Apparent shift in Mg concentrations in standards on a new microprobe

[Joe Boesenberg]

John,

I don't know how varied your experience is with different microprobes, but I have a couple questions for you and the forum. I can see that deadtimes can effect the outcome of the concentration of an element to a certain degree. As I mentioned, I am having issues reproducing Mg values on all of my standards with my new JEOL microprobe on both TAP spectrometers, one with LTAP and the other with small TAP. They all come out low when using Fo100 or Fo97 as the calibration standard (both of these standards are synthetic). I will be running tests to check other TAP elements over the next couple days. 

Pete McSwiggen at JEOL thinks that now that the deadtimes are tuned on my new JEOL my problems should be over. He thinks there is an issue with my Fo97 and possibly the Fo100 standards, which I find very hard to believe. Though the deadtime correction tuning helped, it did not eliminate the low Mg problem. It just reduced the magnitude of the shift a bit. In the past, I have worked on an JEOL 8600, an ARL-SEMQ, and two Cameca SX-100. All four were not tuned for deadtimes beyond the manufacturers setting. If I remember correctly the SX-100s had flat deadtimes of 3ms for all spectrometers. I have been using the Fo97 as the calibration standard and using the same San Carlos Olivine (NMNH) and Lunar Crater Augite (NMNH), both Jarosewich standards, as reference standards for the last 28 years (on the ARL, and both Cameca probes). All three standards are very homogeneous. I have never had this issue on the past three probes.

Could the tuning of the deadtimes on my new JEOL and changing of instruments from the Cameca to the JEOL change the compositions of my standards by a full 0.50 wt% MgO in the San Carlos (nominal 49.4 wt% MgO, but now analyzing around 48.9) or 0.2 wt% MgO in the Augite (nominal 17.35 wt% MgO, now analyzing around 17.15) while using the identical calibration setup?

I have seen in the past round robin sessions where different labs get different compositions for the same reference standard material, but always assumed this was caused by the use of different standards in the setup routine. Is this a similar case?

What is your opinion?

P.S. I realized I had a pure MgO standard in my collection, and as you recommended, I used it as the calibration standard last night. I analyzed the Fo100, the Fo97, my San Carlos olivine, my Lunar Crater augite, a synthetic MgAl2O, a synthetic MgCr2O4, and the pure MgO standard as unknowns. I fixed the compositions for all of the standards and only analyzed Mg to minimize any secondary problems. Analysis of the pure MgO, came out low (average of 99.1 total). I suspect there was some adsorbed water under the carbon coat, so Pete recalculated it back to 100% (multiplied the results by 1.009). The results for the other standards came out surprising to me (after correction). The Fo100 came out high (58.26 instead of 57.3 MgO). The Fo97, the San Carlos, and the Augite all came out nearly perfect, but the MgAl2O4 came out low (27.96 instead of 28.33) and the MgCr2O4 came out very low (20.05 instead of 20.96). I am amazed the strict synthetics do not reproduce nearly as well as the natural samples, but I don't understand why. Shouldn't the synthetics come out on the same trend between zero and the pure MgO by default? 

[JonF]

In your "PS" bit, you say that you standardised on pure MgO and remeasured that same MgO as an unknown but came back with 99.1% totals?

Have you checked that the MgO standard was set to use MgO as the standard?
Or that there were no bad points in the standard data set?

I assume the same beam conditions were used for both standard and unknown, and there were no major changes in ambient air pressure or temperature in the lab between standardising and measuring the MgO as an unknown?

[Joe Boesenberg]

Forgot to mention, I did the analysis through the JEOL software, not PfE. I am brand new to PfE and am still using to use the basic features. With the new probe I am trying to learn PfE, the JEOL software and the Bruker EDS software simultaneously. It is a but overwhelming. So there was no setting to do.

There were no bad points, but as I mentioned, I am pretty sure the standard had water trapped under the carbon coat. Same conditions for standard and sample. Whole run was done in less than two hours, and no change in the room conditions. The only proviso is I don't know where the MgO standard was made, so I don't know about its quality.         

[Probeman]

I don't know how varied your experience is with different microprobes, but I have a couple questions for you and the forum. I can see that deadtimes can effect the outcome of the concentration of an element to a certain degree. As I mentioned, I am having issues reproducing Mg values on all of my standards with my new JEOL microprobe on both TAP spectrometers, one with LTAP and the other with small TAP. They all come out low when using Fo100 or Fo97 as the calibration standard (both of these standards are synthetic). I will be running tests to check other TAP elements over the next couple days. 

Me?  I started on an ARL SEMQ but purchased and performed acceptance testing on two Cameca instruments.

https://smf.probesoftware.com/index.php?topic=924.msg9758#msg9758

First question: when you analyze Fo100 against Fo97 do they agree with each other at high precision with high accuracy?

The real question on standards here is: "what is the truth"?

In my book I would trust a high purity single crystal synthetic MgO, Mg2SiO4 and/or Mg2Al2O4, assuming that they has been properly characterized for trace elements and gave consistent results from XRD.  Based on the FIGMAS round robin tests we did on such materials in 2023 or so, a group of us (on both JEOL and Cameca instruments) got excellent results analyzing MgAl2O4 using MgO and Al2O3 as primary standards:

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

even at 120 nA!  With dead times properly calibrated using the constant k-ratio method of course...

Pete McSwiggen at JEOL thinks that now that the deadtimes are tuned on my new JEOL my problems should be over. He thinks there is an issue with my Fo97 and possibly the Fo100 standards, which I find very hard to believe. Though the deadtime correction tuning helped, it did not eliminate the low Mg problem. It just reduced the magnitude of the shift a bit. In the past, I have worked on an JEOL 8600, an ARL-SEMQ, and two Cameca SX-100. All four were not tuned for deadtimes beyond the manufacturers setting. If I remember correctly the SX-100s had flat deadtimes of 3ms for all spectrometers. I have been using the Fo97 as the calibration standard and using the same San Carlos Olivine (NMNH) and Lunar Crater Augite (NMNH), both Jarosewich standards, as reference standards for the last 28 years (on the ARL, and both Cameca probes). All three standards are very homogeneous. I have never had this issue on the past three probes.

I would tend to agree. Especially at count rates below 30 kcps.

A couple of points though, first regarding dead time calibrations: the "manufacturers settings" are merely a starting point, especially if one is attempting to push accuracy below 2% relative. Especially on a Cameca EPMA which has dead time constants roughly twice the magnitude of JEOL EPMA instruments.  But of course it depends on the beam currents used, in other words count rates. The dead time effect is very dependent on count rates.  And the dead times change as ones detectors age over time.

On a large TAP crystal because of the relatively low sin theta of Mg Ka, it is easy to get count rates over 30 kcps at typical beam currents. We already know those count rates can be problematic using a traditional linear dead time correction:

https://academic.oup.com/mam/article-abstract/29/3/1096/7165464

Here is what Anette found on her JEOL iHP200F instrument:

Ti Ka dead times, JEOL iHP200F, UBC, von der Handt, 07/01/2022
Sp1    Sp2    Sp3    Sp4    Sp5
PETJ    LIFL    PETL  TAPL    LIFL
1.26  1.26    1.27    1.10    1.25        (usec) optimized using constant k-ratio method (six term expression)
1.52  1.36    1.32    1.69    1.36        (usec) JEOL engineer using traditional method

Second, we already know that the San Carlos olivine is a crap standard because it includes a wide range of compositions as documented by John Fournelle and Will Nachlas.  Even Gene Jarosewich who performed the original wet chemistry on this material commented in his reports that one must average intensities from many different grains in order to obtain an accurate measurement!

As for Fo97, where was this material obtained?  You mentioned it was synthetic, but this sounds like a "doped" (impure) material and therefore it cannot be trusted. Yes, one could characterize it using a technique which is known to be accurate but then there's the question of homogeneity. Therefore my suggestion to utilize instead *high purity* end member synthetics.

Could the tuning of the deadtimes on my new JEOL and changing of instruments from the Cameca to the JEOL change the compositions of my standards by a full 0.50 wt% MgO in the San Carlos (nominal 49.4 wt% MgO, but now analyzing around 48.9) or 0.2 wt% MgO in the Augite (nominal 17.35 wt% MgO, now analyzing around 17.15) while using the identical calibration setup?

I need to see averages and standard deviations to judge these numbers. But depending on the difference in the count rates between the primary and secondary standard it could make a difference.  One way to mitigate this dead time dependency is to count at very low count rates, say 10 nA. Of course this means counting longer to obtain the necessary statistics for accuracy.

In fact, because most labs have not properly calibrated their dead time constants (e.g., have not utilized a non-linear dead time correction using the constant k-ratio calibration method), a beam current of 10 nA was recommended by Will Nachlas and Aurelien Moy for the MgAl2O4/MgO/Al2O3 round robin measurements.

For what it's worth, I would say is that if your count rates on your standards (for the high concentration elements), are below 30 kcps, the problem is probably not the dead time calibrations. Which is why Will and Aurelien suggested using 10 to 20 nA beam currents for these round robin tests.

The other thing I would do (which is easy with the single mount provided by Will and Aurelien), is to re-polish and re-coat both the primary and secondary standards at the same time.  I'm sure if you write Will, he would send you a mount to try...

Another point, which is important for Mg Ka because of the large absorption correction, is to check the effective take off angle for each spectrometer. This is done by measuring k-ratios on multiple spectrometers. A statistically significant difference in the k-ratios between spectrometers indicates a geometry problem either in the spectrometer alignments, stage or column.

This was found to be a serious problem at Cal Tech some years ago. First, that the "hot" Bragg crystals were diffracting asymmetrically (so those were swapped out), and second, the electron column was *not centered* in the spectrometer housing! They had to get a new spectrometer housing for their instrument which took JEOL 4 years to replace!

Here is a specification I utilized in both of my instrument acceptance tests at Berkeley and Oregon:

https://smf.probesoftware.com/index.php?topic=369.msg1948#msg1948

If one does not run this test, one will not know.  Here are the results from our SX100 testing at Oregon (see specification 13.3.9 on page 18 here):

https://epmalab.uoregon.edu/reports/Additional%20Specifications%20New.pdf

I have seen in the past round robin sessions where different labs get different compositions for the same reference standard material, but always assumed this was caused by the use of different standards in the setup routine. Is this a similar case?

Yes, this is a known problem that different EPMA labs produce *consistently* different results. Or at least it should be well known! This problem is documented in this recent paper:

https://academic.oup.com/petrology/article-abstract/64/2/egac126/6965244?redirectedFrom=PDF

The problem is that each lab utilizes primary standards that do not agree with one another. One culprit is San Carlos olivine as originally described by John Fournelle.

https://ui.adsabs.harvard.edu/abs/2009AGUFM.V31E2009F/abstract

and more recently here:

https://www.sciencedirect.com/science/article/pii/S0009254122002625

Your difficulty getting these standards to agree with each other is actually a service to our community (in spite of your pain!).  The fact that our (top!) labs obtain results that are consistently difference from each other should alarm all of us greatly.

Again, read the "Barometers behaving badly..." paper linked above.

The solution I think (after you have and re-polished and re-coated your standard materials (together!), and checked your dead times using constant k-ratios, and checked your spectrometer, stage and column geometries using simultaneous k-ratios, is to throw away these problematic materials and use high purity end member compositionally constrained synthetic standard materials.

I describe some of these efforts in this presentation:

https://www.youtube.com/watch?v=mOca7-G4FvQ&ab_channel=ProbeSoftwareInc

This is what Will Nachlas, John Fournelle, Aurelien Moy, Ed Vicenzi (and the FIGMAS group and I) have been advocating for decades, and possibly we are starting to make some small progress in this effort.

P.S. I realized I had a pure MgO standard in my collection, and as you recommended, I used it as the calibration standard last night. I analyzed the Fo100, the Fo97, my San Carlos olivine, my Lunar Crater augite, a synthetic MgAl2O, a synthetic MgCr2O4, and the pure MgO standard as unknowns. I fixed the compositions for all of the standards and only analyzed Mg to minimize any secondary problems. Analysis of the pure MgO, came out low (average of 99.1 total). I suspect there was some adsorbed water under the carbon coat, so Pete recalculated it back to 100% (multiplied the results by 1.009). The results for the other standards came out surprising to me (after correction). The Fo100 came out high (58.26 instead of 57.3 MgO). The Fo97, the San Carlos, and the Augite all came out nearly perfect, but the MgAl2O4 came out low (27.96 instead of 28.33) and the MgCr2O4 came out very low (20.05 instead of 20.96). I am amazed the strict synthetics do not reproduce nearly as well as the natural samples, but I don't understand why. Shouldn't the synthetics come out on the same trend between zero and the pure MgO by default? 

So MgO was the primary standard?  Do you know the Ca concentration and was it freshly polished?

Second, results should not be re-normalized like that! There's a problem there and it needs to be looked into... Second, without variances, it is difficult to evaluate these results. Remember just because a material claims to be stoichiometric and pure, does not mean that it is. That is why Will has been carefully characterizing small amounts of these materials using multiple techniques before purchasing larger amounts for distribution.

And yes, ideally these results should all on a line between a blank standard and the primary standard (ideally the pure end member), if these materials are re-polished and re-coated together and the instrument geometries are correct and the dead times well calibrated. Which is pretty much what we saw with the FIGMAS round robin testing.  When well characterized materials are compared we obtained excellent results. 

That is until each lab goes back to its "favorite" primary standards and then we are consistently different again!

See the use of the Evaluate application in your Probe Software software:

https://smf.probesoftware.com/index.php?topic=340.msg10757#msg10757

In summary:

1. Obtain carefully characterized high purity synthetic materials.

2. Re-polish and re-coat them together.

3. Check your dead time constants using the constant k-ratio method. Adjust the dead time constant until one obtains a constant k-ratio from low count rates to high count rates. This is documented from the Probe for EPMA Help menu.

4. Check your simultaneous k-ratios... do you obtain the same k-ratios within statistics?  If your spectrometers do not produce consistent k-ratios you have a spectrometer, stage or column geometry problem.

5. Note that one can measure constant k-ratios and simultaneous k-ratios if you can measure the same line on multiple spectrometers, say Ti Ka on LiF and PET crystals.  There are many posts about these measurements on this user forum.

6. Have a beer and relax.  This will all get figured out!

[Probeman]

I decided to go through my own data and found a couple of analyses using pure synthetic MgO as a primary standard (largest contaminant we found in that material was Ca at ~200 PPM).

So here is a natural (close to end-member) natural diopside that I developed many years ago (largest contaminant was Fe at ~500 PPM):

St  358 Set   1 diopside (Chesterman)
TakeOff = 40.0  KiloVolt = 15.0  Beam Current = 20.0  Beam Size =    5

St  358 Set   1 diopside (Chesterman), Results in Elemental Weight Percents
 
ELEM:       Si      Al      Fe      Mg      Cr      Ti      Mn      Ca       O   SUM 
   146  25.846    .019    .050  11.129   -.017   -.020    .014  18.377  44.316  99.715
   147  25.813    .024    .058  11.075   -.013   -.017    .009  18.492  44.316  99.756
   148  25.859    .016    .028  11.084    .031   -.040    .077  18.467  44.316  99.838
   149  25.890    .028    .070  11.127    .020    .009    .029  18.462  44.316  99.951
   150  25.831    .015    .058  11.039    .004    .000    .047  18.638  44.316  99.948

AVER:   25.848    .020    .053  11.091    .005   -.014    .035  18.487  44.316  99.842
SDEV:     .029    .006    .016    .038    .021    .019    .028    .095    .000    .108
SERR:     .013    .003    .007    .017    .009    .008    .012    .042    .000
%RSD:      .11   27.38   29.46     .34  401.67 -137.67   79.57     .51     .00

PUBL:   25.925    .016    .047  11.192    n.a.    .000    .000  18.489  44.316  99.985
%VAR:     -.30   27.92   12.68    -.90     ---     .00     .00  (-.01)     .00
DIFF:    -.077    .004    .006   -.101     ---    .000    .000   (.00)    .000
STDS:       14      13     395      12      24      22      25     358     ---

As one can see, the percent variance (%VAR:) from our "accepted" value is -0.90 % relative or -0.1% absolute (DIFF:) . Here's another secondary standard of synthetic Mg2SiO4:

St  273 Set   1 Mg2SiO4 (magnesium olivine) synthetic
TakeOff = 40.0  KiloVolt = 15.0  Beam Current = 20.0  Beam Size =    5

St  273 Set   1 Mg2SiO4 (magnesium olivine) synthetic, Results in Elemental Weight Percents
 
ELEM:       Si      Al      Fe      Mg      Cr      Ti      Mn      Ca       O   SUM 
   136  19.958    .006    .040  34.650   -.007   -.020    .008    .000  45.486 100.122
   137  19.954   -.011   -.022  34.794   -.013    .000    .033    .009  45.486 100.230
   138  19.901    .004    .024  34.639    .015    .000   -.007    .014  45.486 100.076
   139  19.948    .005    .013  34.678   -.004    .023   -.019    .005  45.486 100.135
   140  19.907   -.006   -.008  34.703   -.016    .042    .004    .013  45.486 100.125

AVER:   19.934   -.001    .010  34.693   -.005    .009    .004    .008  45.486 100.138
SDEV:     .027    .008    .025    .062    .012    .024    .019    .006    .000    .057
SERR:     .012    .003    .011    .028    .005    .011    .009    .003    .000
%RSD:      .14-1119.61  252.60     .18 -236.08  264.62  519.51   69.67     .00

PUBL:   19.960    n.a.    n.a.  34.554    n.a.    n.a.    n.a.    n.a.  45.486 100.000
%VAR:     -.13     ---     ---     .40     ---     ---     ---     ---     .00
DIFF:    -.026     ---     ---    .139     ---     ---     ---     ---    .000
STDS:       14      13     395      12      24      22      25     358     ---

This time we're within 0.40% relative or 0.14% absolute. So reasonable accuracy can be achieved using high purity end member synthetics!   

If one cannot attain this sort of accuracy, there is a problem that needs to be determined and corrected.  This is the value of using these extrapolations from high concentrations to lower concentrations.  Such measurements ensure your instrument is operating properly.

[JonF]

Forgot to mention, I did the analysis through the JEOL software, not PfE. I am brand new to PfE and am still using to use the basic features. With the new probe I am trying to learn PfE, the JEOL software and the Bruker EDS software simultaneously. It is a but overwhelming. So there was no setting to do.

There were no bad points, but as I mentioned, I am pretty sure the standard had water trapped under the carbon coat. Same conditions for standard and sample. Whole run was done in less than two hours, and no change in the room conditions. The only proviso is I don't know where the MgO standard was made, so I don't know about its quality.         

I think the first thing I would be doing would be proving to myself that the machine can standardise on a material and re-measure that same material and get 100% totals. If your primary/calibration standards are homogeneous, then this shouldn't be a problem (unless as you say there is an issue with the polish, as on your MgO).

If this comes out ok, then I would start measuring the unknown at increasing concentrations and plotting c/s/nA vs c/s and see how that looks: this will highlight any deadtime issues as you should get a horizontal line. If it starts to dip off at higher count rates, then your deadtime is too low.

What background positions are you using for Mg?

[Probeman]

When attempting to obtain high accuracy from high concentrations to low concentrations (high to low count rates), it is essential to look at all possible sources of error. For example, are the PHA settings properly tuned?

In cases where we are extrapolating from high to low count rates, we need to center the PHA peak in the PHA voltage range and keep the baseline as low as possible.  See here for details:

https://smf.probesoftware.com/index.php?topic=1526.msg11802#msg11802

What background positions are you using for Mg?

Good point. Yes, if the accuracy seems to get worse as one approaches minor concentrations, one should consider problems with the background.

Because JEOL has a smaller Rowland circle than the Cameca, one may need to locate one's off-peak positions further away from the peak, and the background may be more curved.

The above diopside and Mg2SiO4 quant data was acquired on the Berkeley SX51 and here is the Mg peak scanned on a chromite sample for checking off-peak positions:



Zooming in we see this:



Note the secondary lines relatively close to the main emission line. Always a good idea to carefully scan the region around the off-peak positions and the main peak.  The tails of the main emission line need to be avoided for best accuracy.

Another thought, does the relative error away from your "known" concentrations increase or decrease as you approach lower concentrations?  If they increase with decreasing concentration, that would seem to indicate more of a problem with your background measurement, rather than your primary standard.

[Probeman]

It's also worth mentioning that tests performed using the constant k-ratio or simultaneous k-ratios methods (which can be done at the same time), do NOT require that the compositions of the two materials be known. The two materials for the k-ratio could even be unknown, so long as their concentrations (count rates) are sufficiently different, and the materials are beam stable and homogeneous.

All we are trying to do is check that the k-ratios of the two materials are "constant" as the beam current is increased (for the constant k-ratio method) and/or that the same k-ratios (within statistics) are obtained from multiple spectrometers when using the same emission line, independent of the Bragg crystals utilized (for the simultaneous k-ratio method).

[Joe Boesenberg]

John

First, let me say. THANKS to all who responded here on via email. Really appreciate the help. I THINK my problem are the standard blocks. I repolished my 4 main standard blocks in August prior to the new probe and though I left them in my carbon coater under 10-7 torr vacuum for two days, they apparently still did not dry out. I can see water boiling under the carbon coat with the beam on for several standards. I suspect the entire blocks have water which is screwing up the calibration. Not quite sure why Mg (and Si) are primarily effected and not the other non-TAP elements, but maybe the Fo97 and Diopside I used for Mg and Si calibration had substantial trapped water in them. I need to repolish, dry out thoroughly, recoat and retry the analysis.

To answer some of the questions posed in the forum:   

Yes, I have used the Fo100 and Rockport Fayalite to analyze the Fo97 and it comes out fine with good stoichiometry. I don't think I have ever actually analyzed the Fo100 until last week when I found I had MgO (periclase) as a standard.

I recognize that there are issues with San Carlos, but mount I use has chips from a single grain, and though it is not the exact NMNH quoted composition (it has 9.8 wt% FeO instead of 9.55), it is quite homogeneous. I don't use the San Carlos for a primary calibration standard. I use it along with the Lunar Crater Augite as reference standards for oliv and pyx runs. They have nice mid-range compositions that are often in the same range as the unknowns, so they are a good judge of the calibration accuracy.

Fo97 was made at URI in the 90's. I think Steve Carey made it. I came across it at AMNH in the 90's via Charlie Mandeville, who was the probe manager at AMNH (and was one of Steve's former grad students).

All of my analyses were run at 15kv and 20nA and 30-45 sec count times on peak with a point beam. On the JEOL, I have been using 4.1 mm for the negative backgrd and 4.7 mm for the positive background. These are considerably larger than the -1150 and 1150 backgrounds I used on the SX-100 for Mg Ka.

Yes, I used MgO for the first time for the analysis. I forgot I had it. Not sure about its CaO content. Has to be pretty low. Don't see any peak in the Bruker EDS. It was one of the standard blocks I polished in August.

The concern I had/have with the analyses I was getting, was they were all clustered, but none of the analyses were ever over the nominal MgO concentrations.

[John Donovan]

First, let me say. THANKS to all who responded here on via email. Really appreciate the help. I THINK my problem are the standard blocks. I repolished my 4 main standard blocks in August prior to the new probe and though I left them in my carbon coater under 10-7 torr vacuum for two days, they apparently still did not dry out. I can see water boiling under the carbon coat with the beam on for several standards. I suspect the entire blocks have water which is screwing up the calibration. Not quite sure why Mg (and Si) are primarily effected and not the other non-TAP elements, but maybe the Fo97 and Diopside I used for Mg and Si calibration had substantial trapped water in them. I need to repolish, dry out thoroughly, recoat and retry the analysis.

Sounds good. There is nothing wrong with re-polishing and re-coating your standards, ideally in the same mount at the same time. At Oregon we have a small drying oven that sits at 60C all the time and everything goes in there after polishing and cleaning before coating for 5 to 8 minutes to remove any adsorbed water.

Mg Ka is a relatively low energy x-ray so surface prep is important.

Yes, I have used the Fo100 and Rockport Fayalite to analyze the Fo97 and it comes out fine with good stoichiometry. I don't think I have ever actually analyzed the Fo100 until last week when I found I had MgO (periclase) as a standard.

Good. What happens when you analyze Fo100 using MgO as a primary standard?  Assuming they both really are 99.999% pure (and that needs to be checked!).

I recognize that there are issues with San Carlos, but mount I use has chips from a single grain, and though it is not the exact NMNH quoted composition (it has 9.8 wt% FeO instead of 9.55), it is quite homogeneous. I don't use the San Carlos for a primary calibration standard. I use it along with the Lunar Crater Augite as reference standards for oliv and pyx runs. They have nice mid-range compositions that are often in the same range as the unknowns, so they are a good judge of the calibration accuracy.

How do you know this olivine is 9.8wt% FeO.  Is this from another probe measurement? If so, it is worthless because we are testing the EPMA accuracy!

Fo97 was made at URI in the 90's. I think Steve Carey made it. I came across it at AMNH in the 90's via Charlie Mandeville, who was the probe manager at AMNH (and was one of Steve's former grad students).

Again how do we know its composition?  Check your synthetic MgO and Fo100 for traces and use those!  Or get well characterized materials from Will Nachlas or buy them.

Yes, I used MgO for the first time for the analysis. I forgot I had it. Not sure about its CaO content. Has to be pretty low. Don't see any peak in the Bruker EDS. It was one of the standard blocks I polished in August.

But again, you don't know its purity. You need 99.999% purity, then (maybe) you can assume stoichiometry in freshly polished materials. Contact Will Nachlas and buy some materials from his sources.  He would be happy to share what he has found.

MgO, MgAl2O4 and Mg2SiO4 plus a 99.999% SiO2 and Al2O3 to check for blank accuracy.

Note, that if you're trying to characterize trace elements, you can use the EPMA assuming you also run a blank sample to check for trace accuracy. But one cannot use EPMA to determine major element chemistry! That is circular reasoning...

The concern I had/have with the analyses I was getting, was they were all clustered, but none of the analyses were ever over the nominal MgO concentrations.

Yes, that is the problem. But we need to get back to first principles to determine exactly what is going on here.  If I were you and I wanted to know how well my instrument is performing I would get 99.999% pure synthetic MgO, MgAl2O4 and Mg2SiO4.  Also 99.999% SiO2 and Al2O3 for zero blank checks.  I'm going to summarize here one last time:

1. Obtain these high purity synthetic materials. Mount and polish, clean, dry and coat them together.

2. Make sure your PHA settings are "wide open". That is, when your PHA peak (on the highest concentration material, e.g., MgO) is roughly centered in your PHA distribution, set your baseline to 0.5 volts or so and leave the differential mode OFF.  This way, as you go to lower count rates and the PHA peak shifts up, your counter response will remain linear:

https://smf.probesoftware.com/index.php?topic=1466.msg11450#msg11450

3. You already performed dead time calibrations using the constant k-ratio method so you should be good there.

https://smf.probesoftware.com/index.php?topic=1466.msg11102#msg11102

The point being, one should be able to obtain the same k-ratios in say MgAl2O4 (secondary) and MgO (primary) at count rates (beam currents) from 1 kcps to 300 kcps using the logarithmic dead time correction.

4. Measure simultaneous k-ratios on multiple spectrometers,  using two materials with a large absorption correction relative to each other, e.g., MgO and Mg2SiO4:

https://smf.probesoftware.com/index.php?topic=1569.msg12249#msg12249

The point being to stress the absorption correction to determine that all our spectrometers yield the same k-ratio within statistics.  See here to model this in CalcZAF:

https://smf.probesoftware.com/index.php?topic=598.msg12062#msg12062

These simultaneous k-ratio checks test our spectrometer, stage and column geometries.

5. Once these above checks have been performed we are ready to proceed with quantitative tests using these same synthetic materials. For example, measure MgO (60.303 wt% Mg) as the primary standard, next measure Mg2SiO4 (34.554 wt% Mg) and MgAl2O4 (17.084 wt% Mg) as secondary standards.

Of course it doesn't hurt to measure synthetic Al2O3 or SiO2 as blank samples for zero Mg also. The Evaluate application is very helpful for these comparisons:

https://smf.probesoftware.com/index.php?topic=340.msg10757#msg10757

This difficulty you are experiencing is not only a problem in your lab. We see this in every EPMA lab. We are all having a crisis of inter-laboratory reproducibility. We know there's a problem in the microanalysis community because we can see that different labs consistently report different values for the same materials :

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

See Wieser (2023) attached below.

Let's fix it by using the same high purity, homogeneous synthetic standards in our labs and performing the above tests on our instruments!

[crystalgrower]

You can see water boiling off?  This much water?? Seriously??

If you could post a photo of the block then its manufacturer and probable epoxy might be identifiable.  That would give you options for "safe" solvents to try.

Light elements would be more impacted than other materials, whether this is water or oil or some other substance.  We have to ask what would cause a measurable amount of water to adhere to clean polished surfaces.  I suppose that the water-ethanol azeotrope is this stable under vacuum but that is 96% ethanol and 4% water.  Neither is a recommended solvent for mounts.

VERI important to use clean dry HPLC grade solvents.  One litre each of isopropanol and isohexane should last a year.  Use a portion ONCE and discard to a beaker to evaporate in a fume hood.  If you have some already in clean bottles, then adding anhydrous Na2SO4 sodium sulfate will dehydrate sufficiently to be able to clean mounts.  Add the solid until large clear crystals form.  Mush is evidence that water is still present.


 

[Probeman]

We've always just used 190 proof ethanol (not denatured which contains other solvents), rubbing with a Kimwipe until dry, followed by heating in a 60C oven for 5 minutes to remove any adsorbed water, then immediately coating to avoid these issues.

[Joe Boesenberg]

My problems with the Mg and Si concentrations were finally solved over this past weekend. First, new dead time corrections were applied to all five spectrometers (thanks to Scott Boroughs for helping do this) which eliminated about half the concentration shift I was seeing on my analyzed standards. Second, I polished off the carbon coat and spent four days baking the water out of the standard blocks and using a mini vacuum chamber that I could place on a hot plate at 120F. After four days, I recoated and did a full calibration. The new results largely show the proper concentrations I expected. There are still some differences between the concentrations that I get when I analyze olivine versus pyroxene, but I think this is a function of the standards I am using, not anything to do with the mechanics of the probe. Thanks to everyone for the advice.


In response to Crystal Grower, yes, you can see water boiling under the carbon coat when the beam is  on the standard. This is a fairly common issue, since most of the people I know still polish with a slurry of water and grit, not alcohol, which can really smell up a room in a university. You can sometimes evaporate the water source with the beam if it is small. You most often see this on the optical microscope particular when doing relative high magnification work with the BSE or SE running. 

My problem was probably exacerbated by the fact that I polished and carbon coated in August, one of the most humid months of the year along east coast.

Joe

[crystalgrower]

100% isopropanol didn't cause a smell when used for the last manual finish polish with 0.25u alumina.  The polishing station was in the opposite corner from the standard sized fume hood.

Fine grained Rhum anorthosite that was cut under water NEVER dried out even in a vacuum oven at 200C.  The water seeped between the grains faster than you might think.  It sparked like crazy under an electron beam.

There is also a consideration that some sulfide minerals (for example MoS2) are more than slightly soluble in water under pressure.

And I hate to think what a metals mount looks like after the grinding with water:  you can kiss your Fe, Zn, Ag, Cd, Th goodbye.  Ditto for a product mounted into Woods metal.

[Probeman]

Well, it turns out, after checking and double-checking everything mentioned in the above posts regarding dead time calibrations, standard polishing, standard compositions, etc., Joe Boesenberg at Brown is still having issues getting reasonable numbers when extrapolating from high Mg concentrations to low Mg concentrations on both TAP spectrometers. Interestingly, his Al numbers seem to be OK.  His Mg k-ratios always appear to be a few percent lower than expected.

Here is a measurement where he measured an MgCr2O4 synthetic material using synthetic Mg2SiO4 as a primary standard:

ELEM:      Si      Fe      Mg      O      Cr  SUM 
  316    .058    .024  11.482  32.592  54.079  98.235
  317    .055    .033  11.463  32.579  54.079  98.210
  318    .059    .032  11.569  32.652  54.079  98.392
  319    .048    .035  11.502  32.596  54.079  98.260
  320    .054    .019  11.499  32.597  54.079  98.249

AVER:    .055    .028  11.503  32.603  54.079  98.269
SDEV:    .004    .007    .040    .028    .000    .071
SERR:    .002    .003    .018    .013    .000
%RSD:    8.10  23.10    .35    .09    .00

PUBL:    n.a.    n.a.  12.640  33.281  54.079 100.000
%VAR:      ---    ---  -8.99  -2.04    .00
DIFF:      ---    ---  -1.137  -.678    .000
STDS:      226      72    226    ---    ---

As one can see, the totals are low due to the low Mg.

But finally, he was able to obtain an MgO, Al2O3, MgAl2O4 mount from Will Nachlas (from the FIGMAS round robin tests last year), and polished and re-coated it and ran some tests. Here is a measurement of the MgAl2O4, again using synthetic MgSi2O4 as a primary standard:

ELEM:      Si      Fe      Mg      O      Al  SUM 
  341    .037    .007  16.352  44.546  37.932  98.875
  342    .040    .016  16.458  44.622  37.932  99.068
  343    .032    .000  16.553  44.670  37.932  99.187
  344    .034    .009  16.413  44.584  37.932  98.972
  345    .033    .002  16.392  44.567  37.932  98.926

AVER:    .035    .007  16.434  44.598  37.932  99.006
SDEV:    .003    .006    .077    .049    .000    .124
SERR:    .002    .003    .034    .022    .000
%RSD:    9.72  94.27    .47    .11    .00

PUBL:    n.a.    n.a.  17.084  44.984  37.932 100.000
%VAR:      ---    ---  -3.81    -.86    .00
DIFF:      ---    ---  -.650  -.386    .000
STDS:      226      72    226    ---    ---

Again, Mg is low. Then he made a new run, this time using the MgO and Al2O3 as primary standards in the MgO, Al2O3 and MgAl2O4 FIGMAS mount and measuring both Mg and Al in the MgAl2O4 synthetic spinel:

ELEM:      Mg      Al      O  SUM 
  311  16.575  37.920  44.641  99.136
  312  16.641  37.975  44.732  99.348
  313  16.574  37.649  44.398  98.621
  314  16.570  37.686  44.429  98.685
  315  16.602  37.616  44.388  98.606

AVER:  16.592  37.769  44.518  98.879
SDEV:    .030    .166    .158    .341
SERR:    .013    .074    .071
%RSD:      .18    .44    .36

PUBL:  17.084  37.931  44.985 100.000
%VAR:    -2.88    -.43  -1.04
DIFF:    -.491  -.162  -.467
STDS:      411    410    ---

and again a second time after running all three standards again:

ELEM:      Mg      Al      O  SUM 
  326  16.679  38.260  45.011  99.951
  327  16.678  38.102  44.870  99.649
  328  16.575  38.162  44.855  99.592
  329  16.426  37.705  44.351  98.483
  330  16.527  38.003  44.683  99.213

AVER:  16.577  38.046  44.754  99.378
SDEV:    .107    .212    .254    .565
SERR:    .048    .095    .113
%RSD:      .65    .56    .57

PUBL:  17.084  37.931  44.985 100.000
%VAR:    -2.97    .30    -.51
DIFF:    -.507    .115  -.231
STDS:      411    410    ---

So again, Mg appears to be low, but Al seems OK.

Note that Mg is on one spectrometer (SP1), but Al is on another spectrometer (SP3). Therefore we begin to speculate about the spectrometer alignment at higher sin thetas and Pete McSwiggen, whom we had been working with on these issues), had a very interesting idea to test this hypothesis...  more about this in the next post!