We next examined H passivation of Be and Mg in InP and GaAs to examine the effect of changing acceptor and host type. We used Ga(4,5), As(4,5), In(4,5), P(4,5) and H(2,3) for surface H atoms. The core H was H(3,4), with Mg(10,10) and Be(6,6). Bond centred Gaussian orbitals(2,3) were placed in every bond centre of the cluster, and symmetry was constrained to C3v. Clusters were 87 atoms as in the previous section unless specified otherwise.
In all cases the energy difference between the bond-centred site and the anti-bonded site behind the Group V element (either P or As) was normally tiny and much smaller than the expected error in our calculations. Again, in every case the Group II element dropped back to a planar position in order to form three strong bonds with its Group V neighbours, while the H atom formed a single strong bond with its Group V neighbour, which also lay co-planar with its Group III neighbours.
Vibrational mode agreement was not quite as good as that of InP:Be-H, however in all cases the bonds centred LVMs were in reasonable agreement with experiment, and were more accurate than the anti-bonding modes. Previous work on GaAs:Be-H by Briddon et al  produced better vibrational mode agreement, and this was probably due to their choice of cluster. 86-atom clusters have the advantage of being stoichiometric, but then have an inherent cluster dipole moment which will compress the core bond artificially. However atom centred clusters such as the 131-atom cluster have large charges to maintain the correct electron bond populations, and this charging will also effect the local bond lengths. Although much of Patrick's work used smaller 56 atom clusters, these have the advantage of being atom centred and also only having a small cluster charge, leading to more accurate core bonds. For comparison I have included results for both an 87 atom and 132 atom cluster of InP:Mg, and it can be seen that both the BC and AB modes converge on the experimental figure with the increase in cluster size.
In the only case where lower frequency modes are known, GaAs:Be, there is excellent isotopic agreement with the BC site, unlike the AB site. This is the only case theoretically where the Group-II metal A1 mode lies above the H wag mode; in all other cases the lower quoted frequency is that for H wag. This is reflected in the tiny H/D isotopic shift of the 587 cm-1 mode of only 1 cm-1 for GaAs:Be.
These results therefore suggest that both Be and Mg are passivated in InP and GaAs through the addition of a single H atom to the bond centred site. This bonds tightly to the Group-V atom, which moves co-planar with its three Group-III neighbours, while the acceptor atom also moves away to lie co-planar with its Group-V neighbours. As an additional check on the method, test calculations on the same defects in GaN gave good vibrational modes for Mg with H in the AB location, as well as predicting this to be more stable than the BC site, consistent with current theories .