Atomic positions not refining

[giuditta.perversi]

Hi there,
I am trying to refine the pdf of structure with monoclinic symmetry, with a very large unit cell (12, 12,17) and 224 atomic positions. Constraining the atomic positions brings them down to 168 by symmetry.
The issue is that, whereas the initial steps of the refinements work correctly (scale, shape parameters, unit cell and the like) work correctly, as soon as I try to refine the atomic position or thermal parameters of even just one quadruplet of atoms (i.e. three independent parameters) nothing happens. It seems like PDF-gui is not even attempting a shift to refine them and the refined value is always equal to the initial one.
Is this a bug or an intrinsic problem with a large cell of low symmetry?

Thank you in advance for any reply/suggestion.

Best,
Giuditta

[Pavol Juhas]

Hi Giuditta,

PDFgui uses downhill optimization, which may get stuck at
local minima. This can also happen if experimental and simulated
PDFs are too different as small changes in model parameters may
then have no impact on the overall fit quality. I am however
bit surprised that nothing happens even for thermal parameters.
These control the simulated peak widths and thus should have strong
effect on the fit quality. Can you try to use the same isotropic
thermal factor for all atoms and see if it would refine? If that
works, you can gradually make the fit more detailed by using
separate displacement (thermal) parameters for equal-type sites, then
split it further by separating parameters at non-equivalent symmetry
sites and/or allowing anisotropic displacements. These steps should
produce a clear improvement in the fit residuum Rw, otherwise one
cannot say that the more detailed values truly originate from
the data.

Also, please check for other potential pitfalls such as
duplicate atom positions (may happen after symmetry expansion
of fraction-like coordinates, e.g., 1/3), negative
displacement parameters or a very large value of the
delta2 correction factor. Each of these can cause
funny behavior in PDFgui fits.

Hope this helps,

Pavol

--
Dr. Pavol Juhas
Computational Science Initiative
Brookhaven National Laboratory
P.O. Box 5000
Upton, NY 11973-5000

[giuditta.perversi]

Hi Pavol,

thank you very much for your reply.
The trick with the sequential unlocking of the thermal parameters actually worked, just by using them with a fix parameters for every atom type instead of doing the symmetry expansion, and they are now refined together with the delta1 (this is a 400K dataset so delta2 is not refined), cell parameters, scale and Qbroad/damp. Unfortunately, the atomic position still don't cooperate in refining, at all...you can actually see that they stay fixed to the initial value (4 decimal places) even though PDF-gui usually gives outputs with at least 6 decimal places, so it looks like it's not even touching them.
I think the symmetry constraint of the quadruplet of atoms belonging to the same position is correct, I do recognize the various -x+1/2 and the like that you would expect from a Cc monoclinic symmetry, and I can't say it's because of a clear disagreement of model and data because the fit is actually really good. It's just that I need to be able to model the changes in atomic positions if I want any significant data out of this modelling, so I can't call it a day...

Cheers,
Giuditta

[Simon Billinge]

sometimes manually displacing atoms slightly from high symmetry positions, by 0.02 or 0.05 or something, can start them moving.  This will be hard if there are many of them, and if there are symmetry related ones you will have to displace them slightly in the symmetry appropriate way will be a headache, but you could test if it works on a subset of atoms.

S

--
You received this message because you are subscribed to the Google Groups "diffpy-dev" group.
To unsubscribe from this group and stop receiving emails from it, send an email to diffpy-dev+...@googlegroups.com.
To post to this group, send email to diffp...@googlegroups.com.
Visit this group at https://groups.google.com/group/diffpy-dev.
For more options, visit https://groups.google.com/d/optout.

[Giuditta Perversi]

Hi Simon,

thank you for your suggestions. I tried to displace a quadruplet of positions from their cif-provided value in a symmetry coherent fashion and then run the refinement, but it still doesn't do anything at all and it doesn't look like it's attempting a refinement.

I will try and do it with more positions but in theory every quadruplet is independent so I'm not sure it will actually fix it.

Thanks again for helping me with this.

Best,

Giuditta

2017-08-11 10:45 GMT+01:00 Simon Billinge <sb2...@columbia.edu>:

sometimes manually displacing atoms slightly from high symmetry positions, by 0.02 or 0.05 or something, can start them moving.  This will be hard if there are many of them, and if there are symmetry related ones you will have to displace them slightly in the symmetry appropriate way will be a headache, but you could test if it works on a subset of atoms.

S

To unsubscribe from this group and stop receiving emails from it, send an email to diffpy-dev+unsubscribe@googlegroups.com.

To post to this group, send email to diffp...@googlegroups.com.
Visit this group at https://groups.google.com/group/diffpy-dev.
For more options, visit https://groups.google.com/d/optout.

--

You received this message because you are subscribed to the Google Groups "diffpy-dev" group.

To unsubscribe from this group and stop receiving emails from it, send an email to diffpy-dev+unsubscribe@googlegroups.com.