I need to do Rietveld refinement for synchrotron X Ray data which has thousands of profile points. Currently I am using GSAS/EXPGUI but it takes a longer time to converge.
Will sequential GSAS help to refine faster?. Very little information available in the manual and in Rietveld mailing list. Does anybody have an idea how to go ahead with this?
This is an issue with high-resolution data. You can speed things up if your starting model and instrument parameters are close to the final refined values. The sequence of turning on variables to refine is extremely important.
You really have not given enough information in your question for me to give you a useful answer.
I am not entirely sure what you mean by "sequential GSAS", so I cannot give an answer that I am sure addresses your intended question.
Currently I am doing Rietveld refinement using GSAS/EXPGUI. Problem with this is, refinement is going very slow because of huge data points. I came across SEQGSAS.I am new to SEQGSAS. Is it possible to use SEQGSAS to reduce the time consumption for refining such data. How can I speed up refining high resolution data using GSAS? Is there any other way.
Here is a link for
https://wiki-ext.aps.anl.gov/ug11bm/index.php/Sequential_GSAS
SEQGSAS is intended to refine a large number of separate sequential data sets such that the refined parameters of one set can be used as the starting parameters of the next set. This is very useful for refining data collected at a series of temperatures, or for data sets that are gradually changing over time.
I have the impression that you are trying to refine a single scan (consisting of many data points to be sure), in which case SWQGSAS will be of no value to you at all. But if you have many similar scans, then it is useful.
Yes I am refining a single scan data. I was under the impression that even if it is a single scan, SEQGSAS will split the data into several data and refine it one by one. Thank you very much for your valuable suggestion.
Hi Shwetha,
SEQGSAS will not speed up your refinement because it is not designed to split a pattern into pieces and refine them separately. It is designed to refine many datasets that are similar to each-other but with some small variation as Edward said.
I have this same problem, sometimes, with synchrotron data (>20000 points in the pattern) refined in a slow computer. There are 2 things I've done to speed up the process and both are related to Edward's first reply.
You need to speed up the process of getting the closest to final parameters so when you do the final refinement with you whole dataset you just need a few cycles before it converges. But while you find the best parameter you may not need to use all the pattern, just use a small part.
You can start refining a short part of your pattern by changing the data limits in Histogram tab. In general you can safely refine the first half of your pattern. With this you'll save a lot of time in the initial steps of your refinements where things go slow and may go wrong easily because your initial parameter are off the final.
After you have a reasonable fit of the first half of your pattern you can do a refinement of your whole patter, that will be slow but you'll be close to the final answer.
This may not work if you have patterns that show very significant variation of peak shapes with 2theta or strong intensity decay with angle (due to absorption or other effects). So you may observe a significant variation of your profile parameters or your ADPs between the short pattern fit and the final one with the whole pattern, but in general this will get you closer to the final answer in a shorter time.
Alternatively, if you are refining a pattern with broad-enough peaks (I mean you have more than 10 points per FWHM of your narrowest peak) you can generate a dataset that skips some points (e.g. half of them), so you reduce the number of points without reducing your 2theta range.
If you are refining data from nanoparticles in general you have dozens of points per FWHM so skipping 3 out of 4 or such will increase your speed by a factor of four without loosing accuracy.
Working this way you still get the correct values for your parameters, so accuracy is good, but your precision will be smaller than with the whole patterns since the number of observations you are using is lower. Then you do the refinement with all the data to increase precission of your already accurate refined parameters.
I hope this helps.
Leo
- Badges
- Science topic
- Similar topics
- Physical Sciences
- Materials Science