IFeffit list: OK...you're all caught up on this conversation! Hi Dan, You're approaching this the right way, in general. The answer to how you know whether some change in constraints is worth-while involves the following issues: --Does it make the fit better statistically? One way to do this is see if it decreases the reduced chi-square noticeably. --Is there a justification for the constraints you have chosen? Haphazardly changing the way you constrain parameters may occasionally result in a statistical improvement <italic>by chance</italic>. There are, after all, lots of ways you can imagine splitting up parameters. You had better have some logic behind why the parameters <italic>might</italic> split up the way that you are trying. --Are the fitted parameters physically reasonable? It is possible to give Ifeffit enough rope to hang itself, so to speak. By introducing additional guessed parameters, Ifeffit might be able to find a statistically reasonable but completely unphysical solution (a "false minimum"). Remember, Ifeffit's job is not to find the <italic>correct</italic> structure, it is to find the structure that gives the closest match between fit and data. It's <italic>your</italic> job to evaluate whether the resulting structure is physically reasonable or not. Examples of unreasonableness are things like bonds shorter than any known bond, etc.. In some cases Artemis will flag you on this kind of thing, but it really does (and should) remain the fitter's responsibility. Another comment--provide uncertainties on your numbers! I have no idea if the third cumulant the fit found is significantly different from zero. My impression from the reduced chi-squares, though, is that the third cumulant is not helping you, and can probably be safely set back to zero. So at this point I'd say you have pretty good evidence you have the hexagonal phase as opposed to the trigonal phase. In fact, do you still need the two separate E0's in this phase, or was that just because the fit was having trouble with the trigonal model? --Scott Calvin Sarah Lawrence College At 01:19 PM 8/17/2004 -0400, you wrote:
Dan Carter wrote:
Thats great about the R-factor, I was very impressed to see it so
low.
Reduced Chi-square = 382.556284248
R-factor = 0.015036203
(fit I started with--two e0's: one for 1st path/1 for
others)
I'm wasn't positive about how to use the third cumulant. So, I
made a
variable third to guess in that position for the first path (the nearest
neighbor, right). Here are the results:
Reduced Chi-square = 415.039996182
R-factor = 0.014768538
third = 0.000045
(Can you tell if this is a better fit from the Reduced
Chi-square)
If I separate ss's for the first path and the others (leaving
the
third
parameter) I get this:
Reduced Chi-square = 462.878645915
R-factor = 0.014748364
(not much improvement)
One more thing, for the mailing list: How can I find out if it
is
ok to
separate the e0's of the nearest neighbor from the others?