question in determining the TOF calibration constants and how to judge refinement result

kunyang

Hello~~~
I have some question in determining the TOF calibration constants. There are instrument parameter files for north and south bank. We can see three parameters following "ICONS": ICONS 18355.41 -4.12 -14.55. These three parameters should be corresponding to TOF calibration constants.

At the beginning of refinement, we use fitting of standard CeO2 powder to obtain diffractometer constants(t0, t1, t2) and exponential parameters.
The question is:
1. Should I directly use the above three parameters as t0, t1 and t2, like: t0=-14.55, t1=18355.41, t2=-4.12, no refine, always fix them during refinement? Or just use them as original calibration constants and then refine them?
In topas tutorials "TOF calibration / refinement of CeO2", it refined them. http://www.dur.ac.uk/john.evans/ta_tutorials/html/tof/tof.html
But there is another saying in this paper. http://www.isis.stfc.ac.uk/instruments/polaris/data-analysis/refinement-of-profile-parameters-with-polaris-data11478.pdf

2. For exponential parameters, two macro TOF_Exponential(a0, a0v, a1, a1v, wexp, t1, lr) , TOF_Exponential(b0, b0v, b1, b1v, wexp, t1, lr) will be used to determine rise and decay coefficients. How to set original parameters inside macro?
For example,
if we directly choose it from Jedit TreeView, it will be:
TOF_Exponential(@, 100, @, 50, 1, t1, -),
TOF_Exponential(@, 100, @, 50, 4, t1, +).
If according to posted example "TOF_bank2_1.inp" in Topas, it is:
TOF_Exponential(a0, 100, ,0, 1, t1, -), (a1 here is fixed to 0,why?)
TOF_Exponential(b0, 100, b1, 100, 4, t1, +)

We can see that they have been set in different original value. Different original value will make different refinement value, which will effect all the following refinements. How to deal with this in a usual case?

3. I have compared different way (like change original parameters in macros, keep some parameters fixed or refined), after run TA, I found it's difficult to figure out which one is best if according to their profile patterns (they all fitted well). The difference is parameters results in INP files. And there are red LIMIT warnings inside it sometimes.

Does anyone can tell me the usual way to judge a refinement is good or not, especially according to INP results?
By the way, I'm using ISIS TOF data for residual stresses analysis.

Thank you very much!

johnsoevans

I've put some thoughts below. However some of these things you might want to discuss with your instrument scientist.

1. What to fix/refine is your judgement call as in most areas of Rietveld refinement. It would be normal to use a standard like CeO2 to refine instrument constants then fix them for your sample. difc in particular determines the tof to d-spacing conversion so will be 100% correlated with cell parameters. You might find in practice that you can get a marginally better fit to the data if you allow t0/difa to refine when you fit your sample. If you've multiple banks of data you might decide to fix difc for highest resolution bank of your sample and allow difc to refine for other banks. Here you're effectively using your sample as a standard for calibrating all but the highest resolution bank. Depending on how important cell parameter precision is, you might decide to use X-rays and an internal standard to define an accurate cell parameter for your sample. You could then choose to fix the cell and refine difa/difc and zero in your neutron fits. Your instrument scientist ought to be able to advise you on a sensible protocol for the machine you've used.

2. If your refinement is converged and parameters aren't highly correlated then it shouldn't matter what values you start the refinement at. For highly correlated parameters you may find that a range of different values will fit the peak shape of your standard equally well. As you're using this as an instrumental peak shape description, which will be fixed when you refine your actual data, this doesn't matter.

3. Rules for deciding on a good refinement are the same as with any Rietveld. Low agreement factors are important, but the most important is that the data fit well visually and that all the parameters in your model are (a) appropriate and (b) well determined. If you've parameters hitting limits it suggest that this is not the case.

John

kunyang

@John
Thank you so much for the helpful answer and reply soon!
These points are very important for me to make sure if I'm in the right way and to correct it in time. I'm much clearer now. Thank you again. ;)