Geological CPX/OPX thermo/barometers and accuracy of EPMA

[Probeman]

I have recently been contacted by a post-doc asking about accuracy issues in EPMA and how they might relate to problems they are seeing in the P/T results in pyroxene systems. These are excellent questions! The following is part of my response to these questions.

I didn't realize that the Na concentrations for these OPX/CPX thermo/barometers were so low.  Knowing this now, here's the deal with EPMA accuracy (assuming no instrumental problems!):

1. For major concentrations what matters are the standard accuracy and the matrix correction accuracy. I don't think the matrix correction accuracy is a problem with modern software (see below).  But the heterogeneity in the Smithsonian standards is a well known problem. In fact I think Ed Vicenzi has characterized some of these issues in the past.

2. For trace concentrations what matters are the background corrections (and interference corrections if necessary). It is interesting to note that labs using JEOL EPMA instruments (which has roughly an 80% market share) have no way to correct for spectral interferences.  That is unless they utilize software from Probe Software.

Bottom line: the above two points make sense if you consider that the matrix corrections are multiplicative and scales (as does standard accuracy) with concentration (so the smaller the concentration the smaller the absolute accuracy error). That is, a 1% relative error in accuracy is 1 wt% at 100 wt%, but only .1 wt% at a 10 wt% concentration, and only 1 PPM at 100 PPM.

While the background correction is subtractive and is a constant source of error (so the smaller the concentration the larger the accuracy error). That is, a 100 PPM error in the background estimation is a 10% error at a 1000 PPM concentration, but it's a 100% error at 100 PPM and a 1000% error at 10 PPM. This is worth thinking about a bit- it's not obvious. If it was obvious I wouldn't be having to rant about this for so many years!  😁

For minor elements it's a bit of both considerations (standards/matrix and also backgrounds/interferences), but mostly background correction, with some influence from the matrix correction. And also spectral interferences depending on the elements in question. For this system I don't think spectral interferences are an issue, as the only other elements that would interference seriously with Na are Zn and a little from P (but these should elements be present in very low concentrations in most augite compositions, I think?).

Some might suggest a low Na pyroxene standard might fix this accuracy problem and that is the traditional, but wrong approach in my opinion. For highest minor/trace element accuracy in pyroxenes what we should want, is an approximately matrix matched standard with a *zero* Na concentration.  A non-zero minor/trace (and homogeneous) element standard is very difficult to obtain and to characterize.

What is much easier (and better) is a "blank" standard roughly similar to ones unknown (or even a glass with similar elements) that has *zero* (say, less than 1 PPM) Na present. This we can determine easily (e.g., ICP-MS, SIMS) and then our minor/trace accuracy is as good as our measurement precision (think about this, as it's a gift from the science gods).  Basically we are simply testing our background correction accuracy by using a suitable blank standard run as an unknown. This is all spelled out in the Donovan et al., 2011 Amer. Min. paper.

In the Probe for EPMA software, this "blank" correction is applied iteratively during the matrix correction for highest accuracy. It can also be subtracted out after the fact in an Excel spreadsheet manually, as long as the composition doesn't change too much from the blank correction. This is usually the case.

I know this is not what we were taught in grad school, but it's the way forward in this current mess of heterogeneous natural standards. As an example, I couldn't find a pyroxene analysis, but here's an orthoclase analysis with around .8 wt% Na showing all the various matrix corrections in the CalcZAF software:

Elemental Weight Percents:
ELEM:       Na      Si       K      Al      Mg      Ca      Ti      Mn      Fe       P       O       H      Ba   TOTAL
     1    .768  30.520  12.862   8.625    .000    .014    .001   -.005   1.355    .000  45.798    .000    .054  99.991   Armstrong/Love Scott (default)
     2    .764  30.621  12.864   8.571    .000    .014    .001   -.006   1.383    .000  45.798    .000    .054 100.064   Conventional Philibert/Duncumb-Reed
     3    .768  30.419  12.861   8.623    .000    .014    .001   -.005   1.335    .000  45.798    .000    .054  99.869   Heinrich/Duncumb-Reed
     4    .767  30.577  12.863   8.594    .000    .014    .001   -.006   1.356    .000  45.798    .000    .054 100.017   Love-Scott I
     5    .768  30.547  12.863   8.621    .000    .014    .001   -.006   1.355    .000  45.798    .000    .054 100.015   Love-Scott II
     6    .756  30.879  12.866   8.561    .000    .014    .001   -.006   1.375    .000  45.798    .000    .054 100.298   Packwood Phi(pz) (EPQ-91)
     7    .763  30.668  12.862   8.497    .000    .014    .001   -.005   1.343    .000  45.798    .000    .054  99.995   Bastin (original) Phi(pz)
     8    .768  30.561  12.862   8.547    .000    .014    .001   -.006   1.378    .000  45.798    .000    .054  99.978   Bastin PROZA Phi(pz) (EPQ-91)
     9    .767  30.616  12.863   8.554    .000    .014    .001   -.006   1.377    .000  45.798    .000    .054 100.037   Pouchou and Pichoir-Full (PAP)
    10    .764  30.630  12.863   8.570    .000    .014    .001   -.006   1.378    .000  45.798    .000    .054 100.067   Pouchou and Pichoir-Simplified (XPP)

AVER:     .765  30.604  12.863   8.576    .000    .014    .001   -.006   1.364    .000  45.798    .000    .054 100.033
SDEV:     .004    .119    .001    .041    .000    .000    .000    .000    .017    .000    .000    .000    .000    .109
SERR:     .001    .038    .000    .013    .000    .000    .000    .000    .005    .000    .000    .000    .000

MIN:      .756  30.419  12.861   8.497    .000    .014    .001   -.006   1.335    .000  45.798    .000    .054  99.869
MAX:      .768  30.879  12.866   8.625    .000    .014    .001   -.005   1.383    .000  45.798    .000    .054 100.298

As we can see, the variance in the Na matrix corrections are very small (40 PPM), so the main problem with Na concentrations in the cpx/opx barometer is the heterogeneity of the Smithsonian "standard", and the background corrections, where I'm guessing many of us haven't tested our ability to measure zero Na in a roughly similar pyroxene matrix.

I did a quick search in my standard database to the Kakanui augite compositions and here is what I came up with:



So this Dahlheim glass is a close match, but this was a glass produced by Dan Weill at Oregon, so yes, I have a fair amount of it (which I shared with Paul Carpenter at Washington University), but is the Na content truly zero?  I don't know, but it could be characterized using ICP-MS/SIMS... anyone interested in doing this for us?

The other close match to an augite composition is the NIST K-412 glass which nominally contains no Na, but I seem to have reported some 500-600 PPM of Na in this material some decades ago. Is that accurate?  I have no idea.  NIST does not report it in their certificate.  Has anyone out there measured trace elements in K-412 (or K-411) using ICP-MS or SIMS?

[Julien]

Hi John,

I agree with you, the Kakanui augite has some natural variation... We regularly use it as a secondary standard (NEVER as a primary!) for checking our regular pyroxene analyses (along with other secondary standard). We don't rely on the published value, but rather on the reproducibility over time (session after session...). Data quality is somehow ensure by looking at the totals and also the atomic proportion. From the analyses we have, it looks like there are variations of Al and Fe-Mg content. However, it **seems** that the Na-content is quite homogeneous around 1.3 wt% Na2O (or ca. 0.95 wt% Na). I should compile all the data my students and I have one day...

Regarding your second question, if you send me a chip of your glass, I can have it run for LA-ICP-MS one of these days and check the Na-content (along with other traces)... If you have a few other materials that could become great standards, I can also put them in the run! I'm not managing the LA-ICP-MS lab, but my colleague Marcel regularly has a generic setting with almost all lithophile elements. Unfortunately it doesn't seem to perform very well for Na content with detection limits around 25-30 ppm. I think this has to do with isobaric interferences... Hopefully he can improve this; I will check with Marcel when I see him.

Best,

Julien

[sem-geologist]

Just to add my 2 cents to these issues...
We can try to do as precise and accurate analysis (with PfS or without it) especially that we know what to look for. However, other nonetheless important factor is that lots of these P-T calibration curves are produced on analyses of dubious quality... Or at least lacking meticulous methodology. Some discrepancies can be result of unaccounted analytical artefacts and biases. And lots of available P-T calibration curves/equations depends on analysed mineral compositions by EPMA from different kind of P-T experiments. So in the end increasing precision and accuracy of our EPMA results would not lead automagically to better and more sensible P-T recalculations.

Another cent from me is that I appallingly witnessed countless times (theoretically I am geologist BTW, specializing in igneous petrology) when obtained very logic and valid P-T of igneous system at given time-and-space is tried to be interpreted with non-applicable metamorphic petrology mind-set which leads to discarding very valid calculated values of P-T and criticizing model being not good at given range or completely wrong. The first mistreated P-T which comes to my mind is Ridolfi and Renzulli (2012) single amphibole thermo-chemo-barometer, where so far I saw all kind of critique do not recognize that P-T can increase enormously in igneous system without any changes in depth of studied mineral.

Some might suggest a low Na pyroxene standard might fix this accuracy problem and that is the traditional, but wrong approach in my opinion. For highest minor/trace element accuracy in pyroxenes what we should want, is an approximately matrix matched standard with a *zero* Na concentration.  A non-zero minor/trace (and homogeneous) element standard is very difficult to obtain and to characterize.

You opinion is right in so many levels. But differently to the approach with blank standard as standard, as a non-user of PfS I don't use any analytical blind standard but many blind standards for visual setting of single off-peak with a precise slope, which fits all those blind standards of various atom masses off all possible elements in that species of mineral. Often geological minerals tend to be quite extreme zonic (particularly in volcanic environment) and thus it is not so possible to have single matrix-matched blind standard. Also my approach does not require any LA-ICP-MS.

[Probeman]

I agree with you, the Kakanui augite has some natural variation... We regularly use it as a secondary standard (NEVER as a primary!) for checking our regular pyroxene analyses (along with other secondary standard). We don't rely on the published value, but rather on the reproducibility over time (session after session...). Data quality is somehow ensure by looking at the totals and also the atomic proportion. From the analyses we have, it looks like there are variations of Al and Fe-Mg content. However, it **seems** that the Na-content is quite homogeneous around 1.3 wt% Na2O (or ca. 0.95 wt% Na). I should compile all the data my students and I have one day...

Hi Julien,
I would be very interested in data demonstrating the homogeneity/inhomogeneity of Na in the Kakanui augite. Not just within a grain, but from grain to grain (because every lab is using a different grain!).

Regarding your second question, if you send me a chip of your glass, I can have it run for LA-ICP-MS one of these days and check the Na-content (along with other traces)... If you have a few other materials that could become great standards, I can also put them in the run! I'm not managing the LA-ICP-MS lab, but my colleague Marcel regularly has a generic setting with almost all lithophile elements. Unfortunately it doesn't seem to perform very well for Na content with detection limits around 25-30 ppm. I think this has to do with isobaric interferences... Hopefully he can improve this; I will check with Marcel when I see him.

I would very much appreciate this! I will look to see if I can locate the material. I think I set a small amount aside when I sent the bulk of the material to Paul Carpenter.

[BenjaminWade]

Hi all
Just my two cents re: LA-ICP-MS, you wont be able to get down to those Na levels required for testing a blank standard. It has very high background levels in all the tubing and interface, which is why the detection limits are quite poor.

Cheers

[Probeman]

Just my two cents re: LA-ICP-MS, you wont be able to get down to those Na levels required for testing a blank standard. It has very high background levels in all the tubing and interface, which is why the detection limits are quite poor.

Would that still be the case for normal ICP-MS do you think?

[Probeman]

I'm beginning to suspect after talking to a number of colleagues that the main problem with quantification of Na in these thermo/barometers is the Na primary standard.

Many labs are using a natural jadeite. But natural jadeite tends to very inhomogeneous, so why use it?

Well Ed Vicenzi correctly points out that it does have a higher Na concentration of Na than say albite, but when measuring ~1 wt% Na, that shouldn't matter very much. It's still an interpolation. Our lab uses a natural nepheline as a primary Na standard and actually it has a slightly higher Na concentration than our natural jadeite:

St  336 Nepheline (partial anal.)
TakeOff = 40.0  KiloVolt = 15.0  Density =  2.600  Type = silicate  Mount = alkali, carbonate

Analysis by ISE Carmichael (Na, K)
Ca = 750 PPM (EPMA by JJD)
Oxide and Elemental Composition

Average Total Oxygen:       44.418     Average Total Weight%:  100.054
Average Calculated Oxygen:  44.450     Average Atomic Number:   11.037
Average Excess Oxygen:       -.032     Average Atomic Weight:   20.707

ELEM:     Na2O     K2O    SiO2   Al2O3     FeO       O     CaO
XRAY:      ka      ka      ka      ka      ka      ka      ka
OXWT:   16.920   5.610  43.491  33.761    .199   -.032    .105
ELWT:   12.552   4.657  20.329  17.868    .155  44.418    .075
KFAC:    .0735   .0409   .1500   .1333   .0013   .2112   .0007
ZCOR:   1.7068  1.1395  1.3548  1.3408  1.1894  2.1027  1.1090
AT% :   11.299   2.465  14.980  13.705    .057  57.454    .039
24 O:    4.720   1.030   6.257   5.725    .024  24.000    .016

St  316 Jadeite (san benito)
TakeOff = 40.0  KiloVolt = 15.0  Density =  3.300  Type = pyroxene  Mount = hydrous

Contains interstitial albite?
(Na2O = 14.02%, K2O=0.06%, by Flame Photometry, by J. Hampel)
Oxide and Elemental Composition

Average Total Oxygen:       47.486     Average Total Weight%:  100.001
Average Calculated Oxygen:  47.489     Average Atomic Number:   10.684
Average Excess Oxygen:       -.003     Average Atomic Weight:   20.221

ELEM:     SiO2   Al2O3    TiO2     FeO     MnO     MgO     CaO    Na2O     K2O       O
XRAY:      ka      ka      ka      ka      ka      ka      ka      ka      ka      ka
OXWT:   59.453  25.221    .000    .021    .000    .010    .020  15.220    .060   -.003
ELWT:   27.790  13.348    .000    .016    .000    .006    .014  11.291    .050  47.486
KFAC:    .2124   .1007   .0000   .0001   .0000   .0000   .0001   .0660   .0004   .2517
ZCOR:   1.3083  1.3257  1.1884  1.1951  1.2136  1.5263  1.1065  1.7101  1.1524  1.8864
AT% :   20.008  10.003    .000    .006    .000    .005    .007   9.931    .026  60.014
24 O:    8.001   4.000    .000    .002    .000    .002    .003   3.972    .010  24.000

And our nepheline is *much* more homogeneous than our jadeite.  What primary standard does your lab use for Na?

Another possible reason Ed mentioned is the beam sensitivity of say, albite compared to jadeite.  No question jadeite is a less sensitive Na standard. Of course the TDI correction can help in this case, but how much of an effect is this beam sensitivity? Well of course it's going to depend on the beam current and the beam size.

Next we will take a look at some TDI data on these standards if I can find any!

[Anette von der Handt]

Very interesting discussion and the problem of interlaboratory bias, either stemming from unreliable standards and/or the variability in -let's call it - analytical 'rigor" between labs has been a longtime problem.

The problem with MIT data in particular has been occasionally mentioned before. I think I saw it first in a 1996 paper by Yang et al. that states:  "An interlaboratory comparison has been made (Reynolds 1995) including MIT, the Smithsonian Institution in Washington, Lamont Doherty and University of Hawaii. It is the practice in our laboratory to correct microprobe data obtained elsewhere to an MIT reference before making thermobarometric or modeling analyses (see Table 1). Although Grove et al. (1992) neglected to discuss this issue, the Smithsonian data discussed in that paper was corrected before plotting and estimation of crystallization pressure. Failure to do so can result in significant errors, and is most commonly evident as a discrepancy in the pressures estimated from the different equations."
I attached the paper.

I think there are also some small remarks in earlier papers by the Langmuir group at Harvard. We ran into this problem as well for a study from one of my MSc students (Voigt et al., 2017, Lithos) and had to correct the literature data to be able to compare various geobarometers.

I would also point you to the large study by Allison Gale (Gale et al.,  2013 The mean composition of ocean ridge basalts - Gale - 2013 - Geochemistry, Geophysics, Geosystems - Wiley Online Library) that touches upon interlaboratory bias if you haven't seen it yet (see attached table).

[Anette von der Handt]

Here is the table of interlaboratory bias factors, showing only the values for EPMA labs.

[Anette von der Handt]

Regarding the Kakanui Augite:

Huebner and Woodruff (U. S. Geological Survey Open File Report 85-718, Chemical Compositions and Critical Evaluation of Microprobe Standards Available in the Reston Microprobe Facility) state:

"The classical or wet chemical analyses of the Kakanui augite have sums that are slightly high, particularly when the trace elements are included. The preferred analysis (3) incorporates revised values of Al₂0₃, Fe₂0₃, and FeO. The revised Al₂0₃ and total iron are substantiated by Wiggins 1 microprobe values. The preferred analysis can be recalculated to a perfectly stoichiometric pyroxene without any adjustment of the ferrous/ferric ratio!

(Na,Ca,Mn,Sr,Co,Ni,Fe,Mg)_1.000(Mg,V,Ti,Fc,Cr,Al)_1.000(Al,Si)_2.000O₆ The augite from Kakanui is homogeneous with respect to all its major elements PXKA should be an excellent standard and superior known-unknown for major elements in pyroxenes. It has not been used as much as it deserves."

I attached the relevant pages.

John Fournelle looked more recently also into the variability of Kakanui Augite and three other Smithsonian pyroxene standards. He looked at 54 grains in total.

https://ui.adsabs.harvard.edu/abs/2012AGUFM.V23C2827F/abstract

I am sure he can weigh in on the variability of sodium in particular. However, the Kakanui augite should not be used as a primary standard for sodium anyway because of its low abundances.

I would be interested to hear if anyone knows of a homogeneous Jadeite. I recently mounted a Jadeite from Japan (#108468, no further information at hand) and it is god awful.

I am overall happy with my Amelia Albite but I have to re-polish every 6-9 months in my experience to remain happy.

Has anyone experience or thoughts on Sodalite?

[Probeman]

Here is the table of interlaboratory bias factors, showing only the values for EPMA labs.

I note that MIT had the 2nd highest Al, the highest Ca and the 4th lowest Na values. So in this table 1.000 indicates no bias from the average of reported results or from the "accepted" values? 

[Probeman]

I would be interested to hear if anyone knows of a homogeneous Jadeite. I recently mounted a Jadeite from Japan (#108468, no further information at hand) and it is god awful.

My experience as well.

I am overall happy with my Amelia Albite but I have to re-polish every 6-9 months in my experience to remain happy.

I think it would be better than jadeite, but I think nepheline is a better Na primary standard if one can obtain it.  Wondering if pure end member nepheline can be synthesized?

Has anyone experience or thoughts on Sodalite?

Very beam sensitive and inhomogeneous in my experience.

[Probeman]

Very interesting discussion and the problem of interlaboratory bias, either stemming from unreliable standards and/or the variability in -let's call it - analytical 'rigor" between labs has been a longtime problem.

The problem with MIT data in particular has been occasionally mentioned before. I think I saw it first in a 1996 paper by Yang et al. that states:  "An interlaboratory comparison has been made (Reynolds 1995) including MIT, the Smithsonian Institution in Washington, Lamont Doherty and University of Hawaii. It is the practice in our laboratory to correct microprobe data obtained elsewhere to an MIT reference before making thermobarometric or modeling analyses (see Table 1). Although Grove et al. (1992) neglected to discuss this issue, the Smithsonian data discussed in that paper was corrected before plotting and estimation of crystallization pressure. Failure to do so can result in significant errors, and is most commonly evident as a discrepancy in the pressures estimated from the different equations."
I attached the paper.

I think there are also some small remarks in earlier papers by the Langmuir group at Harvard. We ran into this problem as well for a study from one of my MSc students (Voigt et al., 2017, Lithos) and had to correct the literature data to be able to compare various geobarometers.

I would also point you to the large study by Allison Gale (Gale et al.,  2013 The mean composition of ocean ridge basalts - Gale - 2013 - Geochemistry, Geophysics, Geosystems - Wiley Online Library) that touches upon interlaboratory bias if you haven't seen it yet (see attached table).

At least one issue with the MIT EPMA lab is their "procedure" for standardization. Apparently they:

1. Acquire a single point on their primary standard (already I'm starting to worry!).
2. Acquire a single point on their secondary standard (now I'm really worried!)
3. Analyze that single point on their secondary standard, and if they do not obtain the expected composition of the secondary standard...
4. They then edit the primary standard intensity raw data until they do (OK, now the alarm bells in my head are ringing!).

Yes, you read that correctly: they *edit* the raw data intensity of the primary standard...

I'm sure everyone can understand why this "procedure" might be problematic. First we need more than a single point per sampling, to obtain an average and observe the variance. A single point has a zero variance and we won't know if that single point is simply an outlier due to some instrumental or sampling effects.

Second, no one should be editing their standard (or unknown for that matter) raw intensities to obtain an "expected" result! We should modify our models to fit our data, not modify our data to fit our models!   

Third, by "nulling out" the check on accuracy by using a secondary standard, they are now essentially removing their ability to ascertain their accuracy levels. Why even bother using a primary standard?

Consider the following: let's assume they are lucky and the single point measurement of their primary intensity is good, now what if the secondary standard (kakanui augite?) is somewhat inhomogeneous? Or that single grain does not represent the wet chemistry as reported by Gene Jarosewich?  By editing the primary standard intensities, they are normalizing the calibration to that single point on that single grain, whether or not it's representative of the average wet chemistry.

No wonder things are problematic at "Boston Tech".

[BenjaminWade]

Just my two cents re: LA-ICP-MS, you wont be able to get down to those Na levels required for testing a blank standard. It has very high background levels in all the tubing and interface, which is why the detection limits are quite poor.

Would that still be the case for normal ICP-MS do you think?

Hey sorry for the delay. Traditional solution ICP-MS will give you better DL than laser, but it requires a clean instrument and clean acids. Most of the sodium sits in the tubing and on the back of the interface cones and lenses.  What kind of Na level are you needing to characterize a "blank" sample? I am assuming something along the lines of single digit ppm, ideally sub ppb like the Ti in quartz.

To give you an idea if our instrument is really clean (and hasn't been sucking seawater through it any time recently), we can do single digit ppb measurements. However this is on a diluted sample (perhaps 500x to 1000x diluted), which would correct back up to a sample that has ~0.5-1ppm Na. The samples need to be diluted otherwise the major components of the solution cause plasma loading and signal suppression. For better counting statistics you may get away with a smaller dilution if you wanted to get as precise a measurement as you can. However it would depend on how much signal suppression (plasma loading) you get from the major elements in a pyroxene matrix (ie mostly Fe,Mg,Si), and thus the required internal standard correction back to 100% might be a large correction introducing more uncertainty.

I guess one thing to bear in mind is that of course this would be a bulk analysis as well, so would be impossible to determine microscale heterogeneity using solution ICP-MS.

Cheers

[Probeman]

This sounds very useful.

The ultimate detection limit for EPMA has got to be around ~1 to 2 PPM (or worse).

For example when we did our trace element study (Donovan et al., 2011, Amer. Min.) we tuned up 5 spectrometers for an aggregate Ti analysis and got a t-test detection limit of around 2 to 3 PPM at high beam currents and long counting times.

And Ti (Ka) is pretty easy compared to many emitters and SiO2 is a relatively low Z matrix, so anything else under 2 PPM is probably going to essentially be a big fat zero (blank) for EPMA.

If only someone somewhere could grow high purity synthetic (end member) nepheline (NaAlSiO4) and/or leucite  (KAlSi2O6) synthetic crystals!  The web seems to disagree on the exact formulas, so I guess I should ask, are synthetic end members of Na and K aluminum silicates possible?