Constrained Simultaneous Fitting: Tutorial gives wrong answer

[butlerpd]

I note that the final result in 5.x version is given as ~29 +/- 0.3 Å (with 12% Log-Normal polydispersity) which is quite a bit larger to what the 4.x tutorial gets following the same protocol ( ~21 +/- 0.3 Å -with 15% Log-Normal polydispersity) though still a reasonable number. However the error bar is given as NaN which seems like a red flag?

On the other hand, 5.x gives the thickness as ~1.5 +/- 0.8 Å (with 36% Gaussian polydispersity). Which has many problems, not least of which is that it is outside the resolution of SANS really, is unrealistically thin for a surfactant layer and is an order of magnitude smaller than for the 4x tutorial (~10.7 +/- 0.8 Å thick - with 32% Gaussian polydispersity). Moreover, a 30% PDI on the thickness is unreasonably high for a surfactant layer. Those are typically much more monodisperse than the micelle size for obvious reasons. Since the corresponding screen shot is not given in 5.x hard to tell if the 1.5 A is a typo in the tutorial or reflecting a deeper problem. I do note however that the sum or radius and thickness is nearly identical at 30.5 vs 31.7A. This may suggest that in fact it is neither a problem with the fit engine or a typo but that the fitting is over determined in some way?

A strong hint is that even before adding the polydispersity the DREAM correlation plots are showing that scale and radius are perfectly correlated (and that scale 1 and 3 are correlated) This could explain the NaN on the uncertainty and the fact that the result may be random due to not appropriately constraining the parameters? Further the residuals look quite bad it seems .... so maybe it is the model that is incorrect?

Whatever the case we need to figure it out as a tutorial telling people that a 1A surfactant layer is a good answer is a bit ... problematic?

[smk78]

Pursuant to the discussions in SasView/sasview#1533 (comment), also check that constraints in these tutorials adhere to slave = master.