The problem of later thermal donors is a very different one. Here, experiment suggests a common core that undergoes small perturbation with increasing oxygen content. The standard model is one of oxygen aggregation along the 110 plane, resulting in a defect with a fixed core and two tails of Oi stretching out along 110 .
There have been a total of 16 different thermal donors observed. In this discussion I will assume TD3 is the di-y-lid and hence has four oxygen atoms in it. Additionally I will assume that this lies at the core of all following thermal donors; these assumptions do not drastically alter any of the models should they later be revised. Analysis of oxygen loss from solution by Newman et al  suggests an average loss of ten oxygen atoms from solution per thermal donor, at 500C.
There are a number of possible formation models, but these can be roughly divided depending on whether they assume a serial development of thermal donors or allow isomers. The argument also varies depending on whether the TDs grow through Oi addition or dimer addition, which is currently not clear . However standard Oi migration alone cannot account for the quantity of higher order TDs formed , and TD formation from TD2 to TD6 has been shown to have a formation barrier of 1.7 eV , suggesting that dimers are the dominant active species rather than Oi.