Local-field and excitonic effects in the calculated optical properties of semiconductors from first-principles B. Arnaud and M. Alouani Institut de Physique et Chimie des Mate ´riaux (IPCMS), UMR 7504 du CNRS, Universite ´ Louis Pasteur, 23 rue du Loess, 67037 Strasbourg, Cedex, France Received 11 September 2000; published 7 February 2001 The recently developed GW approximation GWAbased on the all-electron full-potential projector aug- mented wave method is used to study the local-field LFand electron-hole excitation effects in the optical properties of small-, medium-, and large-band-gap semiconductors: Si, InP, AlAs, GaAs, and diamond. It is found that while the use of the GWA energies instead of local-density approximation LDAeigenvalues has a tendency to align the calculated structures in the optical spectra with their experimental counterparts, the LF effects do not change these peak positions but systematically reduce the intensities of the so-called E 1 and E 2 structures in all the optical spectra. Taking into account the electron-hole interaction, shifts the theoretical oscillator strength towards lower photon energies and thereby improves considerably the comparison with experiment. It is also shown that the LDA static dielectric constant, a ground-state property, is considerably improved when the LF effects are included. On the other hand, as expected, the static dielectric function obtained using the GW quasiparticle energies, and including the LF effects, is underestimated for all the semiconductors. Including the excitonic effects in the theory is expected to correct this discrepancy with experiment. DOI: 10.1103/PhysRevB.63.085208 PACS numbers: 71.10.-w, 71.15.Mb, 71.20.Nr I. INTRODUCTION The electronic structures of semiconductors and insulators are now well described by means of ab initio methods based on the density-functional theory within the local-density approximation 1 in conjunction with the so-called GW ap- proximation of Hedin. 2 In this approximation the self-energy operator is given as a product of the Green’s function G times the screened Coulmb interaction W. The excited states obtained with this approach are in good agreement with angle-resolved photoemission experiments. 3–5 However, the one-electron description of the optical properties of materials based on the knowledge of the GW electronic structure is not satisfactory. In particular, 1when the GW energies are used the peak positions are much higher in energy than the experimental ones, 2the relative intensity of the so-called E 1 and E 2 peaks is not reproduced with the one-electron theory, i.e., the E 1 peak is underestimated by as much as 50% of the observed value and the E 2 peak is somewhat larger, and 3calculations ignoring excitonic effects, but including local-field LFeffects, reduce the intensities of both E 1 and E 2 peaks. 6 On the other hand, the local-density approximation LDA description of the optical spectra is not satisfactory either, since 1the peak positions are much lower in energy than the experimental ones due to the underestimation of the en- ergy band gap and 2the static dielectric constant, which can be obtained from a functional derivative of the electron density with respect to the total Kohn-Sham potential evalu- ated at the ground state, hence a ground-state property, is overestimated by the LDA calculation. 7–13 However, some of this overestimation is primarily due to the neglect of the local-field effects, and as it can be seen in this paper, the inclusion of the LF effects improves somewhat the agree- ment with experiment. It was always assumed that the inclusion of the electron- hole excitations in the interaction of light and matter is the missing ingredient for an adequate comparison of the theo- retical and experimental optical spectra. Model calculations have somehow qualitatively confirmed this assumption. 14,15 However, it is only recently that ab initio pseudopotential PPcalculations, 16–18 incorporating the electron-hole inter- action into the dielectric function, have been able to deter- mine the importance of this interaction and make a realistic comparison with experiment. To achieve this goal, the so- called Bethe-Salpeter equation has been solved for a range of semiconductors and insulators using the same approach as for the model calculations. 14 The outcome of this hard work was quite an achievement, and led to a good agreement of the optical spectra of Si, Ge, GaAs, diamond, and LiF with experiment. 16–18 Those calculations clearly show that the in- clusion of the two-particle effects in the dielectric function, i.e., the interaction of the electron, promoted from a valence band to a conduction band, with the hole left behind, is in- deed an important ingredient for the description of the opti- cal spectra. In this paper, we are also motivated by the same old prob- lem, i.e., computing correctly the optical spectra without any adjustable parameter. But, instead of using the most popular ab initio pseudopotential method, we use an all-electron method. The calculation becomes, of course, much more complicated due to the complexity of the basis set; neverthe- less, the advantages are well worth the effort. We do not have to pseudoize the valence electron in the atomic region. This is a plus over the PP method, since for localized d electrons the optical matrix elements can be computed with- out any approximation. Even, for semiconductors, it is not clear whether the matrix elements calculated using a PP ap- proach are not necessarily as accurate as those obtained us- ing an all-electron theory. As it can be seen later, the dielec- tric functions of Si and GaAs are reproduced by two different PHYSICAL REVIEW B, VOLUME 63, 085208 0163-1829/2001/638/08520814/$15.00 ©2001 The American Physical Society 63 085208-1