Re: PDFgui, amorphous supercell and avoiding unphysical periodic interactions
Pavol Juhas
> I am using PDFgui with amorphous supercells, and would like to know if
> there is a way to override the automatic calculation of atom density.
> When I simulate the supercell of size (15 x 18 x 14 Angstroms) in the
> standard way in PDFgui there is obviously peaks in the PDF from this
> supercell (I would like the PDF up to r = 20A). So I have embedded the
> supercell in the center of an empty larger supercell (45 x 54 x 42 A)
> to erradicate these lattice peaks. However, this structure now results
> in a lower atom density. Is it possible to override the automatic
> calculation of the atom density, such that I can input the correct
> number?
Hi Claire,
There is no such feature in PDFgui, but after 15 minutes of tinkering
I figured out how to fool the program into using higher densities.
The trick is to add extra dummy atoms into the supercell and to
set them with huge negative thermal factors, for example
U11 = U22 = U33 = -10. The dummy atoms are included in density
evaluation, but because of their huge negative thermals, they
do not contribute to the PDF. Basically, any pair that includes
a dummy atom gives negative width for the Gaussian, that models
its contribution to the PDF, and such contributions are all ignored.
In the future we plan to use new PDF calculator, that will allow
user defined PDF baselines.
Hope this helps,
Pavol
--
Dr. Pavol Juhas
Applied Physics & Applied Mathematics
Columbia University
Room 200 Mudd Building, MC 4701
500 West 120th Street
New York, NY 10027
riza dervisoglu
From your answer, I derive that we can use this trick also for simulating a differential PDF for specific element. Is this right?
Regards,
Riza
Riza Dervisoglu
Ph.D. Student
Grey Solid State Chem.Res. Group
Department of Chemistry
SUNY at Stony Brook
Stony Brook, NY 11794-3400
Simon
>
> The trick is to add extra dummy atoms into the supercell and to
> set them with huge negative thermal factors, for example
> U11 = U22 = U33 = -10. The dummy atoms are included in density
> evaluation, but because of their huge negative thermals, they
> do not contribute to the PDF. Basically, any pair that includes
> a dummy atom gives negative width for the Gaussian, that models
> its contribution to the PDF, and such contributions are all ignored.
>
would put in a weakly scattering atom (e.g., H for x-rays) and give it
a large positive thermal factor (e.g., 10x or 50x too big) so the
correlations involving the H were smeared out. We would then allow
the occupancy to vary. This works because the density, rho_0, is the
number density, not the scattering length weighted density, so the H
contributes fully to rho_0, but only v. weakly to the calculated
pattern. Depending on how big you make your cell you may get
occupancies of hundreds on that atom. By only adding one atom it
doesn't slow the computation too much.
I didn't know that giving a negative thermal factor caused PDFgui to
ignore the atom in the PDF calculation but not in the rho_0
calculation. That is a good trick then, though could cause some
unexpected behavior if the Uiso goes negative during a refinement????
All of these are ugly work-arounds. Having the next generation PDF
calculators is something we are looking forward to. Maybe having the
option of a manually set rho_0 in PDFgui is something we can think of
for a future release.
A final word though: The PDF baseline is, we now understand, a rather
rich and interesting beast. You can read more about it in a
forthcoming paper in Acta Crystallographica A (Farrow and Billinge, to
be published). The current version is on ArXiv in the cond-mat
section.
S
--
Prof. Simon Billinge
Columbia University
Room 200 Mudd, MC 4701
New York, NY 10027
Tel: (212)-854-2918 (o) 851-7428 (lab)
Condensed Matter and Materials Science
Brookhaven National Laboratory
P.O. Box 5000
Upton, NY 11973-5000
(631)-344-5387
email: sb2896 at columbia dot edu
home: http://nirt.pa.msu.edu/
Pavol Juhas
> Hi Pavol,
>
> From your answer, I derive that we can use this trick also for simulating a
> differential PDF for specific element. Is this right?
Hi Riza,
Yes, you can use this for calculating partial PDFs, but these should
be rather set by changing "Included Pairs" filter. For example,
to exclude contributions from atom 7 in a 12 atom structure use
Included Pairs 1:6-1:6, 8:12-8:12
or equivalently
Included Pairs all-all, !7-!7
This partial will have lower density, accounting for missing
contributions from atom 7. If you add partial PDF from
atom 7 (Included Pairs 7-all), they should add to the total
PDF including correct baseline.
Another way is to use the ugly Uiso trick and set Uiso(7) = -10.
In this case the peak amplitudes are the same as for partial PDF
set by Included Pairs, but the baseline would be lower accounting
for number density of the complete structure.
Pavol Juhas
...
> I didn't know that giving a negative thermal factor caused PDFgui to
> ignore the atom in the PDF calculation but not in the rho_0
> calculation. That is a good trick then, though could cause some
> unexpected behavior if the Uiso goes negative during a refinement????
I just found this yesterday. Basically, I recalled that when Uij
values are left at zero, the calculated PDF has no peaks and shows
only the baseline. There should not be problems with Uiso going
negative during refinements, because small thermals cause narrow tall
peaks with presumably high contributions to the Rw cost. Of course,
for dummy atoms the negative thermals should not be refined at all.
Pavol
--
Dr. Pavol Juhas
Applied Physics & Applied Mathematics
Columbia University
Simon Billinge
>> I didn't know that giving a negative thermal factor caused PDFgui to
>> ignore the atom in the PDF calculation but not in the rho_0
>> calculation. That is a good trick then, though could cause some
>> unexpected behavior if the Uiso goes negative during a refinement????
>
> I just found this yesterday. Basically, I recalled that when Uij
> values are left at zero, the calculated PDF has no peaks and shows
> only the baseline. There should not be problems with Uiso going
> negative during refinements, because small thermals cause narrow tall
> peaks with presumably high contributions to the Rw cost. Of course,
> for dummy atoms the negative thermals should not be refined at all.
>
PDFFIT so I wouldn't completely rule it out.
S
--
Prof. Simon Billinge
Applied Physics & Applied Mathematics
Columbia University