The first of the alternatives is that the di-y-lid is TD2. The formation energy of TD1 is initially 1.2 eV and later reverts to 1.7 eV. In this case the 1.2 eV would be the trimer migration barrier, slightly lower than that of the dimer. Thus the 1.2 eV represents an initial pool of trimers present in the as-grown material that migrate to form TD1; once these are exhausted the rate determining step becomes trimer formation and the barrier for this is 1.7 eV, i.e. the dimer migration barrier.
The trimer then migrates until it reaches Oi, at which point the strain energy is sufficient to force it to restructure, giving either TD1 or TD2. The structure for TD1 is unclear, but may involve either a single y-lid or an oxygen `square' structure such as the 4O flanked square structure discussed above. TD2 is the di-y-lid. Now the di-y-lid and TD1 can restructure into their isomeric electrically inactive form, which is probably either two dimers separated by a bond centre, or a linear chain of four Oi atoms.
We then need to explain why TD1/2 gradually change into TD3, which becomes dominant. There are two possibilities here. The first is that TD1/2 are mobile (possibly only one of the two species) and encounter Oi, implying TD3 is a 5O species. However it seems impossible to design a 5O structure that has the observed C2v symmetry without putting O on the C2 axis, which is in breach of ENDOR results.
The second possibility is that TD1/2 are immobile, and so TD3 forms through dimer addition to TD1/2. We next need to explain the predominance of TD3. If TD1/2 are formed through rapidly migrating trimers encountering Oi, this should happen very quickly, since the Oi concentration is so high. However if TD3 relies on TD1/2 trapping a dimer, this requires the meeting of two much lower concentration species. The dimer is probably also not migrating as fast as the trimer. Hence we would expect TD1/2 to form much more rapidly than TD3, and there to be no difference in the formation rate of TD3 and higher order TDs. We would still expect TD3 to be more predominant than any other higher order TD species since it is the first of these (there should be a tailed distribution of these depending on the binding energy of dimers to each species). This is a flaw in this model since it appears to predict more TD1/2 than TD3.
A possible way around this would be that TD1/2 were unstable and regularly broke up into two dimers. However the kinetic analysis of Newman et al  suggests that dimers form through Oi migration. If TD1/2 broke up into dimers, under this scheme, they would encounter Oi to form trimers, which would encounter Oi to form TD1/2 once again, and so dimer formation would be a self-catalysed reaction that did not require Oi diffusion, in contradiction with their results.
An alternative is that TD1/2 are mobile, but for some reason on meeting Oi do not form a thermal donor species. This seems unlikely, since no other pure O species are observed apart from quartz precipitates after long time anneals. It seems unlikely that Oi would not bind to TD1/2 but a dimer would. It cannot be that these 5O species break up into a trimer and a dimer, for the kinetic arguments given above. Another more unusual model that appears consistent is to postulate that as well as TD1/2 trapping dimers to form TD3, there is a parallel formation stage as they migrate (possibly very slowly) to Oi, and form 5Oi species which are the first step towards quartz precipitates.
Annealing studies at 450C (see Figure 6.11b) show that TD2 forms in the first few hours, but then tails off with longer anneal time until it practically disappears after 10-15 hours. Thereafter there is only trace TD2 signal, however the TD3 signal continues to increase in line with the trimers; i.e. there is no observed correlation between the TD1/2 trace and TD3. This loss of TD1/TD2 signal with annealing time presents a problem to any scheme which requires a serial transformation from dimer trimer TD1/2 TD3, since there should always be some sort of correlation. It eliminates the model proposed here in all the forms discussed above, as well as any scheme that requires a serial progression through TD1/2 to higher order TDs.
The only possible way to salvage this model is if TD1/2 are becoming unstable for some reason (e.g. Fermi level movement, see below), and they are by-passed in the formation process thereafter, e.g. through dimer/trimer aggregation. However this seems unlikely given the low concentration of dimer and trimer centres.