Credibility of different calculational schemes for defects in semiconductors: their power and their limits

Two theoretical methods for calculating the electronic structure of point defects are reviewed critically: the empirical tight-binding approximation and local-density theory. The physical basis of the tight-binding approximation is recalled and the dispersion of tight-binding results is analysed for the case of vacancies in silicon and gallium arsenide. Some other points are also discussed such as the validity of the orbital removal method and the pinning of impurity states. The effects of correlation are then analysed: there are two main effects-multiplet splitting, whose local-density estimate is found to provide good agreement with experimental EPR data for vacancies in silicon, and shift in one-particle excitation energies with respect to local-density eigenvalues which is impossible to estimate accurately at the moment. It is concluded that both methods lead to errors of several tenths of an electron-volt as regards the prediction of the position of defect energy levels but nevertheless can provide useful information concerning other properties. Possible ways of improvement are outlined.