule Eq. (3) and the general form of transverse conductivity Eq. (17), we obtain a bound on the transverse optical gap of electronic insulators:

EgT≤ℏne2mϵ0(ϵ−1)=ℏωpϵ−1,(18)E_g^T\le\hbar\sqrt{\frac{ne^2}{m\epsilon_0(\epsilon-1)}}=\frac{\hbar\omega_p}{\sqrt{\epsilon-1}}, \tag*{(18)}

where ωp=ne2/(mϵ0)\omega_p=\sqrt{ne^2/(m\epsilon_0)} is the plasma frequency for the electron density nn. Similarly, a bound on the longitudinal gap can be derived from Eq. (16):

EgL≤ℏωp1−ϵ−1.(19)E_g^L\le\frac{\hbar\omega_p}{\sqrt{1-\epsilon^{-1}}}. \tag*{(19)}

By definition, EgT,LE_g^{T,L} must exceed or equal to the spectral gap Δ\Delta. Therefore, an upper bound on EgT,LE_g^{T,L} is also an upper bound on the spectral gap.

Since the static dielectric constant ϵ\epsilon is usually greater than 1, the bound on the longitudinal gap is larger than that of the transverse gap. This is consistent with the fact that the longitudinal mode is associated with density fluctuation and therefore costs additional electrostatic energy due to the long-range Coulomb interaction. Since electromagnetism waves are transverse, the optical conductivity of solids as experimentally measured corresponds to σT\sigma^T, and the optical gap corresponds to the transverse gap EgTE_g^T. On the other hand, the longitudinal gap can be measured by electron energy loss spectroscopy or inelastic X-ray scattering, which probe the dynamic structure factor.

Remarkably, our gap bound given by Eqs. (18) and (19) can be both practically useful. Figure 1 and Figure 2 show the calculated gap bound and the experimentally measured gap for the transverse and the longitudinal gap, respectively. The two parameters here, the electron density and the dielectric constant, are given in the supplemental material. For the transverse gap, we used experimental data from optical measurements, while we used the plasmon energy measured by electron energy loss spectroscopy for the longitudinal gap. For the transverse gap bound, the most remarkable case is cubic boron nitride (c-BN) for which the gap bound given by Eq. (10) is 20.1 eV, while the measured direct gap is 14.5 eV [17]. Therefore, the actual direct gap reaches 72 % of the gap bound! Besides c-BN, the measured gap in silicon, diamond, and silicon carbide (3C-SiC) reaches 45, 40, 38 % of the gap bound, respectively. The longitudinal gap bound (19) can work even better than the transverse gap bound for some materials. For diamond, the measured longitudinal gap is EgL=34.0E_g^{\mathrm{L}} = 34.0 eV, yielding 81 % of the longitudinal gap bound 42.1 eV. The measured longitudinal gap in c-BN, silicon, 3C-SiC reaches 72, 52, 55 % of the gap bound. It is remarkable that both of our gap bounds (18), (19) merely determined by electron density and dielectric constant can be fairly tight for real materials.

Image not extracted

Figure 1. Universal relation between optical gap (transverse gap), dielectric constant ε\varepsilon, and the electron density nn. The black line is the transverse gap bound given by Eq. (18). The gap used here is taken from optical measurements. The data points shown in red are topological insulators.

Image not extracted

Figure 2. Universal relation between longitudinal optical gap, dielectric constant ε\varepsilon, and the electron density nn. The black line is the longitudinal gap bound given by Eq. (19). The gap used here is the plasma frequency measured in electron energy loss spectroscopy.

We now discuss various circumstances in which the gap bound (18) is (nearly) saturated. The first case is insulating states where electrons are strongly localized, i.e., quantum fluctuation in electron position is small. In the classical limit, the ground state is obtained by minimizing the sum of potential energy and electron-electron interaction energy

Hc=∑iV(xi)+∑i,jU(xi−xj)H_c = \sum_i V(x_i) + \sum_{i,j} U(x_i-x_j)

with VV the potential and UU the two-body interaction.

with VV the potential and UU the two-body interaction. When electrons are slightly displaced from their equilibrium position by δxi\delta x_i, the change in the energy takes a quadratic form

Hc=Ec+∑i,jKijδxiδxj/2.H_c = E_c + \sum_{i,j} K_{ij}\delta x_i\delta x_j/2.

Diagonalizing KK, we obtain the normal modes xα′=∑icαiδxix'_\alpha= \sum_i c_{\alpha i}\delta x_i (with ∑icαicβi=δαβ\sum_i c_{\alpha i}c_{\beta i} = \delta_{\alpha\beta}) and the spring constant KαK_\alpha:

Hc=∑αKαxα′2/2.H_c = \sum_\alpha K_\alpha x_\alpha^{\prime2}/2.

Correspondingly, we can rewrite the total kinetic energy ∑ipi2/(2m)\sum_i p_i^2/(2m) with the momentum conjugate to the normal modes, pα′=∑icαipip'_\alpha= \sum_i c_{\alpha i}p_i, as ∑αpα′2/(2m)\sum_\alpha p_\alpha^{\prime2}/(2m). Then we obtain a collection of independent harmonic oscillators, one for each normal mode,

H=∑ipi22m+Hc≈∑αpα′22m+12Kαxα′2+⋯(20)H = \sum_i \frac{p_i^2}{2m} + H_c \approx\sum_\alpha\frac{p_\alpha^{\prime2}}{2m} + \frac{1}{2}K_\alpha x_\alpha^{\prime2} + \cdots \tag*{(20)}

The frequency for each mode is ωα=Kα/m\omega_\alpha= \sqrt{K_\alpha/m}, leading to the energy gap of ℏωα\hbar\omega_\alpha after quantization.

To determine the polarizability of such a system, let us consider the application of a uniform static electric field E0E_0, which couples to electron displacements: −eE0∑iδxi-eE_0\sum_i\delta x_i. Then, each normal mode is displaced by δxα′=zαeE0/Kα\delta x'_\alpha= z_\alpha eE_0/K_\alpha with zαe=∑icαiez_\alpha e = \sum_i c_{\alpha i}e the effective charge of a normal mode α\alpha, and the resulting polarization is given by P=e∑iδxi/V=(e2E0/V)∑αzα2/KαP = e\sum_i\delta x_i/V = (e^2E_0/V)\sum_\alpha z_\alpha^2/K_\alpha, hence the polarizability is

χ0=e2V∑αzα2Kα=1V∑αe2fαmωα2,(21)\chi^0 = \frac{e^2}{V}\sum_\alpha\frac{z_\alpha^2}{K_\alpha} = \frac{1}{V}\sum_\alpha\frac{e^2 f_\alpha}{m\omega_\alpha^2}, \tag*{(21)}

with fα≡zα2f_\alpha\equiv z_\alpha^2 the oscillator strength.

Since ∑αfα=N\sum_\alpha f_\alpha= N with the number of electrons NN and Kα>0K_\alpha> 0, we have ω02≤n2/(ω02)\omega_0^2 \le n^2/(\omega_0^2), where ω0\omega_0 is the lowest oscillator frequency among all normal modes that couple to the electric field, i.e., those satisfying fα≠0f_\alpha\ne0. On the other hand, the optical gap is Eg=ℏω0E_g = \hbar\omega_0. Therefore, we obtain an inequality relation between the optical gap and polarizability: Eg≤ℏne2/(mχ0)E_g \le\hbar\sqrt{ne^2/(m\chi^0)}, recovering our bound (10). Note that this semiclassical result applies to both crystalline and disordered systems with localized electrons. From this derivation, it is clear that the bound is saturated when only a single mode couples to the electric field. This is consistent with inequality (8) relating optical weights to the energy gap.

As an illustrative example, let us consider a system of interacting electrons in three dimensions, which forms Wigner crystal commensurate with a lattice potential. In this case, because of the lattice translational symmetry, the normal mode can be labeled by wavevector q\mathbf{q}. Writing the Fourier transform of the displacement of electrons from the equilibrium position as uq\mathbf{u}_{q}, the low-energy Hamiltonian is given by

H=m2∑q∣u˙q∣2+(ω02δαβ+ωp2qαqβq2)u−qαuqβ,(22)H = \frac{m}{2}\sum_{q} \lvert\dot{\mathbf{u}}_{q}\rvert^2 + \left(\omega_0^2\delta_{\alpha\beta}+\omega_p^2\frac{q_\alpha q_\beta}{q^2}\right)u_{-q\alpha}u_{q\beta}, \tag*{(22)}

where ω0\omega_0 is an oscillator frequency associated with the lattice potential and ωp2/q2\omega_p^2/q^2 represents the Coulomb interaction between the charge density ρq∝iq⋅uq\rho_q \propto i\mathbf{q}\cdot\mathbf{u}_q. This system has two types of normal modes at long wavelength, a transverse mode with uq⊥q\mathbf{u}_q \perp\mathbf{q}, and a longitudinal mode with uq∥q\mathbf{u}_q \parallel\mathbf{q}. The transverse gap (or optical gap) and the longitudinal gap (or plasmon energy) at q→0q \to0 are respectively given by

EgT=ℏω0,EgL=ℏω02+ωp2(23)E_g^T = \hbar\omega_0,\qquad E_g^L = \hbar\sqrt{\omega_0^2+\omega_p^2} \tag*{(23)}

In this case, we can verify both the transverse and the longitudinal gap bounds are saturated, consistent with the fact that the external electric field couples to only a single mode (see Supplemental Materials for details).

Next, we consider the opposite limit of nearly free electron systems. In the absence of electron-electron interaction, an approximate relation between the dielectric constant and the energy gap has been discussed in the early literature of semiconductor research [18]. With certain simplifications, it was shown that χ0≈(ℏωp/Eg)2\chi^0 \approx(\hbar\omega_p/E_g)^2 to the leading order in Eg/EFE_g/E_F with the Fermi energy EFE_F. Interestingly, this approximate result also leads to the saturation of our inequality (10).

Last but not least, two-dimensional electron gas in a magnetic field provides an interesting example of topological insulating states that saturate the bound, namely (integer or fractional) quantum Hall states. The Hamiltonian of this system takes the form H=∑i(p−eA)2/(2m)+∑ijU(ri−rj)H = \sum_i(\mathbf{p}-e\mathbf{A})^2/(2m) + \sum_{ij} U(r_i-r_j), with A=(By,0)\mathbf{A}=(By,0). Due to translational invariance and the quadratic form of kinetic energy, the center-of-mass motion is decoupled from the other degrees of freedom, and defines a harmonic oscillator with cyclotron frequency ωc=eB/m\omega_c=eB/m. Since the uniform electric field only couples to the center-of-mass coordinate, we conclude by the same analysis as above that this system exactly saturates the bound. This is consistent with the Kohn theorem [19], which states that the optical absorption occurs only at ω=ωc\omega=\omega_c, leading to the saturation of Eq. (8) and hence Eq. (10).

Our inequality implies that highly-polarizable electronic materials must have a small energy gap. This makes physical sense: the dielectric constant describes the polarizability of the material due to the virtual excitation of electron-hole pairs across the excitation gap [20]. Indeed, in narrow gap semiconductors with massive Dirac dispersion, as the gap ∣Δ∣\lvert\Delta\rvert decreases, the dielectric constant diverges as ∼log⁡∣Δ∣\sim\log\lvert\Delta\rvert in three dimensions, 1/∣Δ∣1/\lvert\Delta\rvert in two dimensions, and 1/Δ21/\Delta^2 in one dimension. The asymptotic behavior in each case satisfies the inequality. Importantly, our work shows that the gap associated with transverse excitations (which couple to the electromagnetic wave) must vanish when the electronic dielectric constant ϵ\epsilon diverges, but the longitudinal gap can still be finite. Indeed, when the system approaches Coulomb gas in three dimensions and the dielectric constant diverges, the transverse gap bound approaches zero, consistent with the presence of low-lying excitations in a metal. However, the plasmon excitation remains gapped, saturating the longitudinal gap bound (19).

In many systems with a small energy gap, such as narrow-gap topological insulators shown in Figure 1, the electric susceptibility is large and its value is mostly determined by low-energy states near the Fermi level. On the other hand, our derivation of the universal bound on the energy gap Eq. (10) employs the full spectral weight of all electrons in the system, which includes core electrons. Alternatively, we can construct an effective theory involving only low-energy states, and replace the full spectral weight W0W^0 in Eq. (9) with the effective spectral weight Weff0W^0_{\mathrm{eff}}. Since the number of electrons that participate is smaller in the effective theory, we can then expect a tighter bound for narrow gap systems. On this point, we can use the measured plasmon energy EgLE_g^L to define an effective plasma frequency ωpeff\omega_p^{\mathrm{eff}} associated with the effective carrier density: EgL≡ℏωpeff/1−ϵ−1E_g^L \equiv\hbar\omega_p^{\mathrm{eff}}/\sqrt{1-\epsilon^{-1}}. Then, replacing the plasma frequency with ℏωpeff\hbar\omega_p^{\mathrm{eff}}, the bound on optical gap becomes EgT≲ℏωpeff/ϵ−1=EgL/ϵE_g^T \lesssim\hbar\omega_p^{\mathrm{eff}}/\sqrt{\epsilon-1}=E_g^L/\sqrt{\epsilon}. Interestingly, this inequality between EgTE_g^T, EgLE_g^L and ϵ\epsilon is remarkably tight for many materials (see Supplemental Materials). For example, the transverse gap of KCl, silicon, c-BN reaches 82%, 87%, and (nearly) 100% of the bound by EgLE_g^L and ϵ\epsilon, respectively.

In our discussion, we have not included the effect of phonons. However, due to electron-phonon coupling, the phonons can significantly contribute to the static dielectric constant, ϵ(0)\epsilon(0). Instead, the dielectric constant in our theory should be understood as the electron contribution to the dielectric constant, which is often denoted as ϵ(∞)\epsilon(\infty). For example, near the phase transition to a ferroelectric phase, the static dielectric constant ϵ(0)\epsilon(0) can diverge. However, this does not necessarily imply the suppression of electronic gap through the bound (10). What is suppressed is rather the energy of optical phonons, whose softening leads to ferroelectric distortion. Generalization of our theory to include the phonon effect will be forthcoming.

We thank Raffael Resta, Ivo Souza, and Richard Martin for their comments on the manuscript. We are grateful to an anonymous reviewer for a stimulating suggestion. This work was supported by the U.S. Army Research Laboratory and the U.S. Army Research Office through the Institute for Soldier Nanotechnologies, under Collaborative Agreement Number W911NF-18-2-0048. YO is grateful for the support provided by the Funai Overseas Scholarship. LF was partly supported by Simons Investigator Award from the Simons Foundation.

Supplemental Materials

Figures and table for the gap bounds

Image not extracted

Figure 3. Measured transverse optical gap and transverse gap bound.

Image not extracted

Figure 4. Measured longitudinal gap and longitudinal gap bound.

Image not extracted

Figure 5. Approximate inequality relation between the longitudinal gap (plasmon frequency) EgLE_g^L, the transverse gap (optical gap) EgTE_g^T, and the dielectric constant ϵ\epsilon.

Materialϵ(∞)\epsilon(\infty)nn [10−3 m−310^{-3}\,\mathrm{m}^{-3}]EgTE_g^T [eV]EgT,mE_g^{T,m} [eV]EgT/EgT,mE_g^T/E_g^{T,m}EgLE_g^L [eV]EgL,mE_g^{L,m} [eV]EgL/EgL,mE_g^L/E_g^{L,m}Comment
c-BN4.46 [21]1.02 [17]14.5 [17]20.110.72130.4 [22]42.480.716EgTE_g^T is direct gap.
Si12.0 [17]0.700 [17]4.18 [17]9.380.44616.7 [23]32.440.515EgTE_g^T is direct gap.
Diamond5.7 [17]1.06 [17]7.1 [24]17.620.40334.0 [25]42.060.808EgTE_g^T is direct gap.
LiCl2.78 [26]0.589 [27]8.6 [28]21.360.403-35.62-EgTE_g^T is from the lowest absorption peak
LiF1.96 [26]0.739 [27]12.6 [29]32.580.387-45.61-EgTE_g^T is an excitonic gap
3C-SiC6.38 [17]0.965 [17]6.0 [17]15.730.38121.8 [30]39.730.549EgTE_g^T is direct gap.
AlN4.93 [17]0.958 [17]6.19 [17]18.340.338-40.72-EgTE_g^T is direct gap. ϵ(∞)=ϵ∥(∞)\epsilon(\infty)=\epsilon_{\parallel}(\infty)
NaCl2.34 [26]0.624 [31]7.9 [28]25.340.31215.5 [23]38.770.400EgTE_g^T is from the lowest absorption peak
KCl2.19 [26]0.578 [31]7.8 [28]25.870.30114.1 [32]38.290.368EgTE_g^T is from the lowest absorption peak. EgLE_g^L is the first prominent peak in EELS.
LiBr3.17 [26]0.914 [27]7.2 [28]24.090.299-42.90-EgTE_g^T is from the lowest absorption peak
KF1.85 [26]0.736 [27]9.8 [28]34.540.284-46.98-EgTE_g^T is from the lowest absorption peak
MgO2.94 [17]1.07 [17]7.67 [17]27.520.27922.3 [23]47.210.472EgTE_g^T is an excitonic gap
NaF1.74 [26]0.804 [31]10.6 [28]38.700.274-51.05-EgTE_g^T is from the lowest absorption peak
RbF1.96 [26]1.03 [27]9.5 [28]38.380.248-53.73-EgTE_g^T is from the lowest absorption peak
NaBr2.59 [26]0.869 [27]6.7 [28]27.450.244-44.18-EgTE_g^T is from the lowest absorption peak
KBr2.34 [26]0.758 [27]6.7 [28]27.930.24013.2 [32]42.730.309EgTE_g^T is from the lowest absorption peak. EgLE_g^L is the first prominent peak in EELS.
KI2.62 [26]0.818 [27]5.8 [28]26.390.22011.8 [32]42.720.276EgTE_g^T is from the lowest absorption peak. EgLE_g^L is the first prominent peak in EELS.
RbBr2.34 [26]0.892 [27]6.6 [28]30.300.218-46.35-EgTE_g^T is from the lowest absorption peak
NaI2.93 [26]0.950 [27]5.6 [28]26.050.215-44.58-EgTE_g^T is from the lowest absorption peak
RbI2.59 [26]0.918 [27]5.7 [28]28.210.202-45.40-EgTE_g^T is from the lowest absorption peak
α\alpha-GaN5.2 [17]1.66 [17]3.48 [17]23.360.149-53.27-EgTE_g^T is A-exciton. ϵ(∞)=ϵ⊥(∞)\epsilon(\infty)=\epsilon_{\perp}(\infty).
ZnO3.75 [17]1.60 [17]3.44 [17]28.300.12218.2 [33]54.810.332ϵ(∞)=ϵ∥(∞)\epsilon(\infty)=\epsilon_{\parallel}(\infty).
GaAs10.9 [17]1.42 [17]1.52 [17]14.080.10815.8 [34]46.380.341EgTE_g^T is direct gap.
InN8.4 [17]1.80 [17]1.95 [17]18.340.106-53.15-EgTE_g^T is direct gap.
Ge16.0 [17]1.41 [17]0.898 [17]11.400.07916.0 [35]45.590.351EgTE_g^T is direct gap.
SnTe40 [36, 37]1.61 [17]0.36 [17]7.550.048-47.75-EgTE_g^T is direct gap. ϵ(∞)\epsilon(\infty) varies among Ref. [36, 37] around 40–50.
Sb2_2Te3_351.0 [17]1.69 [17]0.28 [17]6.830.041-48.75-EgTE_g^T is direct gap. ϵ(∞)=ϵ⊥(∞)\epsilon(\infty)=\epsilon_{\perp}(\infty).
Bi2_2Te3_385.0 [17]1.90 [17]0.21 [38]5.580.038-51.49-ϵ(∞)=ϵ⊥(∞)\epsilon(\infty)=\epsilon_{\perp}(\infty).
InAs12.4 [17]1.48 [17]0.418 [17]13.370.03113.9 [39]47.040.295EgTE_g^T is direct gap.
Bi2_2Se3_329.0 [17]1.89 [17]0.22 [40]9.650.023-51.98-ϵ(∞)=ϵ⊥(∞)\epsilon(\infty)=\epsilon_{\perp}(\infty).

Table I. Parameters used to calculate the gap bound in Figure 1. ϵ(∞)\epsilon(\infty) is the optical dielectric constant, nn is the electron density, EgTE_g^T is the optical gap, EgLE_g^L is the measured plasma frequency in electron energy loss spectroscopy (EELS), and EgT/L,mE_g^{T/L,m} is the gap bound calculated from Eqs. (18), (19).

Dielectric constant in Wigner crystals

Classical description of Wigner crystal

In the main text, we discussed a scenario of strongly localized electrons to (nearly) saturate our gap bound. Such a scenario is expected to be realized in, for example, c-BN, where the possible maximum kinetic energy of the system is estimated to be ℏ2n2/3/(2m)∼3.9 eV\hbar^2 n^{2/3}/(2m) \sim3.9\,\mathrm{eV}, which is much smaller than the direct gap 14.5 eV14.5\,\mathrm{eV}.

Motivated by this scenario, we consider a low-density interacting electron gas with a uniform positive charge background in three dimensions. Then the interacting energy will dominate over the kinetic energy, and thus electrons are expected to be strongly localized.

The energy of this system is described by

H=∑im2r˙i2+12∑i≠je24πϵ0∣ri−rj∣.(24)H = \sum_i \frac{m}{2}\dot{r}_i^2 + \frac{1}{2}\sum_{i\ne j}\frac{e^2}{4\pi\epsilon_0|r_i-r_j|}. \tag*{(24)}

where the second term is the Coulomb interaction, rir_i is the position of ii-th electron and e(<0)e(<0) is the charge of an electron. When the system forms Wigner crystal, the electrons are localized at ri=Rir_i=R_i so that the interaction energy is minimized. Then we can expand the Coulomb interaction term near the minima and treat them as the harmonic oscillators. RiR_i forms a periodic lattice, and the corresponding reciprocal lattice vectors are denoted by GG. Below we consider the case where each unit cell contains one electron. The number of unit cells and the volume of the system are denoted by NN and VV, respectively.

To derive the phonon dispersion, let us consider the displacements of the electrons:

ri=Ri+ui.(25)r_i = R_i + u_i. \tag*{(25)}

Then the Hamiltonian up to the second order in the displacements is given by

H=E0+14∑i≠j(ui−uj)αKαβ(Ri−Rj)(ui−uj)β(26)H = E_0 + \frac{1}{4}\sum_{i\ne j}(u_i-u_j)_\alpha K_{\alpha\beta}(R_i-R_j)(u_i-u_j)_\beta \tag*{(26)}
Kαβ(R)=e2∂2∂rα∂rβU(r+R)∣r=0,(27)K_{\alpha\beta}(R) = e^2\left.\frac{\partial^2}{\partial r_\alpha\partial r_\beta}U(r+R)\right|_{r=0}, \tag*{(27)}

where E0E_0 is the energy without any displacement. In terms of the Fourier representation, defining ui=(1/V)∑qeiq⋅Riuqu_i=(1/\sqrt{V})\sum_q e^{iq\cdot R_i}u_q, we have

H=E0+∑q(m2∣u˙q∣2+12u−q(K0−K(q))uq)≡E0+∑qH(q).(28)\begin{aligned} H = E_0 + \sum_q\left(\frac{m}{2}|\dot{u}_q|^2+\frac{1}{2}u_{-q}(K_0-K(q))u_q\right) \\ &\equiv E_0+\sum_q H(q). \tag*{(28)} \end{aligned}

Here, K0K_0 and K(q)K(q) are defined as

K0≡∑R≠0K(R)=e24πϵ0∑R≠0∂2∂rα∂rβ(1∣r+R∣)∣r=0(29)K_0\equiv\sum_{R\ne0}K(R)=\frac{e^2}{4\pi\epsilon_0}\sum_{R\ne0}\left.\frac{\partial^2}{\partial r_\alpha\partial r_\beta}\left(\frac{1}{|r+R|}\right)\right|_{r=0} \tag*{(29)}
K(q)≡∑R≠0e−iq⋅RK(R)=e24πϵ0∑R≠0∂2∂rα∂rβ(e−iq⋅R∣r+R∣)∣r=0(30)K(q)\equiv\sum_{R\ne0}e^{-iq\cdot R}K(R)=\frac{e^2}{4\pi\epsilon_0}\sum_{R\ne0}\left.\frac{\partial^2}{\partial r_\alpha\partial r_\beta}\left(\frac{e^{-iq\cdot R}}{|r+R|}\right)\right|_{r=0} \tag*{(30)}

We note that we cannot naively interchange the sum and the derivative in the expression for K0K_0, because the sum over RR without the derivative will diverge. However, we can still rewrite them by subtracting a constant in the derivative as

K0=e24πϵ0∂2∂rα∂rβ∑R≠0(1∣r+R∣−1∣R∣)∣r=0(31)K_0=\left.\frac{e^2}{4\pi\epsilon_0}\frac{\partial^2}{\partial r_\alpha\partial r_\beta}\sum_{R\ne0}\left(\frac{1}{|r+R|}-\frac{1}{|R|}\right)\right|_{r=0} \tag*{(31)}
K(q)=e24πϵ0∂2∂rα∂rβ∑R≠0(e−iq⋅R∣r+R∣)∣r=0(32)K(q)=\left.\frac{e^2}{4\pi\epsilon_0}\frac{\partial^2}{\partial r_\alpha\partial r_\beta}\sum_{R\ne0}\left(\frac{e^{-iq\cdot R}}{|r+R|}\right)\right|_{r=0} \tag*{(32)}

In this form, the quantity under the derivative does not diverge. To calculate KK and K0K_0, we first recall the Ewald summation formula [41]:

∑R≠0e−iq⋅R∣r+R∣=4πNV∑G(e−(G+q)2/(4α2)+i(G+q)⋅r∣G+q∣2)+∑R≠0e−iq⋅R∣r+R∣erfc⁡(α∣r+R∣)−1−erfc⁡(α∣r∣)∣r∣,(33)\sum_{\mathbf{R}\ne0}\frac{e^{-i\mathbf{q}\cdot\mathbf{R}}}{\lvert\mathbf{r}+\mathbf{R}\rvert}=\frac{4\pi N}{V}\sum_{\mathbf{G}}\left(\frac{e^{-(\mathbf{G}+\mathbf{q})^2/(4\alpha^2)+i(\mathbf{G}+\mathbf{q})\cdot\mathbf{r}}}{\lvert\mathbf{G}+\mathbf{q}\rvert^2}\right)+\sum_{\mathbf{R}\ne0}\frac{e^{-i\mathbf{q}\cdot\mathbf{R}}}{\lvert\mathbf{r}+\mathbf{R}\rvert}\operatorname{erfc}(\alpha\lvert\mathbf{r}+\mathbf{R}\rvert)-\frac{1-\operatorname{erfc}(\alpha\lvert\mathbf{r}\rvert)}{\lvert\mathbf{r}\rvert}, \tag*{(33)}

where erfc⁡(x)\operatorname{erfc}(x) is the complementary error function defined as

erfc⁡(x)≡2π∫x∞e−t2 dt,(34)\operatorname{erfc}(x)\equiv\frac{2}{\sqrt{\pi}}\int_x^\infty e^{-t^2}\,\mathrm{d}t, \tag*{(34)}

and α(>0)\alpha(>0) is a parameter to be chosen so that the convergence of these sums will be fast enough.

We note that this formula diverges when q=0\mathbf{q}=0 due to G=0\mathbf{G}=0 term in the first term. However, this will not be a problem when one plugs this formula into Eq. (31) because of the subtraction. More specifically, when q=0\mathbf{q}=0, the term with G=0\mathbf{G}=0 vanishes because of the subtraction. In other words, only when q=0\mathbf{q}=0, one need to exclude G=0\mathbf{G}=0 term from the sum. This is the only difference between K0K_0 and K(q)K(\mathbf{q}) in the q→0\mathbf{q}\to0 limit.

Then one can easily calculate K0−K(q)K_0-K(\mathbf{q}) in q→0\mathbf{q}\to0 limit as

K0,αβ−Kαβ(q→0)=ne2ϵ0qαqβq2(35)K_{0,\alpha\beta}-K_{\alpha\beta}(\mathbf{q}\to0)=\frac{ne^2}{\epsilon_0}\frac{q_\alpha q_\beta}{q^2} \tag*{(35)}

with n≡N/Vn\equiv N/V, and hence the Hamiltonian for q→0\mathbf{q}\to0 limit is given by

H(q→0)=m2∣u˙q∣2+12u−qαne2ϵ0qαqβq2uqβ(36)H(\mathbf{q}\to0)=\frac{m}{2}\lvert\dot{u}_{\mathbf{q}}\rvert^2+\frac{1}{2}u_{-\mathbf{q}\alpha}\frac{ne^2}{\epsilon_0}\frac{q_\alpha q_\beta}{q^2}u_{\mathbf{q}\beta} \tag*{(36)}

Therefore, the dispersion for the longitudinal and transverse phonon in q→0\mathbf{q}\to0 limit is given by

ωL(q→0)=ωp=ne2mϵ0(37)\omega^L(\mathbf{q}\to0)=\omega_p=\sqrt{\frac{ne^2}{m\epsilon_0}} \tag*{(37)}
ωT(q→0)=0(38)\omega^T(\mathbf{q}\to0)=0 \tag*{(38)}

Pinned Wigner crystal

When the Wigner crystal is pinned by a harmonic potential with frequency ω0\omega_0, the Hamiltonian is

H(q→0)=m2(∣u˙q∣2+(ω02+ωp2qαqβq2)u−qαuqβ),(39)H(\mathbf{q}\to0)=\frac{m}{2}\left(\lvert\dot{u}_{\mathbf{q}}\rvert^2+\left(\omega_0^2+\omega_p^2\frac{q_\alpha q_\beta}{q^2}\right)u_{-\mathbf{q}\alpha}u_{\mathbf{q}\beta}\right), \tag*{(39)}

and the resulting dispersion at q→0\mathbf{q}\to0 will be

ωL(q→0)=ω02+ωp2,(40)\omega_L(\mathbf{q}\to0)=\sqrt{\omega_0^2+\omega_p^2}, \tag*{(40)}
ωT(q→0)=ω0.(41)\omega_T(\mathbf{q}\to0)=\omega_0. \tag*{(41)}

Optical response and dielectric constant

Let us calculate the optical response for the pinned Wigner crystal within the classical treatment. We consider only q→0\mathbf{q}\to0 mode. Since each mode behaves as the harmonic oscillator with a frequency ωL/T\omega_{L/T}, the response of the polarization P=neu=ϵ0χE\mathbf{P}=ne\mathbf{u}=\epsilon_0\chi\mathbf{E} to the electric field with frequency ω\omega is given by

χL/T(ω)=ωp2ωL/T2−ω2.(42)\chi^{L/T}(\omega)=\frac{\omega_p^2}{\omega_{L/T}^2-\omega^2}. \tag*{(42)}

Then the dielectric function is given by

ϵ(ω)=1+χT(ω)=ωL2−ω2ωT2−ω2(43)\epsilon(\omega)=1+\chi^T(\omega)=\frac{\omega_L^2-\omega^2}{\omega_T^2-\omega^2} \tag*{(43)}

This is consistent with another expression for the dielectric constant, ϵ(ω)=(1−χL(ω))−1\epsilon(\omega)=(1-\chi^L(\omega))^{-1}.

References

  1. [1]W. Kohn, Theory of the Insulating State, Physical Review 133, A171 (1964).DOI
  2. [2]D. J. Thouless, M. Kohmoto, M. P. Nightingale, and M. den Nijs, Quantized Hall Conductance in a Two-Dimensional Periodic Potential, Physical Review Letters 49, 405 (1982).DOI
  3. [3]Q. Niu, D. J. Thouless, and Y.-S. Wu, Quantized Hall conductance as a topological invariant, Physical Review B 31, 3372 (1985).
  4. [4]M. Z. Hasan and C. L. Kane, Colloquium: Topological insulators, Reviews of Modern Physics 82, 3045 (2010).
  5. [5]R. Resta, The insulating state of matter: a geometrical theory, The European Physical Journal B 79, 121 (2011).DOI
  6. [6]Y. Onishi and L. Fu, Fundamental bound on topological gap (2023), arXiv:2306.00078 [cond-mat].DOI
  7. [7]R. Kubo, Statistical-Mechanical Theory of Irreversible Processes. I. General Theory and Simple Applications to Magnetic and Conduction Problems, Journal of the Physical Society of Japan 12, 570 (1957).DOI
  8. [8]I. Souza, T. Wilkens, and R. M. Martin, Polarization and localization in insulators: Generating function approach, Physical Review B 62, 1666 (2000), publisher: American Physical Society.DOI
  9. [9]J. J. Hopfield, Sum Rule Relating Optical Properties to the Charge Distribution, Physical Review B 2, 973 (1970).DOI
  10. [10]V. Ambegaokar and W. Kohn, Electromagnetic Properties of Insulators. I, Physical Review 117, 423 (1960).DOI
  11. [11]R. Kubo, The fluctuation-dissipation theorem, Reports on Progress in Physics 29, 255 (1966).DOI
  12. [12]R. Eykholt, Extension of the Kubo formula for the electrical-conductivity tensor to arbitrary polarizations of the electric field, Physical Review B 34, 6669 (1986), publisher: American Physical Society.pubmed.ncbi.nlm.nih.gov/9939306
  13. [13]L. D. Landau, J. S. Bell, M. Kearsley, L. Pitaevskii, E. Lifshitz, and J. Sykes, Electrodynamics of continuous media, Vol. 8 (elsevier, 2013).
  14. [14]M. Altarelli, D. L. Dexter, H. M. Nussenzweig, and D. Y. Smith, Superconvergence and Sum Rules for the Optical Constants, Physical Review B 6, 4502 (1972).DOI
  15. [15]D. Y. Smith and E. Shiles, Finite-energy f-sum rules for valence electrons, Physical Review B 17, 4689 (1978).DOI
  16. [16]C. B. Clark, Coulomb Interactions in the Uniform-Background Lattice Model, Physical Review 109, 1133 (1958).DOI
  17. [17]O. Madelung, Semiconductors: data handbook (Springer Science & Business Media, 2004).DOI
  18. [18]D. R. Penn, Wave-Number-Dependent Dielectric Function of Semiconductors, Physical Review 128, 2093 (1962).DOI
  19. [19]W. Kohn, Cyclotron Resonance and de Haas-van Alphen Oscillations of an Interacting Electron Gas, Physical Review 123, 1242 (1961).DOI
  20. [20]S. M. Girvin and K. Yang, Modern condensed matter physics (Cambridge University Press, 2019).DOI
  21. [21]M. I. Eremets, M. Gauthier, A. Polian, J. C. Chervin, J. M. Besson, G. A. Dubitskii, and Y. Y. Semenova, Optical properties of cubic boron nitride, Physical Review B 52, 8854 (1995).pubmed.ncbi.nlm.nih.gov/9979875
  22. [22]M. Jaouen, G. Hug, V. Gonnet, G. Demazeau, and G. Tourillon, An EELS and XAS Study of Cubic Boron Nitride Synthesized under High Pressure - High Temperature Conditions, Microscopy Microanalysis Microstructures 6, 127 (1995).DOI
  23. [23]R. F. Egerton, Electron energy-loss spectroscopy in the TEM, Reports on Progress in Physics 72, 016502 (2008).DOI
  24. [24]S. Logothetidis, J. Petalas, H. M. Polatoglou, and D. Fuchs, Origin and temperature dependence of the first direct gap of diamond, Physical Review B 46, 4483 (1992).pubmed.ncbi.nlm.nih.gov/10004200
  25. [25]L. A. Bursill, J. L. Peng, and S. Prawer, Plasmon response and structure of nanocrystalline diamond powder, Philosophical Magazine A 76, 769 (1997), publisher: Taylor & Francis _eprint: https://doi.org/10.1080/01418619708214209.
  26. [26]N. W. Ashcroft and N. D. Mermin, Solid State Physics (Holt, Rinehart and Winston, New York, 1976).
  27. [27]B. Bendow, P. D. Gianino, Y.-F. Tsay, and S. S. Mitra, Pressure and Stress Dependence of the Refractive Index of Transparent Crystals, Applied Optics 13, 2382 (1974), publisher: Optica Publishing Group.
  28. [28]K. Teegarden and G. Baldini, Optical Absorption Spectra of the Alkali Halides at 10^3\circK, Physical Review 155, 896 (1967), publisher: American Physical Society.
  29. [29]D. M. Roessler and W. C. Walker, Electronic spectrum of crystalline lithium fluoride, Journal of Physics and Chemistry of Solids 28, 1507 (1967).
  30. [30]J.-M. Constantini and J. Ribis, Analysis of Plasmon Loss Peaks of Oxides and Semiconductors with the Energy Loss Function, Materials 16, 7610 (2023), number: 24 Publisher: Multidisciplinary Digital Publishing Institute.
  31. [31]J. R. Rumble, CRC Handbook of Chemistry and Physics, 2nd 104th ed. (CRC Press/Taylor & Francis, Boca Raton, FL., 2023).
  32. [32]A. Akkerman, T. Boutboul, A. Breskin, R. Chechik, A. G. Gibrekhterman, and Y. Lifshitz, Inelastic Electron Interactions in the Energy Range 50 eV to 10 keV in Insulators: Alkali Halides and Metal Ox- ides, physica status solidi (b) 198, 769 (1996). — eprint: https://onlinelibrary.wiley.com/doi/pdf/10.1002/pssb.2221980222.
  33. [33]M. R. S. Huang, R. Erni, H.-Y. Lin, R.-C. Wang, and C.-P. Liu, Characterization of wurtzite ZnO using valence electron energy loss spectroscopy, Physical Review B 84, 155203 (2011).DOI
  34. [34]M. Horák and M. Stöger-Pollach, The Čerenkov limit of Si, GaAs and GaP in electron energy loss spectrometry, Ultramicroscopy 157, 73 (2015).DOI
  35. [35]Z. Poursoti, W. Sun, S. Bharadwaj, M. Malac, S. Iyer, F. Khosravi, K. Cui, L. Qi, N. Nazemifard, R. Jagannath, R. Rahman, and Z. Jacob, Deep ultraviolet plasmonics: exploiting momentum-resolved electron energy loss spectroscopy to probe germanium, Optics Express 30, 12630 (2022).DOI
  36. [36]N. Suzuki and S. A. S. Adachi, Optical Properties of SnTe, Japanese Journal of Applied Physics 34, 5977 (1995), publisher: IOP Publishing.DOI
  37. [37]R. B. Schoolar and J. R. Dixon, Optical Properties of Tin Telluride in the Visible and Infrared Regions, JOSA 58, 119 (1968), publisher: Optica Publishing Group.DOI
  38. [38]S. A. Nemov, Y. V. Ulashkevich, A. A. Rulimov, A. E. Demchenko, A. A. Allahkhah, I. V. Sveshnikov, and M. Dzhafarov, On the Band Structure of Bi2Te3, Semiconductor 53, 603 (2019).DOI
  39. [39]M. Kundmann, Study of semiconductor valence plasmon line shapes via electron energy-loss spectroscopy in the transmission electron Tech. Rep. LBL-26617, 6340092 (1988).DOI
  40. [40]G. Martinez, B. A. Patiño, J. M. Hakl, M. Potemski, Y. S. Hor, A. Materna, S. G. Strzelecka, A. Hruban, O. Caha, J. Novak, A. Dubroka, C. Drasar, and M. Orlita, Determination of the energy band gap of Bi2Se3, Scientific Reports 7, 6891 (2017), number: 1 Publisher: Nature Publishing Group.DOI
  41. [41]J. M. Ziman, Principles of the theory of solids (Cambridge university press, 1972).

Paper details

vlnc:W_3r8v9e6n
Version
v1 · latest
Published
Fresh — added to the Hub within the last 30 days
Added
License
CC BY 4.0
Usage

views · 9

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