Abstract
Plasmons depend strongly on dimensionality: while plasmons in threedimensional systems start with finite energy at wavevector q = 0, plasmons in traditional twodimensional (2D) electron gas disperse as \(\omega _p \sim \sqrt q\). However, besides graphene, plasmons in real, atomically thin quasi2D materials were heretofore not well understood. Here we show that the plasmons in real quasi2D metals are qualitatively different, being virtually dispersionless for wavevectors of typical experimental interest. This stems from a broken continuous translational symmetry which leads to interband screening; so, dispersionless plasmons are a universal intrinsic phenomenon in quasi2D metals. Moreover, our ab initio calculations reveal that plasmons of monolayer metallic transition metal dichalcogenides are tunable, long lived, able to sustain field intensity enhancement exceeding 10^{7}, and localizable in real space (within ~20 nm) with little spreading over practical measurement time. This opens the possibility of tracking plasmon wave packets in real time for novel imaging techniques in atomically thin materials.
Introduction
Plasmons are quantum collective motions of electrons in solids arising from the longrange Coulomb interaction. Since these propagating collective modes are strongly affected by boundary/geometric effects, plasmon modes can be tuned by assembling nanostructured materials with repeated patterns, or by selectively exciting modes that only exists on the surface of metals, such as the surface plasmon polaritons (SPPs)^{1,2,3,4}. In addition, there has been great interest in plasmons in atomically thin (one to a few atomic layer) crystalline metals such as graphene^{5,6,7,8,9}. Graphene plasmons are highly tunable and display the typical dispersion relation^{10} \(\omega _p \sim \sqrt q\) of an ideal 2D electron gas for q → 0, and can be observed with techniques ranging from electronenergy loss spectroscopy (EELS) to nanoimaging of interference patterns generated with atomicforce microscopy (AFM) tips^{1,2,4,11}.
Besides graphene, plasmons in other atomically thin quasi2D metals have been reported^{12,13,14}, some displaying intriguing dispersion relations. Previous ab initio studies have shown that plasmons on monolayer metallic transition metal dichalcogenides (TMDs) disperse as \(\omega _p \sim \sqrt q\) at small q and can have high energy (~1 eV) that are nearly dispersionless for a large range of wavevectors (q ~ 0.1 Å^{−1} to 0.3 Å^{−1})^{15,16,17,18}. Similar findings were found for other quasi2D metals, such as borophene^{19}. The previous studies however have basically overlooked the origin and significance of this dispersionless behavior. As will become clear in the present paper, dispersionless quasi2D plasmons could find a variety of exciting applications, as flat dispersion relations can translate into realspace localization of plasmon wave packets using practical excitations. Moreover, it is important to unravel what is the physical origin of these dispersionless plasmons and to assess whether they are (or can be controllably made to be) sufficiently long lived for technological applications^{20}.
Here, we demonstrate rigorously why the plasmon dispersion relation in real, quasi2D metals is qualitatively different from that in idealized 2D metals (i.e., a homogeneous electron gas), in metallic slabs, and of the SPPs, tending to flatten at relatively large wavevectors for any quasi2D crystals regardless of their chemical composition or nature of the energy bands crossing the Fermi energy. The localfields screening arising from interband transitions in real quasi2D crystalline metals^{21,22,23,24,25,26,27,28}, which is not present in the 2D electron gas model, is the physics behind these dispersionless intrinsic plasmons. This behavior is universal for any atomically thin quasi2D crystalline metal as long as there is a wellseparated band (or set of bands) either completely occupied or unoccupied near the Fermi energy (as shown in Fig. 1a), a condition that is not fulfilled in doped graphene.
Results
Universal dispersionless plasmons in real quasi2D metals
To shed light on this universal flat plasmon dispersion relation, we depict in Fig. 1 the computed plasmon dispersion relation from first principles for monolayer TaS_{2}. To perform the calculation, with atomistic localfield effects, we have to develop an algorithm to compute plasmons for arbitrarily small wavevectors, which heretofore was a bottleneck for ab initio calculations (see Methods section). As expected from a homogeneous 2D electron gas picture, we find that plasmons in monolayer TaS_{2} in Fig. 1b follows a \(\sqrt q\) dispersion relation up to a wavevector of q ~0.05 Å^{−1}; however, the dispersion relation deviates significantly from this \(\sqrt q\) relation afterwards (it becomes flattened), with an onset that is large compared to the wavevector of visible light but still very small compared to the Brillouin zone size of ~1.1 Å^{−1}. This deviation is related to the broken continuous translational symmetry in real crystals (which only have discrete crystal translational symmetry). A homogeneous 2D electron gas with continuous inplane translation symmetry can only have a single partially occupied electronic band that extends to infinite q, and only the intraband electronic transitions contribute to the system’s dielectric screening. On the other hand, both intraband and interband electronic transitions contribute to screening in a real quasi2D metal such as monolayer TaS_{2}, and, as we argue here, the interband transitions are the reason behind the appearance of virtually dispersionless plasmons. Indeed, the plasmon dispersion relation of monolayer TaS_{2} obtained by only considering intraband transitions (blue curve in Fig. 1b) follows a \(\sqrt q\) dispersion relation up to much larger wavevectors^{16}.
To gain some fundamental insights, we derive a general microscopic analytic expression for the plasmon dispersion relation for a quasi2D metal, going beyond the 2D homogenous electron gas model. Our approach is to recast rigorously the plasmon excitations of a real quasi2D material into contributions from two components: (1) an effective 2D electron gas (which is defined to consist of only those states of the system in the bands crossing the Fermi level); and (2) a nonlocal polarizable medium that incorporates the effects of the interband transitions in the real material (see Fig. 1a). Under the condition of long plasmon wavelengths, we derive (see SI section), by solving for the zeros of the total dielectric function, the following expression for the plasmon frequency,
where the symbolic notation \(\left[ {\frac{n}{{m^ \ast }}\left( {\mathbf{q}} \right)} \right]_{{\mathrm{eff}}}\) is a function that is a qdependent generalization of the ratio of the carrier density n by the effective mass \({m^ \ast }\) of an anisotropic 2D system (see Methods section), which only depends on the nature of the band (or set of bands) crossing the Fermi energy. The plasmon frequency is further dependent on an effective 2D dielectric constant, ε_{eff}(q), which includes contributions to the screening due to all the intrinsic interband transitions (indicated by the green arrows in Fig. 1a) of the quasi2D material, excluding the intraband contributions from the set of bands crossing the Fermi energy, as well as due to any external environment (such as a substrate). As demonstrated below, ε_{eff}(q) is responsible for the flat dispersion relation discussed above.
The fundamental idea here is that the effective dielectric function ε_{eff}(q) from the interband transitions of the material essentially has the same analytical expression and behavior as the dielectric function derived in the context of quasi2D semiconductor screening (see, e.g., Ref. ^{28}), and it allows us to understand and determine, from Eq. (1), plasmon dispersion relations in quasi2D metals using established models for quasi2D semiconductors^{21} together with information from ab initio calculations^{23,26,27,28,29,30}. In particular, the semiconductorlike interband effective 2D dielectric screening ε_{eff}(q) of a monolayer material is not a constant, but increases linearly from unity with wavevector at small q, since very longwavelength excitations effectively only experience screening from the vacuum or substrate (see Fig. 1c).
We show that, for long wavelength excitations, \(\left[ {\frac{n}{{m^ \ast }}\left( {\mathbf{q}} \right)} \right]_{{\mathrm{eff}}}\) does not depend on q, while \(\varepsilon _{{\mathrm{eff}}}\left( {\mathbf{q}} \right)\) can be approximated as \(\varepsilon _{{\mathrm{eff}}}\left( q \right) \approx \frac{{1 + \varepsilon _s}}{2} + \rho _0q\), where ε_{s} is the dielectric constant of the insulating substrate and ρ_{0} is a screening length intrinsic to the quasi2D material under consideration arising from the interband transitions (see Methods section). In this limit,
Equation (2) is one of the key results of this work. It shows that plasmons in real quasi2D metals display a universal tendency to flatten for wavevectors above a critical wavevector q_{c} of the order of \(\sim \frac{{\varepsilon _s + 1}}{{2\rho _0}}\), resulting in a plasmon frequency of \(\omega _p^{{\mathrm{flat}}} = \sqrt {\frac{{2\pi {\mathrm{e}}^2}}{{\rho _0}}\left[ {\frac{{n}}{{{m}^ \ast }}} \right]_{{\mathrm{eff}}}}\) that is independent of the substrate dielectric constant ε_{s}, although importantly the substrate dielectric constant can dramatically change the value of q_{c} and the dispersion relation for q < q_{c}. We emphasize here that this phenomenon arises without considering interactions with photons, and hence is not a plasmonpolariton effect. Moreover, Eq. (2) holds in practice even for larger plasmon wavevectors, since the effective ratio \(\left[ {\frac{n}{{m^ \ast }}} \right]_{{\mathrm{eff}}}\left( q \right)\) decreases for larger values of q due to a typical reduction of the quasiparticle band velocity away from the Fermi surface, while the effective interband screening ε_{eff}(q) also saturates for larger q^{28}. Since these two effects partially cancel each other, the main features predicted by Eq. (2)−that the plasmon dispersion relation flattens at larger q and has an asymptotic energy that does not depend on the substrate–holds from our ab initio calculations for all quasi2D materials studied here up to q ~0.3 Å^{−1}.
The plasmon dispersion relation in atomically thin quasi2D metals is in fact conceptually different from that of metallic slabs with thickness large compared to interatomic distance, which support on each surface (within the classical dielectric response framework) a surface plasmonpolariton dispersion relation that flattens at a wavevector \(q_c \sim \frac{{\omega _p^{{\mathrm{bulk}}}}}{{\sqrt 2 c}}\), where \(\hbar \omega _p^{{\mathrm{bulk}}}\) is the bulk plasmon energy^{31,32} (Fig. 2a). In contrast, as we demonstrated above, systems with atomically thin thickness support dispersionless quasi2D plasmons with a very different q_{c} ~ 1/ρ_{0}: the 2D screening length ρ_{0} in these systems (ρ_{0}~25 Å for monolayer metallic TMDs) is not directly related to the thickness of the material nor the bulk plasmon energy^{28}; it is a new length scale that emerges due to the quantum nature of these systems, which depends primarily on the energy for the interband transitions and the inplane polarizability of the orbitals involved in these transitions. In addition, as mentioned above, neither retardation effects in the Coulomb interaction nor the hybridization of electronic transitions with the transverse electromagnetic fields are included in our calculations, so conventional hybrid plasmonpolariton modes cannot be responsible for the dispersionless plasmons in atomically thin metals. Further, the dispersionless quasi2D plasmons here are also qualitatively and quantitatively different from plasmons in conventional thin metallic slabs, which support surface plasmons with an energy relation \(\frac{{\hbar \omega _p^{{\mathrm{bulk}}}}}{{\sqrt 2 }}\sqrt {1 \pm e^{  qd}}\), where the positive (negative) sign corresponds to a mode of antisymmetric (symmetric) relative oscillations of the electron density on the opposite surfaces, and d is the thickness of the slab^{31,33,34} (Fig. 2b). For materials such as bulk metallic TMDs, the bulk plasmon energy is of the order of 20 eV; a classical (and inappropriate) dielectric description of a metallic film of atomically thin thickness (d ~ 5Å) would yield a surface plasmon energy and a critical wavevector ~1/d for dispersionless surface plasmons that are both over one order of magnitude larger than the values found in our ab initio calculations for monolayer TaS_{2} (Fig. 2c). These differences highlight that the dispersionless plasmons in quasi2D materials given by Eq. (1) are conceptually different from classical surface plasmons or hybrid plasmonpolariton modes, and they have instead their origin in the broken translational symmetry in atomically thin metals.
We illustrate the accuracy of our expression in Eq. (1) by computing, from first principles^{35,36}, the plasmon dispersion relation of monolayer TaS_{2} on a number of substrates spanning a range of ε_{s} ~ 2 to 10 (Fig. 3b). Our novel algorithm for fine sampling of transitions near E_{F} enables the ab initio calculation of the polarizability at arbitrarily small wave vectors q and without requiring mappings to model Hamiltonians, solving an important bottleneck in previous ab initio calculations (see Methods section). We compare our direct ab initio results to those from Eq. (1)–that is, by computing \(\left[ {\frac{n}{{m^ \ast }}({\mathbf{q}})} \right]_{{\mathrm{eff}}}\) and ε_{eff}(q) for a suspended monolayer TaS_{2}, while generalizing ε_{eff}(q) for the system on different substrates using the Keldysh model but based on ab initio parameters without any additional fitting (see Methods section). The excellent agreement in Fig. 3b confirms the idea that plasmons in quasi2D metals can be understood as an effective 2D electron gas that experiences an additional screening (due to interband transitions and substrates) of a form of that is characteristic of quasi2D semiconductors. Figure 3b also highlights that the plasmon group velocity can be changed by roughly an order of magnitude with a suitable choice of substrate, in addition to the tunable plasmon energies.
Lifetime of quasi2D plasmons
We now address the fundamental, and technologically important, question of what the lifetime of these collective excitations is. Plasmons are strongly damped when they are inside the Landau damping region, wherein they can decay into free electronhole pairs^{37}. Previous calculations^{16} on monolayer metallic TMDs employed densityfunctional theory (DFT) to compute the Landau damping regions and showed that plasmons in such systems coexist with singleparticle excitations for q ≳ 0.15 Å^{−1} (that is, they are inside Landau damping regions for such large q plasmons), leading to a conclusion of short lifetimes. However, as static DFT is a groundstate theory, it does not yield the needed accurate quasiparticle energies to determine the correct, relevant Landau damping regions^{35,36} and hence plasmon lifetimes. By means of firstprinciples GW calculations for the quasiparticle excitations, we can accurately evaluate the Landau damping regimes in quasi2D materials. Here, we focus on the monolayer metallic TMDs, as these materials are predicted (see Fig. 3) to have large plasmon energies^{16,24} and are intrinsically metallic and stable above their chargedensitywave^{38} and/or ferromagnetic^{39} transition temperature, if any. We find that, in these systems, DFT underestimates the separation between the halffilled d shell bands and the rest of the occupied bands by almost 200 meV and does not yield the correct threshold for singleparticle interband excitations. Our quasiparticle GW calculations, on the contrary, show that the plasmon dispersion relation is well outside of the Landau damping regions, and coupling to finiteq interband transitions (electronhole pair creations) are not a limiting factor for the plasmon lifetime for the materials studied here. The plasmons only reach the continuum of intraband transitions (thus having a decay channel to electronhole pairs) after q ~ 0.4 Å^{−1} (Fig. 3b). Hence, monolayer metallic TMDs represent potential systems to implement the concept of lossless metals to support longlived, dispersionless plasmons^{40}. Other layered metals displaying a favorable electronic structure with interband transitions sufficiently separated in energy may also be good candidates^{14}.
Plasmons not in the Landau damping regions may also decay to free electronhole pairs, through a higher order process, by additionally emitting or absorbing a phonon. While a firstprinciples calculation of this higherorder decay channel is quite involved, it can be estimated from a density of final states for decay, D(q), and an effective plasmonphonon coupling matrix element \(\bar g\) (see Methods section). We compute D(q) and estimate \(\bar g\) for graphene and monolayer TaS_{2}, both quantities from first principles. Even though D(q) is about two orders of magnitude larger for monolayer TaS_{2} than for graphene due to the larger band velocity in graphene, \(\left {\bar g} \right^2\) is about 20 times larger for graphene due to the stronger chemical bonds. The product \(D({\boldsymbol{q}})\overline g\), which determines the plasmon decay rate via phonons, is roughly the same for both systems. We plot in Fig. 4 the calculated qresolved plasmon spectral function for monolayer TaS_{2}. We obtain a plasmon lifetime τ ~ 2 ps from phononmediated decays, comparable to the lifetime of graphene plasmons measured in recent experiments^{41}. More importantly, because plasmons in monolayer metallic TMDs originate mostly from electron orbitals of the innermost transition metal layer, we expect that the plasmon wavefunction does not extend too much outside the structure, and hence plasmons in monolayer metallic TMDs are more robust against substrate phonons than plasmons in graphene. While our estimate of the lifetime is an upper bound for an intrinsic sample, extrinsic effects such as ripples and defects, which needs to be considered in applications, can also be partially reduced with substrate engineering and encapsulation^{42}.
Slow plasmon wave packets
We now highlight how these finiteq, dispersionless quasi2D plasmons can be used in experiments to create novel localized plasmon wave packets or use them for novel spectroscopic applications. We first consider a setup where plasmons are excited with an ultrafast laser pulse of duration T which is electromagnetically coupled to the sample through an AFM tip, commonly used to image plasmons^{1,2,4,11}. We model this external excitation from the AFM tip as an oscillating point charge with time dependence \(\rho _{{\mathrm{ext}}}\left( {{\mathrm{\Delta }}t} \right) \sim \sin \omega_0t \; e^{  \frac{{{\mathrm{\Delta }}t^2}}{{2T^2}}}\) at ~4 Å above the topmost S atoms and compute the induced charge density as a function of space and time on a monolayer TaS_{2} for \(\hbar \omega _0\sim 1\) eV and T ~ 80 fs. We find the resulting induced charge density moves very slowly through the material: after Δt = 1 ps, the wave packet is still localized around a narrow (25 nm thick) ring with a radius of ~120 nm relative to the AFM tip (see inset of Fig. 5a). In addition, we can deduce that these plasmons are still welldefined excitations by computing the figure of merit L/λ ~ v_{g}τ/λ ~ 60, where L is the plasmon propagation length and λ its wavelength, using the above discussed plasmon lifetime τ of ~2 ps and plasmon group velocity v_{g} ~ 0.6 Å/fs (see Methods section).
Slow plasmon wave packets excited such a way should allow one to use pumpprobe timing information as a way to probe the electronic structure of complex heterostructures, since nanoimaging techniques employing AFM tips that image plasmon standing waves^{1,2,4,11} can be adapted to measure plasmon wave packets on quasi2D metals. For instance, a slow plasmon wave packet created at time t_{1} can bounce back from the edge of the qusi2D metal and eventually reach the AFM tip at time t_{2}; the time difference t_{2 }– t_{1} is a function not only of the distance the plasmon traveled, but also of the plasmon group velocity, which is very sensitive to the local screening environment (see Fig. 3b). Hence, compared to typical plasmon nanoimaging techniques, timeresolved plasmon microscopy should have a very high sensitivity to changes in the local and nonlocal electronic structure and can therefore detect changes in the substrate supporting a quasi2D metal and the presence of point defects and interfaces even if these features are not directly on the monolayer material.
Electric field enhancement
Nearly flat plasmon dispersion relations also allow for a high degree of confinement of the electromagnetic radiation: our firstprinciples calculations show that the resulting field intensity enhancement on both sides of a monolayer TaS_{2} is extraordinarily large−exceeding 10^{7} for an external field generated by a thin disk placed 10 Å under the bottom S layer (modeling a defect or nanostructured pattern) and oscillating with a timedependent charge density at a frequency of \(\hbar \omega _0 =\) 0.86 eV (see Fig. 5b and Methods section for details). This giant field intensity enhancement is indeed originating from the strongly localized and slowly moving plasmons. The enhancement obtained at a lower frequency of \(\hbar \omega _0 =\) 0.16 eV, wherein the plasmon group velocity is larger by a factor of ~20×, is only of the order of 10^{4}. But more importantly, as illustrated in Fig. 5, the field enhancement can be experienced in a wide spatial region on top of the monolayer material−only possible as the enhancement originates from the intrinsic quasi2D plasmons of the 2D materials as opposed to from Mie scattering. This would allow for much easier accessing of these plasmoninduced field enhancements in experiments for biosensing, singlemolecular spectroscopy, and catalytic applications, among others. Moreover, we envision that excitations of these slowly moving plasmons with light in larger metamaterials would open the possibility of creating arrays of coupled localized bosonic excitations in extended systems, enabling the monitoring of plasmonplasmon interactions and scattering processes largely evasive to detailed and systematic studies up to now.
Discussion
In summary, our ab initio calculations and effective model analysis show that plasmons in atomically thin quasi2D metals follow a unique dispersion relation that flattens for larger wavevectors, owing to the characteristics of quasi2D screening and broken translational symmetry in real atomically thin crystalline materials. We moreover show that monolayer metallic TMDs are especially suited to explore these nearly dispersionless plasmons, as they can support such collective excitations with long lifetime. The high tunability of these plasmons with varying substrates, as well as the flexibility to create plasmon wave packets with different group velocities, opens the possibility of exciting applications such as tracking plasmon wave packets in real time for timeresolved plasmonic imaging and supporting giant field enhancement (exceeding 10^{7}) in extended quasi2D metals over wide spatial regions.
Methods
New computational method for plasmon dispersion calculations
The key quantity required to obtain the plasmon dispersion is the proper, timeordered polarizability matrix, which is given within the ring approximation by
where V_{xtal} is the crystal volume, m and n denote band indices, k and q are wavevectors in the Brillouin zone, and f is a FermiDirac occupation factor. The dielectric matrix ε is related, within the RPA, to the proper polarizability by \(\varepsilon _{{\mathbf{GG}}^{\prime} }\left( {{\mathbf{q}},\omega } \right) = \delta _{{\mathbf{GG}}^{\prime} }  v\left( {{\mathbf{q}} + {\mathbf{G}}} \right)\chi _{{\mathbf{GG}}^{\prime} }^0\left( {{\mathbf{q}},\omega } \right)\), where v(q + G) is the Coulomb potential in reciprocal space and G is a reciprocal lattice vector.
In most ab initio codes, one typically needs to employ a uniform MonkhorstPack grid to sample all the k points for the virtual transitions involved in building χ^{0}. Within this approach, the plasmon wavevector q is also typically made commensurate with the MonkhorstPack grid used to sample the k points. This leads to a serious computational challenge, as the number of k points N_{k} increases as \(\left( {\frac{1}{q}} \right)^2\) for quasi2D systems. Note that calculations on semiconductors do not exhibit this difficulty due to the presence of a gap. Here, we use an algorithm based on importance sampling to compute the polarizability matrix at arbitrarily small wavevectors, at a constant computational cost and with negligible and controllable error. First, we separate the proper polarizability matrix into intraband and interband contributions. The interband contribution, while may involve a large number of bands, can be easily sampled on a coarse kpoint grid, since there is no Fermi surface to be resolved. The challenge is to efficiently sample the intraband contribution to χ^{0}, which we denote by \(\chi _{{\mathrm{intra}}}^0\).
In our new scheme to efficiently compute \(\chi _{{\mathrm{intra}}}^0\), we sample only the set of k points that are involved with an intraband transition. We first distribute a set of special anchor points {k^{a}} along the different pieces of the Fermi surface, in a way that: (i) v(k^{a})·q < 0, where v(k) is the band velocity at k (this will later ensure that the product of the occupation factors f_{mk + q}(1−f_{nk}) is nonzero); and (ii) their spacing along the direction perpendicular to q is fixed as 1/N_{a}. Note that N_{a} is not the total number of anchor points, but the maximum number of anchor points that a closed and concave Fermi surface could have. We choose N_{a} to be large enough to finely sample transitions along the Fermi energy, around N_{a} ~ 60; however, note that N_{a} does not depend on the magnitude of q, but on the size of the Fermi surface. We then distribute the actual k points that will go into the calculation of \(\chi _{{\mathrm{intra}}}^0\) in pairs around the anchor points, i.e., as \(\left\{ {{\mathbf{k}}_i = {\mathbf{k}}_i^a \pm \frac{{\mathbf{q}}}{2}} \right\}_i\), so that each pair of k points will sample a transition along the various pieces of the Fermi surface.
We then perform the calculation of \(\chi _{{\mathrm{intra}}}^0\) with these special set of k points and by including only intraband transitions. This is enforced by keeping only transitions between states \(\left. {m{\mathbf{k}} + {\mathbf{q}}} \right\rangle\) and \(\left. {\left {n{\mathbf{k}}} \right.} \right\rangle\) with a large overlap, \(\left {\langle {u_{m{\mathbf{k}} + {\mathbf{q}}}{\mathrm{}}u_{n{\mathbf{k}}}}\rangle } \right^2 \ge \frac{1}{2}\). We then compute interband transitions keeping only transitions such that \(\left {\langle {u_{m{\mathbf{k}} + {\mathbf{q}}}{\mathrm{}}u_{n{\mathbf{k}}}}\rangle } \right^2 < \frac{1}{2}\). Since these interband transitions do not sample the Fermi surface, they can be computed with two independent set of 6×6×1 kpoint grids shifted by the plasmon wave vector q. Note that, both for the interband and intraband calculations, new set of wavefunctions need to be calculated depending on the wavevector q. The kpoint grids can be chosen so that the qdependent wavefunctions only depend on the valence states, so they can be efficiently recomputed. However, the approximation of only sampling the intraband transitions with set of pairs of kpoints around the Fermi surface is only valid for small values of q. For values of q larger than 1/60 of the reciprocal lattice vector, we compute the polarizability with an uniform 60×60×1 kpoint grid for transition involving wellseparated occupied and unoccupied bands which are both ~2 eV from the Fermi energy, and employing a uniform 6×6×1 kpoint for the remaining transitions. Altogether, we can accurately and efficiently compute the polarizability matrix from first principles for all relevant wavevectors.
Finally, we obtain the plasmon excitations by finding the peaks of the loss function \(L\left( {{\mathbf{q}},\omega } \right) =  {\mathrm{Im}}\left[ {\frac{1}{{\varepsilon _m\left( {{\mathbf{q}},\omega } \right)}}} \right] =  {\mathrm{Im}}\, \varepsilon _{00}^{  1}\left( {{\mathbf{q}},\omega } \right)\), where \(\varepsilon _m\left( {{\mathbf{q}},\omega } \right) = 1/\left[ {\varepsilon _{00}^{  1}({\mathbf{q}},\omega )} \right]\) is the macroscopic dielectric function. Finding the poles of \({\mathrm{Im}}\, \varepsilon _{00}^{  1}({\mathbf{q}},\omega )\) typically requires a very dense sampling of different frequencies ω around the plasmon. We avoid this by first decomposing \(u({\mathbf{q}}) + iv({\mathbf{q}}) = 1/\varepsilon _{00}^{  1}({\mathbf{q}},\omega )\). Both u(q) and v(q) are smooth, realvalued functions near the peak of the loss function, so they can be accurately interpolated to give the loss function \(L\left( {{\mathbf{q}},\omega } \right) = \frac{{v({\mathbf{q}})}}{{u({\mathbf{q}})^2 + v({\mathbf{q}})^2}}\), the maximum of which we associate with a plasmon energy ω_{p}. The real and imaginary parts of the G=G’=0 components of the inverse dielectric matrix, as well as the real part of the macroscopic dielectric matrix, are shown in Fig. 6. While the absolute values of dielectric matrices reported are not directly observables (i.e., they depend on the stacking spacing used in our supercell calculations, as well as broadening parameters), the poles of the dielectric matrices are not sensitive to these details, and are associated with plasmon excitations.
GW quasiparticle band structure
Our meanfield DFT calculations are performed with the Quantum ESPRESSO package^{43}. We compute the quasiparticle band structure for monolayer TaS_{2} in the 2H phase, using the ab initio GW method^{35} with the BerkeleyGW package^{36}, and using similar convergence parameters as we employed in previous studies on semiconducting monolayer TMDs^{15}. A few notable differences are: (i) we perform the calculations without resorting to any plasmonpole model by computing the fully frequencydependent dielectric matrix and selfenergy, (ii) we include all unoccupied bands in the calculation of the polarizability matrix and selfenergy, and (iii) in order to sample a very dense kpoint grid, we use the nonuniform neck subsampling (NNS) method^{44}.
Microscopic analysis and dispersion relation of quasi2D plasmons
We identify plasmon collective excitations with peaks in the loss function \(L\left( {{\mathbf{q}},\omega } \right) =  {\mathrm{Im}}\frac{1}{{\varepsilon _m\left( {{\mathbf{q}},\omega } \right)}} =  {\mathrm{Im}}\, \varepsilon _{00}^{  1}\left( {{\mathbf{q}},\omega } \right)\), where \(\varepsilon _m({\mathbf{q}},\omega ) = \frac{1}{{\varepsilon _{00}^{  1}({\mathbf{q}},\omega )}}\) is the macroscopic dielectric function, and \(\varepsilon _{00}^{  1}\) is the G = G’ = 0 component of the inverse dielectric matrix. The condition for a plasmon peak is that Re[det ε] = 0. Following the previous discussion, we can separate the proper polarizability into an interband contribution and an intraband contribution and assume for simplicity here that there is a single partially occupied band that is responsible for the intraband transitions.
The real part of the intraband contribution to the proper RPA polarizability, in the long wavelength limit and at zero temperature, is given by
where u_{k}(r) is the periodic part of the Bloch function for the band crossing the Fermi energy with wavevector k, and E_{F} is the Fermi energy.
Since we are interested in plasmon solutions, we know that we evaluate \(\chi _{{\mathrm{intra}}}^0\) at ω = ω_{p} ≫ E_{k + q} − E_{k}, the reason being that \(\omega _p \sim \sqrt q\) but E_{k + q} − E_{k} scales like q in the long wavelength limit. If we define \(\rho _{\mathbf{k}}\left( {\mathbf{r}} \right) \equiv u_{\mathbf{k}}^ \ast \left( {\mathbf{r}} \right)u_{{\mathbf{k}} \,+\, {\mathbf{q}}}\left( {\mathbf{r}} \right)\) and assume that we are in the long wavelength limit, we can rewrite \(\chi _{{\mathrm{intra}}}^0\) in a basisset independent way for ω near ω_{p} as
where we define
L_{z} is the length of the simulation supercell along the normal direction of the plane of the quasi2D material, FS denotes an integration along the Fermi surface, v_{k} is the band velocity for a wavevector at the Fermi surface, and \(\left[ {\frac{n}{{m^ \ast }}\left( {\mathbf{q}} \right)} \right]_{{\mathrm{eff}}}\) is a qdependent function that is a generalization of the ratio of the carrier density n to the effective mass \({m^ \ast }\).
The real part of dielectric matrix for frequencies near ω_{p} can now be written as
where \(\varepsilon _{{\mathrm{inter}}}\left( {\mathbf{q}} \right): = I  v\left( {\mathbf{q}} \right)\chi _{{\mathrm{inter}}}^0\left( {\mathbf{q}} \right)\) is the interband contribution to the dielectric matrix, which we may safely assume is purely real in the frequency range of interest. Using Sylvester’s determinant identity, the condition for a plasmon peak is equivalent to
and W_{inter} is the Coulomb potential screened by the interband transitions.
In the long wavelength limit, if the charge density ρ is localized in the monolayer, we can approximate
where G_{z} is a reciprocallattice vector along the confined direction of the material.
As done in the context of quasi2D semiconductors^{28}, we define an effective quasi2D dielectric function as
and finally arrive at the expression for the plasmon frequency,
Comparison of firstprinciples calculations with analytical model
In order to assess the accuracy of the closedform plasmon dispersion relation used in the model in Eq. (1), we need to compute separately both the interband effective dielectric function, ε_{eff}(q), and the effective ratio \(\left[ {\frac{n}{{m^ \ast }}\left( {\mathbf{q}} \right)} \right]_{{\mathrm{eff}}}\) for a suspended monolayer TaS_{2}. We evaluate ε_{eff}(q) from first principles using the expression in the previous Methods section, while the ratio \(\left[ {\frac{n}{{m^ \ast }}\left( {\mathbf{q}} \right)} \right]_{{\mathrm{eff}}}\) is obtained by a numerical calculation of the plasmon dispersion relation including only intraband transitions also from firstprinciples. For q → 0, this is equivalent to the integral written in the previous Methods section, but it is easier to evaluate in practice.
Next, we obtain the substratedependent plasmon dispersion for monolayer TaS_{2}. To do so, we first fit ε_{eff}(q) obtained for a suspended monolayer TaS_{2} to the fully qdependent expression for the Keldysh model^{21},
where \(\eta _1 = \frac{1}{2}\log (\varepsilon + \varepsilon _s)  \frac{1}{2}\log (\varepsilon  \varepsilon _s)\), \(\eta _s = \frac{1}{2}\log (\varepsilon + 1)  \frac{1}{2}\log (\varepsilon  1)\), ε_{s} is the substrate dielectric constant, and ε and d are effectively fitting parameters intrinsic to the quasi2D material that were obtained from the suspended calculation (for instance, d does not in general correspond to the thickness of the quasi2D material). Note that, for small q and large ε, this expression is commonly approximated as \(\varepsilon _{{\mathrm{eff}}}\left( q \right) \approx \frac{{1 \,+\, \varepsilon _s}}{2} + \rho _0q\), where \(\rho _0 \approx \frac{{d\varepsilon }}{2}\) whose value requires explicit firstprinciples calculation.
For TaS_{2}, we obtain that a fit with ε = 5.52 and d = 10.3 Å reproduces the ab initio curve for ε_{eff}(q) up to 0.4 Å^{−1}. We then obtain the values of ε_{eff}(q) for different substrates by varying ε_{s}, which highlights the versatility of the model in Eq. (1). We take values of ε_{s} from experiment (\(\varepsilon _s^{{\mathrm{PTFE}}} \sim 2\), \(\varepsilon _s^{{\mathrm{hBN}}} \sim 4.5\)), with the exception of WSe_{2}, for which we compute \(\varepsilon _s^{{\mathrm{WSe}}_2} \sim 15\) from first principles. The accuracy of the resulting closeform dispersion relations is depicted in Fig. 3b of the main text.
Framework for estimating plasmon lifetime
We consider the decay rate of a plasmon with momentum q to an arbitrary final state f using Fermi’s golden rule,
where M is a coupling matrix element, and here we use atomic Rydberg units \((\hbar = 2m_e = e^2/2 = 1)\).
We consider the case of the decay of a plasmon outside the Landau damping region, such as plasmons in monolayer TMDs (see Fig. 3b), so that a direct decay of a plasmon to an electronhole pair cannot conserve both energy and momentum. Under these conditions, we need to consider decay processes that involve the emission of a phonon, i.e., in the lowtemperature limit. There are three kinds of decay processes to consider that are firstorder in the electronphonon coupling matrix: (1) the direct decay of a plasmon to the electronic ground state by emission of a phonon, (2) the decay of a plasmon to another plasmon by emission of a phonon, and (3) the decay of a plasmon to an electronhole pair plus a phonon. Process (1) is not relevant for plasmons close to the dispersionless regime, where \(\hbar \omega _p\left( {\mathbf{q}} \right)\sim 1\,{\mathrm{eV}}\), since the phonon energy in monolayer TMDs is smaller by at least one order of magnitude. Process (2), while possible, has a small amplitude, since there is a tight constraint on momentum conservation because of the plasmon dispersion. Therefore, we focus on process (3), which could in principle be quite different between monolayer TMDs and graphene.
We label the final states as follows: the emitted phonon has a wavevector Q, branch index λ and frequency \({\mathrm{\Omega }}_\lambda ^{{\mathrm{ph}}}({\mathbf{Q}})\), while the final electronhole pair consists of an electron with band index c and wavevector k + q − Q, and a hole with band index v and wavevector k, any combinations which satisfy wavevector conservation. We write the decay rate associated with this process as
where \({\it{\epsilon }}\)_{nk} is the quasiparticle energy for an electron or hole in band n and wavevector k.
The coupling matrix element M between an initial plasmon state \(\left. {{\mathbf{q}}} \right\rangle\) and a final state \(\left. {f} \right\rangle\) mediated by the electronphonon interaction Hamiltonian H_{ep} can be expressed as
with
where N_{k} is the number of k points in the Brillouin zone, g_{nmλ}(k, Q) is the electronphonon matrix element, and a^{†} and c^{†} denote the creation operator for a phonon and an electron, respectively. We write the initial plasmon state within the TammDancoff approximation in the basis of free electronhole pairs, \(\left. {{\mathbf{q}}} \right\rangle = \mathop {\sum }\limits_{v^{\prime} c^{\prime} {\mathbf{k}}^{\prime} } A_{v^{\prime} c^{\prime} {\mathbf{k}}^{\prime} }^{\mathbf{q}}\left. {c^{\prime} {\mathbf{k}}^{\prime} + {\mathbf{q}},v^{\prime} {\mathbf{k}}^{\prime} } \right\rangle\), where \(A_{v^{\prime} c^{\prime} {\mathbf{k}}^{\prime} }^{\mathbf{Q}}\) is an expansion coefficient and \(\left. {\left {c^{\prime} {\mathbf{k}}^\prime + {\mathbf{q}},v^\prime {\mathbf{k}}^\prime } \right.} \right\rangle\) denotes a free electronhole pair of Bloch states excited above the ground state, and we write the final state as \(\left. {\left f \right.} \right\rangle = \left. {c{\mathbf{k}} + {\mathbf{q}}  {\mathbf{Q}},v{\mathbf{k}},{\mathbf{Q}}\lambda } \right\rangle\) as a free electronhole pair plus a phonon. The normalization of the initial plasmon state requires that \(A^{\mathbf{q}} \sim N_k^{  \frac{1}{2}}\), so that \(M\left( {{\mathbf{q}} \to {\mathbf{Q}},\lambda ,v,c,{\mathbf{k}}} \right) \sim N_k^{  1}\left[ {g_{cc\lambda }\left( {{\mathbf{k}} + {\mathbf{q}},  {\mathbf{Q}}} \right)  g_{vv\lambda }\left( {{\mathbf{k}},  {\mathbf{Q}}} \right)} \right]\).
A fully firstprinciple evaluation of Γ for a real quasi2D material with multiple bands is challenging, as it involves the plasmon wavefunction and the fully banddependent and wavevectordependent electronphonon coupling matrix elements. At this point, to make the calculation trackable, we approximate all this dependence into an effective electronphonon coupling matrix element \(\bar g\), which is estimated from typical values of \(\left {g_{mn}\left( {{\mathbf{k}},{\mathbf{Q}}} \right)} \right\) found for each material. This give us an orderofmagnitude estimate of the plasmonphonon coupling matrix element. Note that this approximation is partially justified by the fact that we are dealing with metallic systems, as the electronphonon coupling matrix elements are reasonably smooth and do not diverge for vanishing phonon wavevectors.
This also motivates us to define a momentumintegrated jointdensity of states (MIJDOS) as
which is a measure of how many electronhole pairs can be created with an energy \({\upomega}\), regardless of the momentum.
We then arrive at the following simplified expression to estimate the plasmon decay rate,
where Ω^{ph} is a typical phonon frequency associated with \(\bar g^2\). For monolayer metallic TMDs, we simply set Ω^{ph} ≈ 0.
We compute from first principles the momentumintegrated JDOS using DFT and GW calculations and compute the electronphonon matrix elements using densityfunctional perturbationtheory calculations for graphene and monolayer TaS_{2}. The DFT and GW calculations are performed with the Quantum ESPRESSO^{43} package and the BerkeleyGW^{36} package, respectively. We find that the MIJDOS decreases as a function of energy for monolayer materials for \(\hbar \omega \,\gtrsim\, 0.3\,{\mathrm{eV}}\). At the dispersionless region, we find that the monolayer TMD \({\text{MI}}{\hbox{}}{\text{JDOS}}\left( {\hbar {\upomega} = 1\,{\mathrm{eV}}} \right) \approx 0.1\,\left( {{\mathrm{eV}}} \right)^{  1}\) per unit of TaS_{2}. For graphene doped at E_{F} = 0.5eV, on the other hand, we find that \({\text{MI}}{\hbox{}}{\text{JDOS}}\left( {\hbar {\upomega} = 1\,{\mathrm{eV}}} \right) \approx 0.001\left( {{\mathrm{eV}}} \right)^{  1}\) per two carbon atoms.
We estimate \(\bar g^2\) by computing the average of the electronphonon coupling matrix elements over all possible momenta for states at the halffilled band. For monolayer TaS_{2}, the values of \(\left {g_\lambda ({\mathbf{k}},{\mathbf{q}})} \right^2\) range up to 0.035 eV^{2} with an average of ~0.0005 eV^{2}, the latter of which we assign to \(\bar g^2\). For doped graphene, the values of \(\left {g_\lambda ({\mathbf{k}},{\mathbf{q}})} \right^2\) range up to 0.26 eV^{2} with an average of ~0.01 eV^{2}. Using these results for \(\bar g^2\) together with the above calculated MIJDOS, we obtain the plasmon linewidth given in Fig. 4.
The value of the plasmon lifetime \(\tau = (2\Gamma {\mathrm{/}}\hbar )^{  1} \sim 2\,{\mathrm{ps}}\) allows us to estimate the figure of merit L/λ = v_{g}τ/λ, where L is the plasmon propagation length, λ is its wavelength, and v_{g} the group velocity. For the wave packets used in the simulations to produce Fig. 5, we have that q ~ 0.3 \(\AA^{  1}\) and v_{g }~ 0.6 Å/fs, so L/λ ~ 60. In fact, the plasmons should be welldefined excitations (i.e., L/λ ≥ 1) provided that their lifetimes are longer than ~35 fs.
Electric field enhancement
We compute the electric field enhancement from first principles by computing the total electric field \(E_{{\mathrm{tot}}}\left( {{\mathbf{q}} + {\mathbf{G}},\omega } \right) = \mathop {\sum }\nolimits_{{\mathbf{G}}^\prime } \varepsilon _{{\mathbf{G}},{\mathbf{G}}^\prime }^{  1}\left( {{\mathbf{q}},\omega } \right)E_{{\mathrm{ext}}}({\mathbf{q}} \,+ {\mathbf{G}}^\prime ,\omega )\) given an external longitudinal field E_{ext} . One challenge to compute the total electric field in real space and from first principles is that one needs to perform a Fourier transform of a very sharp function, which depends on the matrix ε^{−1}(q, ω). However, the dielectric matrices are only evaluated for a discrete number of wavevectors. We address this by writing the interacting RPA polarizability matrix close to a plasmon state and for a fixed frequency ω_{0} in terms of a spectral representation,
where q_{0} is the plasmon wavevector corresponding to the frequency ω_{0}, i.e., ω_{p}(q_{0}) = ω_{0}, v_{g} is the plasmon group at the wavevector q_{0}, η ≡ τ^{−1}/v_{g}, where τ is the plasmon lifetime (the results do not change for τ ranging from 1 to 10 ps), and C_{G,G′}(q) is an Hermitian matrix that can be fit from the ab initio inverse dielectric matrix. Note that the field enhancement in real space, which depends on the Fourier transform of χ_{G,G′}(q, ω_{0}) over its spatial directions, gets enhanced for slower plasmons.
Because of the mismatch in their wavelengths, plasmons in quasi2D extended materials will not directly couple to light, and thus require a nanostructure with a length scale comparable to that of the wavelength of the plasmon to be excited. This is modeled here as an infinitesimally thin disk with a radius of 10 Å. Because the total field still depends considerably on the geometry of this nanostructure, we define the external field as the longitudinal field created by this nanostructure, which is assumed to be generated by an oscillating charge density uniformly distributed over the disk. We compute the field intensity enhancement as the ratio of the intensity of the total electric field \(\left {E_{{\mathrm{tot}}}\left( {\mathbf{r}} \right)} \right^2\) to the maximum intensity of the external field \(\left {E_{{\mathrm{ext}}}^{{\mathrm{max}}}} \right^2\) generated by this nanostructure. This definition is likely a lower bound to the field enhancement \(\left {E_{{\mathrm{tot}}}\left( {\mathbf{r}} \right)} \right^2/I_{{\mathrm{light}}}\) defined with respect to the intensity I_{light} of an external planewave light source, since the nanostructure can also be engineered to enhance the intensity \(\left {E_{{\mathrm{ext}}}^{{\mathrm{max}}}} \right^2/I_{{\mathrm{light}}}\).
Data availability
The data that support the findings of this study are available from the corresponding authors upon reasonable request.
Code availability
Ab initio calculations were performed with the BerkeleyGW software package^{36}. The new algorithms proposed here and the post processing scripts that support the findings of this study are available from the corresponding authors upon reasonable request.
References
 1.
Fei, Z. et al. Gatetuning of graphene plasmons revealed by infrared nanoimaging. Nature 487, 82–85 (2012).
 2.
Ni, G. X. et al. Ultrafast optical switching of infrared plasmon polaritons in highmobility graphene. Nat. Photon. 10, 244–247 (2016).
 3.
Basov, D. N. & Fogler, M. M. Plasmonic imaging is gaining momentum. Science 357, 132–133 (2017).
 4.
AlonsoGonzález, P. et al. Acoustic terahertz graphene plasmons revealed by photocurrent nanoscopy. Nat. Nanotechnol. 12, 31–35 (2016).
 5.
Chen, J. et al. Optical nanoimaging of gatetunable graphene plasmons. Nature 487, 77–81 (2012).
 6.
Lundeberg, M. B. et al. Tuning quantum nonlocal effects in graphene plasmonics. Science 357, 187–191 (2017).
 7.
Brar, V. W., Jang, M. S., Sherrott, M., Lopez, J. J. & Atwater, H. A. Highly confined tunable midinfrared plasmonics in graphene nanoresonators. Nano Lett. 13, 2541–2547 (2013).
 8.
Woessner, A. et al. Highly confined lowloss plasmons in graphene–boron nitride heterostructures. Nat. Mater. 14, 421–425 (2014).
 9.
Ju, L. et al. Graphene plasmonics for tunable terahertz metamaterials. Nat. Nanotechnol. 6, 630–634 (2011).
 10.
Hwang, E. H. & Das Sarma, S. Dielectric function, screening, and plasmons in twodimensional graphene. Phys. Rev. B 75, 205418 (2007).
 11.
Hu, F. et al. Imaging the localized plasmon resonance modes in graphene nanoribbons. Nano Lett. 17, 5423–5428 (2017).
 12.
Novoselov, K. S., Mishchenko, A., Carvalho, A. & Castro Neto, A. H. 2D materials and van der Waals heterostructures. Science 353, aac9439 (2016).
 13.
Zhang, Y., Rubio, A. & Lay, G. Le Emergent elemental twodimensional materials beyond graphene. J. Phys. D. 50, 053004 (2017).
 14.
Gjerding, M. N., Pandey, M. & Thygesen, K. S. Band structure engineered layered metals for lowloss plasmonics. Nat. Commun. 8, 15133 (2017).
 15.
Cudazzo, P., Gatti, M. & Rubio, A. Localfield effects on the plasmon dispersion of twodimensional transition metal dichalcogenides. New J. Phys. 15, 125005 (2013).
 16.
Andersen, K. & Thygesen, K. S. Plasmons in metallic monolayer and bilayer transition metal dichalcogenides. Phys. Rev. B 88, 155128 (2013).
 17.
Groenewald, R. E., Rösner, M., Schönhoff, G., Haas, S. & Wehling, T. O. Valley plasmonics in transition metal dichalcogenides. Phys. Rev. B 93, 205145 (2016).
 18.
Sen, H. S., Xian, L., da Jornada, F. H., Louie, S. G. & Rubio, A. Ab initio modelling of plasmons in metalsemiconductor bilayer transitionmetal dichalcogenide heterostructures. Isr. J. Chem. 57, 540–546 (2017).
 19.
Huang, Y., Shirodkar, S. N. & Yakobson, B. I. Twodimensional boron polymorphs for visible range plasmonics: a firstprinciples exploration. J. Am. Chem. Soc. 139, 17181–17185 (2017).
 20.
Basov, D. N., Fogler, M. M. & Garcia de Abajo, F. J. Polaritons in van der Waals materials. Science 354, aag1992 (2016).
 21.
Keldysh, L. V. Coulomb interaction in thin semiconductor and semimetal films. JETP Lett. 29, 658–661 (1979).
 22.
Hybertsen, M. S. & Louie, S. G. Ab initio static dielectric matrices from the densityfunctional approach. I. Formulation and application to semiconductors and insulators. Phys. Rev. B 35, 5585–5601 (1987).
 23.
Cudazzo, P., Attaccalite, C., Tokatly, I. V. & Rubio, A. Strong chargetransfer excitonic effects and the BoseEinstein exciton condensate in graphane. Phys. Rev. Lett. 104, 226804 (2010).
 24.
Cudazzo, P., Tokatly, I. V. & Rubio, A. Dielectric screening in twodimensional insulators: Implications for excitonic and impurity states in graphane. Phys. Rev. B 84, 1–7 (2011).
 25.
Qiu, D. Y., da Jornada, F. H. & Louie, S. G. Optical spectrum of MoS_{2}: manybody effects and diversity of exciton states. Phys. Rev. Lett. 111, 216805 (2013).
 26.
Chernikov, A. et al. Exciton binding energy and nonhydrogenic Rydberg series in monolayer WS_{2}. Phys. Rev. Lett. 113, 076802 (2014).
 27.
Latini, S., Olsen, T. & Thygesen, K. S. Excitons in van der Waals heterostructures: the important role of dielectric screening. Phys. Rev. B 92, 245123 (2015).
 28.
Qiu, D. Y., da Jornada, F. H. & Louie, S. G. Screening and manybody effects in twodimensional crystals: monolayer MoS_{2}. Phys. Rev. B 93, 235435 (2016).
 29.
Berkelbach, T. C., Hybertsen, M. S. & Reichman, D. R. Theory of neutral and charged excitons in monolayer transition metal dichalcogenides. Phys. Rev. B 88, 045318 (2013).
 30.
Rösner, M., Şaşıoğlu, E., Friedrich, C., Blügel, S. & Wehling, T. O. Wannier function approach to realistic Coulomb interactions in layered materials and heterostructures. Phys. Rev. B 92, 085102 (2015).
 31.
Economou, E. N. Surface plasmons in thin films. Phys. Rev. 182, 539–554 (1969).
 32.
Raether, H. Excitation of Plasmons and Interband Transitions by Electrons. Vol. 88 (SpringerVerlag, 1980).
 33.
Ritchie, R. H. Plasma losses by fast electrons in thin films. Phys. Rev. 106, 874–881 (1957).
 34.
Stern, E. A. & Ferrell, R. A. Surface plasma oscillations of a degenerate electron gas. Phys. Rev. 120, 130–136 (1960).
 35.
Hybertsen, M. S. & Louie, S. G. Electron correlation in semiconductors and insulators: band gaps and quasiparticle energies. Phys. Rev. B 34, 5390–5413 (1986).
 36.
Deslippe, J. et al. BerkeleyGW: a massively parallel computer package for the calculation of the quasiparticle and optical properties of materials and nanostructures. Comput. Phys. Commun. 183, 1269–1289 (2012).
 37.
Yan, H. et al. Damping pathways of midinfrared plasmons in graphene nanostructures. Nat. Photon. 7, 394–399 (2013).
 38.
Ugeda, M. M. et al. Characterization of collective ground states in singlelayer NbSe2. Nat. Phys. 12, 92–97 (2015).
 39.
van Loon, E. G. C. P., Rösner, M., Schönhoff, G., Katsnelson, M. I. & Wehling, T. O. Competing Coulomb and electron–phonon interactions in NbS2. npj Quant. Mater. 3, 32 (2018).
 40.
Khurgin, J. B. & Sun, G. In search of the elusive lossless metal. Appl. Phys. Lett. 96, 181102 (2010).
 41.
Ni, G. X. et al. Fundamental limits to graphene plasmonics. Nature 557, 530–533 (2018).
 42.
Raja, A. et al. Dielectric disorder in twodimensional materials. Nat. Nanotechnol. 14, 832–837 (2019).
 43.
Giannozzi, P. et al. QUANTUM ESPRESSO: a modular and opensource software project for quantum simulations of materials. J. Phys. Condens. Matter 21, 395502 (2009).
 44.
da Jornada, F. H., Qiu, D. Y. & Louie, S. G. Nonuniform sampling schemes of the Brillouin zone for manyelectron perturbationtheory calculations in reduced dimensionality. Phys. Rev. B 95, 035109 (2017).
Acknowledgements
This work was supported by the Center for Computational Study of ExcitedState Phenomena in Energy Materials (C2SEPEM) funded by the U.S. Department of Energy, Office of Basic Energy Sciences under Contract No. DEAC0205CH11231 at Lawrence Berkeley National Laboratory, as part of the Computational Materials Sciences Program, which provided for theory development, code implementation, and calculations. Computational resources were provided by the National Energy Research Scientific Computing Center (NERSC), which is supported by the Office of Science of the U.S. Department of Energy under Contract No. DEAC0205CH11231, and the Extreme Science and Engineering Discovery Environment (XSEDE), which is supported by the National Science Foundation under Grant No. ACI1053575. We acknowledge financial support from the European Research Council (ERC2015AdG694097). The Flatiron Institute is a division of the Simons Foundation. The authors thank D. Basov, D.Y. Qiu, and H.S. Sen for helpful discussions.
Author information
Affiliations
Contributions
S.G.L. and A.R. conceived the research. S.G.L. supervised the research and the new formalisms developed. F.H.J. developed the methods/formalisms and performed the DFT and GW studies. L.X. performed DFT calculations. All authors contributed to the analyses of results and writing of the paper.
Corresponding authors
Ethics declarations
Competing interests
The authors declare no competing interests.
Additional information
Peer review information Nature Communications thanks Jacob Khurgin and the other, anonymous, reviewer(s) for their contribution to the peer review of this work.
Publisher’s note Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
Rights and permissions
Open Access This article is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons license, and indicate if changes were made. The images or other third party material in this article are included in the article’s Creative Commons license, unless indicated otherwise in a credit line to the material. If material is not included in the article’s Creative Commons license and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this license, visit http://creativecommons.org/licenses/by/4.0/.
About this article
Cite this article
da Jornada, F.H., Xian, L., Rubio, A. et al. Universal slow plasmons and giant field enhancement in atomically thin quasitwodimensional metals. Nat Commun 11, 1013 (2020). https://doi.org/10.1038/s41467020148268
Received:
Accepted:
Published:
Further reading

Modulation of electric field on lowfrequency plasmons of carbon nanotubes
Physica E: Lowdimensional Systems and Nanostructures (2021)

Densityindependent plasmons for terahertzstable topological metamaterials
Proceedings of the National Academy of Sciences (2021)

Theoretical Prediction of TwoDimensional Materials, Behavior, and Properties
ACS Nano (2021)

Synthesis of Wafer‐Scale Graphene with Chemical Vapor Deposition for Electronic Device Applications
Advanced Materials Technologies (2021)

Firstprinciples study of ultrafast dynamics of Dirac plasmon in graphene
New Journal of Physics (2021)
Comments
By submitting a comment you agree to abide by our Terms and Community Guidelines. If you find something abusive or that does not comply with our terms or guidelines please flag it as inappropriate.