Local vs Average structure deviations

[Myrsini K.]

Dear diffpy-users,

I am doing a 3-phase PDF-refinement of a synchrotron’s data of intercalated FeSe superconductor @300K named LiPyFeSe and I have some questions (sorry for the extensive text that follows):
1) The dominant 1st phase (I4/mmm) I've used for the PDF-refinements had exported from the Rietveld refinement from GSAS-II, BUT when I do the PDF-refinement up to 20 Angstrom (=average structure) I get higher Rw values than when I do it up to 10 Angstrom  (=local), see attached .png file including the different curves. Shouldn't the average pattern at PDF gave me well fits as those I get from Rietveld for the average structure? (1-10Ang:Rw=13.1%, 1-20Ang:Rw=18.1%)???
2) In these refinements I get deviations at 3rd or even 2nd decimal at Lattice Parameters (L.P) compared to Rietveld fits. Should I have to fix L.P? and refine them again at the end? I saw at previous post that it is normal to have such deviations but what is the limit of these deviations, between Rietveld & PDF?
3) I also PDF-refine Uiso values for Fe,Se of the majority phase constrained with the symmentry, but U33 of Fe gave me 0.08, isn't this value high enough? and what does it mean for my data (better fit with this value)?  
4) Last but not least, Uiso for the 3rd phase (alpha-Fe) gave me big value(0.04) and a better fit. Should I trust the value of a minority phase of my data? and why it improves my fit? Also,  L.P of alpha-Fe increases quite a lot (+0.03). Generally, the 3rd phase is ~5% should I fix L.P & Uiso values I get from Rietveld of this minority phase or should I trust refined values from the PDF fits? 

Thank you in advance.
Best regards,
KAITATZI Myrsini
----------------------------------------------------------
PhD Candidate,
Institute of Electronic Structure and Laser (IESL),

Foundation for Research and Technology - Hellas (FORTH),

Quantum Materials & Magnetism Lab

----------------------------------------------------------

[Mikkel Juelsholt]

Hi Kaitatzi Myrsini

Good questions! 

1. Yes and no. If you have major differences between the local structure and the average, then yes, we would expect lower Rw when refining the local structure only. However few things to keep in mind! You are refining the structural parameters in both cases, so in principle, the fit could modify the structure obtained from the Rietveld refinement to fit the local structure well. In that case the Rw could be lower for the 10 Å fit. If you have little or no differences between the average and local structure, then it is completely normal to get higher Rw when you fit fewer points. In general, PDF fits are always worse than Rietveld fits because the Fourier transform put the noise on all points in the PDF.

For your other questions, the most important thing to remember about L.P. is that any L.P. obtained from both PDF and Rietveld refinement are never correct! The only case where you can get absolute values for L.P. is if you have an internal standard (such as LaB6 or Si) mixed into the sample. The L.P. values are correlated with a bunch of experimental factors, such as sample-to-detector distance. In Rietveld, we refine a separate parameter, zero error, for example, but in PDFgui we do not have such parameters. So basically, always refine L.P. in PDFgui. The deviations are because of these experimental factors. So doing this kind of direct comparison between PDF and Rietveld is almost meaningless. It is good to do it, but deviations are difficult to use for a whole lot. (This also goes for comparison between different Rietveld refinements of diffraction patterns. Unless you have an internal standard, then deviations could easily just be differences from the measurements themselves)

The limits for deviations are not easy to say. Mostly because in the PDF the atomic positions and L.P. will correlate, as opposed to Q-space where they are not correlated. The most important thing to do is to ensure that the fitted model is physically reasonable. 

The same goes for U factors, unless you have neutron diffraction data to high Q, then do not trust the values of U factors in both PDF and Rietveld refinements. They correlate with a bunch of different things and are usually fixing a bunch of stuff not described anywhere in your model. Neutron scattering is usually the only data that allows for the refinement of anisotropic u factors.

U factors are temperature sensitive. I do not know what temperature you are measuring at. A u factor of 0.08 at room temperature is very high for a Fe atom. I would, first of all, see how much worse the fit becomes when you just use Uiso. If it becomes much worse, it must be because the model wants to broaden those PDF peaks. I would then check if it could be from stacking faults, a split atom site or something similar that causes the electron density to be smeared out. 

Regarding the minority phase, I think there is only one way to find out. If you fix the Uiso to something reasonable, does the fit gets much worse? If yes, then we need to look at why it does that.

Also remember that you have a instrumental contribution to the PDF peak width, so make sure to get a standard and refine Qdamo and Qbroad.

Cheers Mikkel