>Mathcad is unable to converge to a solution<<br> ____________________________
1. Start by the rules of the fitters.
2. Read less papers/books not oriented.
3. Meditate the example below Marlett, it makes clear that experiments or collected data need a sufficient number of valid points. Some weight function could be added but not shown.
4. For a greater audience, therefore greater collaboration: "Save as 11" or lower versions. Also, provide real data as raw as possible and for the interest of the general readers: better source the data.
I read more pages of the book, void of "working matters". If the same subject would have been written by a good Mathcader, it would resume in few pages and the corresponding working tools, that now you are trying to get by visiting this forum. This work sheet has no red. Again, there are too many "Dose response", that some very particular trick like MLE don't generally apply. Some function just can't be fitted by any method, only by hand. LM [Lvenberg-Marquardt] succeeds more than it fails but might fail miserably. If you jump ship/sheep, hard to follow for yourself and the collabs [Ref: probit]. A final fit requires some decision making that no maths can automate, how can you declare the truth out of the lies ! What I mean here is that no matter and how much maths some data are statistically more correct than other ones, therefore in the last comparison of your short data set, if the two first point are + true than the other ones (and that makes sense), then your model is for the birds ... like butter: you can't have it both ways, the butter and the $ of the butter. In other words, MLE is down the drain for the two first points considering they would be + true than the other ones progressively degrading to non sense.
jmG
