Arxiv:1807.05896V1 [Cond-Mat.Quant-Gas] 16 Jul 2018 + + + + Are Determined by Direct Coulomb Interactions Between Yellow (Γ6 ⊗Γ7 ) and Green (Γ6 ⊗Γ8 ) Series

Interactions between Rydberg excitons in Cu2O

Valentin Walther,1 Sjard Ole Krüger,2 Stefan Scheel,2 and Thomas Pohl1 1Department of Physics and Astronomy, Aarhus University, Ny Munkegade 120, DK 8000 Aarhus C, Denmark 2Institut fürPhysik, UniversitätRostock, Albert-Einstein-Straße 23, D-18059 Rostock, Germany Highly-excited states of excitons in cuprous oxide have recently been observed at a record quantum number of up to n = 25. Here, we evaluate the long-range interactions between pairs of Rydberg excitons in Cu2O, which are due to direct Coulomb forces rather than short-range collisions typically considered for ground state excitons. A full numerical analysis is supplemented by the van der Waals asymptotics at large exciton separations, including the angular dependence of the potential surfaces.

I. INTRODUCTION is of van der Waals type with a van der Waals coefficient that is found to follow a simple scaling law which was pre- viously used for Rydberg state interactions of alkaline13 Excitons play an important role for the optical prop- 14 erties of many semiconductors. Composed of an electron and alkaline earth atoms . From our calculations, we and a hole bound by their Coulomb attraction, excitons determine the corresponding scaling coefficients, provid- may be considered as artificial atoms that feature a series ing easy access to precise values of Rydberg-exciton van of energy levels very similar to that of simple one-electron der Waals coefficients in Cu2O for future studies of many- atoms. Their relatively low exciton binding energies com- body effects or nonlinear optical phenomena due to in- bined with additional effects such as phonon coupling1 or teractions between highly-excited excitons. crystal inhomogeneities, however, render the observation The article is organized as follows. After outlining the of excited exciton states inherently difficult. Cuprous determination of Rydberg exciton wave functions from the semiconductor band structure in Sec. (II), we de- oxide (Cu2O) stands out in this respect, as it features a comparably large Rydberg energy of ∼ 86 meV, which scribe our calculations of the direct Coulomb pair inter- together with the narrow absorption lines provides well- action in Sec. (III). The obtained potential energy curves suited conditions for exciting excitonic Rydberg states. are discussed in Sec. (IV), where we present the pertur- 2 bative calculation of the van der Waals interactions and In fact, recent measurements on Cu2O semiconductors have demonstrated the preparation of highly-excited Ry- summarize our results for the van Waals coefficients for a dberg excitons with record-breaking principal quantum broad range different excitonic Rydberg states. Finally, numbers of up to n = 25. This discovery has sparked re- implications, limitations and potential applications of the newed theoretical and experimental interest in the field results are discussed in Sec. (V). of excitons, ranging from excitonic spectra in magnetic3 and electric4,5 fields as well as non-atomic scaling laws6 to the breaking of all antiunitary symmetries7 and the II. SINGLE EXCITON STATES onset of quantum chaos8. A further particular appeal of such Rydberg states stems from their strong mutual interactions, which, as Bulk Cu2O is a semiconductor with a cubic crystal demonstrated for cold atomic systems9, can lead to en- structure of the point group Oh and a direct band gap 2 10 of Eg = 2.17208 eV at the center of its Brillouin zone . hanced optical nonlinearities of the material . In con- + Without spin, the uppermost valence band has Γ5 sym- trast to ground-state excitons whose low-energy inter- + + actions can often be described in terms of zero-range metry, which is split into an upper Γ7 - and a lower Γ8 - collisions11,12, the interaction between Rydberg excitons band by the spin-orbit interaction. These two bands can become important already on much larger length are separated by a corresponding spin-orbit splitting of scales and lead to an exciton blockade2 that prevents the ∆ = 131 meV and can be described by an effective band 15 optical excitation of two excitons within typical distances Hamiltonian derived in Ref. . + of several µm. Under such conditions the relevant inter- Together with the lowest Γ6 conduction band, these actions are no longer dominated by exchange effects11 but valence bands form two excitonic series: the so-called

arXiv:1807.05896v1 [cond-mat.quant-gas] 16 Jul 2018 + + + + are determined by direct Coulomb interactions between yellow (Γ6 ⊗Γ7 ) and green (Γ6 ⊗Γ8 ) series. The optical the excitons. transition from the excitonic vacuum to the s-excitons In this work, we determine the interaction between Ry- is dipole forbidden for both series due to the positive parity of both the conduction and the valence band. The dberg excitons in Cu2O. Our calculations account for the dipole-dipole coupling between energetically close exciton series of interest to this work is the yellow series whose pair states that is induced by their direct Coulomb inter- p-exciton resonances are located below the band gap and 2 action and dominates the overall interaction at the large have been observed experimentally . 2 distances relevant under experimental conditions of Ry- We determine the exciton binding energies, EK,nl, dberg exciton blockade. Asymptotically, the interaction and wave functions, Ψ(˜ K, k), from the nonparabolic 2 momentum-space Wannier equation

" 2 2 # ~ (αK + k) 2 + Th |βK − k| Ψ(˜ K, k) 2me (1) 2 Z ˜ 0 e 3 0 Ψ(K, k ) ˜ + 3 d k 0 2 = E Ψ(K, k) 8π ε0εr |k − k |

16 as described in Ref. . Here, the hole’s dispersion Th is obtained from an angular average over the hole disper- FIG. 1. a) Sketch of a pair of excitons i and j, consisting of 15 (i),(j) (i),(j) sion derived from the valence band Hamiltonian of Ref. electrons at re and holes at rh . The center of mass and ε0 denotes the vacuum permittivity, while εr ≈ 7.5 coordinates are indicated by Ri,j , the exciton separation by R . The coordinate system is aligned with zˆ. b) Potential is the static relative permittivity of Cu2O. Furthermore, ij k and K denote the relative and center-of-mass (COM) energy surfaces centered around n = 15p with corresponding momentum of the electron-hole pair, respectively, and van-der-Waals curves. The thin lines are obtained from a numerical diagonalization, whereas the colored lines show the α = m /M and β = m /M denote the mass of the elec- e h asymptotic results for the diﬀerent families of M with |M| = 0 tron (me) and the hole (mh) in units of the total exciton (black), |M| = 1 (red), |M| = 2 (green). mass M = me + mh.

The nonparabolicity of the hole dispersion Th plays an important role for the bound state properties and III. RYDBERG EXCITON INTERACTION yields the leading contribution to the excitonic quan- POTENTIAL tum defect16. Its eﬀect on the center-of-mass dynamics with momentum K can, however, be neglected as long The pairwise interaction between excitons is given by as K π/al, where al is the lattice constant. This ap- the sum proximation is well justiﬁed because the momentum of e2 1 1 the optical photon that generates the exciton is much V (ij) = + (i) (j) (i) (j) 4πε0εr smaller than π/al. Therefore, we can separate the rela- |re − re | |rh − rh | tive and COM part of the exciton wave function, whose ! (4) 1 1 real-space representation can consequently be written as − − (i) (j) (i) (j) |re − rh | |re − rh |

1 iK·R of mutual Coulomb interactions between the electron and ΨK,nlm(R, r) = √ e ψnlm(r), (2) (i) (i) V hole of one exciton at respective positions re and rh , respectively, and another electron-hole pair at positions (j) (j) re and rh . Within a multipole expansion, the inter- with corresponding energies action can be rewritten13 as a series of inverse powers of the exciton COM distance Rij = |Ri − Rj| 2 2 ~ K e2 ∞ V (r , r ) EK,nl = + E0,nl. (3) (ij) X lL i j 2M V = l+L+1 , (5) 4πε0εr l,L=1 Rij

Moreover, R = αre + βrh and r = re − rh, as illustrated where Fig. 1a, and ψnlm(r) denotes the bound-state wave func- L (−1) 4π l L tion obtained from the extended Wannier equation with VlL(ri, rj) = p rirj (6) the standard quantum numbers n, l and m. (2l + 1)(2L + 1) s As we have assumed rotational symmetry and ne- X l + L l + L Y (ˆr )Y (ˆr ) glected the non-parabolic COM dispersion, the excitonic l + m L + m lm i L−m j m states are degenerate with regard to the magnetic quan- (7) tum number m. The anisotropy of the valence band can be included in this calculation and would lead to and Ylm denotes the spherical harmonics defined with further splitting of states with l ≥ 2. The size of this respect to the distance vector Rij. splitting depends on the momentum-space extension and While the interaction between ground-state excitons11 scales roughly with n−3. The same is true for exchange- can be often estimated from first-order perturbation the- splitting of the S-excitons and both effects are neglected ory, by evaluating Coulomb scattering matrix elements in this work, as they are of minor importance to the Ry- based on Hartree-Fock states for pairs of interacting ex- dberg states of interest. citons, such an approximation17 becomes inapplicable for 3 excitonic Rydberg states whose large polarizability2,9,18 such that the dipole-dipole interaction only induces a requires a non-perturbative treatment of the Coulomb weak far off-resonant coupling to other exciton pair interactions. In this regime, the exciton interaction pre- states. Due to the aforementioned interaction blockade dominantly stems from the virtual dipole-dipole coupling of exciton excitation, this condition can be satisfied in 2 between exciton bound states while exchange effects are previous Cu2O experiments . We can thus apply degen- negligibly small. This is typically the case for exciton erate second-order perturbation theory in the form of an distances19 effective operator q q !2 2 2 e2 Vˆ (ij)|αihα|Vˆ (ij) Rij 2 · hri i + hrj i . (8) ˆ X 11 11 HvdW = 3 4πε0εrRij δ |αi∈M/ (11) Note that this condition also ensures convergence of µ X C the above multipole expansion, Eq. (5), which for suf- = 6 |µihµ| R6 ficiently large distances is predominantly determined by µ ij the dipole-dipole contribution l = L = 1, such that whose action is restricted to the degenerate subspaces 18 2 ! M = {|sisji} of fixed l and M at energy E¯ . Here (ij) e rirj 3(ri · Rij)(rj · Rij) ¯ V ≈ 3 − 5 (9) δ = 2E − Eα is the Försterdefect, while the two-exciton 4πε0εr R R ij ij eigenstates |µi are now independent of the distance Rij but can still be composed of several pair states |si, sji. We proceed by expanding the resulting Hamiltonian As shown in Fig. 1(b) for n = 15, the van der Waals for the two interacting excitons in a pair product ba- interaction potential obtained in this way provides an sis |si, sji = |nilimi, njljmji composed of the single- excellent description of our numerical results already for exciton states ψ (r ) and ψ (r ), discussed ni,li,mi i nj ,lj ,mj j R & 2.5 µm. in Sec. II. The adiabatic Born-Oppenheimer potentials Figure 2 and Tab. I summarize our results for the van are then obtained by diagonalizing the resulting internal- der Waals interaction between Cu2O Rydberg excitons state Hamiltonian for a given exciton distance Rij. Its with angular momenta l = 0 (s), l = 1 (p) and l = 2 (d). diagonal elements are given by E + E while the 0,nili 0,nj lj The simplest asymptote is that of two s-excitons. With (ij) 0 0 off-diagonal coupling terms hsi, sj|V |si, sji are calcu- only one asymptotic state |n00, n00i, Eq. (11) reduces to lated using Eq. (9). We choose a quantization that is standard non-degenerate perturbation theory. For higher aligned with Rij, such that the total angular momentum angular momenta, l > 0, however, the degenerate pair M = mi + mj is conserved and remains a good quan- states get mixed by the interaction as given in Tab. (I). tum number for the two-exciton states in the presence of The results are invariant with respect to the sign of M, interaction. reflecting the correponding symmetry of the exciton pair. The numerical diagonalization then yields potential en- This leaves a total of 2l + 1 different |M|-states for ergy surfaces Uµ(Rij) and associated two-exciton states a given l and 2l + 1 − |M| states within each of the |µ(Rij)i. Examples of the resulting interaction curves are (l, |M|)-manifolds, which are indicated by different col- shown in Fig. 1(b) for exciton-pair states around the 15p ors in Fig. 1(b). The depicted van der Waals coefficients asymptote for different values of M. The relevant values and associated eigenstates have been obtained by diag- of n, l and m are dictated by the band symmetry and the onalizing Eq. (11) in each (l, |M|)-subspace. While the chosen excitation scheme as well as the frequency and po- result of this calculation may in general depend on the larization of the involved excitation lasers. The polariza- precise value of the principal quantum number n through tion of the laser that drives the Rydberg state transition the corresponding coupling strengths to other pair states defines another axis that generally can have a finite angle and their relative energy separation, the obtained eigen- with the chosen quantization axis aligned along the dis- states turn out to be virtually independent of n (cf. stan- tance vector Rij, such that the optical coupling strength dard deviations given in Tab. (I)). can depend on the orientation of the exciton pair through The van der Waals interaction rapidly increases with the state composition of the two-exciton state |µ(Rij)i, the principal quantum number n. This is due to the as discussed below. quadratic increase of the dipole matrix elements for tran- sitions between Rydberg states and the decreasing level spacing, such that δ ∼ n−3, which overall results in an IV. EXCITONIC VAN DER WAALS 11 increase of the van der Waals coefficient as C6 ∼ n . INTERACTIONS Similar to the behavior of atomic systems13, our numerical results can be well described by the slightly modified The interaction potential and associated two-exciton scaling relation states assume a simple form for large distances Rij where 11 1 2 C6(n) = n c0 + c1n + c2n , (12) (ij) 0 0 hsi, sj|V |s , s i E0,n l + E0,n l − E0,n0 l0 − E0,n0 l0 i j i i j j i i j j whose coefficients ci depend on the angular numbers and (10) are given in Tab. I. As shown in Fig. 2, this simple ex- 4

c0 6 c1 6 c2 6 |M| composition of ns − ns asymptote 2π [mHz µm ] 2π [mHz µm ] 2π [mHz µm ] ×101 ×102 ×103 n00 0 | n00 i -2.046 -0.672 0.125 |M| composition of np − np asymptote n11 2 | n11 i 1.257 3.641 -0.666 1 √1 (| n10 i − | n11 i) 5.853 8.372 -1.503 2 n11 n10 1 √1 (| n10 i + | n11 i) -3.574 0.680 -0.160 2 n11 n10 n1−1 n11 0 (−0.252 ± 0.003)(| n11 i + n1−1 )+ 8.159 11.549 -2.067 n10 (0.934 ± 0.001) | n10 i 1 n11 0 √ (| n1−1 i − ) -3.456 -0.205 0.006 2 n11 n1−1 n1−1 n11 0 (0.661 ± 0.001)(| n11 i + n1−1 )+ -4.371 0.608 -0.143 n10 (0.356 ± 0.004) | n10 i |M| composition of nd − nd asymptote n22 4 | n22 i 6.247 4.067 -0.719 3 √1 (| n21 i + | n22 i) -2.201 -1.936 0.481 2 n22 n21 3 √1 (| n21 i − | n22 i) 10.906 7.237 -1.317 2 n22 n21 n20 n22 n21 2 −0.429(| n22 i + | n20 i) + 0.795 | n21 i 14.881 9.862 -1.807 2 √1 (| n20 i − | n22 i) 1.536 0.650 -0.018 2 n22 n20 n20 n22 n21 2 0.562(| n22 i + | n20 i) + 0.607 | n21 i -4.005 -3.752 0.889 n2−1 n22 n20 n21 1 0.218(| n22 i − n2−1 ) − 0.673(| n21 i − | n20 i) 17.480 11.564 -2.125 n2−1 n22 1 (−0.465 ± 0.001)(| n22 i + n2−1 )+ 4.604 2.582 -0.373 n20 n21 (0.533 ± 0.001)(| n21 i + | n20 i) n2−1 n22 1 (0.533 ± 0.001)(| n22 i + n2−1 )+ -3.559 -3.905 0.954 n20 n21 (0.465 ± 0.001)(| n21 i + | n20 i) n2−1 n22 n20 n21 1 −0.673(| n22 i − n2−1 ) − 0.218(| n21 i − | n20 i) -2.017 -2.208 0.567 n2−2 n22 0 −0.082(| n22 i + n2−2 )+ 18.386 12.153 -2.234 n2−1 n21 n20 0.451(| n21 i + n2−1 ) − 0.762 | n20 i n2−2 n22 0 (−0.222 ± 0.001)(| n22 i − n2−2 )+ 5.559 3.240 -0.501 n2−1 n21 0.671(| n21 i − n2−1 ) n2−2 n22 0 (0.343 ± 0.004)(| n22 i + n2−2 )+ -0.146 -1.315 0.434 n2−1 n21 (−0.448 ± 0.003)(| n21 i + n2−1 )+ n20 (−0.603 ± 0.001) | n20 i n2−2 n22 0 (0.613 ± 0.002)(| n22 i + n2−2 )+ -3.861 -4.031 0.956 n2−1 n21 (0.311 ± 0.003)(| n21 i + n2−1 )+ n20 (0.236 ± 0.004) | n20 i n2−2 n22 0 −0.671(| n22 i − n2−2 )+ n2−1 n21 (−0.222 ± 0.001)(| n21 i − n2−1 ) -3.736 -3.639 0.850

TABLE I. Various asymptotes listed by quantum numbers l, M with corresponding approximate asymptotic wavefunctions and 11 2 van der Waals coefficients C6(n) = n (c0 + c1n + c2n ) as obtained from fitting for principal quantum numbers n = 12–25. The given errors are standard deviations calculated from the numerically obtained eigenfunctions. pression permits an accurate determination of the van der larger exciton distances. However, a comparison with our Waals interaction for the depicted range 12 . n . 25. numerical results [Fig. (3(b)] shows that the agreement The van der Waals coefficients of the np − np asymp- remains good even at relatively small exciton separations totes for n = 11, however, show slightly larger deviations. of ∼ 1µm, comparable to what is also required for n = 12 This is due to the np + np → (n − 1)d + (n + 1)d coupling in the absence of the Försterresonance. channel, which becomes near resonant around n = 11 as An interesting and often relevant situation arises when shown in Fig. 3(a). As the denominator in Eq. (11) goes an external field introduces an axis that is not parallel to through a minimum, the resulting van der Waals inter- the intermolecular axis Rij. Examples include electric action is enhanced, while the validity of Eq. (11) requires and magnetic fields as well as a tilted excitation laser, 5

FIG. 3. At n = 11 a near Försterresonance in the channel np + np → (n − 1)d + (n + 1)d leads to outlying points in the otherwise quite homogeneous range of C6(n). b) The Försterdefect δ of the given channel passes zero near n = 11. Taking the repulsive part of the interaction as an example, the numerical solution (blue dots) is compared with the long- range asymptote with |M| = 0 (black), |M| = 1 (red), |M| = 2 (green) for n = 11 (b) and n = 12 (c). The color code denotes the relative p-component of each asymptote. Despite of the near resonance at n = 11, their difference is small as long as R & 1µm since the dominant channels fall into the van der Waals regime.

coupling, given by the their overlap with the optically active pair state, becomes a function of θ (Fig. 4).

1.0 1 2 0.8 3 4

0.6 5 p

a 6 l r

e 0.4 v o

0.2

0.0 0 2 FIG. 2. C values for a) the s − s, b) p − p and c) d − d 6 interaction angle asymptotes. The ﬁts are for extrapolation for n ≥ 12. Colored lines denote the families of M with |M| = 0 (black), |M| = 1 (red), |M| = 2 (green), |M| = 3 (blue), |M| = 4 (brown). FIG. 4. Overlap of the optically active pair state φa with the 2 asymptotic molecular eigenstates |µi, O = |hφa|µi| . Here, + each deﬁning a new axis zˆlab. Without loss of generality, for illustration, we chose |φai = |n11, n11i for σ -light and we assume that the molecular axis lies in the (x, z)-plane the eigenstates of the p−p-asymptote (labeled as listed in Tab. of the laboratory frame, such that the two zˆ-axes span I). Note that antisymmetric states 2 and 5 do not couple to the interaction angle θ20. A general transformation of the excitation laser. the states between the frames is given by

mol X l ∗ 0 lab |nlmi = d 0 (θ) |nlm i , (13) mm V. DISCUSSION m0

l where dmm0 (θ) denotes elements of the lowercase Wigner In summary, we have evaluated the interaction between 21 d-matrix . While it is often advantageous to express Rydberg excitons in Cu2O semiconductors and provided the external field in the molecular frame, we illustrate an expression that, together with the tabulated parame- the angular dependence by evaluating the optical cou- ters, facilitates a simple and yet accurate determination pling strengths of the p − p asymptotic pair states in the of the resulting van der Waals interaction for a broad laboratory frame. For a definite laser polarization, only range of Rydberg states. Such van der Waals interactions certain pair states are optically active and the optical may be responsible for the recently observed2 excitation 6 blockade of excitons in Cu2O. The highest lying exciton One major difference between the typical scales of state reported in these experiments (n = 25) covers a Rydberg states of excitons and atomic Rydberg states 4 million times larger volume than the 2p-exciton state, stems from the effective electron and hole masses as 2 owing to the ∼ n scaling of the exciton radius. Such well as the dielectric constant, εr, of the semiconductor. a large radius entails an even higher enhancement of the Both factors tend to decrease the binding energy and polarizability as ∼ n7, such that electrostatic interactions lead to a decrease of the Rydberg constant by a factor 2 become relevant at exciton separations where exchange ν = µX /(µAεr), where µX and µA denote the reduced effects are negligible. mass of the excitonic and atomic system, respectively. The importance of long-range dipole interactions for On the other hand, the excitonic radius is increased by −1 Rydberg excitons is connected to the way they are cre- a factor (εrν) . Therefore we expect the van der Waals 4 5 5 ated by optical excitation. Shifts of the Rydberg pair- coefficient to increase as ∼ εrµA/µX . Accordingly, the state energy due to exciton-exciton interactions can in- van der Waals coefficients as calculated in the present hibit the simultaneous generation of Rydberg excitons work exceed those of typical atomic Rydberg states with 13,14 within a certain radius once they exceed the width of comparable quantum numbers by 5 orders of magni- the corresponding exciton line. For strong interactions tude. This also opens the search for other suitable semi- 22,23 and sufficiently narrow excitation lines, this excitation conductor systems with Rydberg states , each featur- blockade effect thus ensures that excitons are only cre- ing different interaction properties and additional rich 24 ated at distances where van der Waals interactions domi- physics . nate. This may open up a new regime where strong inter- Rydberg excitons thus suggest promising avenues action effects become observable at very low densities of to studies of strong interaction effects in confined 25 10 excitons, which therefore interact over long distances in a geometries , optical nonlinearities or nonclassical light 26,27 quasi-static fashion, as opposed to short-range collisional generation at ultralow exciton densities. The results interactions that determine the behaviour of ground- of the present work provide simple yet accurate interac- state excitons. The accurate knowledge of van der Waals tion potentials for future theoretical explorations of these interactions between Rydberg excitons, as provided by perspectives. the present work, enables quantitative theoretical stud- VI. ACKNOWLEDGEMENT ies of this blockade effect. This in turn would also make it possible to estimate the importance of other mecha- We would like to thank the authors of Ref.20 for shar- nisms such as interactions with the free charges of po- ing their Rydberg potential software “pairinteraction”, tentially forming electron-hole plasmas4 and to thereby parts of which we used in the exact diagonalization. We determine their relative contribution to the nonlinear op- are grateful to the DFG SPP 1929 GiRyd for financial tical response of the semiconductor. support.

1 F. Schweiner, J. Main, and G. Wunner, Phys. Rev. B 93, 12 F. Tassone and Y. Yamamoto, Phys. Rev. B 59, 10830 085203 (2016). (1999). 2 T. Kazimierczuk, D. Fröhlich, S. Scheel, H. Stolz, and 13 K. Singer, J. Stanojevic, M. Weidemüller, and R. Côté, M. Bayer, Nature 514, 343 (2014). Journal of Physics B: Atomic, Molecular and Optical 3 S. Zielińska-Raczyńska, D. Ziemkiewicz, and G. Cza- Physics 38, S295 (2005). jkowski, Phys. Rev. B 95, 075204 (2017). 14 R. Mukherjee, J. Millen, R. Nath, M. P. A. Jones, and 4 J. Heckötter, M. Freitag, D. Fröhlich, M. Aßmann, T. Pohl, Journal of Physics B: Atomic, Molecular and Op- M. Bayer, M. A. Semina, and M. M. Glazov, Phys. Rev. tical Physics 44, 184010 (2011). B 95, 035210 (2017). 15 K. Suzuki and J. C. Hensel, Phys. Rev. B 9, 4184 (1974). 5 S. Zielińska-Raczyńska, D. Ziemkiewicz, and G. Cza- 16 F. Schöne,S.-O. Krüger,P. Grünwald, H. Stolz, S. Scheel, jkowski, Phys. Rev. B 94, 045205 (2016). M. Aßmann, J. Heckötter,J. Thewes, D. Fröhlich, and 6 J. Heckötter, M. Freitag, D. Fröhlich, M. Aßmann, M. Bayer, Phys. Rev. B 93, 075203 (2016). M. Bayer, M. A. Semina, and M. M. Glazov, Phys. Rev. 17 V. Shahnazaryan, I. A. Shelykh, and O. Kyriienko, Phys. B 96, 125142 (2017). Rev. B 93, 245302 (2016). 7 F. Schweiner, J. Main, and G. Wunner, Phys. Rev. Lett. 18 T. G. Walker and M. Saffman, Phys. Rev. A 77, 032723 118, 046401 (2017). (2008). 8 M. Aßmann, J. Thewes, D. Fröhlich, and M. Bayer, Nature 19 R. J. L. Roy, Canadian Journal of Physics 52, 246 (1974), Materials 15, 741 EP (2016), article. https://doi.org/10.1139/p74-035. 9 M. Saffman, T. G. Walker, and K. Mølmer, Rev. Mod. 20 S. Weber, C. Tresp, H. Menke, A. Urvoy, O. Firstenberg, Phys. 82, 2313 (2010). H. P. Büchler, and S. Hofferberth, Journal of Physics B: 10 V. Walther, R. Johne, and T. Pohl, Nature communica- Atomic, Molecular and Optical Physics 50, 133001 (2017). tions 9, 1309 (2018). 21 E. Wigner, Group theory and its application to the quantum 11 C. Ciuti, V. Savona, C. Piermarocchi, A. Quattropani, and mechanics of atomic spectra (Academic Press, 1959). P. Schwendimann, Phys. Rev. B 58, 7926 (1998). 7

22 A. Chernikov, T. C. Berkelbach, H. M. Hill, A. Rigosi, 26 G. Mu˜noz-Matutano, A. Wood, M. Johnson, X. Vidal Y. Li, O. B. Aslan, D. R. Reichman, M. S. Hybertsen, and Asensio, B. Baragiola, A. Reinhard, A. Lemaitre, J. Bloch, T. F. Heinz, Physical review letters 113, 076802 (2014). A. Amo, B. Besga, M. Richard, and T. Volz, ArXiv e- 23 P. T. Greenland, S. A. Lynch, A. F. G. van der Meer, prints (2017), arXiv:1712.05551 [cond-mat.mes-hall]. B. N. Murdin, C. R. Pidgeon, B. Redlich, N. Q. Vinh, 27 M. Khazali, K. Heshami, and C. Simon, Journal of Physics and G. Aeppli, Nature 465, 1057 EP (2010). B: Atomic, Molecular and Optical Physics 50, 215301 24 G. Wang, A. Chernikov, M. M. Glazov, T. F. Heinz, (2017). X. Marie, T. Amand, and B. Urbaszek, Rev. Mod. Phys. 90, 021001 (2018). 25 S. O. Kr¨uger and S. Scheel, Phys. Rev. B 97, 205208 (2018).