I recently posted here:
https://smf.probesoftware.com/index.php?topic=155.msg9457#msg9457
about the need to acquire (and treat) ones standard samples, similar to their unknown samples, when using duplicate elements in aggregate mode. Basically, if you disable quantification for an element channel in your unknown sample, you should also disable quantification for the corresponding element channel in your standard sample. Because in aggregate mode the software basically adds up all the photons for the on and off-peak measurements for all channels that are the same element, emission line (though not necessarily the same Bragg crystal!), and also the same beam energy of course.
But I noticed that Ben Wade's sample provides a nice example of improving sensitivity by using duplicate elements in aggregate mode so I started a new topic to discuss this. Let's start with my initial observation with his glass sample, that the variance in the totals (where all elements except Cl are specified as fixed concentrations, so all the variance comes from the Cl measurements), before the aggregate mode is turned on (see Analytical | Analysis Options menu) as shown here is around :
ELEM: Cl Cl Cl Cl Cl SUM
XRAY: (ka) (ka) (ka) (ka) (ka)
229 .03827 .03803 .03709 .03733 .03756 100.367
230 .03833 .03919 .03795 .03770 .03694 100.369
231 .03773 .03516 .03829 .03928 .03818 100.368
232 .03869 .03898 .03822 .03723 .03735 100.369
233 .03840 .03895 .03783 .03689 .03848 100.369
234 .03791 .03882 .03796 .03869 .03805 100.370
235 .03952 .03721 .03832 .03809 .03766 100.369
236 .03761 .03892 .03753 .03841 .03867 100.370
237 .03797 .03810 .03787 .03667 .03878 100.368
238 .03903 .03849 .03795 .03692 .03833 100.369
AVER: .03834 .03819 .03790 .03772 .03800 100.369
SDEV: .00060 .00122 .00037 .00087 .00060 .00103
SERR: .00019 .00039 .00012 .00028 .00019
%RSD: 1.56001 3.19181 .97764 2.30551 1.59035
STDS: 545 545 545 545 545
Note that the variance on the total is 0.00103 or around 10 PPM and the variances on the Cl channels range from 0.00037 to 0.00122 (or roughly 4 to 12 PPM). Now with the aggregate mode on, we get this result:
ELEM: Cl Cl Cl Cl Cl SUM
XRAY: (ka) (ka) (ka) (ka) (ka)
229 .03760 .00000 .00000 .00000 .00000 100.217
230 .03781 .00000 .00000 .00000 .00000 100.217
231 .03806 .00000 .00000 .00000 .00000 100.217
232 .03799 .00000 .00000 .00000 .00000 100.217
233 .03809 .00000 .00000 .00000 .00000 100.217
234 .03816 .00000 .00000 .00000 .00000 100.217
235 .03828 .00000 .00000 .00000 .00000 100.217
236 .03811 .00000 .00000 .00000 .00000 100.217
237 .03799 .00000 .00000 .00000 .00000 100.217
238 .03818 .00000 .00000 .00000 .00000 100.217
AVER: .03803 .00000 .00000 .00000 .00000 100.217
SDEV: .00020 .00000 .00000 .00000 .00000 .00015
SERR: .00006 .00000 .00000 .00000 .00000
%RSD: .51712 .0000 .0000 .0000 .0000
STDS: 545 0 0 0 0
Now the variance on the total decreased to 0.00015 or about 1.5 PPM! While the variance on the aggregated Cl channels is now 0.00020 or 2 PPM!
So, yes this is exactly what one would expect by aggregating our duplicate channels. Now we can ask how does this affect our detection limits? Well, starting again with the unaggregated Cl channels we get single point detection limits as seen here:
Detection limit at 99 % Confidence in Elemental Weight Percent (Single Line):
ELEM: Cl Cl Cl Cl Cl
229 .00100 .00204 .00086 .00114 .00094
230 .00101 .00206 .00086 .00113 .00094
231 .00100 .00206 .00086 .00111 .00093
232 .00101 .00206 .00086 .00113 .00093
233 .00100 .00206 .00086 .00113 .00093
234 .00101 .00203 .00086 .00113 .00093
235 .00100 .00206 .00087 .00113 .00094
236 .00101 .00204 .00086 .00113 .00093
237 .00101 .00205 .00086 .00114 .00093
238 .00100 .00208 .00086 .00113 .00093
AVER: .00100 .00205 .00086 .00113 .00093
SDEV: .00000 .00001 .00000 .00001 .00000
SERR: .00000 .00000 .00000 .00000 .00000
And t-test values shown here:
Detection Limit (t-test) in Elemental Weight Percent (Average of Sample):
ELEM: Cl Cl Cl Cl Cl
60ci .00023 .00052 .00015 .00026 .00019
80ci .00036 .00081 .00023 .00041 .00030
90ci .00048 .00107 .00031 .00055 .00040
95ci .00059 .00133 .00038 .00068 .00049
99ci .00084 .00191 .00055 .00097 .00071
So single point detection limits from 8 to 20 PPM and t-test detection limits (at 99% confidence) from 5 to 19 PPM for the average of all points. This indicates that the sample is quite homogeneous in Cl.
Now with the aggregated Cl channels, we get this for single line detection limits:
Detection limit at 99 % Confidence in Elemental Weight Percent (Single Line):
ELEM: Cl Cl Cl Cl Cl
229 .00048 .00000 .00000 .00000 .00000
230 .00048 .00000 .00000 .00000 .00000
231 .00047 .00000 .00000 .00000 .00000
232 .00047 .00000 .00000 .00000 .00000
233 .00047 .00000 .00000 .00000 .00000
234 .00047 .00000 .00000 .00000 .00000
235 .00048 .00000 .00000 .00000 .00000
236 .00048 .00000 .00000 .00000 .00000
237 .00047 .00000 .00000 .00000 .00000
238 .00048 .00000 .00000 .00000 .00000
AVER: .00047 .00000 .00000 .00000 .00000
SDEV: .00000 .00000 .00000 .00000 .00000
SERR: .00000 .00000 .00000 .00000 .00000
Now our single line detection limit is down around 5 PPM (compared to 8 to 20 PPM with the unaggregated data). And here is the aggregated t-test results:
Detection Limit (t-test) in Elemental Weight Percent (Average of Sample):
ELEM: Cl Cl Cl Cl Cl
60ci .00009 --- --- --- ---
80ci .00014 --- --- --- ---
90ci .00018 --- --- --- ---
95ci .00023 --- --- --- ---
99ci .00032 --- --- --- ---
Which is now down to around 3 PPM for a 99% confidence t-test (compared to 5 to 19 PPM for the unaggregated channels). So we can see that by utilizing duplicate elements and the aggregate mode in Probe for EPMA we can significantly improve our trace element sensitivities.
By the way, Ben's conditions were 20 keV and 80 nA with an on-peak counting time of 360 seconds and 180 seconds on each off-peak. He also utilized the TDI correction during the on-peak acquisition which showed an increase in the Cl intensities over that time of between 1 and 2%. Which makes sense since Cl is a negative ion and would tend to migrate to the surface from the effects of subsurface charging. Thus increasing the emitted intensity over time due to the shallower emission depth.
Pretty darn interesting dataset actually. I hope Ben will feel free to chime in with his own observations...
Hi John
Yes thanks for sorting that out John, I was definitely confused as to what was going on until you pointed out I needed to disable the same channel in the primary standard as well...
To give you some background on the relatively high beam conditions for what might seem like relatively high Cl concentrations, that glass shown was just one of many analysed, others having down to single digit Cl concentrations. It has been shown many times now that measuring Ti in quartz can get really nice DL via EPMA with aggregation on large xtals and blank subtraction, so I was just wondering how low we could go with Cl as well.
We have a group interested in low level halogens in minerals (F,Cl,Br,I), and although I can't help them too much for low level Br and I via EPMA, I could have a go at Cl given I have four large and one regular PET xtal to sum. As with everything the main problem has been finding secondary standards with certified Cl concentrations to test against....which of course don't really exist. There are a bunch of the common glass standards (NIST, USGS basalt glasses, MPI-DING etc) which have Cl numbers on them, but can vary quite a bit both inter technique (ie SIMS, Noble Gas, LA-ICP-MS, EPMA) and intra technique. There have been some attempts at creating halogen glass standards and I have kindly been given a few of them from researchers, but haven't had a chance to analyse them yet. Also they are commonly in high concentration as to use as primary standards, so not great for checking your low level accuracy.
So it was a suck it and see. I ran spots at relatively moderate conditions (15kV, 40nA, 20um defocussed, 360 on peak, 180 off peak using alternating On/Off peak mode), and at more extreme conditions (20kV, 80nA, 20um defocussed, 360 on peak, 180 off peak using alternating On/Off peak mode). Individual spot errors halved with the higher conditions, and the TDI correction range was identical between both ranging from <1 to 2%. This is probably due to my highly defocussed beam, I am sure there would be a difference if I used a more focussed beam.
One problem I had is that for some glasses I was underestimating the concentration, others were OK. One of the glasses that has very low level Cl returned negative concentrations. If I plotted Cr concentration against the LO/HI background ratio there was a positive correlation, and it turns out there is a Cr Ka(II) peak right where I had set my low background (see plots below)....whoops. Given a lot of these were basalt glasses as well, some had high Cr. Its not a huge interference and was subtracting anywhere from 5-25 ppm from the peak measurement, so not much but enough to give negative concentrations in one of the glasses and affect some of the other low level ones.
I did do wavelength scans in the initial setup, but obviously:
1) at nowhere near long enough dwell times to pick up the small interference, and;
2) I didn't do wavelength scans on all the glasses...only on BCR2G which has low Cr so wouldn't have seen it anyway.
(https://smf.probesoftware.com/gallery/318_16_09_20_9_48_40.jpeg)
(https://smf.probesoftware.com/gallery/318_16_09_20_9_49_30.jpeg)
I also played around with blank subtraction, with the obvious problem what do I have that I know has either zero Cl in it, or that I know definitely what the Cl values is. The answer was nothing, so I just tried a couple of candidates, namely Astimex almandine garnet and San Carlos olivine. Below is a compilation table of the 20kV test results
(https://smf.probesoftware.com/gallery/318_16_09_20_9_50_10.jpeg)
Turns out that San Carlos olivine ?probably? has some very small amount of Cl in it, despite being quite variable. The Astimex garnet was pretty much right on zero, and as a result there isn't much difference between the uncorrected data and the blank (garnet) corrected data. If I trust the SIMS value for the ML3B-G glass and use that as a blank corrector, I get Cl concentrations identical to when I use the Astimex garnet.
Anyway, I now have some other standards to try, I am also going to go back to 15kV and use higher beam current. Currently all these runs have been done using natural tugtupite as a primary standard. Its internal variance is quite good (0.22% RSD on raw CPS/nA), however who knows how accurate the "certified" value is... As mentioned I do now have a halogen glass which I am going to use as a primary standard and see how it compares. Given the current state of the available standards I think accuracy is always going to be the biggest issue over precision.
Cheers
Quote from: BenjaminWade on September 16, 2020, 09:52:59 PM
To give you some background on the relatively high beam conditions for what might seem like relatively high Cl concentrations, that glass shown was just one of many analysed, others having down to single digit Cl concentrations. It has been shown many times now that measuring Ti in quartz can get really nice DL via EPMA with aggregation on large xtals and blank subtraction, so I was just wondering how low we could go with Cl as well.
Hi Ben,
Absolutely. When one requires ultimate sensitivity increasing geometric efficiency is the way to go. Right now larger Bragg crystals and running the same element on multiple spectrometers using aggregate mode is the only options we have. Of course running at higher beam energies can also increase sensitivity (while reducing spatial resolution), assuming the emission line you are observing is sufficiently energetic to be emitted at those depths.
As an example of improving the Ti sensitivity using multiple spectrometers and aggregate mode I attach below a recent publication from one of our students here in Oregon (yes, the fire smoke is really bad, but starting to improve a little yesterday) on Ti in quartz.
Quote from: BenjaminWade on September 16, 2020, 09:52:59 PM
We have a group interested in low level halogens in minerals (F,Cl,Br,I), and although I can't help them too much for low level Br and I via EPMA, I could have a go at Cl given I have four large and one regular PET xtal to sum. As with everything the main problem has been finding secondary standards with certified Cl concentrations to test against....which of course don't really exist. There are a bunch of the common glass standards (NIST, USGS basalt glasses, MPI-DING etc) which have Cl numbers on them, but can vary quite a bit both inter technique (ie SIMS, Noble Gas, LA-ICP-MS, EPMA) and intra technique. There have been some attempts at creating halogen glass standards and I have kindly been given a few of them from researchers, but haven't had a chance to analyse them yet. Also they are commonly in high concentration as to use as primary standards, so not great for checking your low level accuracy.
Yeah finding a good Cl primary standard that is also not beam sensitive is tough. We have utilized a synthetic apatite grown at the University of Nice, but it is definitely not robust under the beam.
As for secondary standards, I will say it again (while everyone rolls their eyes), but I think that trying to find such trace level standards is a waste of our collective time. Even if we had such a "doped" material how well do we know its accuracy or homogeneity? It's always an open question. Instead, I think what we should be looking for (and/or synthesizing) are matrix matched synthetic materials that contain a zero concentration of the element(s) in question. That is, below the detection limit of the EPMA, and for most elements this is around 1 PPM or lower.
If the standard concentration is below the instrumental detection limit, we know that it's a zero concentration which is accurate to within our measurement precision.To take the example of your Cl sensitivity measurements that I posted above, where you have a three sigma detection limit of 3 PPM (t-test) for the 5 aggregated spectrometers, that means (if you had a known zero concentration of Cl in (ideally) a similar matrix matched standard, you would now be able to claim a trace level *accuracy* of 3 PPM for all your measurements.
We literally cannot do better than that on the EPMA! :D
For more information on the blank correction see here:
https://smf.probesoftware.com/index.php?topic=29.0
Also Donovan et al., 2011:
https://epmalab.uoregon.edu/pdfs/3631Donovan.pdf
So yes, we need a robust accurate primary standard with a high concentration of the element (matrix matched is nice, but not required as the matrix corrections can deal with that), and we also need an accurate zero concentration standard (in this case a matrix matched standard is important, because the shape of the continuum is determined by the major and minor element concentrations), to test our trace level accuracy. Everything else is (iterative) interpolation between these two endpoints.
Quote from: BenjaminWade on September 16, 2020, 09:52:59 PM
One problem I had is that for some glasses I was underestimating the concentration, others were OK. One of the glasses that has very low level Cl returned negative concentrations. If I plotted Cr concentration against the LO/HI background ratio there was a positive correlation, and it turns out there is a Cr Ka(II) peak right where I had set my low background (see plots below)....whoops. Given a lot of these were basalt glasses as well, some had high Cr. Its not a huge interference and was subtracting anywhere from 5-25 ppm from the peak measurement, so not much but enough to give negative concentrations in one of the glasses and affect some of the other low level ones.
Yeah, subtle off-peak interferences that are not discovered in a "normal" sensitivity wavescan are troublesome. And of course, as you know, this is why for ultimate accuracy when using off-peak measurements, the multi-point background (MPB) is the way to go. As I have said before, it's like combining your quant acquisition with a high precision wavescan!
For those who haven't tried the MPB method here is a topic on the technique:
https://smf.probesoftware.com/index.php?topic=131.0
The full peer reviewed paper on multi-point background (and the related "shared" background) method is here:
https://search.proquest.com/openview/208992e799d48f657d3508a6ef11ca1c
Hi John
Thanks for the paper, I will have a read of it.
I do agree with everything you said re: secondary standards. With regards to having a repository of synthetic high purity matrix matched minerals though, I feel like the situation isn't going to get better anytime soon though. Would take a real concerted effort from the whole analytical community, probably across techniques as well as I can see the application of these kind of materials in other techniques as well such as SIMS and LA-ICP-MS.
I assume these are these kind of things discussed at FIGMAS. One day hopefully in my lifetime I will be allowed to fly again...and one day this century hopefully the university will give us back our travel budget so I can attend an M&M in person.
Cheers
No worries.
I also agree that getting a large number of high purity synthetic mineral standards will require "concerted effort" by the microanalysis community. But if we don't make an effort to start creating global standards, we are simply giving up. Imagine if in the 18th century we had decided to not bother with creating global meter and kilogram standards?
But it's not so hard to get specific standards in many cases. Let's start with some relatively simple materials that already are produced on industrial scales, e.g., the MgO, Al2O3, and MgAl2O4 materials we've already mentioned, plus other materials mentioned such as a synthetic Mg2SiO4 and SiO2. Then there's synthetic TiO2. Crystalgrower mentioned quite a few others that are already commercially available.
We just have to start with a few such industrially produced materials that already available and go on from there.
Hi John
Hmm probably a discussion for a different thread (like the community k-ratio thread!), but I am definitely interested. Sounds like it would be a bit like the G-Probe but with more focus on microbeam standards rather than LA-ICP-MS.
Cheers
I'm trying to aggregate elements - but it says I can't as I'm using different backkground modes.
For one of the spectrometers a multi-point polynominal seems to work best, whereas for the others a linear background seemed to work best.
Quote from: Ben Buse on October 19, 2023, 03:00:00 AM
I'm trying to aggregate elements - but it says I can't as I'm using different backkground modes.
For one of the spectrometers a multi-point polynominal seems to work best, whereas for the others a linear background seemed to work best.
I'm trying to think of a reason why that would matter, but I'm sure there's a good reason for this. Can you send me the MDB file so I can take a look?
Are you using the "shared bgd" option for one of the elements? Actually I think the reason is that the MDB background is determined iteratively, so that becomes difficult to combine with a normal off-peak background. Can you use another off-peak background on that spectrometer?
Quote from: Ben Buse on October 19, 2023, 03:00:00 AM
I'm trying to aggregate elements - but it says I can't as I'm using different backkground modes.
For one of the spectrometers a multi-point polynominal seems to work best, whereas for the others a linear background seemed to work best.
We did not forget about this post, but Aurelien and I needed to go over the code changes to support a modification regarding the error handling you encountered. In fact the error trapping for this change (see below) went from one line to about 20 lines!
Yes, you are correct. Probe for EPMA does not allow one to aggregate duplicate elements that use different background models.
The reason for this is clear if one considers how would one add say, off-peak and MAN or, off-peak and MPB bgds intensity data, as their background intensity arrays are very different. Any attempt to do this results in this error message (that you saw):
(https://smf.probesoftware.com/gallery/1_25_10_23_1_45_42.png)
By the way, if anyone gets an error message and wants more information about it, please use the <alt> <Print Screen> key combination to capture the currently active window to the clipboard, then paste that into an e-mail with your question. This makes it much easier for us to track down any issues.
One way to deal with this is to simply disable the quant on the duplicate elements with a different background acquisition method. Here we disabled quant for the Ti Ka MPB element from the Elements/Cations dialog and the results are shown here:
Un 56 Si metal, Results in Elemental Weight Percents
SPEC: Si O
TYPE: SPEC SPEC
AVER: 100.000 .000
SDEV: .000 .000
ELEM: Ti Al Ti Ti Ti
BGDS: LIN LIN MULT LIN LIN
TIME: 240.00 240.00 --- .00 .00
BEAM: 544.25 544.25 --- .00 .00
AGGR: 3 ---
ATWT_S 47.900 26.982 47.900 47.900 47.900
ATWT_U 47.900 26.982 .000 47.900 47.900
ELEM: Ti Al Ti-D Ti Ti SUM
XRAY: (ka) (ka) (ka) (ka) (ka)
2299 -.00023 -.00051 --- .00000 .00000 99.9993
2300 -.00046 -.00246 --- .00000 .00000 99.9971
2301 -.00025 -.00284 --- .00000 .00000 99.9969
2302 -.00072 -.00239 --- .00000 .00000 99.9969
2303 -.00017 -.00325 --- .00000 .00000 99.9966
2304 -.00045 -.00162 --- .00000 .00000 99.9979
2305 -.00011 -.00240 --- .00000 .00000 99.9975
2306 -.00027 -.00189 --- .00000 .00000 99.9979
2307 -.00014 -.00221 --- .00000 .00000 99.9977
2308 -.00049 -.00242 --- .00000 .00000 99.9971
2309 -.00037 -.00253 --- .00000 .00000 99.9971
2310 -.00044 -.00170 --- .00000 .00000 99.9979
AVER: -.00034 -.00218 --- .00000 .00000 99.9975
SDEV: .00018 .00070 --- .00000 .00000 .00071
SERR: .00005 .00020 --- .00000 .00000
%RSD: -52.285 -31.960 --- .0000 .0000
STDS: 802 807 --- 0 0
But after looking things over we realized that there could be an exception to this. Specifically when aggregating off-peak and MPB intensity data, and that is by converting the MPB data to off-peak data by using the one of the off-peak bgd fit options in the Elements/Cations dialog:
(https://smf.probesoftware.com/gallery/1_25_10_23_1_46_22.png)
After doing this for both your standards and unknowns, you can aggregate both off-peak and MPB elements as seen here:
Un 56 Si metal, Results in Elemental Weight Percents
SPEC: Si O
TYPE: SPEC SPEC
AVER: 100.000 .000
SDEV: .000 .000
ELEM: Ti Al Ti Ti Ti
BGDS: LIN LIN EXP LIN LIN
TIME: 240.00 240.00 .00 .00 .00
BEAM: 544.25 544.25 .00 .00 .00
AGGR: 4
ELEM: Ti Al Ti Ti Ti SUM
XRAY: (ka) (ka) (ka) (ka) (ka)
2299 -.00179 -.00051 .00000 .00000 .00000 99.9977
2300 -.00098 -.00246 .00000 .00000 .00000 99.9966
2301 -.00071 -.00284 .00000 .00000 .00000 99.9965
2302 -.00095 -.00239 .00000 .00000 .00000 99.9967
2303 -.00080 -.00325 .00000 .00000 .00000 99.9960
2304 -.00088 -.00162 .00000 .00000 .00000 99.9975
2305 -.00067 -.00240 .00000 .00000 .00000 99.9969
2306 -.00079 -.00189 .00000 .00000 .00000 99.9973
2307 -.00072 -.00221 .00000 .00000 .00000 99.9971
2308 -.00079 -.00242 .00000 .00000 .00000 99.9968
2309 -.00078 -.00253 .00000 .00000 .00000 99.9967
2310 -.00095 -.00170 .00000 .00000 .00000 99.9973
AVER: -.00090 -.00218 .00000 .00000 .00000 99.9969
SDEV: .00030 .00070 .00000 .00000 .00000 .00050
SERR: .00009 .00020 .00000 .00000 .00000
%RSD: -33.066 -31.960 .0000 .0000 .0000
STDS: 802 807 0 0 0
Though course you're not using MPB fitting strictly speaking, but only adding the intensities from the high side MPBs and the low side MPBs. Note though that one can select say, an exponential fit and adjust the curvature term.
The latest version of Probe for EPMA now supports this method. By the way, we also updated the error message shown above to provide more information:
(https://smf.probesoftware.com/gallery/1_25_10_23_1_46_39.png)
Update PFE from the Help menu as usual...