On Tue, Mar 10, 2009 at 6:23 PM, Sébastien Morin <sebastien.morin.1@xxxxxxxxx> wrote:
Hi Ed, PY is next to me and we discussed this issue together. This seems very strange since values of the AICc criterion (extracted during model selection) are not equal for models with or without these 10e-19 Rex... If this is in s-1, then the criterion should be almost equal as a Rex of 10e-19 s-1 would be negligible.
For this value, see below. If Rex really is zero between 2 nested models where the Rex parameter is the only difference, the AIC values should be different by 2! The reason is because as: AIC = chi^2 + 2k, if Rex = 0, then the other parameters must be the same (unless of a model or optimisation failure) and the chi-squared values equal. Hence as the parameter number difference k_diff = 1, then AIC_diff = 2.
Moreover, if using the latex_mf_table.py script, the Rex values are reported and associated to the 800 MHz field strength with values in s-1 between 1 and 6... These values correspond to those we would obtain by multiplying the Rex values in the xml results file by (2 pi frq)**2... Hence, I would think values extracted in the xml file, that is values one gets with the command "value.display('Rex')", are in the field independent format obtained from Rex/omega**2. Are we wrong ?
I just checked the code in the math_fns package, and the chemical exchange is handled internally as Rex/omega**2. Looking at the XML dumping code, this just dumps the contents of the spin container directly to file, hence the value is Rex/omega**2. Therefore the comment in the XML results file is incorrect. Would you be able to file a bug report for this? Then the fixes can be tracked. It would be better to have two reports, one for the latex_mf_table.py script and one for the XML results file. Cheers, Edward