News:

:) To post in-line images, login and click on the Gallery link at the top

Main Menu

Matrix corrections vs matrix matched standards

Started by Brian Joy, August 08, 2022, 06:52:47 PM

Previous topic - Next topic

Brian Joy

Quote from: Probeman on August 08, 2022, 10:50:53 AM
Proof once again that we really do not require matrix matched standards.

Again, for most silicates and oxides the problem is *not* our matrix corrections. Instead it's our instrument calibrations, especially dead time calibrations, and of course having standards that actually are the compositions that we claim them to be.

This is an outrageous claim considering that you haven't strayed outside the system SiO2-Al2O3-MgO-CaO.  No transition metals??  Try analyzing some Fe-rich silicates using arbitrary but well-characterized standards.  Or how about garnet with widely varying Cr2O3 content (0-15 wt% or so)?  Or spinels of widely varying composition?  Or ilmenite using Fe2O3 and TiO2 as respective Fe and Ti standards?  Moving to some more complicated minerals, have you tried analyzing pollucite for Si and Al using anorthite as a standard?  Or how about analyzing for fluorine in bastnaesite, parisite, and synchysite using any common, well-characterized fluorine standard.  While this last group of minerals may not be familiar-looking, they are often quite abundant in REE deposits.

Also, keep in mind that I've pointed out that the Armstrong phi(rho*z) model is mathematically incorrect:

https://smf.probesoftware.com/index.php?PHPSESSID=25d1a37e49cefb24befa2c07d52156cd&topic=1430.0

Alternatively, click here
Brian Joy
Queen's University
Kingston, Ontario
JEOL JXA-8230

Probeman

#1
Quote from: Brian Joy on August 08, 2022, 06:52:47 PM
Quote from: Probeman on August 08, 2022, 10:50:53 AM
Proof once again that we really do not require matrix matched standards.

Again, for most silicates and oxides the problem is *not* our matrix corrections. Instead it's our instrument calibrations, especially dead time calibrations, and of course having standards that actually are the compositions that we claim them to be.

This is an outrageous claim considering that you haven't strayed outside the system SiO2-Al2O3-MgO-CaO.  No transition metals??  Try analyzing some Fe-rich silicates using arbitrary but well-characterized standards.  Or how about garnet with widely varying Cr2O3 content (0-15 wt% or so)?  Or spinels of widely varying composition?  Or ilmenite using Fe2O3 and TiO2 as respective Fe and Ti standards?  Moving to some more complicated minerals, have you tried analyzing pollucite for Si and Al using anorthite as a standard?  Or how about analyzing for fluorine in bastnaesite, parisite, and synchysite using any common, well-characterized fluorine standard.  While this last group of minerals may not be familiar-looking, they are often quite abundant in REE deposits.

Well I just showed that all 10 matrix corrections give essentially the same results for this system in spite of the rather large (~40%) matrix extrapolations, so outrageous?  Probably not.   :)   In fact I find it invigorating that we can apparently perform such large matrix corrections with such high accuracy.  Physics is cool.

But again I repeat: for *most* silicates and oxides...  exceptions abound and make our jobs interesting.  For high atomic number systems we have other physical issues to deal with in the matrix corrections (e.g., backscatter corrections), and I will be publishing a new paper on this soon so stay tuned.

As for the transition metals, based on the k-ratio measurements I showed previously, I would say we seem to be able to extrapolate from pure Ti to TiO2 quite well, thank-you very much.  I also have data for Fe and Mn Ka if you are interested.

That said, there are certainly "black holes" in the periodic table and I certainly agree that peak shift and shape issues can be problematic possibly for certain alumino-silicates as John Fournelle has documented. This is an area that should definitely be investigated further.

But my own k-ratio tests along with Anette's instrument seem to reveal that disagreement between our individual spectrometers contributes a larger error, than the error for many systems from our matrix corrections. Not to mention some nonsensical resistance from some of us, to utilizing a more precise form of the dead time correction, which again contributes yet another source of inaccuracy...  which I suspect is just one more reason some analysts retreat to utilizing "matrix" matched standards of uncertain composition.

Hey, if ones instrument is poorly calibrated, one has no choice but to minimize extrapolations in count rate and matrix.   ;D

Quote from: Brian Joy on August 08, 2022, 06:52:47 PM
Also, keep in mind that I've pointed out that the Armstrong phi(rho*z) model is mathematically incorrect:

https://smf.probesoftware.com/index.php?PHPSESSID=25d1a37e49cefb24befa2c07d52156cd&topic=1430.0

Alternatively, click here.

I wouldn't call it mathematically incorrect.  Just as with the PAP correction, the exponent was adjusted to match an empirical data set, which I agree isn't an ideal solution, but it does seem to help with accuracy and as noted above, the results seem to agree quite well with all the other modern phi-rho-z equations.

Curious, did you contact Armstrong about this and ask his opinion?
The only stupid question is the one not asked!

Brian Joy

Quote from: Probeman on August 08, 2022, 07:50:55 PM
Well I just showed that all 10 matrix corrections give essentially the same results for this system in spite of the rather large (~40%) matrix extrapolations, so outrageous?  Probably not.   :)

I repeat: for *most* silicates and oxides...  exceptions abound and make our jobs interesting. 

As for the transition metals, based on the k-ratio measurements I showed previously, I would say we seem to be able to extrapolate from pure Ti to TiO2 quite well, thank-you very much.

This does not address the question I asked.  You have not tested "*most* silicates and oxides," only a very few.

Quote from: Probeman on August 08, 2022, 07:50:55 PM
I wouldn't call it mathematically incorrect.  Just as with the PAP correction, the exponent was adjusted to match an empirical data set, which I agree isn't an ideal solution, but it does seem to help with accuracy and as noted above, the results seem to agree quite well with all the other modern phi-rho-z equations.

Curious, did you contact Armstrong about this and ask his opinion?

The units do not work; this is due in part to the use of fractional exponents (courtesy of Packwood and Brown).  If the units do not work, then the math does not either.  I've laid out the issue pretty clearly.
Brian Joy
Queen's University
Kingston, Ontario
JEOL JXA-8230

Probeman

Quote from: Brian Joy on August 08, 2022, 08:04:58 PM
Quote from: Probeman on August 08, 2022, 07:50:55 PM
Well I just showed that all 10 matrix corrections give essentially the same results for this system in spite of the rather large (~40%) matrix extrapolations, so outrageous?  Probably not.   :)

I repeat: for *most* silicates and oxides...  exceptions abound and make our jobs interesting. 

As for the transition metals, based on the k-ratio measurements I showed previously, I would say we seem to be able to extrapolate from pure Ti to TiO2 quite well, thank-you very much.

This does not address the question I asked.  You have not tested "*most* silicates and oxides," only a very few.

It's our working hypothesis based on a number of systems we have examined so far.  Let's follow the data and see how far it takes us. Sound reasonable?

Quote from: Brian Joy on August 08, 2022, 08:04:58 PM
Quote from: Probeman on August 08, 2022, 07:50:55 PM
I wouldn't call it mathematically incorrect.  Just as with the PAP correction, the exponent was adjusted to match an empirical data set, which I agree isn't an ideal solution, but it does seem to help with accuracy and as noted above, the results seem to agree quite well with all the other modern phi-rho-z equations.

Curious, did you contact Armstrong about this and ask his opinion?

The units do not work; this is due in part to the use of fractional exponents (courtesy of Packwood and Brown).  If the units do not work, then the math does not either.  I've laid out the issue pretty clearly.

Again, I'm just curious. Did you contact Armstrong and show him your work? What did he say?
The only stupid question is the one not asked!

Brian Joy

Quote from: Probeman on August 08, 2022, 08:11:55 PM
Quote from: Brian Joy on August 08, 2022, 08:04:58 PM
Quote from: Probeman on August 08, 2022, 07:50:55 PM
Well I just showed that all 10 matrix corrections give essentially the same results for this system in spite of the rather large (~40%) matrix extrapolations, so outrageous?  Probably not.   :)

I repeat: for *most* silicates and oxides...  exceptions abound and make our jobs interesting. 

As for the transition metals, based on the k-ratio measurements I showed previously, I would say we seem to be able to extrapolate from pure Ti to TiO2 quite well, thank-you very much.

This does not address the question I asked.  You have not tested "*most* silicates and oxides," only a very few.

It's our working hypothesis based on a number of systems we have examined so far.  Let's follow the data and see how far it takes us. Sound reasonable?

Your goal should be to prove that your hypothesis is incorrect.  This is what science is all about.  Whenever I come up with an idea that I really like, I try my best to trash it.  If it survives whatever tests I throw at it, then it might be a good idea.  Or it might be that I just haven't come up with the right test.

Quote from: Probeman on August 08, 2022, 08:11:55 PM
Quote from: Brian Joy on August 08, 2022, 08:04:58 PM
Quote from: Probeman on August 08, 2022, 07:50:55 PM
I wouldn't call it mathematically incorrect.  Just as with the PAP correction, the exponent was adjusted to match an empirical data set, which I agree isn't an ideal solution, but it does seem to help with accuracy and as noted above, the results seem to agree quite well with all the other modern phi-rho-z equations.

Curious, did you contact Armstrong about this and ask his opinion?

The units do not work; this is due in part to the use of fractional exponents (courtesy of Packwood and Brown).  If the units do not work, then the math does not either.  I've laid out the issue pretty clearly.

Again, I'm just curious. Did you contact Armstrong and show him your work? What did he say?

Feel free to show John Armstrong my post and ask him to reply, as I do not want the discussion to remain private:

https://smf.probesoftware.com/index.php?topic=1430.msg10536#msg10536

My position is very well thought out and easily defensible; I don't really need anybody's opinion on it.  Does the model even produce accurate phi(rho*z) curves?  You've seen for yourself that it doesn't after I wrote you some code to calculate these curves (with automated truncation).
Brian Joy
Queen's University
Kingston, Ontario
JEOL JXA-8230

Probeman

#5
Quote from: Brian Joy on August 08, 2022, 08:40:52 PM
Quote from: Probeman on August 08, 2022, 08:11:55 PM
Quote from: Brian Joy on August 08, 2022, 08:04:58 PM
Quote from: Probeman on August 08, 2022, 07:50:55 PM
Well I just showed that all 10 matrix corrections give essentially the same results for this system in spite of the rather large (~40%) matrix extrapolations, so outrageous?  Probably not.   :)

I repeat: for *most* silicates and oxides...  exceptions abound and make our jobs interesting. 

As for the transition metals, based on the k-ratio measurements I showed previously, I would say we seem to be able to extrapolate from pure Ti to TiO2 quite well, thank-you very much.

This does not address the question I asked.  You have not tested "*most* silicates and oxides," only a very few.

It's our working hypothesis based on a number of systems we have examined so far.  Let's follow the data and see how far it takes us. Sound reasonable?

Your goal should be to prove that your hypothesis is incorrect.  This is what science is all about.  Whenever I come up with an idea that I really like, I try my best to trash it.  If it survives whatever tests I throw at it, then it might be a good idea.  Or it might be that I just haven't come up with the right test.

No kidding.  That's exactly what we are doing by testing these matrix corrections to their limits by utilizing end-member (pure metal and simple oxide) primary standards!  So far we're seeing very promising results.  But we will continue to test our hypothesis.

Quote from: Brian Joy on August 08, 2022, 08:40:52 PM
Quote from: Probeman on August 08, 2022, 08:11:55 PM
Quote from: Brian Joy on August 08, 2022, 08:04:58 PM
Quote from: Probeman on August 08, 2022, 07:50:55 PM
I wouldn't call it mathematically incorrect.  Just as with the PAP correction, the exponent was adjusted to match an empirical data set, which I agree isn't an ideal solution, but it does seem to help with accuracy and as noted above, the results seem to agree quite well with all the other modern phi-rho-z equations.

Curious, did you contact Armstrong about this and ask his opinion?

The units do not work; this is due in part to the use of fractional exponents (courtesy of Packwood and Brown).  If the units do not work, then the math does not either.  I've laid out the issue pretty clearly.

Again, I'm just curious. Did you contact Armstrong and show him your work? What did he say?

Feel free to show John Armstrong my post and ask him to reply, as I do not want the discussion to remain private:

https://smf.probesoftware.com/index.php?topic=1430.msg10536#msg10536

My position is very well thought out and easily defensible; I don't really need anybody's opinion on it.  Does the model even produce accurate phi(rho*z) curves?  You've seen for yourself that it doesn't after I wrote you some code to calculate these curves (with automated truncation).

All I've pointed out is that the Armstrong model agrees very well with the other modern phi-rho-z models. That is exactly why I showed results from all the other 10 models. If you have a problem with the Armstrong model then you shouldn't use it. 

Again, the point here isn't the Armstrong matrix correction specifically.  It's simply that we are seeing greater accuracy issues with the instrument dead time corrections, the spectrometer alignments and crystal diffraction symmetry and as demonstrated by Weiser et al., our standard compositions.

I am merely saying, let's focus our attention on the largest sources of error first, and improve them if possible.  We've already demonstrated for several systems (both high and low emission energies) that the matrix corrections are in essential agreement, while our dead time corrections, spectrometer k-ratios and standards are not.

We think this is a reasonable course of action and are pursuing our investigations.
The only stupid question is the one not asked!

sem-geologist

Brian,
I see You are active there and here with concerns.  :D Actually I am concerned about similar things too and I am happy for your constructive criticism. Probably without it I would be to afraid to criticize some status quo stuff. I just want to point that not only non-existent pile-up correction (which probeman keeps calling the "dead-time" calibration), but also some nasty 2nd, 3rd or 4th order emission lines of overlooked and not corrected overlaps (especially on the background) had produced some classical matrix matched standard cargo cultism. Don't understand me wrong, this new pile-up correction log equation (or as probeman calls "dead-time calibration") is one of the most important recent advancements (I am looking forward to see the log equation for which I have my own physical model and explanation why it works and why it stops working at very high counting rates, and it does not require physical contradictions). However, even without that, if using manufacturers recommended dead-time values (the real blanking timing, that is 3.3µs for all of Cameca WDS) and staying at count rate where pile-ups do not exceed 0.5% of registered pulses (basically below 10kcps on Cameca instruments) I came to conclusion that matrix matched standards can be ditched in most of the cases. Having no builtin logarithmic pile-up correction equations in Peaksight in some very rare cases I practice something what could be called "counting-rate matched simple standard measurements".

There are X-ray lines which requires matrix matched standards such as i.e. Sulphure Ka due to line shift (although if offseting the S Ka position of unknown to the correct for unknown position the sulfate-sulfide S Ka standards can be nearly interchangeable). But line shifts are actually only the half of the problem. Sulfosalts in general are difficult beast: common extreme pace of oxidation of surface is very often overlooked and I know some laboratories uses commonly Chalcopyrite and Pyrite as main standards for sulphure (Those are often included in most of buy-able sets of mineral standards), which are a very terrible choose  (I need often to deal with users coming with terrible measurement protocols designed elsewhere). Unless the unknown sample of sulfides had spent exactly same time in the same environment as standards (and we still need to suppose that unknown and CuFeS or FeS2 had oxidized at the same pace) there will be kind of correct matrix matched S-standard hunt (my bitter experience from first years of my probe job, when I was not questioning any already settled practices). Also, who re-polishes the standard-sets every week? From sulfides, AFAIK, only ZnS is resilient to oxidation in some sane degree (can withstand half year without re-polishing), and quantification works correctly if unknown is just freshly polished and ZnS is used as S Ka standard. But there comes the common problem: often user who own the samples with unknowns of sulfides is lazy and comes to lab with samples polished two weeks or few months ago... and then there is surprise Pikachu face why analytical totals are closing only to 96%. Because of laziness at least for sulfides people leaned-on "matrix matched standards" (which in correct form should be actually called "surface oxidation matched standards"). When using oxidation-resistant sulfide standard and re-polishing the unknown sulfides same or a day before probe session - the need for matrix matched standard just is no more.

So Metal vs Oxides has very similar problem of oxidation like sulfides, just a bit less severe. At least from my experience if I use metal oxide as standard, metals as unknowns always produces lower sums than expected - never the larger, which taking into consideration of oxidation of metal surface makes complete sense (also there is these O Ka peaks in WDS and EDS if looked for). The effect can be minimized by re-polishing. But perfect metal surface preparation system would require polishing, drying and transportation under the inert gas. Fortunately many metals have this feature of initial oxidated layer of the metal preventing/slowing down further oxidation, which lots of sulfides do not posses.

Then there is another case with complex minerals and substances which in particularly with large diff XTALs leaves no space for precise background measurement (but believe me some minerals with 30+ major and minor elements will cause a hell even with standard (small) sized diff crystals). And when if people can't get right measurements using classical approaches they turn to matrix matched standards (which have the similar troubles with background, thus the problem is partly eliminated). Ironically while from one side PfS development in its own way uncovers redundancy of matrix matched standards (i.e. by this pile-up/"dead-time" callibration) on the other hand PfS-implemented multi-point-background for sure will keep some to matrix matched standard preferences (I am fiercely opposed to the MPB and see it not only highly redundant and unnecessary, but also dangerous as results could insert some biases to large sets of inter institutional measurements). However with proper background measurement (single background with ultimately universally precise slope (single and same slope for very different density materials at same background position, often it is possible to make it without any absorption edge in between it and the peak!), which I find using HussariX software) - simple standards just work without any need of matrix matched ones. Sometimes The complex minerals are so complicating the measurements of background that it is better to switch some measurements to 2nd order lines or if applicable use the higher resolution lower intensity equivalent on other diff XTALs. (i.e. Si Ka measure on PET instead of overcrowded TAP). Fortunately recently even those cases got much easier to deal with starting with Cameca Peaksight 6.5, where negative interference corrections just works (that is background position interference with other peaks can be itteratively corrected).

F is problematic if measured on TAP, as there is F energetic dispersion depending from the type of mineral, and so TAP is too high resolution, but going to PC0 allows to use single F rich substance universally. So, what is the problem with Bastnesite if F is measured with PC0 and F is calibrated on pure F-Flogopite or pure F-Topaz? Well due to strong and wide REE M lines and if using MPB or Two (left-right) backgrounds - the situation would be just pure doom. But single background with precise slope and few precise interference corrections makes it work without any issues. As for lower atomic numbers probably some of matrix matched standards would find its niche there, I have not much experience there.

As conclusion, Every time I come across any inherited need for matrix matched standard for common rock forming or accessory minerals I can find the hidden overlooked reason behind, which often originated either due to unaccounted sample oxidation, or poor measurement practices (or both). Often it is that no one cared to look deeper as using matrix matching standard was the least-effort approach for short-term to get the predicted results. People are keen to take protocols which are known just to work, even if it is not the best or more tedious way in bigger picture. Ditching matrix matched standards is so much time saving in the long term as keeping the limited number of simple standards updated is more efficient and easier than standard base with all kind of combinations of matrices.

Probeman

Quote from: sem-geologist on August 09, 2022, 06:51:15 AM
I just want to point that not only non-existent pile-up correction (which probeman keeps calling the "dead-time" calibration), but also some nasty 2nd, 3rd or 4th order emission lines of overlooked and not corrected overlaps (especially on the background) had produced some classical matrix matched standard cargo cultism. Don't understand me wrong, this new pile-up correction log equation (or as probeman calls "dead-time calibration") is one of the most important recent advancements (I am looking forward to see the log equation for which I have my own physical model and explanation why it works and why it stops working at very high counting rates, and it does not require physical contradictions). However, even without that, if using manufacturers recommended dead-time values (the real blanking timing, that is 3.3µs for all of Cameca WDS) and staying at count rate where pile-ups do not exceed 0.5% of registered pulses (basically below 10kcps on Cameca instruments) I came to conclusion that matrix matched standards can be ditched in most of the cases. Having no builtin logarithmic pile-up correction equations in Peaksight in some very rare cases I practice something what could be called "counting-rate matched simple standard measurements".

Hi SG,
I appreciate your support and comments.  Thank-you.

Yes, I agree completely that much of the reason people feel the need to resort to matrix matched standards is due to a multitude of factors, among them poorly calibrated dead time constants, uncorrected/unrecognized spectral interferences, and spectrometer/crystal misalignments. I probably should write up a summary of all of these instrument calibration issues someday, for example picoammeter non-linearity could contribute to errors measuring ones standards and unknowns at different beam currents... the list goes on.

I will share the log dead time expression soon as we are currently writing it up now, but it suffices to say that the multi-term expression, described in this post, gives almost identical results to the log expression:

https://smf.probesoftware.com/index.php?topic=1466.msg11032#msg11032

And as you also point out, by relying on matrix matched standards we have introduced another error and that is: do we really know exactly what are the compositions of our natural standard materials?  We already know from studies by John Fournelle and Ed Vicenzi that these natural materials routinely utilized by many labs are heterogeneous and full of inclusions.  This is exactly the impetus for our FIGMAS efforts with Will Nachlas and Aurelien Moy to identify high purity synthetic mineral candidates for eventual wide scale (global) distribution to microanalysis labs.

https://smf.probesoftware.com/index.php?topic=1415.0

As you can tell, I am usually not fastidious when it comes to nomenclature, but whether it's called dead time or pulse pileup or photon coincidence I also agree this is an area that most labs are not paying much attention to.  Frankly I never gave dead time calibrations much attention until John Fournelle, Anette von der Handt and I started performing these "constant k-ratio" measurements and we started seeing problems at moderate to high count rates on our respective instruments:

https://smf.probesoftware.com/index.php?topic=1466.msg11025#msg11025

Hopefully both Cameca and JEOL will implement these improved dead time corrections in their OEM software soon. As you point out this is especially important for Cameca instruments with their rather long dead times constants.  The good news is that the current version of Probe for EPMA now contains *four* dead times corrections!   :o  They are: the traditional (single term) expression, the Willis (1992) (two term) expression, the new six term expression, and now the logarithmic expression.



Quote from: sem-geologist on August 09, 2022, 06:51:15 AM
There are X-ray lines which requires matrix matched standards such as i.e. Sulphure Ka due to line shift (although if offseting the S Ka position of unknown to the correct for unknown position the sulfate-sulfide S Ka standards can be nearly interchangeable). But line shifts are actually only the half of the problem.

Yes, exactly. We always adjust our sulfur peak positions  based on the average oxidation state of our volcanic glasses and just use pyrite as the primary standard as it is about a 10% effect in the case of most basaltic glasses.  This peak shift effect could also be dealt with using an APF correction as well. 

Quote from: sem-geologist on August 09, 2022, 06:51:15 AM
Sulfosalts in general are difficult beast: common extreme pace of oxidation of surface is very often overlooked and I know some laboratories uses commonly Chalcopyrite and Pyrite as main standards for sulphure (Those are often included in most of buy-able sets of mineral standards), which are a very terrible choose  (I need often to deal with users coming with terrible measurement protocols designed elsewhere). Unless the unknown sample of sulfides had spent exactly same time in the same environment as standards (and we still need to suppose that unknown and CuFeS or FeS2 had oxidized at the same pace) there will be kind of correct matrix matched S-standard hunt (my bitter experience from first years of my probe job, when I was not questioning any already settled practices). Also, who re-polishes the standard-sets every week? From sulfides, AFAIK, only ZnS is resilient to oxidation in some sane degree (can withstand half year without re-polishing), and quantification works correctly if unknown is just freshly polished and ZnS is used as S Ka standard. But there comes the common problem: often user who own the samples with unknowns of sulfides is lazy and comes to lab with samples polished two weeks or few months ago... and then there is surprise Pikachu face why analytical totals are closing only to 96%. Because of laziness at least for sulfides people leaned-on "matrix matched standards" (which in correct form should be actually called "surface oxidation matched standards"). When using oxidation-resistant sulfide standard and re-polishing the unknown sulfides same or a day before probe session - the need for matrix matched standard just is no more.

So Metal vs Oxides has very similar problem of oxidation like sulfides, just a bit less severe. At least from my experience if I use metal oxide as standard, metals as unknowns always produces lower sums than expected - never the larger, which taking into consideration of oxidation of metal surface makes complete sense (also there is these O Ka peaks in WDS and EDS if looked for). The effect can be minimized by re-polishing. But perfect metal surface preparation system would require polishing, drying and transportation under the inert gas. Fortunately many metals have this feature of initial oxidated layer of the metal preventing/slowing down further oxidation, which lots of sulfides do not posses.

I agree completely about these surface oxidation (and often overlooked) effects.  This is the primary reason I have insisted to Will and Aurelien that we mount these FIGMAS standards in an acrylic mount that can be easily re-polished and re-coated. For example:

https://smf.probesoftware.com/index.php?topic=172.msg8991#msg8991

Now with silicates and oxides this is not so much a problem, but as you point out for metals and sulfides we find it necessary to re-polish and re-coat every few months or whenever we get sulfide work.  That said, we often avoid some of these issues by running our standards and unknowns at 20 or 25 keV which reduces surface sensitivity.  But for those performing low voltage work, re-polishing of standards (and unknowns) just prior to analysis is essential.

Quote from: sem-geologist on August 09, 2022, 06:51:15 AM
Then there is another case with complex minerals and substances which in particularly with large diff XTALs leaves no space for precise background measurement (but believe me some minerals with 30+ major and minor elements will cause a hell even with standard (small) sized diff crystals). And when if people can't get right measurements using classical approaches they turn to matrix matched standards (which have the similar troubles with background, thus the problem is partly eliminated). Ironically while from one side PfS development in its own way uncovers redundancy of matrix matched standards (i.e. by this pile-up/"dead-time" callibration) on the other hand PfS-implemented multi-point-background for sure will keep some to matrix matched standard preferences (I am fiercely opposed to the MPB and see it not only highly redundant and unnecessary, but also dangerous as results could insert some biases to large sets of inter institutional measurements). However with proper background measurement (single background with ultimately universally precise slope (single and same slope for very different density materials at same background position, often it is possible to make it without any absorption edge in between it and the peak!), which I find using HussariX software) - simple standards just work without any need of matrix matched ones. Sometimes The complex minerals are so complicating the measurements of background that it is better to switch some measurements to 2nd order lines or if applicable use the higher resolution lower intensity equivalent on other diff XTALs. (i.e. Si Ka measure on PET instead of overcrowded TAP). Fortunately recently even those cases got much easier to deal with starting with Cameca Peaksight 6.5, where negative interference corrections just works (that is background position interference with other peaks can be itteratively corrected).

F is problematic if measured on TAP, as there is F energetic dispersion depending from the type of mineral, and so TAP is too high resolution, but going to PC0 allows to use single F rich substance universally. So, what is the problem with Bastnesite if F is measured with PC0 and F is calibrated on pure F-Flogopite or pure F-Topaz? Well due to strong and wide REE M lines and if using MPB or Two (left-right) backgrounds - the situation would be just pure doom. But single background with precise slope and few precise interference corrections makes it work without any issues. As for lower atomic numbers probably some of matrix matched standards would find its niche there, I have not much experience there.

I totally appreciate the single point slope background and it is implemented in PFE as you can see here for both sides of the peak:



Note the Slope (Hi) and Slope (Lo) methods. 

As for the MPB method, I think that it can be useful, but can also be abused like most things. However, according to Mike Jercinovic and others, it is essential for high accuracy trace elements in complex matrices. Frankly I personally prefer the MAN method where the background is calibrated at the emission line position. This method is accurate to a few hundred PPM or better in most silicates and oxides and one can further improve accuracy using a quantitative blank correction.  See Donovan et al., (2011) for the blank correction and Donovan et al., (2016) for details on the MAN correction. I like particularly that one can obtain higher sensitivity in less time than off-peak methods.  It's sort of like magic!   8)

Quote from: sem-geologist on August 09, 2022, 06:51:15 AM
As conclusion, Every time I come across any inherited need for matrix matched standard for common rock forming or accessory minerals I can find the hidden overlooked reason behind, which often originated either due to unaccounted sample oxidation, or poor measurement practices (or both). Often it is that no one cared to look deeper as using matrix matching standard was the least-effort approach for short-term to get the predicted results. People are keen to take protocols which are known just to work, even if it is not the best or more tedious way in bigger picture. Ditching matrix matched standards is so much time saving in the long term as keeping the limited number of simple standards updated is more efficient and easier than standard base with all kind of combinations of matrices.

I could not agree more with you.  Thanks for sharing your thoughts on this.
The only stupid question is the one not asked!

Brian Joy

Quote from: sem-geologist on August 09, 2022, 06:51:15 AM
As conclusion, Every time I come across any inherited need for matrix matched standard for common rock forming or accessory minerals I can find the hidden overlooked reason behind, which often originated either due to unaccounted sample oxidation, or poor measurement practices (or both). Often it is that no one cared to look deeper as using matrix matching standard was the least-effort approach for short-term to get the predicted results. People are keen to take protocols which are known just to work, even if it is not the best or more tedious way in bigger picture. Ditching matrix matched standards is so much time saving in the long term as keeping the limited number of simple standards updated is more efficient and easier than standard base with all kind of combinations of matrices.

Keep in mind that I did not necessarily advocate for matrix matching – John was the one who provided the title for the topic.  My main point was that he had not demonstrated "proof" that the dead time correction is the most important consideration in quantitative analysis; this is his hypothesis, but he has not tested it sufficiently.  If you are using a "modern" matrix correction ZA model, such as PAP, XPP, X-Phi, or PROZA96 in conjunction with the MAC30 mass absorption coefficients, then the issue of matrix matching is not so critical.  But try analyzing pollucite using anorthite as Si and Al standard while using the Armstrong model and FFAST MACs (which many people combine) and see what you get.  Or try analyzing various ilmenites using any model with Fe2O3 and TiO2 as standards and then recalculate for Fe2O3 – you're likely to see some negative numbers, possibly due to problems with the fluorescence correction (see Evans et al., CMP 152:149-167).  I shouldn't have mentioned the REE fluorocarbonates, as the issues with them are more complex.  Using LDE1, I can't get away with using a fixed slope – I can't remember what the situation is with PC0.  Of course, interference from Ce Mz must be taken into account.

I analyze a lot of sulfides and sulfosalts, and I agree with your points about oxidation; hydration can be a problem as well.  I also see growth of acanthite on Ag-bearing standards and growth of covellite(?) on tetrahedrite.  Also, I never use high purity metal standards unless I've polished them very recently – the presence of the oxide film results in high analytical totals, for instance when analyzing sulfides or alloys (the inverse of the problem you were describing when analyzing metals using oxides).

Going back to my initial comment and noting that no matrix correction model is perfectly accurate, a certain degree of matrix matching is unavoidable if only to avoid excessive matrix corrections.  For instance, many people use MgF2 as a standard when analyzing for fluorine in apatite (which can only be analyzed accurately in prismatic section – TDI cannot be used).  Since F Ka is absorbed strongly by both oxygen and calcium, noting that it is energetic enough to ionize the Ca L subshells, this will lead to an absorption correction somewhere in the range of 150% (very roughly).  Applying matrix corrections of this magnitude is never good practice.
Brian Joy
Queen's University
Kingston, Ontario
JEOL JXA-8230

Probeman

#9
Quote from: Brian Joy on August 09, 2022, 06:15:26 PM
Keep in mind that I did not necessarily advocate for matrix matching – John was the one who provided the title for the topic.  My main point was that he had not demonstrated "proof" that the dead time correction is the most important consideration in quantitative analysis; this is his hypothesis, but he has not tested it sufficiently.

Going back to my initial comment and noting that no matrix correction model is perfectly accurate, a certain degree of matrix matching is unavoidable if only to avoid excessive matrix corrections. 

You need to re-read what I actually wrote.  I never said the dead time correction is the most important consideration in quantitative analysis (it was in fact SEM Geologist that wrote: "this new pile-up correction log equation (or as probeman calls "dead-time calibration") is one of the most important recent advancements"). 

I've always mentioned that dead time calibrations, along with spectrometer alignments (effective takeoff angle) and standard compositions (along with matrix corrections) are all under appreciated by some labs, and together these issues conspire (in labs where attention is not being paid to these issues) to push analysts into utilizing "matrix matched" standards, I suspect in order to avoid dealing with these instrumental details.

However, our initial testing of these candidate materials continues and we are still in early days, but based on the data so far, we are seeing surprisingly good results, as we have shown already:

https://smf.probesoftware.com/index.php?topic=1466.msg11076#msg11076

Note that the correction for Al Ka from Al2O3 to spinel is over 140%. That certainly impressed me to see ~1% relative accuracy over a range from 5 nA to 120 nA...

And I'll say it again, of course nothing is perfectly accurate, the point is we should focus our attention on those areas where the largest errors are seen, and the specifics of those relative errors will indeed depend on the particular details of the physics as I have stated many times previously.

I'm just pleased to see an effort (FIGMAS) is being made to identify, procure and distribute high purity synthetic minerals, which in addition to simply being great standards for all sorts of applications, will also allow us to test these various instrumental calibrations more easily.

I think we can all appreciate that.
The only stupid question is the one not asked!

Probeman

#10
When it was initially proposed to utilize MgO, Al2O3 and MgAl2O4 as the first materials for the FIGMAS round robin (primarily because these synthetic materials were readily available, cheap and of high purity), some questioned whether it would be worthwhile to measure k-ratios for such large matrix extrapolations. But there are good reasons for attempting such an effort:

1. Maybe one would not typically utilize such non-matrix matched standards, but that is exactly the point of testing an extreme matrix extrapolation: let's perform the test and see how good/bad we do!

2. It doesn't really matter what k-ratios we actually end up with.  The more important question is: can we obtain the same k-ratios (within statistics) on all spectrometers on our instruments, and again, can we obtain the same k-ratios (again, within statistics), compared to other instruments?

3. Maybe we will (likely) discover that on some matrix systems we will do rather better than expected, and on other systems do worse than expected?  Let's perform the experiment!

So, we've already seen some examples of the accuracy for this MgO-Al2O3-MgAl2O4 system here:

https://smf.probesoftware.com/index.php?topic=1466.msg11076#msg11076

But let's look a bit more systematically at the quant results by exporting the calculated compositions for all the beam current measurements from 5 to 120 nA using the CalcZAF "Standard format" export using this menu in Probe for EPMA:



This export format exports the k-ratios and also the "published" value, so we can perform quantitative analyses using different matrix correction models (and different MAC values).  This format is somewhat similar to the "binary" k-ratio format utilized by Pouchou and Bastin to tune their PAP and XPP phi-rho-z models, but the CalcZAF "standard" format allows one to look at results for materials with three or more elements. Examples of both formats are provided with the CalcZAF/Standard open source application.  See here for more details:

https://smf.probesoftware.com/index.php?topic=1256.0

These binary k-ratio databases were also utilized by myself, Philippe Pinard and Hendrix Demers for our 2019 paper on using non-linear alpha factors to perform fast quant analyses using k-ratios generated from Penepma Monte Carlo calculations:

https://epmalab.uoregon.edu/publ/high_speed_matrix_corrections_for_quantitative_xray_microanalysis_based_on_monte_carlo_simulated_kratio_intensities.pdf

Anywho, once this k-ratio data is exported from Probe for EPMA, we can import it into CalcZAF using this menu:



The output we obtain calculates the composition and relative errors from the "published values, and also calculates an average and standard deviation as seen here for Mg Ka and Al Ka:

Using the Armstrong phi-rho-z we obtain:

"Mg Average"                    .985950
"Mg StdDev"                     .009610

"Mg Minimum"                    .971185
"Mg Maximum"                    1.00052
"Al Average"                    .991880
"Al StdDev"                     .006217
"Al Minimum"                    .985132
"Al Maximum"                    1.00262

Using PAP we obtain:

"Mg Average"                    .987174
"Mg StdDev"                     .009548

"Mg Minimum"                    .972463
"Mg Maximum"                    1.00157
"Al Average"                    1.00198
"Al StdDev"                     .006207
"Al Minimum"                    .995280
"Al Maximum"                    1.01274

So around 1% relative accuracy and sub percent precision using the MgO and Al2O3 as primary standards and MgAl2O4 as a secondary standard measured at beam currents from 5 to 120 nA!

Using the logarithmic dead time correction expression of course.   :D
The only stupid question is the one not asked!