[Ifeffit] Questions for Artemis fit

huyanyun at physics.utoronto.ca huyanyun at physics.utoronto.ca
Tue Aug 19 13:28:53 CDT 2014


Hi Scott,

Thank you for giving me detailed examples. I will go with your  
suggestion and try to reduce free parameters.

Best,
Yanyun
Quoting Scott Calvin <scalvin at sarahlawrence.edu>:

> Hi Yanyun,
>
> On Aug 19, 2014, at 12:05 PM, huyanyun at physics.utoronto.ca wrote:
>>
>> But my question is, for each path there are a set of Path Parameters,
>> does that mean we have to fit this set of path parameters independent
>> from those used in other paths?  For instance, if I am going to
>> include 30 paths in my fitting, do I need to guess about 150
>> (30*5=150) parameters (except guess the same 'enot' for all paths)?
>
> No. One of the great principles of Ifeffit (and thus Artemis) is  
> that the path parameters do not have to be the same as the fit  
> parameters.
>
> As a simple example, it's good to start with the model that all  
> paths have the same value for E0. So you can guess a single  
> parameter and then use that parameter for the E0 for all paths.
>
> As another simple example, a cubic crystal might be modelled as  
> having a uniform thermal (or Vegard's law) expansion. Then there  
> could be a single guessed parameter indicating the fractional  
> expansion (called, for example, alpha) while the delr for each path  
> could then be entered as alpha*reff.
>
> Much more complicated constraint schemes are possible, but for fits  
> with large number of paths, the number of guessed parameters is  
> almost always much much lower than the number of path parameters.
>
>>
>> As shown on the fit Log file, the Correlation value between two fit
>> parameters is the bigger the better, or the smaller the better?
>
> Not necessarily. Correlations are provided to help understand the  
> relationship between guessed parameters, but there's nothing  
> inherently wrong with a high correlation...the uncertainty  
> associated with that relationship is already represented in the  
> uncertainties reported with the fit. In other words, it is more  
> direct to focus on getting the uncertainties to be lower, rather  
> than the correlations.
>
> --Scott Calvin
> Sarah Lawrence College
>
>
> _______________________________________________
> Ifeffit mailing list
> Ifeffit at millenia.cars.aps.anl.gov
> http://millenia.cars.aps.anl.gov/mailman/listinfo/ifeffit






More information about the Ifeffit mailing list