R-free Refinements

marksenn

well someone has to be the first person to poss a question on this new forum....

I am looking to do a R-free refinement in Topas of a single crystal data set. In practice this will involve:
1) loading only a random subset of reflections into topas and performing a refinement.
2)calculating the R value of the subset of relections for the model
3) calculating the R value for all reflections base on the model calculated in 1).

I could do this by editing the input hkl files but their must be some clever way to do it in topas.

any suggestions welcome.

Thanks,

Mark

johnsoevans

Mark,

As you posted the first question I guess I should post the first answer (or non-answer)!

I don't know an automatic built in way of doing this.

If I was doing it I would work from command line mode and create a small program or script to autodelete lines from the hkl file. You could then have a .bat file to rapidly select reflections then run tc. Something like:

choosehkl <full.hkl >strip.hkl
tc file.inp "macro filename {strip.hkl} "
tc file.inp "macro filename {full.hkl} "

You'd need something like RAW(filename) in your input file (see e.g. http://topas.dur.ac.uk/topaswiki/doku.php?id=automating_multiple_refinements).

The ideas aren't very different to the "bootstrap errors" methods in topas so there may be a cleverer way of doing this.

Sorry it's not more helpful.

John

alan-coelho

Mark,
I will ensure that calculating R-Free is possible for the next version. At present the only work around I can think of is parsing the hkl file and removing a percentage of reflections as John has mentioned. It is also similar to how bootstrap errors are calculated again as John mentioned.

Does the following sound correct:

1 temporarily remove a certain percentage of yobs data
2 perform refinement to convergence
3 keep refined parameter values
4 include temporialy removed yobs data
5 recalculate R-values

Steps 4 and 5 would be post refinement and can be tacked on without difficulty

Do you know which R-values is requited; ie. is it R-Bragg, Rwp etc... or is it simply all of them.

Also, how many times would you like to perform the procedure; an option is probably necessary, something like:

xdd_scr...
r_free_percent_exclude 10
r_free_cycles 100

For output I think the out_prm_vals_on_convergence would do the job.

marksenn

Dear John and Alan,

Thanks for your answers. I will try John's suggestion.

R-free can be calculated for either R or Rw. It would be good to be able to monitor the R-free vs R over the course of the refinement, but steps 1-5 sound good to me. I don't think an option to perform multiple cycles is necessary, but people working with small data sets might find it useful.

Thanks for your help,

Mark

brawsky

Thank you for the info ^^"
__________________________

Believe the business.gclub is a growth business in the world soon called soon gclub will rule the world.

tattoo997

Thank you.