Hi all
I am posting on an issue that I have noticed for a long time now, but never done anything about it. I am hoping that its an issue that others have, and are able to explain it to me. Namely why some "standards" just never sit on a MAN fit. To clarify I am not talking about:
1) Standards that have trace amounts of an element thus plot above MAN fit; or
2) Standards in which the element sits on a hole in the background continuum; or
3) Standards that have the wrong z-bar value.
I am talking about standards that consistently plot wayyyy below the MAN fit and constantly have to be removed from the MAN fit. You might ask why would you analyse these standards then. Well they are standards that I calibrate on in typical silicate runs, namely for elements such as Sr, Zr, Ba etc (ie Celestite, Zircon, Barite etc). I have posted below some MAN fits of some select elements that illustrate my problem. Note, although these are excel plots, the data used is the direct export of the MAN fit graph (via Export Dialog on right click of MAN plot in PfE). The reason I have put them in excel is, in PfE there is no way that I can display the troublesome standards on the sample plot without including them in the MAN fit regression.
(https://smf.probesoftware.com/gallery/318_13_04_20_9_24_49.jpeg)
(https://smf.probesoftware.com/gallery/318_13_04_20_9_25_37.jpeg)
(https://smf.probesoftware.com/gallery/318_13_04_20_9_25_56.jpeg)
(https://smf.probesoftware.com/gallery/318_13_04_20_9_26_15.jpeg)
What you will notice is that the vast majority of the standards (that I have analysed) that display this problem of low returned CPS, all fluoresce under an electron beam, some a lot more than others. The standards use to construct the MAN fits nominally do not fluoresce under the beam. Without putting up MAN plots of every element, below is another plot of calculated wt% of all analysed elements in the "problem" standards, and you will note for the vast majority they are -ve due to this MAN fit problem, up to an over -1000ppm for some elements
(https://smf.probesoftware.com/gallery/318_13_04_20_9_32_01.jpeg)
Is anyone able to explain to me why this is happening? My understanding is that the ONPK CPS/nA used to create the MAN fits have already run through the ZAF correction routine, which is why we have the "Correct for Continuum Absorption" tickbox to display with and without.
Other than the obvious of just not using MAN fits for analysing these minerals, has anyone for example analysed for low level elements in zircon using MAN and noticed any problems? Would a separate MAN fit using only fluorescent minerals need to be constructed? It is of course not a huge problem for normal MAN work, as I just exclude these from the MAN fit, but if I ever wanted to use the MAN method to analyse low levels of some of these elements in zircon, going by these plots it would be a problem.
Hoping that it is something obvious I am missing.
Cheers
Hi Benjamin.
Few weeks ago, I had a similar observation of one standard being significantly below the MAN curve. It is posted in the topic on the MAN background. The standard was synthetic CaTh(PO4)2 which was prepared by my colleague. And I also observed very significant photoluminescence under the electron beam.
(https://smf.probesoftware.com/gallery/1783_12_06_19_3_30_41.jpeg)
Best regards, Jakub Haifler
Jakub and Ben,
Nice work. You have caused me to re-examine the correction for continuum matrix effects and we should work on this together. But first a little history.
When we first came up with the MAN background correction it was just a simple linear fit vs. Z-bar to deal with the fact that we had some fixed monochrometers permanently tuned to specific emission lines which could not be detuned off-peak to measure the background (necessity is a mother!). This was back in the mid 1980s. About 8 years later I was at an M&M conference and was explaining to John Armstrong how we were doing this mean atomic number (MAN) background method, and he said to me: "Are you doing a correction for continuum absorption to the standard intensities?", and I said "uh, no". But it made perfect sense that one should do that since the continuum *generated* in each standard will depend only on the average Z (Kramer's Law), but the continuum actually *emitted* by each standard will depend on the composition (Beer's Law). So as soon as I got home I added a continuum absorption into the code.
Because continuum x-rays are emitted somewhat differently than characteristic x-rays, the first thing I tried was a simple continuum absorption correction from Heinrich modified by Mycklebust, but that actually made the MAN fit worse! It was a very primitive calculation. So then I tried using just the absorption correction term from the normal matrix correction calculations, since the energy of these continuum emissions are essentially the same energy as the characteristic emission! Hey, you do what you have to! :D
That actually worked pretty well and in almost every case produced a better fit to the MAN intensities, so we decided to just utilize that. That was 25 years ago! :o
Fast forward to 2016, when Jared Singer and I wrote this up and we included John Armstrong as a co-author to thank him for his suggestion to add a correction to the continuum intensities for the MAN fit:
https://epmalab.uoregon.edu/publ/A%20new%20EPMA%20method%20for%20fast%20trace%20element%20analysis%20in%20simple%20matrices.pdf
So here are data from a test run I did recently to look at backscatter coefficients on some standard materials, but I had included a few analyzed elements just for fun. So looking at one of these emission lines, here are the Zr La MAN intensities measured on a number of standards, *without* any correction to the continuum x-rays:
(https://smf.probesoftware.com/gallery/1_05_07_19_9_03_18.png)
Next, here are the same intensities but corrected for absorption (only) using the absorption correction term from the matrix correction physics for each standard:
(https://smf.probesoftware.com/gallery/1_05_07_19_9_03_38.png)
That is a significant improvement over doing no correction at all to the MAN standard intensities! And that's where things stood until now. But now you guys are pointing out (correctly I should add!) that we really should also be looking at continuum fluorescence, because the continuum can fluoresce other characteristic x-rays in the MAN standards (and of course the unknowns as well!), that might (if we are unlucky), be the same energy as our analytical emission line and be detected by the spectrometer. And correct me if I am wrong, but it seems to me that we don't need to worry about backscatter loss or stopping power calculations since this is just x-ray physics.
Now I don't have any code laying around to do a proper treatment for continuum absorption and fluorescence, but thinking back to my early efforts, I decided to try the fluorescence correction term for each standard (yes, I know it shouldn't work), along with the absorption correction term, from the normal matrix correction calculations, and here is what we get:
(https://smf.probesoftware.com/gallery/1_05_07_19_9_03_55.png)
Well, I'll be darned. That's even a better fit!
OK, let's get really crazy and just for fun, let's throw in the atomic number correction term as well, and now we get this:
(https://smf.probesoftware.com/gallery/1_05_07_19_9_04_14.png)
Dang, that's even better! I realize that this makes no sense at all from a physics perspective, but please try it out yourselves and please let me know what you find. And if someone can come up with a proper continuum absorption/fluorescence correction, I would be thrilled to try it out!
But I think it's important to keep in mind that all these corrections to the MAN background are quite small, and also they are essentially normalized out when we perform the actual MAN correction. Because although we correct each MAN standard intensity for the matrix effects on the emitted continuum by multiplying the MAN standard intensities times the matrix correction (prior to the MAN fit regression), we then divide the unknown MAN intensity (derived from the MAN standard fit), by the unknown matrix correction!
For those reading all this that haven't thought about this stuff as much as Jakub and Ben have, the steps are:
1. Correct each MAN (continuum) standard intensity for the matrix correction physics for that standard (or whatever physics you want to use).
2. Calculate the fit coefficients for these corrected continuum intensities as a function of the average atomic number of each standard.
3. During the (unknown) sample matrix iteration, calculate the average Z of the unknown sample.
4. Determine the background intensity from the MAN standard fit coefficients based on the current unknown sample average atomic number.
5. De-correct the calculated unknown sample MAN intensity by the matrix correction (or whatever physics you want) for the current unknown composition.
6. Subtract the de-corrected unknown sample MAN intensity from the measured on-peak sample intensity
7. Repeat steps 3 to 6 until the MAN calculation converges...
It's so simple! >:(
I think the first step is to look at the x-ray physics and see if we can try and understand what is it about these compositions that cause them to fall below the general trend of the MAN fit. It could be fluorescence, but it could also be related to how we calculate average atomic number! For example, we already know that we *cannot* utilize mass fractions for calculating average Z for backscatter coefficients, as shown here:
https://smf.probesoftware.com/index.php?topic=1111.msg8249#msg8249
Remember to log in to see the pdf attachment. I'll be presenting on this at M&M next month. I've played around with using a Z fraction weighting for calculating average atomic number for MAN plots and haven't found anything that works as well as plain old mass fraction weighting, but maybe it's a combination of absorption/fluorescence and average Z weighting for the continuum issues we are seeing...
Try looking at the A/Z ratios for the elements in these compounds, and take a look at the "alternative" Zbar calculations in the Output | Calculate Alternative Zbars menu in the Standard application.
And finally if we want to utilize the MAN background correction for high Z materials, please remember that accuracy can be problematic. Which is why for best accuracy one should always combine the MAN correction with the blank correction using a suitable blank standard. For zircon a synthetic zircon works excellently.
By the way, the code from Heinrich/Myklebust to calculate continuum absorption is in the CalcZAF open source on Github, which is currently not utilized for the MAN fit, but I paste it here:
Sub ZAFGetContinuumAbsorption(continuum_absorption() As Single, zaf As TypeZAF)
' Calculate continuum absorption correction (modified Heinrich from Myklebust) for all emitters
ierror = False
On Error GoTo ZAFGetContinuumAbsorptionError
Dim i As Integer
ReDim m7(1 To MAXCHAN1%) As Single
ReDim h(1 To MAXCHAN1%) As Single
ReDim gsmp(1 To MAXCHAN1%) As Single
ReDim gstd(1 To MAXCHAN1%) As Single
' Calculate standard (pure element) absorption
For i% = 1 To zaf.in1%
If zaf.il%(i%) <= MAXRAY% - 1 Then
h!(i%) = 0.0000012 * (zaf.eO!(i%) ^ 1.65 - zaf.eC!(i%) ^ 1.65)
gstd!(i%) = (1# + h!(i%) * zaf.mup!(i%, i%) * zaf.m1!(i%)) ^ 2
' Modify intensity using depth production and anisotropy from Small and Myklebust
gstd!(i%) = gstd!(i%) * 1.15 - 0.15 * 1# / gstd!(i%)
End If
Next i%
' Calculate sample absorption
For i% = 1 To zaf.in1%
If zaf.il%(i%) <= MAXRAY% - 1 Then
m7!(i%) = ZAFMACCal(i%, zaf)
gsmp!(i%) = (1# + h!(i%) * m7!(i%) * zaf.m1!(i%)) ^ 2
' Modify intensity using depth production and anisotropy from Small and Myklebust
gsmp!(i%) = gsmp!(i%) * 1.15 - 0.15 * 1# / gsmp!(i%)
End If
Next i%
' Create continuum absorption correction factors
For i% = 1 To zaf.in1%
If zaf.il%(i%) <= MAXRAY% - 1 Then
continuum_absorption!(i%) = gsmp!(i%) / gstd!(i%)
End If
Next i%
Exit Sub
' Errors
ZAFGetContinuumAbsorptionError:
MsgBox Error$, vbOKOnly + vbCritical, "ZAFGetContinuumAbsorption"
ierror = True
Exit Sub
End Sub
Note as has been pointed out, we should probably perform a rigorous treatment for both continuum absorption and fluorescence!
Hi Ben,
Well I've been looking into the code and testing some data and found out why the Heinrich and Myklebust continuum absorption correction wasn't working as well as it should have. Now the Heinrich continuum absorption and the Armstrong phi-rho-Z absorption corrections both give similar results for the MAN curves, with the Heinrich method slightly better.
The data is a probe run I recently did looking at some standards with large differences in their A/Z elements, originally to look at the backscatter coefficients, but I also acquired Si Ka and Zr La just for fun. So here are the correction factors for the Armstrong absorption correction for these two elements:
MAN fit data for Si ka, SP2 LPET, 15 keV
MANStd MANAss MANSet Npts Z-bar Cps AbsCorr
522 1 1 1 21.9300 9.52304 1.34430
524 2 1 2 23.9842 8.90178 1.48430
526 3 1 3 26.0000 8.46901 1.64735
528 4 1 4 28.0000 7.66998 1.83075
529 5 1 5 28.9580 9.10173 1.95332
532 6 1 6 32.0000 7.31487 2.35743
540 7 1 7 39.8400 16.8615 1.27072
542 8 1 8 41.7280 13.3753 1.34197
547 9 1 9 46.6880 12.6544 1.56866
550 10 1 10 50.0000 14.1195 1.74366
552 11 1 11 52.0000 11.7216 1.85512
574 12 1 12 73.8284 26.8303 1.36576
710 13 1 13 25.5960 9.40190 1.69769
729 14 1 14 71.1680 23.5443 1.46016
731 15 1 15 73.1560 18.1162 1.51554
732 16 1 16 41.0880 13.9311 1.53672
MAN fit data for Zr la, SP3 LPET, 15 keV
MANStd MANAss MANSet Npts Z-bar Cps AbsCorr
514 1 1 1 14.0000 3.45187 1.49583
16 2 1 2 67.2231 11.8851 1.28334
19 3 1 3 50.8415 8.95550 1.65873
20 4 1 4 67.2231 11.3864 1.28334
263 5 1 5 18.6926 5.84859 1.15176
272 6 1 6 20.0138 5.93751 1.20616
273 7 1 7 10.5798 3.41837 1.17087
274 8 1 8 19.4692 5.67127 1.17913
275 9 1 9 18.0833 5.36072 1.12748
376 10 1 10 26.7484 6.25967 1.23218
386 11 1 11 62.7264 11.6543 1.17626
522 12 1 12 21.9300 8.60179 1.02954
524 13 1 13 23.9842 7.90280 1.10178
526 14 1 14 26.0000 8.42461 1.19110
528 15 1 15 28.0000 8.20279 1.28806
529 16 1 16 28.9580 8.72435 1.35356
532 17 1 17 32.0000 7.71448 1.57259
547 18 1 18 46.6880 12.8653 1.16002
550 19 1 19 50.0000 12.3656 1.26612
710 20 1 20 25.5960 8.06984 1.22105
731 21 1 21 73.1560 15.1411 1.15042
732 22 1 22 41.0880 12.7766 1.14610
And here are the Heinrich absorption correction factors:
MAN fit data for Si ka, SP2 LPET, 15 keV
MANStd MANAss MANSet Npts Z-bar Cps AbsCorr
522 1 1 1 21.9300 9.52304 1.41239
524 2 1 2 23.9842 8.90178 1.58663
526 3 1 3 26.0000 8.46901 1.79544
528 4 1 4 28.0000 7.66998 2.05681
529 5 1 5 28.9580 9.10173 2.14897
532 6 1 6 32.0000 7.31487 2.58038
540 7 1 7 39.8400 16.8615 1.30489
542 8 1 8 41.7280 13.3753 1.38678
547 9 1 9 46.6880 12.6544 1.64365
550 10 1 10 50.0000 14.1195 1.80516
552 11 1 11 52.0000 11.7216 1.88858
574 12 1 12 73.8284 26.8303 1.35475
710 13 1 13 25.5960 9.40190 1.87763
729 14 1 14 71.1680 23.5443 1.48133
731 15 1 15 73.1560 18.1162 1.53529
732 16 1 16 41.0880 13.9311 1.62064
MAN fit data for Zr la, SP3 LPET, 15 keV
MANStd MANAss MANSet Npts Z-bar Cps AbsCorr
514 1 1 1 14.0000 3.45187 1.66857
16 2 1 2 67.2231 11.8851 1.30786
19 3 1 3 50.8415 8.95550 1.75943
20 4 1 4 67.2231 11.3864 1.30786
263 5 1 5 18.6926 5.84859 1.21173
272 6 1 6 20.0138 5.93751 1.28797
273 7 1 7 10.5798 3.41837 1.24194
274 8 1 8 19.4692 5.67127 1.24153
275 9 1 9 18.0833 5.36072 1.17560
376 10 1 10 26.7484 6.25967 1.29463
386 11 1 11 62.7264 11.6543 1.19327
522 12 1 12 21.9300 8.60179 1.04781
524 13 1 13 23.9842 7.90280 1.13826
526 14 1 14 26.0000 8.42461 1.25239
528 15 1 15 28.0000 8.20279 1.38910
529 16 1 16 28.9580 8.72435 1.43865
532 17 1 17 32.0000 7.71448 1.66847
547 18 1 18 46.6880 12.8653 1.18220
550 19 1 19 50.0000 12.3656 1.28135
710 20 1 20 25.5960 8.06984 1.30032
731 21 1 21 73.1560 15.1411 1.14379
732 22 1 22 41.0880 12.7766 1.17511
You'll notice that they track quite well, with the Heinrich absorption factors being somewhat larger.
Now when we apply the Armstrong absorption factors to this data set, we get a average fit deviation of 11.1 for Si Ka, and 9.19 for Zr La. When the Heinrich factors are applied we get 11.0 for Si Ka and 9.0 for Zr La. Not a big change, but an improvement. Obviously we will need to test on many other data sets to know better, but we are working on a new version of Probe for EPMA with additional options in the GUI for toggling these calculations on and off. More on that tomorrow.
But I do have a physics question. You mentioned the possibility that there might be some sort of a fluorescence correction needed for the MAN curve fit, at least for certain MAN standard compositions, but I'm not sure how that physics would work. I can understand that we can get additional characteristic fluorescence from continuum production. For example at 15 keV about 1% of the Fe Ka produced in pure Fe metal is from continuum radiation:
https://smf.probesoftware.com/index.php?topic=58.msg5621#msg5621
But how can we get additional continuum radiation production? That is, when the element isn't present so there are no atoms to be fluoresced? Continuum radiation should solely be a product of electron deceleration, no? Are you thinking of a sample absorption edge effect where some of the continuum radiation produced is being absorbed on the way out? But that should be taken care of by the continuum absorption correction I would think.
I've added a MAN fluorescence correction plot option as seen here:
(https://smf.probesoftware.com/gallery/395_06_07_19_6_47_02.png)
But right now it just returns a 1.0 correction factor when using the alternative continuum correction method (it utilizes the characteristic fluorescence correction factors when using the default continuum correction method which is wrong of course but interestingly it does seem to improve the fit- more on this later today). But if you (or anyone else) can think of what fluorescence physics might be involved, we will be happy to code it in so we can test it.
Hi Ben (and Jakub),
I went a little crazy this weekend and we added several new options for testing different MAN fit parameters including switching between the matrix and continuum specific absorption correction methods and also using a Z fraction based average atomic number calculation as shown here:
(https://smf.probesoftware.com/gallery/1_07_07_19_2_23_25.png)
This latest version of Probe for EPMA (12.6.6) will also save these parameters to the MDB file and display them when selected. Don't worry though, the default mode is to still utilize the MAN continuum absorption correction method we've been using all these years (using the characteristic absorption correction for the MAN intensities).
Here is an example of the new output using the Z (electron) fraction average atomic number method for Zr La:
(https://smf.probesoftware.com/gallery/1_07_07_19_2_24_06.png)
This is an improvement over the weight fraction method which gives a average deviation of 9.19. Note benitoite (376) now plots on the fitted line.
These options will give you a chance to try different data sets and report what you are seeing in your MAN curves. Jakub if you'd be interested in trying this Z fraction method for calculating average atomic number see the abstract attached below, which I submitted to M&M 2019 for a presentation in August in Portland. Yes, it's specific to elastic (BSE) scattering, but similar physics should apply to both productions since neither elastic scattering nor continuum are significantly affected by mass (atomic weight).
Edit by John: I've decided to move these global MAN options from the Analysis Options dialog to the MAN dialog, to make it easier to compare the different results graphically. Give me a day to two and I will try to get to it.
Hi Ben,
OK, we just uploaded a new version of Probe for EPMA (12.6.7) which now has all the MAN fit options in the MAN dialog as seen here:
(https://smf.probesoftware.com/gallery/1_08_07_19_7_53_27.png)
This should allow one to evaluate these options on various datasets more easily. If you click the Cancel button the options selected are not saved for quant calculations and instead only displayed. But if you click the OK button, the MAN fit options are saved for quant calculations on samples.
Please let me know what you think. We can work together on further code changes as necessary.
john
Hi John, Jakub, and all
Thankyou for the detailed reply, it has taken me some time to digest and have more of a play with my data using the new features. I think I can conclude at least one thing. That being, as you have surmised I think the effect I have been seeing is almost solely down to how the z-bar is calculated. Attached below are the MAN plots (including the problem elements) of the same elements from the original post (Cr Ka, Na Ka, and Ti Ka) both with the "traditional" default z-bar, and with an z-fraction exponent=0.7 z-bar fit.
(https://smf.probesoftware.com/gallery/318_13_04_20_9_33_47.jpeg)
(https://smf.probesoftware.com/gallery/318_13_04_20_9_34_12.jpeg)
(https://smf.probesoftware.com/gallery/318_13_04_20_9_34_40.jpeg)
(https://smf.probesoftware.com/gallery/318_13_04_20_9_35_07.jpeg)
(https://smf.probesoftware.com/gallery/318_13_04_20_9_35_29.jpeg)
(https://smf.probesoftware.com/gallery/318_13_04_20_9_35_48.jpeg)
As you can see the "problem" photoluminescent standards pull back right onto the MAN fit on both the Cr and Ti plots. Na is a bit messier as expected but arguably the plot is better using the z-fraction method. I have looked at most elements in my run now, and I can say that if I want to include the problem minerals (zircon, barite, etc) in the MAN fit, using the modified z-fraction method gives a vastly superior fit for all elements on PET and LIF, and probably TAP but the data is much noisier.
As such I may have been chasing a red herring regarding fluorescence. As you mention it is hard to conceptualise physically how this could have an effect on the generation of the background continuum at the peak position, I was more thinking that there was a first principle correlation in that the majority of problem standards luminesced under the beam more than the other standards. However I think this is probably just an unhappy coincidence, due to the fact that the problem standards are also compounds comprised of a heavy element with a light element, thus the z-bar recalculation will affect these standards the most (ie Zr/Si, Zr/O, Ca/W, Ca/Mo, Sr/C, Ba/S etc).
So this brings me to my next question, should we be using the exponent method by default? And if so what exponent should I be using? I note in your original post in the other thread you use 0.8, PfE uses 0.7 by default, and then in your post below you use 0.6. I am guessing that was playing around with the value to fit the empirical data the best. It sounds like potentially you are working on a software update to put those parameters into the MAN window to play around with it more easier, will be fun to use.
I guess we are talking about quite small differences in background interpolation, and when measuring minor to major elements its not really a problem anyway. But interesting! Let me know what you think.
Cheers
Hi Jakub
Interestingly I also note in your plot below, that if we recalculate that synthetic cheralite using exponent of 0.7 its z-bar comes down to ~34.6, bringing it much more in line with the MAN fit. Of course the rest of the MAN fit would change as well, so it may well bring it right back onto the line.
cheers
(https://smf.probesoftware.com/gallery/1783_12_06_19_3_30_41.jpeg)
Quote from: BenjaminWade on July 08, 2019, 07:58:21 PM
Hi John, Jakub, and all
Thankyou for the detailed reply, it has taken me some time to digest and have more of a play with my data using the new features. I think I can conclude at least one thing. That being, as you have surmised I think the effect I have been seeing is almost solely down to how the z-bar is calculated. Attached below are the MAN plots (including the problem elements) of the same elements from the original post (Cr Ka, Na Ka, and Ti Ka) both with the "traditional" default z-bar, and with an z-fraction exponent=0.7 z-bar fit.
(https://smf.probesoftware.com/gallery/318_13_04_20_9_33_47.jpeg)
(https://smf.probesoftware.com/gallery/318_13_04_20_9_34_12.jpeg)
[snip]
Hi Ben,
Wow, this is an impressive improvement. Note particularly how standards 618, 617 and 739 change positions between the two average Z methods with the Z fraction method giving excellent results. If you download the latest PFE (v. 12.6.7) I've moved these options all into the MAN plot dialog.
Since this Z fraction averaging seems to deal pretty well with the fitting for these high Z materials, and we can't think of what fluorescence has to do with continuum emission I'm going to remove the fluorescence option from the MAN code later this week, unless anyone has any objections.
It's interesting that both you and Jakub seemed struck by the fact that many/most of the "problem" standards were cathodoluminescent, but I think that was just a coincidence, as I can't think of how band gap emission is related to continuum fluorescence, but what do I know? The only thing I noticed was that applying the characteristic fluorescence correction to the MAN intensities did seem to make a small improvement to the fit, but that was probably just a spurious correlation.
As to the exact exponent to utilize for the Z fraction weighting, yes, I was experimenting with different values, but something between 0.6 and 0.8 seems to give the best results. I think you are correct in that this should be the new default, but I hesitate to change the default behavior of the software to avoid causing concern when people start seeing slightly different numbers. But the software probably should output a warning if one is using the mass fraction averaging method for MAN fitting.
Thanks for pushing me to implement this idea as I've published on it, but never got around to coding it in PFE.
john
Hi John
Yes, many thanks once again for your quick implementation of the Z-fraction into the MAN fit window. I have gone through a bunch of my datasets now, and the improvement to the fit is remarkable for many of the elements, to the point that you could forgive someone if they accused you of fudging the MAN fit to make the standards sit on the same line. Elements on TAP still have their issues but that is due to other factors, even so it does improve these fits as well.
Importantly this has also cleared up a lot of problems I had been having with some sulfide standards always plotting below the MAN fit, which makes sense as they were standards like Grenockite (CdS), Cinnabar (HgS) etc, which you can imagine would benefit from this Z-fraction calculation.
Anyway, good stuff! I will do more testing but I can see myself using the z-fraction as default
Cheers
Hi Ben,
Again, thanks for pointing this out and I'll be further improving the plot/log output later this week for these parameters. I also just got off the phone with Zack Gainsforth (card carrying physicist) and he concurs that photo-luminescence and continuum production is likely to be a 10-7 magnitude effect, so I will remove that option from the software.
It still amazes me to see the amount of improvement the Z^0.7 fraction Zbar has on these MAN fits. Looking a bit more at your data file (thanks!), I note that all the elements seem to benefit from the Z based fit, with the exception of Si Ka and Al Ka (where the fit gets slightly worse), but it should be noted that the other TAP elements all show improvement using the Z fraction Zbar.
It almost seems too good to be true. Here is another great example of a plot of K Ka with mass fraction averaging:
(https://smf.probesoftware.com/gallery/395_10_07_19_12_37_53.png)
and the same using Z fraction averaging:
(https://smf.probesoftware.com/gallery/395_10_07_19_12_38_10.png)
Like "beads on a string". 8)
OK, the MAN dialog has been redesigned to better display the elements and MAN standard assignments.
Here is how it looks now with v. 12.6.8 showing Ben Wade's MAN data for Sr La on a PET crystal:
(https://smf.probesoftware.com/gallery/1_11_07_19_11_43_56.png)
Without the Z fraction Zbar calculation the relative % deviation was 7.49 (compared to 2.01 in the above screenshot).
Ready to download now.
Looks great John. I like the extended box displaying the standards, will save my mouse scrolling finger getting RSI now.
Cheers
Quote from: BenjaminWade on July 11, 2019, 02:56:44 PM
Looks great John. I like the extended box displaying the standards, will save my mouse scrolling finger getting RSI now.
Cheers
RSI? Repetitive Scrolling Irritation? :)
Hi Ben and John,
to be honest, I did not have any special physical theory on that phenomenon. I just wanted to point out that I observed the same behaviour on synthetic cheralite. And actually, the observation of a very strong photoluminescence also agreed. But later, I was thinking about it a little bit. A strong photoluminiscence usually occurs when e.g. quartz is exposed to electron beam. Given that MAN method is applied on quartz, I guess someone would have observed such phenomenon.
Many thanks for the explanation. I will need to find something about the alternative expression of the Z-bar.
Best regards, Jakub Haifler.
Hi Jakub,
No worries.
Actually I blame myself. I published on these electron (or Z) fraction based Z-bar calculations for both elastic scattering and continuum production 20 years ago. At the time I thought the effects would not be so significant for normal EPMA work.
But then people more recently starting utilizing high Z materials in their MAN background fits, which allows for a greater possibility of compounds with very different A/Z ratios. And again, Ben Buse starting the elastic scattering discussion again with his re-discovery of the Z-bar effect when running BSE simulations in Penepma and WinCasino.
Now I realize that these effects are significant enough, especially in the case of high Z elements. So indeed we should be implementing these new Z-bar calculations into our quantitative software. This has been done in Probe for EPMA for the MAN curves, as shown in several posts above.
On the Z fraction elastic scattering Z-bar calculations, I am working with one of our colleagues on modifying perhaps the Pouchou & Pichior backscatter loss equations, and hopefully we'll have something to show later this year. It turns out that everyone in the past just assumed that BSE loss simply scales with mass! How silly of us... :D
But in the meantime you can read the abstract on elastic scattering that I will be presenting at M&M next month in Portland which is attached to this post:
https://smf.probesoftware.com/index.php?topic=1111.msg8249#msg8249
The thing that amazes me is that both continuum production *and* BSE loss are well modeled by applying the Z^0.7 Z fraction average atomic number equations. Two completely different physical processes...
Yesterday at the M&M social Peter Statham reminded me of a paper he sent me in 2016, where he independently derived a fit for continuum production with Z, and where he found a Z exponent of 0.75. Which is quite close to 0.70.
I attach his 2016 paper and my original 2002 papers (BSE and Continuum) below (please login to see attachments). In another cute coincidence of science, please see fig. 3 in both papers!
I also include this screen shot of an Excel spreadsheet with a summary of the Z exponents for Ben Wade's MAN fits to continuum data:
(https://smf.probesoftware.com/gallery/395_05_08_19_7_57_16.png)
If you're at M&M and see me, please feel free to ask me what the color coding represents, though you may indeed figure it out before I explain.
Perhaps this has been posted elsewhere, but I should point out that Stephen Reed has objected to the use of atomic number averaging that is used in place of more traditional mass averaging in the MAN continuum calculation. In his brief comment, Reed focuses on beam electron energy loss within the target (i.e., stopping power), which accounts for continuum production in the target. I've attached the relevant references including the paper by Pouchou and Pichoir from "Electron Probe Quantitation," aka "the green book," as Reed mentions their Figure 4 and associated equations. In their Figure 4 (page 35), Pouchou and Pichoir compare the results of their model for dE/dρs with those of Bethe's model; their discussion of electron deceleration begins on page 34.
Yes, that was the controversy. But Reed was unfortunately wrong.
We already know that the only reason mass averaging was originally utilized in the stopping power calculation is because these equations are expressed in mass normalized terms to be consistent with the absorption correction which is also mass normalized (mass absorption coefficients).
From isotope measurements (Donovan and Pingitore 2002, Donovan et al. 2003), we also know that the physics of EPMA is electrodynamics based, not mass based. The use of mass is simply a holdover from early days of chemistry, when the scale balance was the primary tool of science! (that's only a slight exaggeration!) :D
The better response to Reed's objections is found in the attachment below. But I encourage each of you to try the different average atomic number averaging methods and see which gives the best fit to a combination of compounds and pure elements using your own measurements of continuum intensities. A fairly recent investigation into these effects is found in this abstract:
https://epmalab.uoregon.edu/publ/average_atomic_number_and_electron_backscattering_in_compounds.pdf
By the way, Aurelien Moy even more recently revisited this question and confirmed our own measurements using Monte Carlo modeling from Penepma:
https://www.cambridge.org/core/journals/microscopy-and-microanalysis/article/universal-mean-atomic-number-curves-for-epma-calculated-by-monte-carlo-simulations/6F1C63250D980846ED0A765B10C504DE
Quote from: Brian Joy on September 18, 2021, 07:03:41 PM
In their Figure 4 (page 35), Pouchou and Pichoir compare the results of their model for dE/dρs with those of Bethe's model; their discussion of electron deceleration begins on page 34.
This reminds me of a story John Armstrong related to me many years ago when he happened to meet Hans Bethe at a conference.
The story goes that John introduced himself to Hans telling him that their entire field of EPMA was based on his original equations for electron energy loss. Apparently there was a brief silence at this statement, after which Hans exclaimed "But that was only accurate for hydrogen!".
;D
I moved this discussion to this topic since it more directly deals with the issue of mass vs. electron or Z fraction averaging for continuum intensities.
I also found the topic that Ben Buse started a while ago looking at the averaging issue for backscatter elentrons:
https://smf.probesoftware.com/index.php?topic=1111.0
Basically he found the same problems with mass fraction averaging that we did back in the 2000s.
On a related note I've attached below (please login to see attachments), some isotope measurements of characteristic emissions of stable isotopes of Ni, Cu and Mo compared to the natural abundance materials from 2001. These were actually the first measurements we had made on isotopes in the EPMA, but for some reason I had never published them.
Again as was the case for continuum and backscatter measurements, we found no significant statistical differences in characteristic emission intensities between the enriched stable isotopes and the natural abundance materials.
In calculation of the mean atomic number, how do you justify the use of an arbitrary fractional exponent? What does it actually mean? In your 2002 paper with Nicholas Pingitore, it appears without explanation. If you are trying to construct a model that's physically more realistic, then what, physically, does the fractional exponent represent? Why does it appear to produce an improvement in the fit?
In that same paper, in your plots on p. 434 (Fig. 3), it looks like the curves represent 2nd-degree polynomials. Is there a physical explanation for why continuum intensity should vary with mean atomic number or mass in such a manner? This approach seems not to work for some high mean Z compounds (like cheralite and galena). And why is zircon problematic?
Also, like Reed points out in his earlier comment (from 2000), the atomic number averaging only produces a marginal apparent improvement over mass averaging in the plots shown in Fig. 2 of Pingitore et al. (1999). This also appears to be true in the plots presented in Fig. 3 of your 2002 paper.
Why not use a model such as PAP to predict continuum intensity relative to that of an analyzed reference material that produces only continuum radiation at the wavelength of the X-ray line of interest?
In summary, what I'm saying is this: You expend some effort in creating a new model that you say has a more physically realistic basis, but then, regardless of Reed's criticisms, you render that physical basis null and void by introducing an unexplained, empirical fractional exponent and polynomial fit. Even the relatively complex PAP model is still semi-empirical, noting that two parabolas are used to model φ(ρz) rather than a seemingly more appropriate "surface-centered Gaussian" model.
P.S. I'm not trying to be a jerk, just devil's advocate. In particular, the outliers could contain a lot of useful information. Why are they outliers?
Quote from: Brian Joy on September 19, 2021, 10:29:03 AM
In calculation of the mean atomic number, how do you justify the use of an arbitrary fractional exponent? What does it actually mean? In your 2002 paper with Nicholas Pingitore, it appears without explanation. If you are trying to construct a model that's physically more realistic, then what, physically, does the fractional exponent represent? Why does it appear to produce an improvement in the fit?
I'm sorry, I just noticed this post. Been swamped with other work lately!
The exponent is simply tuned to the data (much like the PAP models themselves!) to provide the best fit. As Aurelien Moy points out in his recent paper, the best fit exponent seems vary slightly with x-ray energy.
As to why a fractional exponent, well it seems to be related to a geometric charge screening effect of the distribution of coulombic charge, e.g., Yukawa Potential. This would imply a Z^0.66 response. We are currently working on this idea...
In the case for a fractional Z exponent for backscatter production, we explain in some of our publications that the fractional exponent seems to relate to the decrease in backscatter production at higher average Z, due to well known screening of the nuclear charge by the increase in inner orbital electrons. Essentially another geometric screening effect.
Quote from: Brian Joy on September 19, 2021, 10:29:03 AM
In that same paper, in your plots on p. 434 (Fig. 3), it looks like the curves represent 2nd-degree polynomials. Is there a physical explanation for why continuum intensity should vary with mean atomic number or mass in such a manner? This approach seems not to work for some high mean Z compounds (like cheralite and galena). And why is zircon problematic?
Outliers can be interesting I agree and I welcome any insight into these, though I will notice that the Monte Carlo models do not show outliers, so I suspect they are simply poor measurements, but certainly worth keeping an eye on.
But keep in mind that the original effort was solely based on the fact that backscatter, continuum (and for that matter characteristic), emissions/productions are overwhelmingly based on electrodynamics. That is already known from physics, and confirmed by the isotope data measurements.
Atomic mass is a best only a rough proxy for these electrodynamic effects, so why not simply exclude mass, since we already know mass has essentially no effect on these productions? In fact, because A/Z generally increases as a function of Z, merely due to reasons of nuclear stability, including neutrons into these electrodynamic calculations, introduces a mass bias into our calculations. As my friends analyzing interstellar dust would say: atomic number is universal, but atomic mass is local.
Quote from: Brian Joy on September 19, 2021, 12:48:14 PM
In summary, what I'm saying is this: You expend some effort in creating a new model that you say has a more physically realistic basis, but then, regardless of Reed's criticisms, you render that physical basis null and void by introducing an unexplained, empirical fractional exponent and polynomial fit. Even the relatively complex PAP model is still semi-empirical, noting that two parabolas are used to model φ(ρz) rather than a seemingly more appropriate "surface-centered Gaussian" model.
We are not committed to the polynomial fit in any way. It just happens that earlier continuum models utilized a straight line fit and the polynomial fit seems to better represent the data. If you can suggest a better method for fitting continuum production as a function of average Z we can certainly try that also.
Quote from: Brian Joy on September 19, 2021, 10:29:03 AM
Also, like Reed points out in his earlier comment (from 2000), the atomic number averaging only produces a marginal apparent improvement over mass averaging in the plots shown in Fig. 2 of Pingitore et al. (1999). This also appears to be true in the plots presented in Fig. 3 of your 2002 paper.
I think we all prefer scientific models that are more physically realistic *and* produce an improvement in our predictions of measurements. Even if the effect is relatively small, it's still an improvement. Shouldn't we welcome improvements in our physical models?
Quote from: Brian Joy on September 19, 2021, 10:29:03 AM
Why not use a model such as PAP to predict continuum intensity relative to that of an analyzed reference material that produces only continuum radiation at the wavelength of the X-ray line of interest?
I think that is an interesting idea. I look forward to seeing your results from this. Just remember, mass doesn't affect any of the emissions/production we observe in the microprobe. Of course if we utilized a 1 MeV electron beam, that would be another story!
I do have to add one comment. Whenever I discuss this atomic mass versus atomic number issue with "card carrying" physicists, they all respond the same way: "Well duh, it's all electrodynamics!". But for some reason in the field of EPMA there is this obsession with atomic mass. I suspect it's just historical inertia as chemists are used to reporting results in mass fractions, because "wet chemistry."
That and also that we started out using some mass normalized expressions (i.e., mass absorption coefficients), even though we know that density is unrelated to EPMA physics, except for considerations of non-infinitely thick specimen geometry of course.
I also have one question about MAN. I use PHA integral as differential mode in most of cases cause more problems than good (in my cases). So the question is: Does your MAN method takes into account 2nd, 3rd, 4th and etc order bremsstrahlung (which is additionally complicated by Ar edge corresponding orders, in case of P10 gas)?
That is the neat thing about calibration curves: they handle all the physics we don't yet understand! :D
Of course the MAN background correction is really a semi-empirical calibration curve because it is based on the physics of Kramer's Law, which assumes that continuum production is primarily an effect of average atomic number, hence the discussion regarding on what basis should average Z be calculated: mass fraction vs. Z fraction.
And we apply a continuum absorption correction to the intensities measured on each standard material (and our unknown), though we find that the modern phi-rho-Z absorption corrections seem to do a better job than continuum specific absorption corrections from decades ago.
Remember, with the MAN correction we are modeling the continuum at a *single* continuum energy, corresponding to the emission energy of the element we are observing. Also we construct a separate calibration curve for each element/x-ray/spectrometer/crystal combination which automatically handles these instrument dependent effects.
The good news is that these MAN calibrations are very easy to do and are very stable over time, so one spends a few minutes once in a while acquiring these MAN curves (which could simply be the primary standards one is utilizing as long as they don't contain the particular element of interest, e.g., pure MgO and TiO2 for Al Ka), and then one obtains better precision measurements in about half the acquisition time. The MAN method is particularly nice when performing quantitative x-ray mapping.
https://pubs.geoscienceworld.org/msa/ammin/article-abstract/101/8/1839/264218/A-new-EPMA-method-for-fast-trace-element-analysis (https://pubs.geoscienceworld.org/msa/ammin/article-abstract/101/8/1839/264218/A-new-EPMA-method-for-fast-trace-element-analysis)
As discussed above it's interesting that the Yukawa potential model yields a Z^0.666 fit. It's also interesting how often this number 0.666 (or ~2/3) shows up so often in physics models.
Probably just a coincidence (as my co-author Andrew Ducharme points out simply because 2 and 3's are common numbers!), but here's another weird physics 2/3 coincidence:
https://en.wikipedia.org/wiki/Koide_formula