There are also other possibilities beyond those considered above. One of these is that the dimer can structurally transform into TD1/2. This is unlikely since it would require four distinct structures for an oxygen pair; the dimer, TD1 and TD2, and the electrically inactive form of TD1/2. In addition, such a model also suffers from the lack of direct correlation between [TD1/2] and the 1012 cm-1 dimer absorption. We have not been able to determine any electrically active double donor structures for the dimer (although structures such as a di-y-lid core could theoretically be produced by only two Oi atoms, the resultant lattice strain would be huge).
It has also been suggested that trimers do not form [143], but the models discussed here all assume they are at least marginally stable. If trimers do not form, it is difficult to explain the 1006 cm-1 LVM. However rather than being a stable defect complex, it is possible the trimer is simple the result of a dimer pushing Oi along for a few steps before moving on separately again.
There are many other ways these arguments may break down, not least that some of the experimental evidence may be misleading; for example the isomeric nature of TD1 and TD2 could actually just be due to very similar inactive forms that vibrate at the same frequencies.
It has been stated at above that it is impossible to generate a C2v structure with an odd number of Oi atoms without putting an oxygen on the C2 axis. To further complicate matters it may be possible to have a structure such as the Snyder-Stavola model where a single oxygen atom lies off the C2 axis, but has a tiny barrier to the symmetrically mirrored structure with the oxygen on the other side of the axis. In this case the structure would in practise flip rapidly between the two structures and would obtain time averaged C2v symmetry, have an odd number of Oi, and yet might not show Oi on the actual C2 axis.
Unfortunately without the ability to accurately compare energies between these different structures there is little more that theory can do for the thermal donor problem at this point, beyond proposing a variety of models, and eliminating those that it can on the basis of the available experimental evidence. The variability in oxygen energies does not appear to be just a problem with AIMPRO but of many DFT codes; Chadi finds that addition of Oi to OV is endothermic in contradiction with experiment [95], and work by Pantelides et al on the Oi diffusion barrier has varied between 1.8 eV and 2.5 eV in recent papers[115,130].
This work proposes many models that seem to be suitable as either TD1, TD2 or TD3. The di-y-lid structure fits TD3 very well in terms of structure, electronic properties, and vibrational modes. Other structures we cannot eliminate include the Snyder-Stavola, a new 4O `flanked square' structure, and even the partially dissociated trimer structure given above. However all of these three would appear to be models for only TD1 and possibly TD2 for the Snyder-Stavola, on symmetry grounds. It is definitely the case that there are a variety of possible models with shallow double donor character that only require trivalent Oi, BC Oi and lattice Si, and do not need to invoke Si self-interstitials.