Recently, Ghiringhelli et al. ͓Phys. Rev. B 73, 035111 ͑2006͔͒ reported on the d-d crystal field and chargetransfer excitations in MnO by using resonant inelastic x-ray scattering ͑RIXS͒. The data were analyzed with the single impurity Anderson model ͑SIAM͒ as well as with the crystal-field model ͑CFM͒, with both models providing very different parameters. The results of their CFM analysis are in conflict with the results obtained by analyzing electron-energy-loss spectroscopy ͑EELS͒ data with the same model.Transition-metal compounds have been a subject of solidstate research for a long time due to their many interesting properties. The most important one is perhaps the occurrence of what is now often called Mott insulation. 1,2 This means insulating behavior in a system with partially filled bands ͑such a system would be metallic in simple band theory͒ due to large on-site Coulomb interactions. The gap in these systems is often called a Mott gap, though matter can be more complicated. The occurrence of superconductivity in the cuprates and of very large magnetoresistance in the manganites has increased the interest in this class of compounds. 2 The low energy electronic excitations in these systems are important for the understanding of their properties. They have been investigated by optical spectroscopy, electronenergy-loss spectroscopy ͑EELS͒ and by resonant as well as nonresonant inelastic x-ray scattering ͑RIXS, NIXS͒.In a recent paper, Ghiringhelli et al. 3 presented RIXS experiments on MnO, a simple prototype transition-metal compound, measuring the d-d crystal field and charge-transfer excitations. If a new method is used to investigate an established system and the data are analyzed in a known framework, two points have to be addressed: What additional information does the new method provide, and what kind of new information does the new analysis yield? In light of this, we would like to comment on the paper by Ghiringhelli et al. 3
as follows:The RIXS technique is by no means as trivial as the authors make it look. A detailed discussion of the problems has, e.g., been given by Platzman and Isaacs. 4 The discussion of Ghiringhelli et al. 3 uses Eq. ͑5͒ of the work of Kotani and Shin 5 which contains considerable simplifications. In order to document the experimental situation, the RIXS data by Ghiringhelli et al. 3 ͑taken from 4 spectra of this work͒ are compared to the EELS data of Fromme et al. 6 and the EELS data of the present work in Table I. While there is ͑not unexpectedly͒ a good agreement between the two sets of EELS data, there are ͑surprisingly͒ deviations with respect to the RIXS data. This is probably due to the fact that the secondorder RIXS process is a quite complicated one, 4 where it is not obvious that its results must agree with those of optical or EELS experiments ͑which may be called first-order techniques͒. This problem is not addressed by Ghiringhelli et al.The second point concerns the data analysis given by Ghiringhelli et al. 3 Their data have been analyzed by the single im...