DTSA II and PFE

[mmengason]

Hi John.  I have a pair of feature requests to make it easier to do combined analysis by DTSA II and PFE.

You have an implementation within Probe for EPMA for performing EDS analysis if you are using a Bruker EDS system.  For my work I want to use DTSA and the specific application of linear least squares fitting that it uses.  It has been shown to be very robust against a lot of problems that affect EDS and the residual spectra is a very help tool as well.  More important for most of your users is that not everyone has a bruker/esprit system.  DTSA and PFE is a way for people to use EDS for majors/minors and WDS for trace who don’t specifically have a bruker/esprit system on their microprobe.
 
So here are my requests:
 
1.   I would appreciate a way to import compositions for analysis on a line-by-line basis.  You currently have a nice feature in your “specified element concentrations” window that allows you to read compositions in from a text file for a given sample number.  But that composition applies to the whole sample.  I have been using PFE to record transects across zoned minerals.  I am using it to record my EDS and I would like to use WDS for trace Ni and Co.  Since I’m offloading all the other elements onto EDS I can give each element its own spectrometer (or more than one for some elements) and count away throughout the whole EDS acquisition time.  But since the composition changes across the sample I can’t really use this correctly.  So I would appreciate a way to read in compositions for each line of an analysis or set of analyses from a text file.

2.   I would appreciate a way to use these entered composition values in your overlap correction algorithm.  I already use PFE to record my standards.  You need to know the concentration of the interfering element which I am providing from DTSA (although this could come from some other technique entirely like mass-spec).  But currently you only allow this to be used if the concentrations come from the Bruker EDS integration.  Allowing the correction based on entered composition would open this feature up to people with other detectors and/or people who want to use DTSA or some other software.

Thanks,
-Michael Mengason

[John Donovan]

You have an implementation within Probe for EPMA for performing EDS analysis if you are using a Bruker EDS system.  For my work I want to use DTSA and the specific application of linear least squares fitting that it uses.  It has been shown to be very robust against a lot of problems that affect EDS and the residual spectra is a very help tool as well.  More important for most of your users is that not everyone has a bruker/esprit system.  DTSA and PFE is a way for people to use EDS for majors/minors and WDS for trace who don’t specifically have a bruker/esprit system on their microprobe.
 
So here are my requests:
 
1.   I would appreciate a way to import compositions for analysis on a line-by-line basis.  You currently have a nice feature in your “specified element concentrations” window that allows you to read compositions in from a text file for a given sample number.  But that composition applies to the whole sample.  I have been using PFE to record transects across zoned minerals.  I am using it to record my EDS and I would like to use WDS for trace Ni and Co.  Since I’m offloading all the other elements onto EDS I can give each element its own spectrometer (or more than one for some elements) and count away throughout the whole EDS acquisition time.  But since the composition changes across the sample I can’t really use this correctly.  So I would appreciate a way to read in compositions for each line of an analysis or set of analyses from a text file.

Hi Michael,
I was wondering why you wanted to use DTSA, but now I see you want to utilize the linear least squares method for extracting the net intensities that is apparently only available in DTSA-2 (though Thermo NSS might utilize this method I think).  I will just first note that the integration of PFE with EDS, is not only with Bruker Esprit, but PFE also supports Thermo NSS (now Path Finder), and soon the JEOL EDS will be integrated seamlessly with PFE also, as soon as JEOL Japan gets me their final EDS API.

It's really too bad you can't utilize Bruker for the net intensity extraction and have to use DTSA manually.  One advantage of not using linear least squares is that, with the Bruker interface at least, they return not only the net intensities, but also the background intensities after peak stripping, so one can also calculate EDS element detection limits as described here:

http://probesoftware.com/smf/index.php?topic=822.msg5265#msg5265

Your suggestion of specifying the EDS elements from DTSA as fixed element compositions in PFE with the WDS elements is a good idea, but the specified element arrays can only be a single fixed composition per sample (or elements (or formulas), by difference, oxygen stoichiometry or stoichiometry to an arbitrary element). So currently the only way this will work is if you use a single point for each sample in PFE. Then you can use the text file loading method for the specified elements and you are all set.

But honestly, the best method for your purposes would be:

1. Get Bruker to implement the linear least squares method, or

2. Get Nicholas to implement a net intensity processing API in DTSA-2. 

On this last point, because PFE can utilize one EDS API for spectrum acquisition and a separate interface for EDS spectrum processing, if Nicholas did implement an EDS spectrum processing API in DTSA-2 you would be all good to go in that you could acquire spectra in PFE using Bruker for each analysis point, then extract the net intensities for the EDS elements using DTSA.  I've attached a generic EDS API pdf below that is the one utilized by Bruker, Thermo (and soon JEOL) for EDS integration with PFE. You should see if he can produce a compatible EDS API for spectrum processing.  The only function that would need to be implemented at this point is the EDSGetNetIntensities (and the EDSInit and EDSClose) function.

You will note that this API also includes an interface for processing EDS spectrum images for synchronized EDS and WDS map quantification.  I am working with Thermo, Bruker and JEOL on this now.  See here for more details:

http://probesoftware.com/smf/index.php?topic=400.msg3702#msg3702

Another advantage of using DTSA to extract the net intensities directly in PFE (as opposed to just the fixed concentrations) is that you would then be using a self consistent matrix correction for all the elements in the measurement (both WDS and EDS).

2.   I would appreciate a way to use these entered composition values in your overlap correction algorithm.  I already use PFE to record my standards.  You need to know the concentration of the interfering element which I am providing from DTSA (although this could come from some other technique entirely like mass-spec).  But currently you only allow this to be used if the concentrations come from the Bruker EDS integration.  Allowing the correction based on entered composition would open this feature up to people with other detectors and/or people who want to use DTSA or some other software.

Sorry, but this won't work.

The quantitative spectral interference correction in PFE works for elements from WDS by WDS, WDS by EDS, EDS by WDS and EDS by EDS, but if you check my interference correction paper here:

http://epmalab.uoregon.edu/publ/Improved%20Interference%20(Micro.%20Anal,%201993).pdf

you will see that the quantitative interference correction method requires that both the interfering *and* the interfered elements be measured together. Why? Because the spectral interference correction needs to iterate the matrix correction as the composition changes in the matrix due to the interference correction itself.

Obviously for small interference corrections where the matrix doesn't change much, this is not a big issue. But for large interference corrections, and also spectral interference corrections of the "cascade" type, e.g., Al-Ti-V-Cr alloys where Ti -> V Ka and V -> Cr Ka. In addition, we also need to iterate "self" interferences such as Mo La <-> S Ka, As Ka <-> Pb La, etc.  The cool thing about an iterated spectral interference correction is that the element order doesn't matter! 

The good news is that if you can get Bruker to implement a linear least squares method, or if you can get Nicholas to implement an API to obtain the EDS net intensities (and ideally the background intensities as well for detection limits calculations), you will have Bruker for EDS acquisition and DTSA for EDS spectrum processing automatically in PFE!

Please let me know if I haven't explained anything clearly enough.  It's complicated!   
john

[Nicholas Ritchie]

It will take time and effort to fully integrate the EDS and WDS world views.  It isn't that they really are very different but the devil is in the details.  Net intensity and background intensity may be the optimal way to enter data in the WDS world but they doesn't work as well in the EDS world where it isn't as obvious what range of channels should represent the peak and the background (1 FWHM/2?/3?).  Furthermore, the uncertainties inherent in the EDS fit process are similar but subtlety different from the simple intensity based models used in WDS.  Simply plugging characteristic peak intensity and background estimates from EDS spectra won't produce the correct detection limits and other uncertainty metrics.

The optimal place to meet up is at the k-ratio level.  Both techniques can produce k-ratios with the associated uncertainties.

However, even at the k-ratio level there are differences.  WDS measures the intensity at one energy (nominally one characteristic x-ray).  EDS measures the intensity of a range of energies and a collection of characteristic x-rays.  The correct matrix correction for EDS is a weighted sum of the individual x-ray matrix corrections.  Most of the time it doesn't matter (K lines for example) but with L and M lines the difference in absorption correction between the L3-M5 and the L2-M4 can be a factor of 2 or more.  To get the right answer, you have to use the weighted sum matrix correction.

It will take both communities time to understand the subtleties associated with the other community. It is worth the effort but it will take time.

[John Donovan]

It will take time and effort to fully integrate the EDS and WDS world views.  It isn't that they really are very different but the devil is in the details.  Net intensity and background intensity may be the optimal way to enter data in the WDS world but they doesn't work as well in the EDS world where it isn't as obvious what range of channels should represent the peak and the background (1 FWHM/2?/3?).  Furthermore, the uncertainties inherent in the EDS fit process are similar but subtlety different from the simple intensity based models used in WDS.  Simply plugging characteristic peak intensity and background estimates from EDS spectra won't produce the correct detection limits and other uncertainty metrics.

The optimal place to meet up is at the k-ratio level.  Both techniques can produce k-ratios with the associated uncertainties.

However, even at the k-ratio level there are differences.  WDS measures the intensity at one energy (nominally one characteristic x-ray).  EDS measures the intensity of a range of energies and a collection of characteristic x-rays.  The correct matrix correction for EDS is a weighted sum of the individual x-ray matrix corrections.  Most of the time it doesn't matter (K lines for example) but with L and M lines the difference in absorption correction between the L3-M5 and the L2-M4 can be a factor of 2 or more.  To get the right answer, you have to use the weighted sum matrix correction.

It will take both communities time to understand the subtleties associated with the other community. It is worth the effort but it will take time.

Hi Nicholas,
Excellent points.  I totally agree that EDS peak fitting, etc. can affect the ultimate measurement uncertainty in EDS, but the (single measurement) detection limits I calculate in PFE are based only on counting statistics (from Goldstein et al.).  That is, they are a *best case* estimate of sensitivity.  For total reproducibility calculations we utilize the t-test calculation (also from Goldestein et al.) which is based on the actual standard deviation of the concentrations (multiple measurements), which would include all sources of imprecision in the measurement(s).

Obviously there are also many effects other than count rates that affect uncertainty in WDS, e.g., spectrometer reproducibilty. In the Goldstein et al. (single measurement) detection limits I am utilizing in PFE, none of these other sources of error are included in the calculation as you point out.  But that doesn't mean that a counting statistics (only) detection limit should not be calculated and is not useful...

So the question remains for Michael and I: can you provide an API that returns the background intensities (and net intensities!) for the purposes of calculating a *counting statistics only* detection limit?  The total sensitivity t-test calculation in PFE already utilizes the actual variance of the concentrations (at least in a homogeneous material!) so this should not be an issue for this API.
john

PS If you are philosophically opposed to providing the background intensities from an EDS spectra, I could live with that. But the bigger question is: can you provide an API that *at least* provides the net intensities, so Michael can process his Bruker acquired EDS analyses on the NIST 8500 using PFE with DTSA, without having to manually export each EDS spectra from PFE, process them manually in DTSA and then manually re-enter the concentrations back in to PFE for integrated EDS and WDS acquisitions?

Dude, I'm just trying to help out your colleague here because he seems to want to utilize DTSA instead of Bruker for obtaining his EDS net intensities...