Oxide” Supplementary Information on ”Giant Rydberg Excitons in Cuprous Oxide” 1 , S. Scheel2 , H. Stolz2 & M. Bayer1,3 T.Supplementary Kazimierczuk1 , D. Fröhlich Information on ”Giant Rydberg Excitons in Cuprous 1 1 2 T. Kazimierczuk , D. Fröhlich , S. Scheel , H. Stolz2 & M. Bayer1,3 1 Experimentelle Oxide” Physik 2, Technische Universität Dortmund, D-44221 Dortmund, Germany SUPPLEMENTARY INFORMATION Institut für Physik, Universität Rostock, D-18051 Rostock, Germany Experimentelle 2, Technische Universität Dortmund, D-44221 Dortmund, Germany T. KazimierczukPhysik , D. Fröhlich , S. Scheel , H. Stolz & M. Bayer 2 1 1 1 2 2 1,3 3 Ioffe Physical-Technical Institute of the Russian Academy of Sciences, 2 Institut für Physik, Universität Rostock, D-18051 Rostock, Germany doi:10.1038/nature13832 St. Petersburg, 194021, Russia 3 Ioffe 1 Physical-Technical Institute of theUniversität Russian Academy of Sciences, St. Petersburg, 194021, Russia Experimentelle Physik 2, Technische Dortmund, D-44221 Dortmund, Germany für Physik, Rostock, D-18051 Rostock, Germany First we present detailsUniversität of the material Cu2 O and its electronic structure with emphasis on the lowest exciton 3 Ioffe Physical-Technical Institute of the Russian Academy of Sciences, St. Petersburg, 194021, Russia series (yellow series). We then describe the experimental setup and give the corrections of the nP resonances First we present details of the material Cu2 O and its electronic structure with emphasis on the lowest exciton due to phonon background, which allows us to extract the P exciton energies. In the next section we discuss series (yellow series). We then describe the experimental setup and give the corrections of the nP resonances details of the theoretical estimate of the dipole blockade efficiency and the blockade volume. due to we phonon background, which allows toand extract the P exciton energies. the nextonsection we discuss First present details of the material Cuus its electronic structure with In emphasis the lowest exciton 2O details of the theoretical estimate of the dipole blockade efficiency and the blockade volume. series (yellow series). We then describe the experimental setup and give the corrections of the nP resonances Cuprous oxide background, which allows us to extract the P exciton energies. In the next section we discuss due to phonon details ofoxide the theoretical estimate of the dipole blockade efficiency and the blockade volume. Cuprous Cu O crystallizes in a cubic lattice (space group 2 Institut 2 b Oh ). The crystal unit cell and the electronic band a E Cu O crystallizes in a cubic lattice (space group 2 Cuprousareoxide structure shown in Fig. S1. The two upmost va- a b unit cell and the electronic band Oh ). The crystal CB E lence bands (Γ+ Γ+ stem from Cu 7 and 8 symmetry) structure are shown in S1. The two(space upmostgroup vaΓ Cu2 O crystallizes in Fig. a cubic lattice 0.45eV 3d electrons, the lowest+ conduction band from Cu CBb a lence (Γ+ and Γ symmetry) stem from Cu Γ The crystal unit cell and the electronic band Oh ).+bands 7 Γ E and 8the next higher conduction 4s (Γ6 symmetry) 0.45eV 3dstructure electrons, lowest conduction from Cu are the shown in Fig. S1. two upmost va− The band Ett=t2.17eV Γ symmetry). Exciband +from O 2p+electrons (Γ + next 8 higher conduction CB symmetry) andΓthe 4slence (Γ bands (Γ7 and stem from Cu 8 symmetry) ΓΓ tation6 from the two valence bands to the two conCu − VB Ett=t2.17eV 0.45eV Δttt=t0.13eV Exciband from O 2p (Γ8 symmetry). 3d electrons, theelectrons lowest conduction band from Cu Γ O duction+bands results in four exciton series, which Γ Γ tation the two valence to theconduction two conk Cu VB and the bands next higher 4s (Γfrom Δttt=t0.13eV 6 symmetry) are named yellow, green, blue and violet series, as − Γ Ett=t2.17eV O duction bandsO results in four(Γ exciton series, which Exciband from 2p electrons 8 symmetry). shown in Fig. S1. We are concerned with exci- Fig. S1: Structure of Cu2 O. (a)Γ Elementaryk cell of are named yellow, green, bluebands and violet as Cu2 O: The Cu tation from the two valence to theseries, two conO-atoms (red VB spheres) are arranged in a Δttt=t0.13eV tons from the lowest energy series (Γ+ to Γ+ tran- Fig. 7 6 S1: Structure of Cu(yellow Elementary of 2 O. (a)Γspheres) shown in bands Fig. results S1. Wein are with exciO Cu-atoms duction fourconcerned exciton series, which bcc lattice, the formcell a fcc sition). The exciton motion can be+divided k in a + into Cu2 O: The O-atoms (red spheres) are arranged tons theyellow, lowestgreen, energyblue series to Γseries, are from named and(Γviolet 6 tran-as lattice. The sub-lattices are shifted relative to each the centre of mass motion and the 7relative mo- bcc the Cu-atoms (yellow spheres) form cell acorfccof by quarter ofofthe body diagonal, which Fig.lattice, S1:one Structure Cu Elementary sition). The exciton motion can be divided 2 O. (a) shown in Fig. S1. We are concerned with into exci- other lattice. The sub-lattices are shifted relative to each tion of electron and hole. The relative motion is + to cubic symmetry Oh . (b) are Schematic elecO-atoms (red spheres) arranged in a Cu2 O: The the the(Γrelative mo- responds tonscentre from of the mass lowestmotion energyand series to Γ+ 7 number 6 tranby one structure quarter ofaround the (yellow body diagonal, which corcharacterized by the principal quantum n other tronic band the center of Brillouin bcc lattice, the Cu-atoms spheres) form a fcc tion of electron and hole. Thecan relative motioninto is sition). The exciton motion be divided to cubic symmetry . (b) Schematic h and the orbital angular momentum quantum number responds ΓO irreducible repzone (Γ-point) in Cu lattice. The sub-lattices shifted relative toeleceach 2 O. Theare i denote characterized the principal quantum numbermon tronic band structure around the center of Brillouin the centre ofbymass motion and the relative 23 the band symmetry Oh .corother by oneofquarter of functions the body for diagonal, which l = 0, 1, 2, . . . (labeled S, P, D, as in hydrogen). Be- resentations and angular numberis Transitions O. The Γ denote irreducible repzone (Γ-point) in Cu 2 i tiontheoforbital electron and momentum hole. The quantum relative motion between the two valence (VB) and responds to cubic symmetry Oh . (b)bands Schematic eleccause of its band structure, Cu2 O has rather unique resentations ofstructure the band functions forcenter symmetry Oh .23 l characterized = 0, 1, 2, . . . (labeled S, P, D, asquantum in hydrogen). Be-n the by the principal number two conduction bands (CB) lead to four exciton tronic band around the of Brillouin properties as compared to other direct-gap semicon- Transitions between the two valence bands (VB) and cause of orbital its bandangular structure, Cu2 O has rather unique green, and violet. and the momentum quantum number series The Γblue irreducible repzone denoted (Γ-point) yellow, in Cu2 O. i denote ductors like e.g. GaAs. One unusual property is that the two conduction bands (CB) lead to four exciton properties as.compared semiconl = 0, 1, 2, . . (labeledtoS,other P, D,direct-gap as in hydrogen). Be- resentations of the band functions for symmetry Oh .23 denoted yellow, green, blue and violet. the upmost valence and the lowest conduction band series ductors like GaAs. One unusual property that Transitions between the two valence bands (VB) and cause of itse.g. band structure, Cu2 O has rather is unique are derived from states of the same ion (see above). Because both bands have the same parity, dipole transithe upmost valence and the lowest conduction band the two conduction bands (CB) lead to four exciton properties as compared to other direct-gap semicontions between these band states are forbidden. Transitions from the ground (fullblue valence band, empty series denoted yellow,state green, and violet. are derived the unusual same ionproperty (see above). ductors likefrom e.g. states GaAs.ofOne is thatBecause both bands have the same parity, dipole transiconduction band) to excitons, however, obtain the additional quantum numbers of the relative motion of tions betweenvalence these band states are forbidden. from the ground state (full valence band, empty the upmost and the lowest conduction Transitions band electron and hole. The total symmetry of the exciton is thus given by the direct product of the symmetries conduction to excitons, however, obtain the additional quantum the parity, relativedipole motion of are derivedband) from states of the23same ion (see above). Because both bandsnumbers have theof same transiof the bands and the envelope : electron and hole. The total symmetry of the exciton is thus given by the direct product of the symmetries tions between these band states are forbidden. Transitions from the ground state (full valence band, empty Γex = Γv ⊗ Γc ⊗ Γenv 23 ofconduction the bands and thetoenvelope band) excitons,: however, obtain the additional quantum numbers of the relative motion of Γv ⊗ Γisc thus ⊗ Γenv electron and hole. The total symmetry ofΓthe exciton given by the direct product of the symmetries ex = 23 of the bands and the envelope : 1 Γex = Γv ⊗ Γc ⊗ Γenv 1 + 7 yellow green blue violet + 6 yellow green blue violet 8 + 6 8 8 yellow green blue violet + 8 + +6 7 + 8 g g SO SO g + 7 + 8 SO 1 W W W. N A T U R E . C O M / N A T U R E | 1 RESEARCH SUPPLEMENTARY INFORMATION Fig. S2: Scheme of the experimental setup. CCD, charge-coupled device camera; P1-P3, Glan polarizers; λ/2, half wave plates; cryostat, Oxford Variox or Oxford Spectromag; VA, variable attenuators. + Γ+ 7 ⊗ Γ6 ⊗ Γ+ for S envelope 1 Γ− for P envelope 4 + + Γ3 ⊕ Γ5 for D envelope . .. . Because of parity, S excitons are dipole forbidden but quadrupole allowed, whereas P excitons are dipole allowed24 . Because of their strong coupling to light, the excitons discussed so far (free excitons) in general have to be considered as polaritons. Polaritons are exciton–photon mixed states25 that move through the crystal with their group velocity vg , which can be derived from their dispersion relation. For the quadrupole allowed 1S excitons of the yellow series the polariton character was clearly demonstrated26–28 . Although our results are well described within the exciton picture, it will be interesting to analyse the P series within a multi-resonant polariton concept. Experimental setup In Fig. S2 we present the schematic setup of our experiment. The setup serves two purposes: (i) classical transmission spectroscopy by use of a white light source and a monochromator, (ii) high resolution spectroscopy by use of single frequency dye lasers. 2 2 | W W W. N A T U R E . C O M / N A T U R E SUPPLEMENTARY INFORMATION RESEARCH We started our experiments by use of a broadband white light source (Energetiq LDLS) in front of the sample and a double monochromator in second order with a CCD camera as a detector (spectral resolution of about 10 µeV). In our first try we detected resonances only up to n = 12. We discovered that broadband white light leads to a quenching of higher excitons, probably due to the fact that, besides the simultaneous excitation of all P excitons from the yellow series, also excitons from the other series (green, blue and violet series, see Fig. S1) and even free carriers beyond the band gap are excited, which leads to multiple interactions, potentially destroying high-n excitons. Inserting cut-off and interference filters in front of the sample brought some improvement (resonances up to n = 13 were detected). After inserting a small double monochromator (Solar MSA-130, bandwidth about 0.1 nm ∼ 300 µeV), we were able to detect resonances up to n = 17 (Fig. S3). Because of the limited spectral resolution of the detection monochromator (about 10 µeV) these resonances are broader than the resonances determined with the laser. The limitation of the bandwidth of the illumination monochromator to about 300 µeV was a possible reason, why only resonances up to n = 17 were resolved. Optical density We then used a single frequency dye laser. Most of the measurements were done using a Co1.5 n=8 white-light herent 899-21 laser with a linewidth of about 5 neV n=9 scanned laser corresponding to 1.2 MHz. We first focused the laser down to about 50 µm spot size. In that way 1.0 we obtained resonances up to n = 25, if the laser power was cut down to about 500 nW. A single frequency laser has a broad band background due 0.5 to ASE (amplified spontaneous emission). Sending the laser through the Solar-monochromator did not 2.1705 2.1710 2.1715 bring any further improvement. We improved the Energy (eV) uniformity of the laser cross-section by inserting a fibre and obtained the best results by use of a 2 mm Fig. S3: Absorption spectra of P excitons of Cu2 O spot size of the laser on the crystal. This rather large from a 34 µm thick sample at 1.2K. Upper trace: spot size has the advantage that laser power levels Spectrum measured using a white light source with of several µW could be used without running into a resolution of 10 µeV. The highest observed resonance n = 17 is marked with an arrow. Lower trace: any detectable exciton blockade effects. For the balSpectrum measured with a single frequency dye laser anced detection system (NewFocus Nirvana 2007) with a line width of 5 neV, n = 25 is the highest resopower levels of > 5 µW were necessary to guaran- nance resolved. The spectra are shifted vertically for tee linearity and low noise operation. The balanced clarity. detection led to a suppression of laser fluctuations to less than 0.1%. A careful alignment of sample and reference beam on the detector was important to achieve sufficient resolution of the small absorption signals. We tried several different beam splitter setups. We finally decided to use a Glan prism as a beam splitter. The laser system allows an automatic tuning of 20 GHz (83 µeV) only. The spectral range from 2P up to 25P, however, corresponds to 27 meV. Besides the electronically controlled thick etalon there are two other rougher tuning elements (Lyot-filter and thin etalon). The tuning of the laser was automated with use of a stepper motor for the micrometer screw of the Lyot-filter and also for the potentiometer of the thin etalon. The wavelength setting was controlled by a high resolution wavemeter (HighFinesse WSU). All tuning and diagnostic elements were implemented in a LabView program, which allowed us to control spectral range, scan speed, etc. For the measurements of the intensity dependence (Fig. S2), it was of importance to stabilize the laser power. For that purpose we used a liquid crystal-based noise eater BEOC LPC (Laser Power Controller in Fig. S2). 3 W W W. N A T U R E . C O M / N A T U R E | 3 RESEARCH SUPPLEMENTARY INFORMATION The bulk of experience gained in optimizing the single color transmission studies could be exploited in setting up the two-color pump-probe experiment. For these experiments we additionally used a second tunable single-frequency dye laser Matisse DS with a line width of 1 neV, corresponding to about 250 kHz band width. One of the two lasers (Matisse) was used to measure transmission through the sample as described above, while the other was focused at the same spot but not detected after passing through the sample. The overlap of the two beams was controlled by use of a 50µm pin-hole. The samples were strain-free mounted in an Oxford Variox cryostat. For the magnetic field measurement we used an Oxford Spectromag 4000 cryostat. The cryostat is fitted with a split-coil superconductive magnet. In the reported measurements a magnetic field of 0.8T was applied in the Faraday configuration (the field is oriented parallel to the optical axis, along which laser beams propagate). Corrections of nP resonances due to phonon background The measured absorption lines were fitted using the formula10–12 : Γn 2 αn (E) = Cn + 2qn (E − En ) Γn 2 2 + (E − En )2 , (S1) where the amplitude Cn is proportional to the oscillator strength, the width Γn is related to the exciton lifetime through τn = πh̄/Γn , the exciton resonance energy En is taken without the phonon background, and the parameter qn describes the line asymmetry. In Table S1 we present the experimentally measured positions of the absorption lines and the exciton energies En after unfolding using Eq. S1. For n = 2 the relative shift from the unfolding amounts to more than 1 meV and for higher-n excitons it strongly decreases similar to the linewidth (i.e., roughly proportional to n−3 ). Description of exciton series by Rydberg formula including quantum defects As shown in the manuscript for the P-excitons, the exciton energies is well described by the Rydberg formula . − Ry n2 In hydrogen, levels of the same principal quantum number n are degenerate with respect to the orbital angular momentum l (neglecting the Lamb shift). In cuprous oxide, however, we find in combination with data recorded earlier29 for a fixed n small, yet systematic splittings between states of different angular momentum (Fig. S4), which arise from a deviation of the electron-hole interaction potential from a pure 1/r-dependence due to the crystal environment. The deviation arises from effects such as the central-cell correction due to band non-parabolicity, the electron-phonon interaction etc. Deviations from the Rydberg formula are known also from the physics of Rydberg atoms: The spectrum of hydrogen-like alkali atoms differs from the spectrum of hydrogen for low angular momentum states because the valence electron interacts with the ionic core. The Pauli repulsion and the dipole polarization of the rump ion’s electron cloud leads to a larger binding energy compared to hydrogen. From atomic physics it is established that the deviations of the energy levels from the Rydberg formula can be accounted for by the concept of quantum defects in which in the Rydberg formula the following replacement is made n → n − δl , where the δl is the quantum defect8 . Also in the context of semiconductors 4 4 | W W W. N A T U R E . C O M / N A T U R E SUPPLEMENTARY INFORMATION RESEARCH Table S1: Energies of yellow excitons in Cu2 O. Data for the P excitons were measured on 34 µm sample at 1.2K. Difference between the positions of the absorption maxima and exciton energies originate from phonon background according to Eq. S1. Note, that since the experimental data are taken at a photon momentum h̄K, the band gap energy at K = 0 is lower from the given value by the kinetic energy ∆E = h̄2 K 2 /2M ≈ 24 µeV, where the exciton mass M = 1.56m0 . Data for S and D series is taken from Ref. 29. 5 W W W. N A T U R E . C O M / N A T U R E | 5 RESEARCH SUPPLEMENTARY INFORMATION the quantum defect concept has been already introduced theoretically30 . Hence, the resonance energies are given by Ry . (S2) En,l = Eg − (n − δl )2 By fitting Eq. S2 to our data for the P excitons we find a quantum defect for the P series δP = 0.23, which describes the resonance energies of states with n > 10 extremely well. Estimation of dipole blockade radius A rough estimate of the exciton blockade volume can be obtained from the absorbed laser energy, given by the absorbed laser power times the exciton lifetime for which we use an upper boundary of 1 ns, based on the linewidth data. A pronounced reduction of the absorption is obtained already for 100 µW absorbed power, resulting in an absorbed energy of about 100 femtoJoule. With an exciton energy of more than 2 eV we obtain a number of about 300 000 excitons within the laser excitation spot, having a diameter of about 2 mm, corresponding to one exciton in an volume of 300 µm3 . From this we derive a blockade radius of about 5 µm. 2 Energy difference (meV) We stress that this estimation of the quantum defect is rather crude due to limited data about other exciton series. Further studies are required to assess whether in the case of excitons quantum defect can be proven to be independent on n. S D F 1 0 -1 3 4 5 6 7 n Fig. S4: Effect of quantum defect on exciton series. The plot presents the differences between known energies of S, D, as well as F excitons and P excitons of the same principal number n. Data for S and D series were taken from Ref. 29. Solid lines represent n−3 fits revealing the dependences of the quantum defects on exciton angular momentum: This estimate of the dipole blockade vol- δS −δP = 0.17, δD −δP = −0.23, and δF −δP = −0.16. ume can be done more quantitatively either based From the independent fit to the P exciton energy seon the experimentally measured saturation under ries δP ≈ 0.23. strong laser excitation or based on theoretical calculations of the dipole-dipole interaction. In case of the experiment-based estimate we use an expression for the blockade efficiency Sn derived in the same way as Eq. 1 of the manuscript: ηα0 Vblockade T1 , Sn = Aexc h̄ω (S3) where α0 ≈ 20 mm−1 is the peak absorption without the blockade, Vblockade is the volume around the exciton unavailable for absorption, Aexc ≈ 3 mm2 is the laser spot area, h̄ω ≈ 2.1 eV is the energy of the exciton and T1 is the n-dependent exciton lifetime, assuming that the linewidth is radiatively limited. The lower bound for T1 is given by the exciton linewidth: T1 ≥ Γ2h̄n ≈ 7 · 10−14 s · n3 , using the experimental data. The factor η = 0.5 accounts for the reflection losses on the cryostat windows and the crystal surface. For simplicity we assume here a constant value of α0 . This simplification corresponds to the assumption that the deviation from the power law for highest-n excitons (as shown in Fig. 2 in the manuscript) is due to 6 6 | W W W. N A T U R E . C O M / N A T U R E SUPPLEMENTARY INFORMATION RESEARCH local inhomogeneity of the crystal. By solving the above equation for Vblockade we get: Vblockade = Sn Aexc h̄ω 1.3 · 106 mW · µm3 ≤ Sn . ηα0 T1 n3 (S4) 3 Blockade volume The second approach to estimate the blockade volume is based on an analysis of the dipoledipole interaction between two excitons at a distance R. The dipole-dipole interaction may be discussed either in the resonant or nonresonant regime and can be computed either from perturbation theory or from an exact diagonalization of the static dipole-dipole interaction potential. The former method can be used to obtain the scaling of the interaction with the principal quantum number n: At large spatial separations between excitons, the dipole-dipole interaction contributes to nondegenerate second-order perturbation theory ∝ |Vdd |2 /∆, where ∆ is the energy difference between neighboring dipole-allowed states. The strongest contribution comes from coupling with the states of the same n, which would be degenerate for a perfect 1/r potential. As mentioned, lifting of the degeneracy between those states can be described using the quantum defect concept, according to which the resonance energy is given by Eq. S2. V blockade ( μm ) From inserting the fitted values for Sn (see manuscript) into this formula we obtain the blockade volumes shown in Fig. S5 as function of the principal quantum number. The blocked volume increases drastically with n by more than two orders of magnitude from n = 12 upwards. 10 3 10 2 10 1 experiment theory 12 14 16 n 18 20 22 24 Fig. S5: Estimate of the blockade volume. Red squares give experimental data for blockade volume Vblockade calculated from the Sn values, solid line represents results of theoretical calculation. For n = 24 the corresponding critical radius Rc of the blockade volume amounts to about 10 µm. The energy difference between neighboring dipole-coupled states scales as n−3 , both in case of the same n and different n states. The dipole moments increase as n2 , similar to the size of a Rydberg state. The dipole-dipole interaction potential Vdd is quadratic in the dipole moments and scales with the third power of the distance between interacting excitons: Vdd ∝ (p1 p2 ) /R3 , where p1 , p2 are the dipole moments of the interacting excitons. Hence, the nonresonant van der Waals interaction takes the form: |Vdd |2 n4 /R3 EvdW (n) ∝ ∝ ∆ n−3 2 ∝ n11 . R6 (S5) The proportionality constant C6 (n = 1) can be calculated by adding up the contributions from all dipolecoupled states. On the other hand, for closer separation of the excitons, i.e., when the dipole-dipole interaction becomes comparable or larger than the energy spacing between the exciton levels, the interaction contains also resonant contributions proportional to Vdd . By the above arguments, this means that this Förster-type interaction takes the form: EF (n) ∝ Vdd ∝ n4 /R3 . (S6) 7 W W W. N A T U R E . C O M / N A T U R E | 7 RESEARCH SUPPLEMENTARY INFORMATION In this case the proportionality constant C3 (n = 1) can be calculated by diagonalizing the dipole interaction within the subspace of dipole states considered as degenerate. As discussed in the manuscript, we define a blockade radius by comparing the dipole-dipole interaction strength with the linewidth of the excitonic line, which scale falls into the Förster regime. The C3 coefficient for the interaction of two particular P states can be computed from the eigenvalues of the (static) dipole-dipole interaction potential in the two-exciton basis containing the nearest dipole-allowed states. In our calculation we include S and D states with the same and adjacent n. We neglect any retardation effects that might contribute at larger separations. Taking into account that the excitons reside in a host medium with a static permittivity of = 7.5 reduces the C3 coefficient from its nominal free-space value. Thus, we estimate the C3 coefficient for the resonant interaction between two excitons in Rydberg states as C3 /n4 ≈ 6 · 10−4 µeV µm3 . Consequently, the expected blockade volume is: 4 C3 ≈ 3 · 10−7 µm3 · n7 . Vblockade = π 3 Γn /2 (S7) The results of these calculations are shown by the open symbols interpolated by a solid line in Fig. S5. Here we take into account the influence of a single exciton only onto the absorption. For the van der Waals form of the interaction the dependence looks similar. The calculations reproduce the increase of the blocked volume with increasing principal quantum number quite well. The calculated volumes are comparable to the experimentally estimated ones even though they tend to be somewhat smaller by a factor of 2 compared to the estimates from experiment. This is still a reasonable agreement, given the simplicity of our estimates done on purpose for transparency of the model, without trying to enforce agreement. For n = 24 the radius of the blockade volume would be about 10 µm, almost an order of magnitude larger than the exciton Bohr radius of this state, suggesting an enormous strength of the dipolar interactions. Exclusion of alternate explanations for the weak absorption of high-n excitons As discussed in the main text, the absorption strength for high-n excitons is weaker than the extrapolation of the dependence for low-n excitons, which follows the n−3 scaling expected for isolated P excitons. We have attributed this to the effect of the dipole blockade. For completeness, we discuss and exclude possible alternative explanations. At first sight, one may attribute the drop to thermal effects. The energy separation between the highest P excitons (n = 25) and the ionization continuum amounts to only about 140 µeV, which is comparable to the thermal energy at 1.2K. Yet their line width is below 3 µeV. This shows that scattering of these Rydberg excitons by phonons and thermal excitation into the ionization continuum is not efficient. In any case thermal scattering would only broaden an absorption line, but not reduce its area. Therefore it can be safely ruled out as origin for the absorption drop. The reduction may also point towards an invalidity of the electric dipole approximation, which is based on the assumption that the product of the light wave number k and the extension of the state wave function r is small compared to unity (kr 1). Usage of this approximation may be challenged due to the large exciton size being comparable to or even considerably larger than the light wavelength. While this issue is worth further investigation, any influence related to it would not depend on the laser intensity, as observed in the experiment. The same argument applies to the temperature-based mechanism discussed above, because the laser power causing bleaching of high-n excitons is well below the power level needed 8 8 | W W W. N A T U R E . C O M / N A T U R E SUPPLEMENTARY INFORMATION RESEARCH to induce any measurable heating effects. By monitoring the crystal transmission under modulated-in-time laser excitation, we verified also that the excitation induced changes of the absorption do not arise from long-lived exciton populations formed after relaxation of optically injected excitons. An example for such a long-lived exciton is the paraexciton which is the spin-triplet configuration of the 1S ground state exciton with lifetimes in the µs-range. Supplementary references 23. Koster, F., G., Dimmock, O., J., Wheeler, G., R. & Statz, H. The Properties of the Thirty-Two Point Groups (The MIT Press, Cambridge, 1963). 24. Agekyan, T., V. Spectroscopic properties of semiconductor crystals with direct forbidden energy gap. phys. stat. sol. (a) 43, 11 (1977). 25. Andreani, C., L. Exciton-polaritons in bulk semiconductors and in confined electron and photon systems in Strong Light-Matter Coupling: from Atoms to Solid-State Systems (World Scientific, Singapore, 2014). 26. Fröhlich, D. et al. Coherent propagation and quantum beats of quadrupole polaritons in Cu2 O. Phys. Rev. Lett. 67 (1991). 27. Fröhlich, D. et al. High resolution spectroscopy of yellow 1s excitons in Cu2 O. Sold State Comm. 134, 139 (2005). 28. Fröhlich, D., Brandt, J., Sandfort, C., Bayer, M. & Stolz, H. High resolution spectroscopy of excitons in Cu2 O. phys. stat. sol. (b) 243, 2367 (2006). 29. Uihlein, C., Fröhlich, D. & Kenklies, R. Investigation of exciton fine structure in Cu2 O. Phys. Rev. B 23, 2731 (1981). 30. Wu, W. & Fisher, J., A. Exchange between deep donors in semiconductors: A quantum defect approach. Phys. Rev. B 77, 045201 (2008). 9 W W W. N A T U R E . C O M / N A T U R E | 9