Community consensus k-ratio database?

[NicholasRitchie]

At QMA-2019, John Donovan said "the only thing we measure is k-ratios.  Everything else we pull out of our ass."  He has a point.  We measure k-ratios and then apply a set of correction factors that have some empirical basis to estimate the composition.  k-ratios are the fundamental characteristic of our business.  For a given beam energy and take-off angle, your instrument should produce the same k-ratios (to within statistics) as mine for the same material.  If they don't then there is cause to investigate.  In fact, if we measure a set of k-ratios and we are able to match them up with k-ratios from a known material, there is no need to apply matrix corrections, we know that we are measuring that material.  This is becoming particularly important as we start to move into domains in which our traditional matrix correction algorithms don't work (like L lines in transition metals...)   We may never be able to develop matrix correction algorithms that work in this domain however, by database lookup we could identify the material if anyone anywhere has entered it into the database.

  So I'm going to propose a (significant, long-term) microanalysis community project:

1. Create an online repository / database of empirical k-ratios
2. Encourage the community to submit k-ratios  and maintain an online record of all the submissions
3. Make the database available to all (through a web form and through a web application programming interface)

Uses:
1. Evaluate matrix correction algorithms against a community consensus database
2. You can make all measurements against the best similar (matched) standard in the database without actually having that material in you lab!! (See below)
     With community consensus, a day could come in which all a lab would need to do standards-based quant against any material would be one standard per element. (All your standards on one or two blocks that never need come out of your instrument..)
3. We can accurately measure the composition of materials for which matrix correction algorithms don't work (if they are in the database)
4. QC your instrument by comparing the k-ratios you measure to community consensus k-ratios
5. Reduce our dependence on matrix correction algorithms (all measurements could be against similar standards)
6. k-ratios never go out of stock or become degraded (like NIST / standards)
7. The range of compositional variation of inhomogeneous standards could be characterized.

Questions:
1.  Is this a good idea?  (Am I overlooking something?)
2.  Is the community interested in participating?  (It will only work if many people contribute)
3.  How do we design such a database?
     How do we curate data?  Do we curate data? (Is community consensus better than curation? Do we implement outlier detection?)
     Do we attribute data to labs/individuals?
     What data needs to be included?  (What is the minimum required, what else would be helpful, what would be icing?)
4.  Where is it hosted? (MAS? NIST? FIGMAS?)
5.  Who develops and maintains it?  (One person, a group (FIGMAS?, MAS working group?, research group???))
6.  What materials would we suggest measuring the k-ratio against?  (Pure elements? What about surface oxidization? What is the most stable?)
7.  Do we need a separate WDS and EDS databases?  (For EDS, should we just archive spectra (unknown and standards?))

Similar (matched) standards -
  The idea here is to "share k-ratios not materials."  Any standard that anyone anywhere had measured an entered into the database could be used in your measurement without having that material in your lab.
  Historically, if we want to make a measurement against Kakanui hornblende (KH), we need a chip or two in our lab.  But this isn't necessary.  If we have a database of k-ratios of KH against Si, Al, Fe, Mg, CaF2, Albite, etc and we make measurements in our lab using Si, Al, Fe, ... We can use the consensus k-ratios to convert our simple Si, Al, Fe... k-ratios into k-ratios relative to KH and then use these to measure our unknown against KH.  (Furthermore, the spread of the k-ratios in the database could provide information about the standard's heterogeneity.)

[Probeman]

At QMA-2019, John Donovan said "the only thing we measure is k-ratios.  Everything else we pull out of our ass."  He has a point.  We measure k-ratios and then apply a set of correction factors that have some empirical basis to estimate the composition.  k-ratios are the fundamental characteristic of our business.  For a given beam energy and take-off angle, your instrument should produce the same k-ratios (to within statistics) as mine for the same material.  If they don't then there is cause to investigate. 

I guess I did say that, over a lunch table I believe!   

In fact, if we measure a set of k-ratios and we are able to match them up with k-ratios from a known material, there is no need to apply matrix corrections, we know that we are measuring that material.

I think this is a very good idea, but as we all know(!), the problem is knowing!

What I mean is this: at that lunch table the original idea I proposed was a "round robin" test, not of compositions, but of k-ratios. Why k-ratios? Because, as stated above so profoundly and eloquently(!), that's what our instruments fundamentally produce. So at the very least we should be testing that our instruments produce the same fundamental measurement within the precision of our measurements.  That's just a place to start.

Now there are some practical difficulties with such k-ratio measurements. For example we have to make a background correction, and there are many situations where such a basic measurement of net intensity can be problematic, e.g., curved backgrounds, complex backgrounds (spectral interferences), even sample sensitivity can be problematic for background determinations (at least for WDS), since the background intensity is usually measured *after* the peak intensity.  But let's assume that at least for major elements, such background issues are 2nd order effects, or at least, solvable. 

Then there's the issue of take-off angle. Nicholas' presentation at QMA-2019 of sources of error comes right into play here. Maybe we can assume that all our WDS spectrometers on our microprobes have the same take off angle, or do they? I don't think so. Have we all performed a simultaneous k-ratio measurement on all our spectrometers for all our Bragg crystals?  As Nicholas pointed out at QMA-2019, at least the EPMA doesn't have a variable working distance and a tilt stage, two sources of take off angle error, and hence k-ratio measurement error that SEMs are all blessed with.

And then there's the question of knowing our knowns!  A lot of people say they have San Carlos olivine or Amelia albite, this or that hornblende, but what do they actually have?  A natural material that has (un)documented variation in composition and a number of various inclusions.

So I whole heartedly agree that the construction of a k-ratio database is a great idea, but we should start by performing a few "quality control" checks on each instrument, before any k-ratios from that instrument are compiled into a database.

Here's a start: let's purchase/synthesize/beg/borrow/steal some simple, synthetic materials that would be a suitable test of of our instruments' ability to make a k-ratio measurement.  Let's start simple, for example just one possibility off the top of my head, let's measure a pure synthetic Al2O3 single crystal, and a pure synthetic spinel (MgAl2O4) single crystal, both available in kilogram quantities and both beam stable.  Let's measure Al Ka in both materials (which is significantly absorbed by Mg in the spinel), and let's see if each of our spectrometers produce the same k-ratios. 

We can bring LiF and PET crystal into play by utilizing other elements, but consider this: for the purposes of a k-ratio comparison between spectrometers or between different instruments, it really doesn't matter what the exact compositions of our two materials are, so long as we are measuring the exact same two compositions. If the two materials are exactly the same compositionally, the k-ratios from all spectrometers, on all instruments (assuming a nominal 40 degrees takeoff), should be the same within precision (let's ignore the differences in effective electron landing energy from different carbon coating thickness, etc. for now, by utilizing significant overvoltages on our measurements!).

On the other hand, obviously, if we picked two materials for which we know exactly their compositions (via considerations of crystallinity, purity, homogeneity, thermodynamics or whatever), we would have a good first candidate for our k-ratio composition database.

This is just a first cut, but I suggest we start thinking about some high purity, beam stable, synthetic single crystal materials which are available already in kilogram quantities, from say laser optics crystal manufacturers for example (just their "cutoff" material as we obtained for our RbTiOPO4 material), and see which ones are good candidates for testing our k-ratio machines.

[Probeman]

  The idea here is to "share k-ratios not materials."  Any standard that anyone anywhere had measured an entered into the database could be used in your measurement without having that material in your lab.
  Historically, if we want to make a measurement against Kakanui hornblende (KH), we need a chip or two in our lab.  But this isn't necessary.  If we have a database of k-ratios of KH against Si, Al, Fe, Mg, CaF2, Albite, etc and we make measurements in our lab using Si, Al, Fe, ... We can use the consensus k-ratios to convert our simple Si, Al, Fe... k-ratios into k-ratios relative to KH and then use these to measure our unknown against KH.  (Furthermore, the spread of the k-ratios in the database could provide information about the standard's heterogeneity.)

Nicholas if I read you correctly, you're basically saying we replace our secondary standards with k-ratios from a community database. So in our labs we'd only be measuring simple primary standards, say pure metals, against our unknown intensities.

I guess my main objection to this is that it eliminates the accuracy check we get from our secondary standards. Are our spectrometers truly linear in their intensity response?  You know dead time, count rate gain...

Also it essentially makes EPMA a calibration curve method.  I've got nothing against calibration curves of course, but I have to admit, that one thing I've always liked about microanalysis is that it's all physics based.

You've given much food for thought!   Whatever happens in the long run, there's certainly nothing wrong with starting some k-ratio measurements on readily available shared pure synthetics (as I posted above), just to see if we're all getting the same k-ratios from our spectrometers/instruments. For a start at least.

[NicholasRitchie]

You have a good point about secondary standards as a QC check.  However, it doesn't have to be the standard that you are using for your measurement.  Plus a secondary standard is only sensitive to certain types of errors - for example if the spectrometer focus is off both the primary and secondary standard will agree even though throughput is poor.

I suspect that in time, the database would become its own QC check.  Labs would evaluate their instrument against consensus k-ratios (on other materials) before submitting new k-ratios (for a new material) - just out of an abundance of caution.  Initially there might be some spurious k-ratios from labs that didn't realize they had a problem but these would quickly become evident.

Initially, maybe we could we consider measuring just one material (possibly Doug Maier's ADM glass) to evaluate our instruments.  When we came to consensus on these k-ratios we could move on from there.

[Probeman]

You have a good point about secondary standards as a QC check.  However, it doesn't have to be the standard that you are using for your measurement.  Plus a secondary standard is only sensitive to certain types of errors - for example if the spectrometer focus is off both the primary and secondary standard will agree even though throughput is poor.

I suspect that in time, the database would become its own QC check.  Labs would evaluate their instrument against consensus k-ratios (on other materials) before submitting new k-ratios (for a new material) - just out of an abundance of caution.  Initially there might be some spurious k-ratios from labs that didn't realize they had a problem but these would quickly become evident.

Initially, maybe we could we consider measuring just one material (possibly Doug Maier's ADM glass) to evaluate our instruments.  When we came to consensus on these k-ratios we could move on from there.

OK. I agree with the QC stuff, but on the ADM glass measurement don't we need two materials for a k-ratio?

[Probeman]

It also occurs to me that having such a k-ratio database would be an ideal data set for training a "microanalysis" neural net for quantitative analysis.

Of course if such a microanalysis neural net eventually became successful, it would essentially mean the end of science (physics) in microanalysis (as we know it), because a neural net is essentially a completely opaque calibration curve. To us humans at least.

[NicholasRitchie]

Certainly my management likes that aspect of the k-ratio database.  It is part of how I sell it to them.  However, I believe the phase space is too large and regardless of how large the database becomes it will remain too small to train a neural network.  Microanalysis will likely remain safe from the mindless AI.

[Probeman]

Certainly my management likes that aspect of the k-ratio database.  It is part of how I sell it to them.  However, I believe the phase space is too large and regardless of how large the database becomes it will remain too small to train a neural network.  Microanalysis will likely remain safe from the mindless AI.

I agree, but it is fun to speculate. After all, at one time no one thought AI could win at chess (and now Go) ever!

[Probeman]

Here's a start: let's purchase/synthesize/beg/borrow/steal some simple, synthetic materials that would be a suitable test of of our instruments' ability to make a k-ratio measurement.  Let's start simple, for example just one possibility off the top of my head, let's measure a pure synthetic Al2O3 single crystal, and a pure synthetic spinel (MgAl2O4) single crystal, both available in kilogram quantities and both beam stable.  Let's measure Al Ka in both materials (which is significantly absorbed by Mg in the spinel), and let's see if each of our spectrometers produce the same k-ratios. 

We can bring LiF and PET crystal into play by utilizing other elements, but consider this: for the purposes of a k-ratio comparison between spectrometers or between different instruments, it really doesn't matter what the exact compositions of our two materials are, so long as we are measuring the exact same two compositions. If the two materials are exactly the same compositionally, the k-ratios from all spectrometers, on all instruments (assuming a nominal 40 degrees takeoff), should be the same within precision (let's ignore the differences in effective electron landing energy from different carbon coating thickness, etc. for now, by utilizing significant overvoltages on our measurements!).

On the other hand, obviously, if we picked two materials for which we know exactly their compositions (via considerations of crystallinity, purity, homogeneity, thermodynamics or whatever), we would have a good first candidate for our k-ratio composition database.

This is just a first cut, but I suggest we start thinking about some high purity, beam stable, synthetic single crystal materials which are available already in kilogram quantities, from say laser optics crystal manufacturers for example (just their "cutoff" material as we obtained for our RbTiOPO4 material), and see which ones are good candidates for testing our k-ratio machines.

OK, it seems we have a lack of excitement for this project so I'm going to start locating some pure synthetic materials, Al2O3, MgO, MgAl2O4, etc. and start making up some mounts (with Julie Chouinard's help!).  They can all be carbon coated at the same time for reproducibility so that should not matter too much as long as our overvoltages are significant.

Question to the community: should we circulate a single mount, or should there be a number of replicate mounts that are circulated?