Relative K-ratio for smoother calibration curves

[Walter]

Hi everyone,

Performing simulations of films with variable thicknesses I've noticed that the K-ratio is not very stable. Probably it could be improved simulating more electrons. But I also noticed that if I use "relative K-ratio" (normalizing the sum of the K-ratio of the components to 1) the results are very smooth.

See the picture where I reported the K-ratios of an Au film (from 10 nm to 10 um) on Cu.

I wonder why this happens (is there a statistical reason?) and if this procedure is acceptable (or do the results will be distorted?) to improve the quality of data to build a calibration curve.

Thank you,

Walter

[jrminter]

Walter, I just saw this on the forum update John sent out.

First, 1000 electrons seems low. The 'simulation alien' in the DTSA-II current release (Lorentz) simulates 3000 trajectories as a default for a simulation at 15 kV. I typically run at least 10,000. These can take a few hours at high kVs. The 'simulation alien' does not give you full control over the parameters. The DTSA-II Jython scripting language gives you much more control over the simulation parameters. I have simulated a wide range of C on Si and C on Cu with DTSA-II and have Jython scripts that can be modified to work. I will need your DTSA-II  "detector configuration" parameters to configure the simulation. If you need help on the parameters see N. Ritchie's instructions here: https://www.cstl.nist.gov/div837/837.02/epq/dtsa2/DTSA-II_Configuration.pdf. I will also need the transitions you want to compute. I am guessing you want the CuKa and AuLa transitions since they are the most intense at 15 kV.   

Hope this helps,
John Minter
Retired from Kodak Analytical Sciences

[jrminter]

Walter, my simulation PC went down and I finally replaced it during the post-Thanksgiving sale. Took me a while to get everything working again. I simulated 10,000 trajectories per thickness using a Jython script. The plot looks like this:



I used a Rmarkdown file to generate the plot. One could extend this to fit a smooth curve to the data.

All the code is on github. I have a Jython script to compute the standard spectra (Au and Cu) and the desired thicknesses. The script uses some library files that wrap some of the operations. All of these are on github. The best part is that one has much more control over the simulation.

Hope this helps...
Best Regards,
John Minter

[Walter]

Hi John

I answer you after a while, but I hope to find you good. I performed the simulation with Jython script as well, using the Kelvin release. But my question was not exactly about that.

To make the question clearer I took the data you posted. As you can see also in you case the points are quite scattered increasing the thickness of the layer. If you normalize the K-ratios to 1 (see the picture I posted) the curve is much smoother.

My question is: probably this normalization does not have any physical meaning but, if I use the inverse of this curve to obtain the thickness of a sample, could it be ok? Or is it better to do not manipulate the data?

Best regards,

Walter