I have been working on the JEOL probe here at the Smithsonian, National Museum of Natural History, using the Probe for EPMA software and am attempting to analyze Ti in Quartz. I have analyzed TiO2 on 3 spectrometers to try and obtain reliable Ti results at low abundance. I was hoping to merge the Ti counts from the three spectrometers and was wondering whether there is a way that Probe for EPMA can do this?
If you have any information or advice on how to do this, that would be greatly appreciated!
It's almost too easy!
Simply go to the Analytical | Analysis Options menu and check the Use Aggregate Intensities For Duplicate Quantitative Elements check box. This way you can globally toggle separate or aggregate calculations for your run.
Note that for best results for Ti in quartz you may also want to avail yourself of the blank correction. You simply need an unknown sample acquired on a quartz standard with a zero or known non zero concentration of Ti. Then assign that sample in the Standard Assignments dialog for each Ti spectrometer and the software will automatically correct for small artifacts in the continuum which can vary from zero to *negative* 30 or 40 PPM on some PET crystals.
See our paper here:
http://probesoftware.com/download/Ti%20in%20Quartz,%20Am.%20Min.%20Donovan,%202011.pdf
Apologies for "necroing" an old post, however I am too going down this path. John, I have a question regarding a wavescan in that paper. How many channels was that wavescan on SiO2 done over? I am aiming to discover if our LPETS/PETS produce similar continuum "holes" at the Ti Ka position.
I notice on your figure it says 220 seconds per point....and if that is split up over 2048 channels, that is a very long wavescan? Or am I getting it wrong? If it is a very long wavescan, do you raster the beam over the period of that wavescan?
Cheers
No apologies necessary! This is exactly what these topics are for!
Yes, you really do need over 200 seconds per wavescan point as these artifacts are "only" around -40 to 40 PPM or so.
I think I ran 200 points per scan, it does take a while! There is a check box in the Peak/Scan Options dialog that allows one to specify a small stage "jog" every n seconds during a long wavescan. It is usually necessary with beam sensitive materials such as quartz. Note that you should also specify an Unknown Count Factor under the Count Times dialog of around 8 or more and check the "Use Alternating On/Off Peak Acquire" option in the Acquisition Options dialog.
Not every crystal will show this artifact. See the attachment below, where spec 2 LPET is the worst (~ -30 PPM magnitude artifact).
Ok awesome. Thanks John. I will give it another go.
So this may sound weird, but for simple matrices I think it will work to simply acquire a rough MAN calibration curve for the samples being investigated (SiO2, TiO2, ZrSiO4, etc) and assuming that one has a well characterized standard suitable for use as a zero (or non-zero) blank, there really is no need to measure off-peak backgrounds at all.
The aggregate and blank corrections discussed here can be turned on and/or off from the Analytical | Analysis Options menu:
(https://smf.probesoftware.com/oldpics/i40.tinypic.com/1sgifn.jpg)
For discussions on utilizing the MAN correction see this thread:
http://smf.probesoftware.com/index.php?topic=4.0
Here are measurements of Ti and Al (on two spectrometers each and aggregated using the Aggregate Intensity feature) *without* the blank correction (just MAN bgd corrected):
ELEM: Ti Ti Al Al Si O
TYPE: ANAL ANAL ANAL ANAL DIFF CALC
BGDS: MAN MAN MAN MAN
TIME: 200.00 .00 200.00 .00
BEAM: 100.12 .00 100.12 .00
AGGR: 2 2
ELEM: Ti Ti Al Al Si O SUM
XRAY: (ka) (ka) (ka) (ka) () ()
56 -.00255 .00000 .00370 .00000 46.7417 53.2571 100.000
57 -.00187 .00000 .00364 .00000 46.7413 53.2570 100.000
58 -.00251 .00000 .00303 .00000 46.7423 53.2572 100.000
59 -.00223 .00000 .00343 .00000 46.7417 53.2571 100.000
60 -.00169 .00000 .00398 .00000 46.7408 53.2569 100.000
AVER: -.00217 .00000 .00356 .00000 46.742 53.257 100.000
SDEV: .00038 .00000 .00036 .00000 .001 .000 .00000
SERR: .00017 .00000 .00016 .00000 .00025 .00005
%RSD: -17.619 .00007 9.98434 .00020 .00119 .00022
Which means since the off-peak intensities were *not* measured, the subsequent precision of the measurement is improved by the SQRT(2), compared to measurements using off-peak backgrounds. In the above case the precision is excellent (3.8 PPM for Ti and 3.6 PPM for Al), but the accuracy errors for the SiO2 blank (determined by ICP-MS and AA to be 1.4 PPM Ti and 15 PPM Al) are showing significant errors as expected for using the MAN correction by itself (note - 21.7 PPM for Ti and 36 PPM for Al).
However, when combined with the "blank" correction, the accuracy is brought to the same level as the precision.
ELEM: Ti Ti Al Al Si O
TYPE: ANAL ANAL ANAL ANAL DIFF CALC
BGDS: MAN MAN MAN MAN
TIME: 200.00 .00 200.00 .00
BEAM: 100.12 .00 100.12 .00
AGGR: 2 2
ELEM: Ti Ti Al Al Si O SUM
XRAY: (ka) (ka) (ka) (ka) () ()
56 -.00024 .00000 .00165 .00000 46.7417 53.2569 100.000
57 .00031 .00000 .00158 .00000 46.7414 53.2567 100.000
58 -.00037 .00000 .00097 .00000 46.7424 53.2570 100.000
59 -.00008 .00000 .00137 .00000 46.7419 53.2569 100.000
60 .00050 .00000 .00193 .00000 46.7409 53.2567 100.000
AVER: .00002 .00000 .00150 .00000 46.742 53.257 100.000
SDEV: .00037 .00000 .00036 .00000 .001 .000 .00000
SERR: .00017 .00000 .00016 .00000 .00025 .00005
%RSD: 1558.68 .00006 23.6681 .00018 .00122 .00022
See attached figures and please let me know if you have questions.
John, thanks so much for your response! That's fantastic that there is such an easy option for doing this. I had one additional question. I ran Ti on three spectrometers but I only want to merge the counts from 2 of the 3 spectrometers. I tried disabling quantification of Ti on the 3rd spectrometer but this doesn't seem to work. Any other suggestions? Thanks so much!
Some of you have asked to see the MAN fits that result in sub 100 PPM accuracy for Ti in quartz, though I didn't do anything special. But I've attached a pdf that shows the MAN fits for Ti and Al both without and with the Aggregate Intensity feature turned on.
Quote from: sbgrocke on October 03, 2013, 11:30:23 AM
I had one additional question. I ran Ti on three spectrometers but I only want to merge the counts from 2 of the 3 spectrometers. I tried disabling quantification of Ti on the 3rd spectrometer but this doesn't seem to work. Any other suggestions? Thanks so much!
Hi Stephanie,
Great question.
It could perhaps be more slick, but since PFE "aggregates" the x-ray intensities for *both* the unks and stds (on and off-peak), you need to also disable quant for that channel for the standards also. When that has been performed you'll see something like this:
Un 4 Rusk sample
TakeOff = 40.0 KiloVolt = 20.0 Beam Current = 100. Beam Size = 10
(Magnification (analytical) = 40000), Beam Mode = Analog Spot
(Magnification (default) = 400, Magnification (imaging) = 800)
Image Shift (X,Y): .00, .00
Number of Data Lines: 29 Number of 'Good' Data Lines: 29
First/Last Date-Time: 08/27/2013 05:44:12 PM to 08/28/2013 12:32:27 AM
WARNING- Using Exponential Off-Peak correction for al ka
WARNING- Using Exponential Off-Peak correction for al ka
WARNING- Using Blank Trace Correction
WARNING- Using Alternating On and Off Peak Acquisition
WARNING- Using Aggregate Intensities for Duplicate Elements
WARNING- Quantitation is Disabled For ti ka, Spectro 5
Average Total Oxygen: 53.254 Average Total Weight%: 100.000
Average Calculated Oxygen: 53.254 Average Atomic Number: 10.805
Average Excess Oxygen: .000 Average Atomic Weight: 20.029
Average ZAF Iteration: 1.97 Average Quant Iterate: 3.00
Oxygen Calculated by Cation Stoichiometry and Included in the Matrix Correction
Element Si is Calculated by Difference from 100%
Un 4 Rusk sample, Results in Elemental Weight Percents
ELEM: Ti Ti Al Al Si O
TYPE: ANAL ANAL ANAL ANAL DIFF CALC
BGDS: LIN LIN EXP EXP
TIME: --- 400.00 400.00 .00
BEAM: --- 100.05 100.05 .00
AGGR: --- 2
ELEM: Ti-D Ti Al Al Si O SUM
XRAY: (ka) (ka) (ka) (ka) () ()
114 --- .00537 .01540 .00000 46.7252 53.2540 100.000
115 --- .00539 .01557 .00000 46.7251 53.2540 100.000
116 --- .00519 .01523 .00000 46.7255 53.2541 100.000
117 --- .00602 .01529 .00000 46.7248 53.2539 100.000
118 --- .00635 .01538 .00000 46.7245 53.2538 100.000
119 --- .00535 .01550 .00000 46.7251 53.2540 100.000
120 --- .00561 .01562 .00000 46.7248 53.2539 100.000
121 --- .00616 .01516 .00000 46.7248 53.2539 100.000
122 --- .00688 .01590 .00000 46.7236 53.2536 100.000
123 --- .00516 .01559 .00000 46.7252 53.2540 100.000
124 --- .00715 .01595 .00000 46.7233 53.2536100.0000
125 --- .00552 .01583 .00000 46.7247 53.2539 100.000
126 --- .00544 .01578 .00000 46.7248 53.2540 100.000
127 --- .00477 .01614 .00000 46.7250 53.2541 100.000
128 --- .00555 .01572 .00000 46.7248 53.2539 100.000
129 --- .00409 .01248 .00000 46.7288 53.2546 100.000
130 --- .00589 .01584 .00000 46.7244 53.2539 100.000
131 --- .00668 .01591 .00000 46.7237 53.2537 100.000
132 --- .00582 .01612 .00000 46.7242 53.2538 100.000
133 --- .00672 .01573 .00000 46.7239 53.2537 100.000
134 --- .00657 .01564 .00000 46.7241 53.2537 100.000
135 --- .00641 .01554 .00000 46.7243 53.2538 100.000
136 --- .00550 .01527 .00000 46.7252 53.2540 100.000
137 --- .00546 .01528 .00000 46.7253 53.2540 100.000
138 --- .00532 .01544 .00000 46.7252 53.2540 100.000
139 --- .00729 .01582 .00000 46.7234 53.2535 100.000
140 --- .00669 .01571 .00000 46.7239 53.2537 100.000
141 --- .00696 .01525 .00000 46.7241 53.2537 100.000
142 --- .00548 .01444 .00000 46.7260 53.2541 100.000
AVER: --- .00589 .01547 .00000 46.725 53.254 100.000
SDEV: --- .00076 .00067 .00000 .001 .000 .00000
SERR: --- .00014 .00012 .00000 .00019 .00004
%RSD: --- 12.9839 4.33293 .00048 .00219 .00041
I think it would be a good idea to take this Ti in quartz discussion in a step by step fashion because it is all rather complicated at first.
So we'll start by acquiring Ti Ka on all 5 spectrometers. We'll get the best detection limits using the PET crystal variety, but this measurement could also be done on all LIF crystals or a mixture of PET and LIF Bragg crystals as long as the standards and unknowns are acquired using the same sample setup (which is the default mode in PFE).
On our instrument we used this setup and acquired in this case 960 seconds on-peak and the same amount of time off-peak. That's over half an hour which is a fairly long time, especially since it is well known that quartz crystal (in contrast to quartz glass) is visibly damaged by the electron beam. Particularly since we are using a 200 nA beam current!
To minimize the effect of prolonged exposure of the beam on the quartz we utilize the "alternating on and off peak" acquisition method. This method is activated by a single mouse click to one's sample setup in the Acquisition Option dialog from the Acquire! window as seen here (note that Ti Ka on an PET crystal is specified for each spectrometer):
(https://smf.probesoftware.com/oldpics/i42.tinypic.com/2rnk6f5.jpg)
What this does is break up the acquisition into a number of cycles that "alternates" between on-peak and off-peak measurements. We can specify the number of cycles utilized using the "Unknown Count Factor" parameter in the Count Times dialog. Since we are counting all the elements the same way, we can simply click and drag all five element rows in the Count Times dialog and then we will see this dialog for the "Selected Elements":
(https://smf.probesoftware.com/oldpics/i43.tinypic.com/vr5xzo.jpg)
By specifying an Unknown Count Factor for all the Ti channels, we are telling the software to alternate between the on and off peak measurements 32 times per data point acquisition. This interval data is saved automatically and can be plotted as seen here where we can see a small but consistent downwards trend in the background corrected intensities.
(https://smf.probesoftware.com/oldpics/i42.tinypic.com/294hpiu.jpg)
The point being that the "delta" between the peak and off-peak is being tracked over time for the most reliable measurement. When this is done we can see the resulting quantitative analysis performed on an SiO2 "blank" standard, which has been previously characterized by ICP-MS as containing 1.42 PPM Ti, as seen in this output:
Un 30 960 sec on SiO2
TakeOff = 40.0 KiloVolt = 20.0 Beam Current = 200. Beam Size = 20
(Magnification (analytical) = 8000), Beam Mode = Analog Spot
(Magnification (default) = 600, Magnification (imaging) = 100)
Image Shift (X,Y): .00, .00
Number of Data Lines: 5 Number of 'Good' Data Lines: 5
First/Last Date-Time: 06/12/2009 04:58:27 PM to 06/12/2009 07:26:32 PM
WARNING- Using Alternating On and Off Peak Acquisition
Average Total Oxygen: 53.255 Average Total Weight%: 99.994
Average Calculated Oxygen: 53.255 Average Atomic Number: 10.804
Average Excess Oxygen: .000 Average Atomic Weight: 20.028
Average ZAF Iteration: 1.00 Average Quant Iterate: 2.00
Oxygen Calculated by Cation Stoichiometry and Included in the Matrix Correction
WARNING- Duplicate analyzed elements are present in the sample matrix!!
Use Aggregate Intensity option or Disable Quant feature for accurate matrix correction.
Un 30 960 sec on SiO2, Results in Elemental Weight Percents
ELEM: Ti Ti Ti Ti Ti Si O
TYPE: ANAL ANAL ANAL ANAL ANAL SPEC CALC
BGDS: LIN LIN LIN LIN LIN
TIME: 960.00 960.00 960.00 960.00 960.00
BEAM: 199.56 199.56 199.56 199.56 199.56
ELEM: Ti Ti Ti Ti Ti Si O SUM
XRAY: (ka) (ka) (ka) (ka) (ka) () ()
266 -.00015 -.00116 -.00256 .00003 -.00006 46.7430 53.2544 99.9935
267 .00076 -.00160 -.00336 .00081 -.00006 46.7430 53.2547 99.9943
268 -.00013 -.00104 -.00324 .00038 -.00002 46.7430 53.2543 99.9932
269 .00026 -.00101 -.00302 .00124 -.00035 46.7430 53.2551 99.9952
270 -.00011 -.00090 -.00265 .00119 -.00022 46.7430 53.2552 99.9955
AVER: .00012 -.00114 -.00297 .00073 -.00014 46.743 53.255 99.9943
SDEV: .00039 .00027 .00035 .00052 .00014 .000 .000 .00100
SERR: .00018 .00012 .00016 .00023 .00006 .00000 .00018
%RSD: 317.644 -23.837 -11.885 71.4915 -97.829 .00000 .00076
STDS: 922 922 922 922 922 0 0
STKF: .5883 .5883 .5883 .5883 .5883 .0000 .0000
STCT: 667.34 1600.07 1901.70 531.93 828.32 .00 .00
UNKF: .0000 .0000 .0000 .0000 .0000 .0000 .0000
UNCT: .00 -.03 -.08 .01 .00 .00 .00
UNBG: .99 2.64 3.42 .79 1.38 .00 .00
ZCOR: 1.1969 1.1969 1.1969 1.1969 1.1969 .0000 .0000
KRAW: .00000 -.00002 -.00004 .00001 .00000 .00000 .00000
PKBG: 1.00119 .99017 .97657 1.00705 .99878 .00000 .00000
Detection limit at 99 % Confidence in Elemental Weight Percent (Single Line):
ELEM: Ti Ti Ti Ti Ti
266 .00072 .00049 .00047 .00081 .00068
267 .00072 .00049 .00047 .00081 .00069
268 .00072 .00049 .00047 .00081 .00069
269 .00072 .00049 .00047 .00080 .00068
270 .00072 .00049 .00047 .00080 .00068
AVER: .00072 .00049 .00047 .00080 .00068
SDEV: .00000 .00000 .00000 .00000 .00000
SERR: .00000 .00000 .00000 .00000 .00000
From the calculated detection limits just above we can see that we are getting quite reasonable detection limits from 0.00047 wt.% (4.7 PPM) to 0.0008 wt.% (8 PPM) so what could we possibly complain about? Well, for one thing, our average concentrations (AVER:) are *larger* from zero (1.42 PPM) than our standard deviations! This means that we have better precision now than accuracy. That is to say we cannot accurately measure zero (1.42 PPM) as well as our sensitivity.
For example, on spectrometer 2 we have a standard deviation of 0.00027 (2.7 PPM) and a quantitative result of 0.00114 (11.4 PPM). Even with 3 sigma statistics we cannot reconcile that, so we have an accuracy problem. Spectrometer 3 shows an even larger accuracy issue.
What is causing that? Well take a look at this slide and we can see that this could be caused by several different factors. But since these are Ti in SiO2 measurements on a PET crystal we know that the "holes" in the continuum can be caused by secondary Bragg diffraction of the PET crystal.
(https://smf.probesoftware.com/oldpics/i42.tinypic.com/zsvf2v.jpg)
The hole in the Ti continuum to the right side of the peak can be avoided, but *not* the smaller hole in the background which is right underneath the Ti Ka peak!
But the point is that we can already make a correction for these continuum artifacts since we have characterized an SiO2 standard containing a zero or known (in this case 1.42 PPM) amount of Ti in SiO2. To illustrate this we will perform an analysis of the same sample, but this time using the intensities from a previous 960 second acquisition as a "blank" correction which is done simply by selecting the sample in the Standard Assignments dialog as seen here:
(https://smf.probesoftware.com/oldpics/i41.tinypic.com/2w6xf8x.jpg)
Note that the blank assignment needs to be specified for each Ti channel (note that we specified 0.000142 (1.42 PPM) as the "blank" level, but we could also have just left the default as zero since the Ti level is so close to zero). Once that is done, we can re-calculate our test sample and we now get the following "blank" corrected results:
Un 30 960 sec on SiO2
TakeOff = 40.0 KiloVolt = 20.0 Beam Current = 200. Beam Size = 20
(Magnification (analytical) = 8000), Beam Mode = Analog Spot
(Magnification (default) = 600, Magnification (imaging) = 100)
Image Shift (X,Y): .00, .00
Number of Data Lines: 5 Number of 'Good' Data Lines: 5
First/Last Date-Time: 06/12/2009 04:58:27 PM to 06/12/2009 07:26:32 PM
WARNING- Using Blank Trace Correction
WARNING- Using Alternating On and Off Peak Acquisition
Average Total Oxygen: 53.258 Average Total Weight%: 100.002
Average Calculated Oxygen: 53.258 Average Atomic Number: 10.805
Average Excess Oxygen: .000 Average Atomic Weight: 20.029
Average ZAF Iteration: 1.00 Average Quant Iterate: 3.00
Oxygen Calculated by Cation Stoichiometry and Included in the Matrix Correction
WARNING- Duplicate analyzed elements are present in the sample matrix!!
Use Aggregate Intensity option or Disable Quant feature for accurate matrix correction.
Un 30 960 sec on SiO2, Results in Elemental Weight Percents
ELEM: Ti Ti Ti Ti Ti Si O
TYPE: ANAL ANAL ANAL ANAL ANAL SPEC CALC
BGDS: LIN LIN LIN LIN LIN
TIME: 960.00 960.00 960.00 960.00 960.00
BEAM: 199.56 199.56 199.56 199.56 199.56
ELEM: Ti Ti Ti Ti Ti Si O SUM
XRAY: (ka) (ka) (ka) (ka) (ka) () ()
266 .00007 .00022 .00053 -.00030 .00042 46.7430 53.2576 100.002
267 .00098 -.00023 -.00027 .00049 .00043 46.7430 53.2579 100.002
268 .00010 .00033 -.00015 .00005 .00046 46.7430 53.2575 100.001
269 .00048 .00037 .00007 .00091 .00013 46.7430 53.2583 100.003
270 .00011 .00047 .00044 .00086 .00026 46.7430 53.2584 100.004
AVER: .00035 .00023 .00012 .00040 .00034 46.743 53.258 100.002
SDEV: .00039 .00027 .00035 .00052 .00014 .000 .000 .00100
SERR: .00018 .00012 .00016 .00023 .00006 .00000 .00018
%RSD: 112.699 117.296 291.993 129.733 41.2622 .00000 .00075
STDS: 922 922 922 922 922 0 0
STKF: .5883 .5883 .5883 .5883 .5883 .0000 .0000
STCT: 667.34 1600.07 1901.70 531.93 828.32 .00 .00
UNKF: .0000 .0000 .0000 .0000 .0000 .0000 .0000
UNCT: .00 .01 .00 .00 .00 .00 .00
UNBG: .99 2.64 3.42 .79 1.38 .00 .00
ZCOR: 1.1969 1.1969 1.1969 1.1969 1.1969 .0000 .0000
KRAW: .00000 .00000 .00000 .00001 .00000 .00000 .00000
PKBG: 1.00333 1.00200 1.00096 1.00389 1.00290 .00000 .00000
BLNK#: 26 26 26 26 26 ---- ----
BLNKL: .000142 .000142 .000142 .000142 .000142 ---- ----
BLNKV: -.00008 -.00123 -.00295 .000471 -.00034 ---- ----{
Note the lines in red color which show the sample assigned as the "blank" correction, the "blank" level (1.42 PPM) and the "blank" value, which is the concentration by which our measurement differs from the known concentration of Ti in our blank standard. Note that spectrometer 3 has a continuum artifact of 0.00295 wt.% (~30 PPM)!
Now we can see that our accuracy is similar to our precision in the AVER: and SDEV: lines above around 4 to 5 PPM. So now can we further improve our statistics by increasing our geometric efficiency? Yes, by "aggregating" our 5 spectrometers into a single "giant virtual" spectrometer. We'll have to aggregate the photons for both our on-peak and off-peak intensities and also for both the standard and unknown (and also of course the blank standard as well).
When we do this, by simply clicking the checkbox seen here (which also shows how every software correction we have specified can be globally turned on or off *without* changing the actual sample assignments for quick comparisons):
(https://smf.probesoftware.com/oldpics/i39.tinypic.com/mcf78l.jpg)
we obtain the following results:
Un 30 960 sec on SiO2
TakeOff = 40.0 KiloVolt = 20.0 Beam Current = 200. Beam Size = 20
(Magnification (analytical) = 8000), Beam Mode = Analog Spot
(Magnification (default) = 600, Magnification (imaging) = 100)
Image Shift (X,Y): .00, .00
Number of Data Lines: 5 Number of 'Good' Data Lines: 5
First/Last Date-Time: 06/12/2009 04:58:27 PM to 06/12/2009 07:26:32 PM
WARNING- Using Blank Trace Correction
WARNING- Using Alternating On and Off Peak Acquisition
WARNING- Using Aggregate Intensities for Duplicate Elements
Average Total Oxygen: 53.257 Average Total Weight%: 100.000
Average Calculated Oxygen: 53.257 Average Atomic Number: 10.805
Average Excess Oxygen: .000 Average Atomic Weight: 20.029
Average ZAF Iteration: 1.00 Average Quant Iterate: 3.00
Oxygen Calculated by Cation Stoichiometry and Included in the Matrix Correction
Un 30 960 sec on SiO2, Results in Elemental Weight Percents
ELEM: Ti Ti Ti Ti Ti Si O
TYPE: ANAL ANAL ANAL ANAL ANAL SPEC CALC
BGDS: LIN LIN LIN LIN LIN
TIME: 960.00 .00 .00 .00 .00
BEAM: 199.56 .00 .00 .00 .00
AGGR: 5
ELEM: Ti Ti Ti Ti Ti Si O SUM
XRAY: (ka) (ka) (ka) (ka) (ka) () ()
266 .00028 .00000 .00000 .00000 .00000 46.7430 53.2572 100.001
267 .00007 .00000 .00000 .00000 .00000 46.7430 53.2571 100.000
268 .00013 .00000 .00000 .00000 .00000 46.7430 53.2571 100.000
269 .00029 .00000 .00000 .00000 .00000 46.7430 53.2572 100.001
270 .00041 .00000 .00000 .00000 .00000 46.7430 53.2573 100.001
AVER: .00024 .00000 .00000 .00000 .00000 46.743 53.257 100.000
SDEV: .00014 .00000 .00000 .00000 .00000 .000 .000 .00023
SERR: .00006 .00000 .00000 .00000 .00000 .00000 .00004
%RSD: 57.1257 .00002 .00002 .00002 .00002 .00000 .00017
STDS: 922 0 0 0 0 0 0
STKF: .5621 .0000 .0000 .0000 .0000 .0000 .0000
STCT: 5529.37 .00 .00 .00 .00 .00 .00
UNKF: .0000 .0000 .0000 .0000 .0000 .0000 .0000
UNCT: .02 .00 .00 .00 .00 .00 .00
UNBG: 9.23 .00 .00 .00 .00 .00 .00
ZCOR: 1.1969 .0000 .0000 .0000 .0000 .0000 .0000
KRAW: .00000 .00000 .00000 .00000 .00000 .00000 .00000
PKBG: 1.00211 .00000 .00000 .00000 .00000 .00000 .00000
BLNK#: 26 26 26 26 26 ---- ----
BLNKL: .000142 .000142 .000142 .000142 .000142 ---- ----
BLNKV: -.00133 .000000 .000000 .000000 .000000 ---- ----
Wow! We got 2.4 PPM +/- 1.4 PPM or less than one sigma within our blank standard of 1.42 PPM Ti.
Please refer to this publication which describes these methods in more detail:
http://probesoftware.com/download/Ti%20in%20Quartz,%20Am.%20Min.%20Donovan,%202011.pdf
Next a look at t-testing for confidence- after all we *only* acquired 5 data points in our average not an infinite number! ;)
Excellent field guide, Probeman!
One comment, one question.
(1) The acquisition option to "use alternate on-off peak acquire" is a great feature now I understand what it does. Really cool. But that term wasn't clear to me. What I thought it would do, was to reduce the spectrometer driving time of an analysis, by having one acquisition start counting on the peak, the next spot in a different location would start with one of the backgrounds, the one after that with the on-peak, etc. This would reduce the length of a single analysis a tiny bit in the case of 1-element-per-spectro, and save a tiny bit of time. Never spent much time thinking further about this, because I didn't see how this would be useful for me. So, that first impression of a feature is important. I wouldn't know if I hadn't read this post, and there are still many other hidden features I don't use. Rather than stumbling over these life-changing options, how can the poweruser or average probe manager find out what else is possible? Guess I'll just ask.
(2) "Unknown Count Time Factor". It is clear what it does, but the "why would I want to use this" is where I got it all wrong. Acquiring standards and unknowns under the same conditions, including counting time, seems like a reasonable principle to me to make data quality transparent. So, I thought no point of using UCTF. In the context of trace element analysis, particularly in combination with the alternate on-off-peak, this value of this function is instantly clear. Can you maybe give us some guidance on the actual number?
Many thanks
Eric
Quote from: ericwgh on November 08, 2013, 03:29:52 PM
Rather than stumbling over these life-changing options, how can the poweruser or average probe manager find out what else is possible? Guess I'll just ask.
I would ask. To be honest, there are lots of features that even I don't know about (or at least have probably forgotten about), but I'm sure someone will be happy to chime in! ;)
Quote from: ericwgh
In the context of trace element analysis, particularly in combination with the alternate on-off-peak, this value of this function is instantly clear. Can you maybe give us some guidance on the actual number?
Good question.
Maybe it's a weird thing, but it has occurred to me that EPMA is one of those fields that quite often will run their standards using different conditions than their unknowns. This can introduce sometimes subtle, sometimes not so subtle biases in the measurements. In fact one of the things that I really like about PFE is that the default acquisition mode is to acquire one's standards using the exact same conditions as their unknowns!
Not just the keV and beam current and beam size, but also the same elements and in the same order! Of course the "quick standard" mode is always just a mouse click away when time is more important than accuracy, but I almost always acquire the first set of standards *not* using the "quick standard" mode, because one, I want to have lots of secondary standard measurements to check accuracy, two, one can easily see if there are off-peak problems because instead of measuring zero, one will observe negative intensities (you did run a full wavescan on all the elements to check for off-peak interferences, right?), and three, if there are any on-peak interferences present, one will likely observe intensities above zero instead of the expected zero intensity. And of course, by then you've already acquired the needed interference calibration intensities on the (hopefully) appropriate standards!
So, if I'm measuring major elements using standards with similar concentrations of those elements, clearly it makes perfect sense to just utilize the default "unknown count time factor" of 1.0. Since I'll normally be acquiring and averaging a number of (3 to 5) acquisitions for each standard (a "sample" being the fundamental PFE unit of acquisition, e.g., an average and its associated variance), we'll have plenty of precision on the standards anyway for major elements.
And let's get this next issue out of the way right now- though I'm always open to discussion of course! :)
When measuring trace elements we will almost always want to utilize a *primary* standard with a high concentration of the element for best sensitivity. If you need convincing on this point, look closely at typical calculations for detection limits, as seen here for example, and note that the highest standard intensity relative to the concentration of the element in the standard (think pure element) is going to give the best detection limit:
(https://smf.probesoftware.com/oldpics/i40.tinypic.com/11gjcat.jpg)
Of course we'll still want to run a secondary standard with a concentration and a matrix similar to our unknowns, or even better, a standard "blank" sample as a check on the accuracy with which we can measure a zero concentration (though PFE works just as well with non-zero "blank" standards- as long as the matrix is similar).
In a related fashion, the acquisition of standards for characterizing trace elements is less important than the background measurement of the unknown, which now dominates the accuracy of the trace element measurement. Also consider that the variance of the standard intensity will often be at the sub percent level, while the variance of the sample on-peak and background trace level measurements will dominate the trace element sensitivity. Also remember that when performing the off-peak measurement we are subtracting the two values so the differential variance will increase (by the square root of two if the intensities are the same). One more reason for considering the method of combining MAN bgd and blank corrections together, for samples with a simple matrix where one has a chance of finding a suitable blank material as discussed here:
http://smf.probesoftware.com/index.php?topic=29.msg237#msg237
And all other measurements will fall between these two extremes.
So to summarize: if acquiring major elements using standards with similar concentrations, we'll probably use unknown count time factors of 1, while minor and trace elements can benefit from unknown count factors of 2, 3, 4 or higher. The Ti in quartz trace element measurements presented above utilized unknown count factors of 32! That is, we counted 32 times longer on the unknown than the standard and we obtained detection limits around a few PPM.
In fact when "load balancing" our spectrometers to avoid any spectrometers sitting "idle" we can also utilize the unknown count time factors to deal with this optimization as seen here in a typical glass setup for majors and traces.
(https://smf.probesoftware.com/oldpics/i41.tinypic.com/9a0qv6.jpg)
The circled interval is where the beam current is increased for the subsequent trace elements, so we're both increasing the unknown count factors and the beam current for our trace elements.
Quote from: Probeman on November 07, 2013, 02:50:04 PM
Next a look at t-testing for confidence- after all we *only* acquired 5 data points in our average not an infinite number! ;)
So continuing our discussion of Ti in quartz, we now turn to dealing with our confidence when we have not performed an infinite number of replicate analyses- though in practice we'll probably never need more than 25 or 30 replicates. But what if we only measured 3 or 5 replicates- how does that affect our confidence? Those interested should read this wiki page before continuing:
http://en.wikipedia.org/wiki/Student's_t-test
In any case, the result is a table which provides a multiplier to adjust our measured standard deviation based on the confidence level we would like to quote (60% confidence, 80% confidence, 95% confidence, etc) and the number of measurements we settled for. Here is the first dozen or so lines in the table which can be printed from the Analytical menu in Probe for EPMA:
# pts d.f. 0.60 0.80 0.90 0.95 0.99
2 1 1.3764 3.0777 6.3138 12.7062 63.6567
3 2 1.0607 1.8856 2.9200 4.3027 9.9248
4 3 .9785 1.6377 2.3534 3.1824 5.8409
5 4 .9410 1.5332 2.1318 2.7764 4.6041
6 5 .9195 1.4759 2.0151 2.5706 4.0321
7 6 .9057 1.4398 1.9432 2.4469 3.7074
8 7 .8961 1.4149 1.8946 2.3646 3.4995
9 8 .8889 1.3968 1.8595 2.3061 3.3554
10 9 .8834 1.3831 1.8331 2.2622 3.2498
11 10 .8791 1.3722 1.8125 2.2281 3.1693
12 11 .8755 1.3634 1.7959 2.2010 3.1058
13 12 .8726 1.3562 1.7823 2.1788 3.0545
14 13 .8702 1.3502 1.7709 2.1604 3.0123
15 14 .8681 1.3451 1.7613 2.1448 2.9768
16 15 .8662 1.3406 1.7531 2.1315 2.9467
You will notice that the 99% confidence interval (CI) column begins to level out to around 3 after a reasonably large number of measurements (# pts) which is consistent with what we know from normal statistics (3 standard deviation is usually quoted as 99% confidence).
But if we have less than 10 or 20 measurements, one can see that the alpha term begins to increase particularly for 99% confidence until when we have only two measurements, the alpha term for 99% confidence is 63.6567. In other words, you'd better measure more points or you won't be able to distinguish between two different but similar compositions with any reasonable degree of confidence.
Probe for EPMA provides these calculations automatically from the Calculation Options dialog accessed from the Analyze! window. That is, when the Calculate Detection Limits And Sensitivity checkbox is checked as seen here:
(https://smf.probesoftware.com/oldpics/i41.tinypic.com/2irm8ih.jpg)
and one has acquired more than a single measurement (you do know that a single measurement has no scientific meaning, don't you?), the software will calculate the t-test detection limits as seen here:
Detection Limit (t-test) in Elemental Weight Percent (Average of Sample):
ELEM: Ti Ti Ti Ti Ti
60ci .00008 .00004 .00019 .00014 .00009
80ci .00013 .00007 .00032 .00022 .00015
90ci .00018 .00009 .00044 .00031 .00021
95ci .00024 .00012 .00057 .00040 .00028
99ci .00039 .00020 .00095 .00066 .00046
Notice first of all that the analytical sensitivities are not shown because they only are useful for major and minor elements, so the software ignores that calculation unless the concentration is over 1 wt%.
Next note that if we are willing to accept a very low level of confidence, we can claim sub-PPM detection! But if we want to be a little more credible and quote 99% confidence intervals we obtain between 2 and 7 PPM detection.
If we now turn on the aggregate feature in the Analytical | Analysis Options menu dialog we get some additional benefit from combining photons from all our spectrometers:
Detection Limit (t-test) in Elemental Weight Percent (Average of Sample):
ELEM: Ti Ti Ti Ti Ti
60ci .00008 --- --- --- ---
80ci .00012 --- --- --- ---
90ci .00017 --- --- --- ---
95ci .00022 --- --- --- ---
99ci .00037 --- --- --- ---
and now obtain a 99% CI of 3.7 PPM and 95% CI of 2.2 PPM and this is *only* for the average of 5 points. We can be assured by "student" that more measurements would further improve our confidence.
Before I forgot, I wanted to post Mike Jercinovic's wonderful comment on the usefulness of the "blank" correction, particularly with regard to the question: "Can we measure a zero concentration?"...
Mike says: "If you can't analyze something, then see if you can analyze nothing... because, if you can't do nothing right, then you can't do anything."
That is, if one doesn't have a trace element standard with an ideal concentration of the element of interest, then try measuring a sample with *none* of the element of interest...
In other words, if you do not obtain a zero concentration (within statistics of course) when measuring a "blank" sample (a well characterized sample with a similar matrix to one's unknown (SiO2, TiO2, ZrSiO4, etc), which contains a zero concentration of the element of interest), then you have need to apply a "blank" correction to improve accuracy.
Of course the blank correction is only the last step in properly optimizing the measurement. That is after obtaining a primary standard with a high concentration of the element of interest (because that improves our standard statistics and overall analytical sensitivity), proper placement of the background measurement positions (and correcting for curvature in the background and possible application of the multi-point background method described in Mike's PPT given at a recent workshop) and careful treatment of possible spectral interferences.
http://epmalab.uoregon.edu/Workshop2/CAMCOR%20meeting%209-07%20Jercinovic.pdf
And I hasten to add, the cool thing about the "blank" correction in Probe for EPMA is, that it calculates the physics necessary to utilize a blank standard that might also contain a *non-zero* concentration of the element of interest.
Here is the abstract John Armstrong and I just submitted to M&M on this trace element topic (see attached).
john
Quote from: Probeman on November 07, 2013, 02:50:04 PM
Please refer to this publication which describes these methods in more detail:
http://probesoftware.com/Ti%20in%20Quartz,%20Am.%20Min.%20Donovan,%202011.pdf
In case anyone would like to try and reproduce the trace calculations shown above in Probe for EPMA for themselves, I've attached the Ti in Quartz MDB file (also the Standard.mdb file since this older probe database does not contain the standard compositions).
Hi John and all
Again necroing an old thread, however I was just wondering peoples thoughts on background fits for Al in quartz due to the tail of the Si peak, as I am again attempting to dabble in Ti/Al in quartz. John your 2010 paper utilises a polynomial regression fit on a detailed WDS scan, however I was wondering if this was all done before you had implemented the multi-point fit into PFE? I guess my question is, is this still necessary, or is a good multipoint fit with a number of points either side and using PFE polynomial fit good enough? I am conducting wavescans at the moment and will try both methods, but was just wondering peoples thoughts.
Cheers
Quote from: BenjaminWade on August 17, 2015, 06:35:16 PM
Hi John and all
Again necroing an old thread, however I was just wondering peoples thoughts on background fits for Al in quartz due to the tail of the Si peak, as I am again attempting to dabble in Ti/Al in quartz. John your 2010 paper utilises a polynomial regression fit on a detailed WDS scan, however I was wondering if this was all done before you had implemented the multi-point fit into PFE? I guess my question is, is this still necessary, or is a good multipoint fit with a number of points either side and using PFE polynomial fit good enough? I am conducting wavescans at the moment and will try both methods, but was just wondering peoples thoughts.
Cheers
Hi Ben,
Quote from: BenjaminWade on August 17, 2015, 06:35:16 PM
Hi John and all
Again necroing an old thread, however I was just wondering peoples thoughts on background fits for Al in quartz due to the tail of the Si peak, as I am again attempting to dabble in Ti/Al in quartz. John your 2010 paper utilises a polynomial regression fit on a detailed WDS scan, however I was wondering if this was all done before you had implemented the multi-point fit into PFE? I guess my question is, is this still necessary, or is a good multipoint fit with a number of points either side and using PFE polynomial fit good enough? I am conducting wavescans at the moment and will try both methods, but was just wondering peoples thoughts.
Cheers
Hi Ben,
No apology necessary. This is old stuff but good stuff- and it's new to some people!
To answer your question: you could use the MPB method instead of an exponential or polynomial fit to deal with the tail of an interfering peak on the off-peak position. The advantage of the MPB method is that it fits the actual shape of the continuum, and if you consider the extended tail of the secondary line as just more "background", then it should all come out fine.
I've got an example here for a weird test I did recently looking at possible oxygen in the carbon coat deposition. Here is the problem:
(https://smf.probesoftware.com/oldpics/i57.tinypic.com/i50y2e.jpg)
Notice that because of the large tails from the Fe La/Lb/Ll lines on the low side of the oxygen peak, I had to use 4 MPBs all on the high side of the oxygen peak and extrapolate using a polynomial function (I know, dangerous) underneath the O ka peak for the background intensity.
By the way I got around 0.3 wt% oxygen (not thin film geometry corrected) for the carbon coated FeS2 and around 0.1 wt% oxygen in the uncoated FeS2 (the variance on both was around 100 PPM). So it seems that maybe we are adding oxygen to the surface with our carbon coater (the venerable Edwards 306A evaporator), but I want to do some more tests on cleaned Ge and Au.
Great, thanks John I will try both methods but will more than likely end up using the MP. I do love a good MP fit, there is something satisfying about it.
Interesting about the carbon coat...I wonder if its the same with ours. Would you be using Ge and Au due to the absence of tails on the low side or because they will be more resistant to potential surface oxidation for your test? or both?
Quote from: BenjaminWade on August 19, 2015, 05:44:16 PM
Great, thanks John I will try both methods but will more than likely end up using the MP. I do love a good MP fit, there is something satisfying about it.
Interesting about the carbon coat...I wonder if its the same with ours. Would you be using Ge and Au due to the absence of tails on the low side or because they will be more resistant to potential surface oxidation for your test? or both?
The possibility of less surface oxidation. But please feel free to report results from your carbon coater as well...
Here's an interesting thing with the alternating on/off peak acquisition mode. I went back to an old Ti in quartz run with 5 aggregate spectrometers and here is the alternating TDI plotted for just the first spectrometer.
(https://smf.probesoftware.com/gallery/1_12_10_16_1_04_52.png)
You might convince someone that there is a downward trend over time, but seems like random statistics to me... Now here is all 5 spectrometers aggregated together:
(https://smf.probesoftware.com/gallery/1_12_10_16_1_05_23.png)
The downward trend is now a little more convincing visually, but it still amounts to almost nothing correction-wise, at least under these conditions. Still nice to see that confirmed rigorously.
john
John,
If you don't mind, and have the info easily available, I'd like to see your PET/LPET high precision wavescans of the Ti ka background on synthetic SiO2. I can find examples of your Spectro 2 scan on the forum, but presumably you have similar wavescans for 3 (or 4?) other spectrometers. Do you have similar from your SX50 instrument also?? I am interested in the frequency and position of the 'holes' on other instruments.
Cheers,
Gareth
Quote from: Gseward on October 21, 2016, 10:45:17 AM
If you don't mind, and have the info easily available, I'd like to see your PET/LPET high precision wavescans of the Ti ka background on synthetic SiO2. I can find examples of your Spectro 2 scan on the forum, but presumably you have similar wavescans for 3 (or 4?) other spectrometers. Do you have similar from your SX50 instrument also?? I am interested in the frequency and position of the 'holes' on other instruments.
Hi Gareth,
Great question!
Yes, I've only seen these "holes" in the continuum on some PET spectrometers, the LPET sometimes shows these effects even more. There's also some issue with the tails of the peaks that are visible when you plot them on a log scale.
It may be dependent on the orientation of the analyzing crystal at the time of manufacturing... at least for PET/LPET crystals. It would be interesting to see some data from JEOL PET crystals.
See files attached below.
john
PS Some of the data above was utilized in my 2011 Amer. Min. paper here:
http://epmalab.uoregon.edu/pdfs/3631Donovan.pdf
At the recent EMAS meeting in Konstantz, Germany (a fantastic meeting by the way), someone asked me if these "holes" in the continuum could be sample related.
I was pretty sure they are only related to the Bragg crystal from secondary diffraction, but it's always good to test ones assumptions. To start, here is a scan 15 keV, 100 nA, 220/sec per point in a synthetic SiO2 on :
(https://smf.probesoftware.com/gallery/395_17_05_17_10_14_53.png)
These artifacts seem to be visible only in the large PET crystals (Mike Jercinovic: do you see these same artifacts in your VLPETs?), but on my sp3 LPET these artifacts are much smaller as seen here:
(https://smf.probesoftware.com/gallery/395_17_05_17_10_15_09.png)
Now as for other matrices I ran Al2O3, Cr2O3 and NiO and they all show the same continuum artifacts, so this artifact is apparently specific to the Bragg crystal as seen here for Al2O3:
(https://smf.probesoftware.com/gallery/395_17_05_17_10_15_29.png)
Actually no, but this is because we sacrificed the sin-theta range below .32000 as the crystal is too big to fully enter the objective port. This was to be able to get the background around UMb. Had we thought about Ti at that time, we could have cut just a little more off of the corner to get down to this wavelength. Next time!
If I recall, Dave Wark found these holes also on standard PET. There may be an orientational aspect to this too, depending on how the crystal is mounted on the holder. Part of why they show up on high count rate crystals is that they are high count rate crystals. You obviously have to have sufficient precision on the WDS scan to see these. This is part of the issue in trace element work is that evaluating the wavelength region at low precision relative to the actual peak counting precision can be really misleading. But of course it takes forever to really do these sorts of high precision scans...hence the use of blanks, the PfE blank correction, and multi-point scanning! Although, as you have shown, for a simple matrix like SiO2 the background may be done pretty well by MAN estimation.
The "holes" were indeed found on both standard and large PET crystals, on JEOL and Cameca instruments. And they certainly are orientation dependent. We ended up swapping our PET crystals that showed the holes with new crystals (oriented differently) that did not show the holes.
I'm currently analyzing some additively manufactured TiO2-SiO2 glasses, and want to use multiple spectrometers with PET crystals to improve the count rates for Ti (samples have less than 1.5 wt% TiO2). After analyzing my standards with 3 PET crystals, I analyzed my TiO2 standard and got totals of 212 wt% - much more than I'd bargained for. I poked around until I found the "aggregate" check-box and all was well. I was subsequently directed to this discussion and wish that I'd visited here first. However, it is heartening to know that, starting from different points, multiple users can negotiate the PFE environment and get to the same place...
Now I'd like to make x-ray maps of the samples to compare with their index of refraction maps. Can I similarly aggregate the counts from multiple spectrometers in Probe Image to improve the Ti images?
Quote from: Falling Man on October 11, 2017, 11:10:29 AM
I'm currently analyzing some additively manufactured TiO2-SiO2 glasses, and want to use multiple spectrometers with PET crystals to improve the count rates for Ti (samples have less than 1.5 wt% TiO2). After analyzing my standards with 3 PET crystals, I analyzed my TiO2 standard and got totals of 212 wt% - much more than I'd bargained for. I poked around until I found the "aggregate" check-box and all was well. I was subsequently directed to this discussion and wish that I'd visited here first. However, it is heartening to know that, starting from different points, multiple users can negotiate the PFE environment and get to the same place...
Hi Rick,
Very cool that you found this on your own! I think the software also prints out a warning if duplicate elements are not aggregated, let me see if I can produce the warning... here it is:
QuoteWARNING- Duplicate analyzed elements are present in the sample matrix!!
Use Aggregate Intensity option or Disable Quant feature for accurate matrix correction.
Quote from: Falling Man on October 11, 2017, 11:10:29 AM
Now I'd like to make x-ray maps of the samples to compare with their index of refraction maps. Can I similarly aggregate the counts from multiple spectrometers in Probe Image to improve the Ti images?
Yes, absolutely. You simply acquire standards in Probe for EPMA (with the duplicate elements) and maps in Probe Image (with the same duplicate element setup). Then you combine these in CalcImage and utilize the same Use Aggregate Intensities for Duplicate Elements checkbox under Analytical | Analysis Options menu.
Here's a very pretty example of trace sulfur acquired by a student at Adelaide using multiple spectrometers:
http://smf.probesoftware.com/index.php?topic=41.msg2180#msg2180
john
Quote from: Probeman on May 17, 2017, 11:22:35 AM
I was pretty sure they are only related to the Bragg crystal from secondary diffraction, but it's always good to test ones assumptions. To start, here is a scan 15 keV, 100 nA, 220/sec per point in a synthetic SiO2 on :
(https://smf.probesoftware.com/gallery/395_17_05_17_10_14_53.png)
Someone asked a great question at the Athens EPMA workshop on whether absorption edges and "negative" peaks have anything in common, and the short answer is "no".
We didn't have time to go into this during the workshop today, but basically "negative peaks" or as some refer to them, "holes" in the background, are due to secondary Bragg diffraction in cases such as the hole under the Ti ka peak on some PET crystals as shown above, or as seen here near the Au Ma line:
https://smf.probesoftware.com/index.php?topic=980.msg8769#msg8769
Here is a citation for the hole near the Au Ma line:
Self, P. G., et al. "Holes in the background in XRS." X‐Ray Spectrometry 19.2 (1990): 59-61.
Absorption edges are quite different looking and are from the mechanism of x-ray absorption for energies just above the ionization energy of an element such as seen here:
https://smf.probesoftware.com/index.php?topic=32.msg1214#msg1214
This absorbing element can be either in the sample or the detector gas!