TOPAS v6 Bug or weighting factor issue.

suneetkale

Hi,

I am trying to refine synchrotron data (.xye files). When I use weighting = 1/Sqrt(Yobs) for the weighting, the program stops responding and crashes after a while.

If I comment the line out, it runs but doesn't correctly fit the prominent phase peaks, I know of two phases in the sample. The peak intensities and position do not match, even after refining the lattice parameters.

If I run the .inp with the prominent phase alone, the positions match after refining lattice parameters. However, the intensities are still an issue.

I have three types of .xye files from Synchrotron - normal 2th.xye, 2th corrected.xye, and 2th.subtracted_bkg.corr.xye
The problem persists no matter which file I use.

Is there a way to mitigate this? I have done 30-odd samples from the same Synchrotron, and they work fine.

johnsoevans

If you have synchrotron xye data then you ought to use the experimental uncertainty for the weighting (presuming it is correct in the xye file) rather than Poisson statistics as implied your 1/Sqrt(Yobs) expression. TOPAS should do this by default with xye files.
Check that there are no strange esds in the data. One way is just to plot esd vs 2-theta. Sometimes having unrealistically small esd's can throw the refinement. Similar things could happen if you use your Poisson equation and have a rogue close-to-zero Yobs. The default weighting schemes in TOPAS try and catch these types of problems.
You could also try getting the refinement started with unit weights (weighting = 1😉. That would tell you if the problem is with the weighting or other aspects of the model.

suneetkale

Thank you @johnsoevans

I think the problem was the weighting factor. Per your suggestion, I plotted the esd vs. 2theta, and those values look reasonable. I then used weighting = 1 to check if there were issues with the model, and it worked fine.
W/O weighting, however, TOPAS doesn't compute the lattice parameters and stops mid-refinement, as seen in the enclosed picture. The background isn't fully refined.

Also, with weighting = 1, the GOF is now 188; I think that's because my r_exp is 0.0319. Is there a way to calculate all the fit values for synchrotron data correctly?

johnsoevans

The weighting = 1 was just to get you going and identify where the problem is. If you want to do things properly (and get reasonable GoF), you'll need to figure out why the proper weighting is affecting your refinements.
The TOPAS standard for tof xye data is:
weighting = If(SigmaYobs < 1, 1, 1/SigmaYobs²);
You'll see that for low SigmaYobs unit weights are applied, Have you tried that type of expression?

suneetkale

johnsoevans

I had tried weighting = 1/( Sin (X Deg / 2) Max (Yobs, 1); and weighting = ( Abs (Yobs-Ycalc) / Abs (Yobs+Ycalc) + 1 ) / Sin ( X Deg / 2);

For weighting = 1/( Sin (X Deg / 2) Max (Yobs, 1); TOPAS shows "Too many ( in 1/( Sin (X 0.01745329251994 / 2) Max (Yobs, 1); in: = 1/( Sin (X 0.01745329251994 / 2) Max (Yobs, 1);"

For weighting = ( Abs (Yobs-Ycalc) / Abs (Yobs+Ycalc) + 1 ) / Sin ( X Deg / 2) seems to work the best. However, the GOF is around 900.

As you suggested, I also tried the default TOF weighting = If(SigmaYobs < 1, 1, 1/SigmaYobs²), but the calculated intensities are almost half off.

I don't understand why it does this. Is using weighting = 1/Sqrt(Yobs); for Synchrotron data incorrect?

rowlesmr

when experimenting with weightings, pay no attention to the GOF, as it's value is necessarily affected by the weighting scheme.

The fact that things are so far off that you need to play with weighting schemes is a little concerning.

Are you able to post an example of normal_2th.xye that doesn't behave nicely, as well as one that does (maybe from a previous experiment)?

suneetkale

Thanks for the advice. I think the Intensity and Angle Dependent Weighting i.e. weighting = 1 / ( Sin (X Deg / 2) * Max (Yobs, 1) ) has worked. There are still issues with intensity matching in peaks; lower angle peaks are underestimated, and higher angle peaks are overestimated.

Sure. Here is the data file for the new data

integrated2th.txt

732kB

This is is the .inp file I used for it with Intensity and Angle Dependent Weighting.

lle13-10cu-intensity-and-angle-dep-weigh.txt

10kB

This data file is from the same instrument, and I used the background subtracted version to run the refinements. And they work very well.

motr-aadl-08-0003h52th.txt

732kB

along with its .inp.

aadl-08-003-sk378-10co.txt

8kB

They are slightly different materials.

deppaa

In integrated2th.txt you have two Yobs values that are 0, and this explains why weighting = 1/Sqrt(Yobs) is not working (a dividing by 0 error) while weighting = 1 / ( Sin (X Deg / 2) * Max (Yobs, 1) ) is working. However, commenting out the user defined weighting-scheme (and allowing TOPAS to just deal with it), your inp.file runs and gives a reasonable looking fit on my computer without any other changes.
You should of course consider to limit the upper 2theta range for your refinement to for example 13.2 to not include the Yobs=0 points. This would also remove some data points where SigmaYobs=0. And maybe double check the reliabilities of the SigmaYobs values in general.
Lastly, you might consider removing the spherical harmonics and size broadening for the Nb2Sn0.9Cu0.1C phase, and use strain broadening in stead, it seems to describe the observed peak shape much better (assuming this makes sense with what you know about the sample of course).

suneetkale

deppaa

Thank you for getting back to me. I tried your suggestion and made three observations:

When I use the weighting formula "If(SigmaYobs < 1, 1, 1/SigmaYobs²)," I get a decent fit, and my fit stats look good. However, the difference curve at a lower angle (main peak) is higher.
When I use the weighting formula "1 / ( Sin (X Deg / 2) * Max (Yobs, 1) )," I get a better difference curve, and the fit stats and fit look about the same.
The default weighting used by TOPAS treats the data as TOF and uses the formula "If(SigmaYobs < 1, 1, 1/SigmaYobs²)" when no weighting is specified.

Also, I have removed the spherical harmonics. It doesn't make sense with the data I have.