Hi Jatin, Jason, All,
On Fri, Nov 16, 2012 at 12:49 PM, Jason Gaudet
Since R_eff is constant, surely all you need to do is multiply R_eff by the uncertainty in alpha? That will be the uncertainty in delta_R and R_fit.
The program is only giving you statistically-derived uncertainty for the values it fits; the user needs to determine how to propagate or otherwise deal with this uncertainty.
As Jason points out, propagating the uncertainty from the variable alpha delta_R = alpha * Reff is straightforward, as delta_R_stderr = alpha_stderr * Reff Ifeffit does not do this calculation for you. For more complicated expressions of variables, the error propagation can be more involved. For un-correlated variables the math is generally tractable. Including correlations can become complicated. I've been trying hard to not make the following comment about every question that comes up on the mailing list, but this is one case where I can't resist. The propagation of errors is much, much better with Larch than Ifeffit, and I would highly encourage anyone interested to try out larch: http://xraypy.github.com/xraylarch/ At this point, the main interface for Larch is a command-line / batch processing program. Sadly, I don't yet have a windows installer for this yet, but hope to have one available soon. Feedback, suggestions, and collaborations would be greatly appreciated and no doubt lead to better analysis. In more detail, for constraint expressions such as "delta_R = alpha * reff", Larch does propagate and report uncertainties for all Path Parameters as well as fitted variables and constraints ("Def" expression in Ifeffit). The propagated uncertainties include the correlation of variables. Currently, the propagation only works for algebraic expressions with operators (+, -, *, /, **), but not when calling functions (even built-in mathematical functions like sin() and sqrt()). That is, for variables 'alpha', 'beta', and 'gamma', errors will be correct for "alpha * constant", "alpha * beta + gamma", "alpha ** 2", and so forth, but "sqrt(beta)" will currently (bug fixed in the past hour!) and the uncertainty will be set to 0. I think I understand the issue, have been in contact with the author of the underlying error propagation library about this, and believe this will be fixable. Anyway, that's all to say that this is yet another reason to start investing in trying out using Larch instead of Ifeffit. --Matt