Use of bkg_file() macro for Complex Background in Pawley vs Rietveld

jmas5

Hi All,

I'm currently trying to fit a dataset, where I've gotten good pawley fits, but am struggling to make the Rietveld work. I think this is an issue with my .inp, as opposed to a scientific one, as the structure is known and I have fit it just fine before. I have gotten good Rietveld refinements on ex-situ data of the same material.

In this case, I am working on operando data, with a significant background (it is the bulk of the pattern). I have a background scan for this (cell with everything except material of interest), and using bkg_file() in the case of the pawley fitting, I am able to a good fit, with the background scaling appropriately.

When I move to a Rietveld refinement, it is like I am unable to refine the scale of the bkg_file and the scale of the calculated pattern independently. I understand why this would not be an issue for the Pawley fit, since the peak intensities are arbitrary and no phase scale is refined. Using Rietveld, I am still able to get a good fit of the background to the data, but in doing so the scale of the phase itself gets very small. If I manually set the scale higher, you can see that the phase's peaks are in the correct place, but the intensity of the background scan becomes enormous (increasing exponentially with 2theta). I'm wondering if there is perhaps something linked in my input file that I don't realise, or can't find, as I've not had any issues using this macro previously.

I've included some screenshots below for reference:

Pawley fit:

Letting the phase scale and background refine, minimises phase:

Setting phase scale to fixed value, background still refining (start vs end):

I've attached the .inp file in question in case this is useful. For the purpose of trouble shooting/trying to set this up, I am only using bank 2 in this case.

shareable-op3-riet-test.txt

124kB

Any advice or suggestions greatly appreciated!

jmas5

Looking at it, I am actually getting a good fit to the phase at higher Q, so I'm wondering if potentially there's some issue with scaling calculated intensity as a function of Q? This is TOF data, I would not be surprised if it was something like this (but unfortunately I've not been able to spot it myself). Thanks again!

jmas5

Further thoughts from a colleague, neither of us are too sure how we would investigate this in the .inp so any advice appreciated:

Having a quick think about it, I wonder if it's got to do with the way in which both scales (background are phase) are being related mathematically in the input file. It's almost like they are not being treated separately as when they are both refined, the scale of the background seems to be more than enough to account for the overall scale not needing the scale to be even present. The opposite occurs when forcing the scale to stay fixed; the scale of the background can't go down potentially due to some mathematical constraint imposed by the phase's scale.

johnsoevans

I've only been able to look at your file quickly as I'm travelling, but check if you have min/max limits on your background scaling - I think in one place you may have hit a max limit of 1. Your file also seemed to contain a high order background polynomial along with the measured background. If you have a good measured background you should only need a small number of extra background terms just to "tweak" things slightly.
In general I'd suggest:

Do the Pawley fit including intensities and your background functions.
Set a scale of 0.0 on the Pawley phase, do a zero-iters calculation and make sure the overall background looks sensible.
Then fix the background and try the Rietveld fit.
Check that in your Rietveld you have the d⁴ scaling for tof data. Check that you haven't got unreasonable values for atomic displacement parameters or quantities like mu which will affect data as a function of q.
With a multibank INP file like this it's very easy to make mistakes. You could try setting up a simple one bank INP file while you are trouble shooting.
I saw you had a fixed bkg polynomial in the file. Just note that if you change the X fitting range these background coefficients won't be correct - they're likely to give rise to a very different calculated background.

jmas5

Hi John,
Thanks for your reply, it’s great to get your thoughts here.
I originally had no min/max limits on the background scalar, I’m not sure what the defaults are but it can definitely scale to >1. This is how a pawley looks with no extra background polynomial, and a defined min 0 max 10 (refined to 1.256). Note the number of frames/counts is not equal between the background scan and the sample scan. For this work we took x times ₂ min long scans on Polaris, summing approx. 1.5h worth of scans in Mantid. The number of scans for the background was clearly not as many as for the sample here, hence the >1 value:

With regards to the many term polynomial background (in addition to the background scan), it is possible to fit the bulk of the pattern without any polynomial at all. However, as the battery is charged, additional amorphous components are produced which contribute to the large hump in the background from around 1-2 angstroms. It takes a good few terms to get a decent fit to this (at least 10), but I wonder if there is possibly some other background function that might do a better job of fitting this with less terms?
No additional polynomial:

5 terms:

10 terms:

15 terms:

19 terms (as per original file, here background file scalar refines to 1.025):

Following your suggestions, the below is a pawley with scale 0.0, 0 iterations run:

I then fix the background polynomial and scalar to their current values, and attempt the Riet (iters no longer at 0), the phase does not manifest at low Q, but does fit everything beyond 3.71 angstroms well (phase scale refined to 0.24481358`_0.006299):

Manually setting scale to 200 gives a good fit to the layered peak at low Q, but again exponential rise in the calculated intensity with increasing Q:

I thought I would also try turning off the additional background polynomial, but that results in the below:

With regards to the next suggestions, ADPs are not relevant here, as I’m not actually refining the structure. I am using the known structure from combined Rietveld refinement of SXRD/NPD (In this case, I am doing everything on the well known and reported C2/m NaNiO2 structure). Keeping scale locked at 200 and allowing mu to refine (from originally 3.95208067_0.0472607982 to now 26.3590697`_0.372391622) results in not fitting any structural peaks, but the background becomes sensible again:

Allowing both scale and mu to refine yields a much more sensible output (mu = 0.243075473_0.0193732411 , scale = 0.0243109662_0.0006219):

Letting the polynomial refine again gives a nicer fit to the complex background at higher Q (which I assume was previously arbitrarily being fit by intensity on the allowed reflections in the pawley):

It seems that the whole issue was due to the lack of mu refining. I’ll now move on to setting this up with all the banks and across the rest of my datasets. Thanks very much for your help here, it is greatly appreciated!
All the best,
James

johnsoevans

I'm glad things have worked out!
You asked about alternative ways to model the amorphous content. You could try just inserting one or two broad peaks at that point. A position, width, intensity and (perhaps) asymmetry of some sort might do as good a job as multiple background terms. There's one method for inserting tof peaks on the wiki.