A critical review of the experimental valence charge density of GaAs

The valence charge and difference densities of GaAs have been calculated without previous refinements of a charge density model using six different data sets of X-ray structure amplitudes published until now. Since the data sets have been measured by means of different experimental methods and due to the different data treatment, the individual structure factors differ on the absolute scale. Furthermore, different temperature factors have been published. In order to bring the data to a common level, we used the same two harmonic temperature factors and the same algorithm for correcting the different sets of experimental data for anomalous dispersion. Because of the non-centrosymmetry of the zinc blende structure, these procedures are not strictly model-independent. A simple bond charge model was used to obtain phases of the structure amplitudes and to perform the above-mentioned corrections. In general, the details of `experimental' charge densities depend sensitively on the balanced ratio among the structure factor moduli used. A smooth density map is only obtained if all F have the same high level of accuracy [δ(F)/F < 1%] and if `outliers' are omitted. Only four of the six data sets describe the covalent bond and the partial charge transfer between neighboring atoms, in qualitative agreement with our expectation based on the results of pseudo-potential calculations. However, some quantitative discrepancies remain, particularly in the height of the charge density maximum between nearest neighbours and in some details outside the bonding region.