Offset (Interpolated) MAN backgrounds

[orlandin]

Hello, all! I was really inspired by the forum posts which resulted in the 'use interpolated off peak backgrounds for MAN fit' (https://probesoftware.com/smf/index.php?topic=987.0), because I don't have an optimized MAN standard setup yet. However, it does not seem to work quite how I would expect. After collecting my standards using normal off-peak backgrounds, I check the box under the Analysis dropdown for 'use off-peak standards', and then check the box in Analysis Options for 'use interpolated off peak'. Then I usually also check the box just under 'use off peak standards' so that the MAN correction is applied because it doesn't seem like you should ever not do that, although it doesn't seem likely to influence the problem below.

When I open my 'Assign MAN fits', what I expect to see is every single element for every single standard since I did not use Quick standards and PFE has a calculated/interpolated value for the background intensity under the peak for everything. But instead, I find a rather sparsely populated MAN curve as if I had run all my inappropriate-for-MAN standards as if they were just MAN standards, with several elements reporting having no MAN standard data at all due to everything having a non-zero concentration of that element.

I'm pretty sure I understood the forum post about how to use this feature to have a long standardization but still retain many of the benefits of MAN backgrounds on unknowns, but now I am not sure about what sequence of buttons to press or boxes to tick in order to get PFE to utilize the interpolated/calculated background values from the off-peak standardization. Thank you!

[John Donovan]

Hi Phil,
You raise an excellent point about using the "interpolated" off-peak MAN feature.

Yes, it makes some sense that one should be able to use all the elements for all the standards for this "interpolated" MAN background curve feature, because as you say, if you're actually using off-peak measurements to calibrate your MAN curves, why not use them all?

Now, as you also say, for normal un-interpolated MAN background calibrations of course we *cannot* utilize standards containing the element of interest, because we wouldn't be measuring the actual background, since we are measuring on-peak.  But if we're using off-peak backgrounds, we could in principle.

But here's the thing, in practice I think it could cause other issues. For example, one might assign their MAN standards using these off-peak interpolated measurements, but then if they uncheck that box, all hell would break loose for those standards that *do* contain the element of interest.

I'll tell you the real reason why we implemented this "interpolated" off-peak MAN method, and that was because Ben Hanson at Corning Glass had a similar problem as you, in that he had standards which weren't quite as well characterized as he thought they were!  Well, at least not as pure as he thought they were.  This happens more often than one might think. Here's an example of undeclared Cl (~200 PPM) in an amphibole standard showing up in the MAN curve for Cl:

https://probesoftware.com/smf/index.php?topic=4.msg5289#msg5289

It turned out that several of Ben's standards had trace amounts of elements that weren't declared in the standard composition database, so of course they showed up as high intensity outliers in the MAN plots.  The same thing can occur when the on-peak position is interfered with as seen here:

https://probesoftware.com/smf/index.php?topic=4.msg5136#msg5136

So basically the idea was to utilize standards that might have undeclared contaminations and/or on-peak interferences that need to be avoided, and by using this "interpolated" off-peak MAN feature we could do just that.  And for these situations it works great.

The other issue that Ben ran into was trying to create an MAN background curve for O Ka.  And as we all know, it's hard to find standards that don't contain oxygen and/or don't have a native oxide layer.  Ge and Au pure metals are the only things I've found that work OK for MAN on O Ka.

So while I understand that you'd like to utilize all your silicate standards for the calibration of the Si Ka background, that's not what it was designed for. I guess I would ask, don't you have a TiO2 standard? Or an MgO standard?  Or an Al2O3 standard? Those would make ideal curves for Si Ka, and other elements as well.

So yes, we will think about this "feature not a bug", and see if we can come up with a way to allow standards with the element of interest to be utilized as an MAN standard, but in the meantime you probably should work on your standards. You won't regret it!  Using the MAN curves for quantitative x-ray mapping is so cool!  No need to acquire off-peak background maps!

By the way, the MAN feature does allow some small concentration of the element of interest to be utilized in the MAN fits. Right now it is defined in the code as seen here:

MANMaximumValue! = 0.01     ' maximum value (wt%) for use as MAN background standard

[orlandin]

Hi John!

Thank you so much for this response! I do think I understand a bit better now why PFE was side-stepping me on this. I am not a coder, but from a user perspective it seems like maybe a stern pop-up warning when un-checking the 'use interpolated off peak' might be enough to get the risk across? If that were ignored or didn't exist, as you say as soon as one looked at their 'Assign MAN Fits' hopefully it would be pretty obvious that something was deeply awry. And even if you ignored that, your results should really make introspection unavoidable?

Yes, I do agree with you that my standard collection will only benefit from more love and attention - right now the only non silicates with reliable documentation that I have is a rutile and some phosphates which find their way into all of my attempts to make MAN backgrounds routine for points and mapping. I have recently inherited this lab, and there is a lot to learn about the decades-old collection here!

That being said, maybe both could be reasonable goals: a better MAN-appropriate collection on my end, and possibly including this as a beneficial PFE feature? The attitude here before I got to this lab seemed to be that MAN was voodoo requiring an expensive special set of otherwise useless standards. Adding a feature like this might make MAN backgrounds more accessible to analysts globally, in addition to all of the other utility that you've pointed out. Looking one post above my question, perhaps this approach could even have a more fundamental benefit when you're willing to take the extra time to use it on a good MAN standard set:

In thinking about this a bit more, it occurs to me that not only is this interpolated off-peak MAN measurement method likely to be as accurate as normal off-peak measurements, and maybe even a little better, since it's based on a regression of multiple standards (and because off-peak interferences are less likely in the simple metal and oxide standards that are usually utilized for the MAN calibration fit), but the typical time savings and better sensitivity of the MAN method will still apply. That is because we are only utilizing off-peak measurements for our MAN standards, not for the unknown samples using MAN (on-peak only) measurements.

[John Donovan]

The attitude here before I got to this lab seemed to be that MAN was voodoo requiring an expensive special set of otherwise useless standards.

I think you meant to say *peer reviewed* voodoo!   

https://pubs.geoscienceworld.org/msa/ammin/article-abstract/101/8/1839/264218

Seriously, yeah, I hear you.  And I must say based on our discussions so far here on the forum (and by email), I'd say you are doing way, way better than anyone without formal training from Probe Software has a right to expect!   

The MAN method is a little unintuitive, but worth it for better precision and saving time. And it is neat that it is a physics based (Kramer's Law) approach that works much better than it has any right to (with a continuum absorption correction). And is great for when running lots of points or quant mapping. And I do appreciate your efforts to make it work with your existing standards.

I'm still thinking about how I can balance all the various considerations (but have an idea) so give me some time to consider things. That said, improving the existing standard collection is the very first thing I did when I started at Berkeley on the microprobe back in 1988.  It will pay off for you and your lab, no question.

Here are some links for inexpensive standard materials:

https://probesoftware.com/smf/index.php?topic=308.0

ESPI is awesome, they have lots of pure single crystal oxides and other compounds.  Perfect for use as both primary and MAN standards.

One more point: if you're utilizing MAN background calibrations for point analyses then yes, you'll want to take some care making these calibrations, especially for higher Z materials. Typically one can expect accuracy levels around 200 PPM for common (low to moderate Z) oxides and silicates. But for quantification of x-ray maps, your precisions are usually much worse than for point analyses, and so the background correction accuracies are nowhere near as critical as for point analyses.  It's simply a question of "wise" statistics, e.g., not spending a lot of time measuring your off-peak backgrounds when measuring major element point analyses.

[orlandin]

Hi John! Thank you for the kind words - you can give Julien Allaz all of the credit for making sure that this clueless former dog biscuit baker got the best mentoring and education he could provide. And there is really is no understating how valuable this forum is for learning how to do new things, and do old things better!

It may be a while before I can explain how MAN works to users without resorting to 'and then the computer models the background', but I love what I do understand about it. For now, I will take your advice and commit to really sorting out and documenting this collection. Thank you again!

[John Donovan]

You are welcome. I congratulate you because running a lab is quite a different animal than just running some software!

As for teaching, I know it's a bit dense on statistics, but reading the Amer. Min. MAN paper linked above will really help you understand the gritty details. For the time being just say to the students: Kramer's Law!

Also I don't know if you know this but for teaching EPMA, John Fournelle (at Wisconsin)and Probeman (at Oregon) have useful PowerPoint files from their EPMA courses available for download:

http://www.geology.wisc.edu/~johnf/g777/

https://epmalab.uoregon.edu/lecture.htm

By the way, I think we've almost got this new MAN option working. Give us another 6 or 8 more hours and we should have something for you to update and try.  But unfortunately I think I'm gonna have to prevent this feature being utilized on older versions of the databases. Because since these flags aren't saved in older databases, it can really mess up the MAN assignments when that MDB database is re-opened. Maybe, maybe not. Still working on it.

[John Donovan]

I am not a coder, but from a user perspective it seems like maybe a stern pop-up warning when un-checking the 'use interpolated off peak' might be enough to get the risk across? If that were ignored or didn't exist, as you say as soon as one looked at their 'Assign MAN Fits' hopefully it would be pretty obvious that something was deeply awry. And even if you ignored that, your results should really make introspection unavoidable?

Hi Phil,

OK, I think have something that will work.  What we did was add a new checkbox to the Analysis Options dialog as seen here along with the original Use Interpolated Off-Peaks For MAN Fit checkbox:



This new checkbox is called Use Interpolated Off-Peak Intensities Intensities for MAN Even If Element Is Present.  That's a mouthful but I think it does what it says.  What this new flag does is basically just ignore the MAN minimum intensity level for MAN standard assignments that is set at 100 PPM. Now the cool thing is that this new MAN fit option actually reveals several interesting problems with using off-peak intensities for determining backgrounds (off-peak or MAN!).  Three separate issues actually I think.  These issues have been described in detail by many of our colleagues, Mike Jercinovic and others, Julien Allaz, Karsten Goemann among them I'm pretty sure.  You've probably seen them yourself as well.

So backing up a bit, when I first tried checking this new flag on some off-peak measured standards, then checked the Use Off-Peak Elements For MAN Fits menu, selected Clear All MAN Assignments, and opened the Assign MAN Fits dialog, I didn't know what I expected to see, but here is the first MAN fit that displayed which is the Ca Ka MAN fit:



Well heck, it's not a great fit but considering that all the circled standards contain at least some Ca, it did a pretty good job. I learn something new every day, thanks to sharp, young guys like you!

Now one could argue that some of those high outliers are a little too much outside of statistics (I should have turned on the error bars option when I captured this image!), but let's not worry about it and go to another emission line, this time K Ka and this time there's definitely a high intensity outlier for the orthoclase standard, which of course contains 15 wt% K:



Now what could be causing this?  Well there are at least two things I can think of that would cause a high outlier for off-peak measurements. First, the absorption edge effect. That is, if one of the off-peak backgrounds is on the other side of an absorption edge it might cause an under estimate or an over estimate of the continuum intensity under the peak, depending on which side of the absorption edge, the emission line of interest is on.  The other reason for a high outlier is that the background is curved and we had utilized a linear fit between the off-peaks. Of course then there is the situation that if our off-peak measurement positions are on the tails of the peak (and a major peak from the element is present), we will again also be over estimating the intensity of the continuum intensity under the emission line of interest. Oh yeah, also if there is an off-peak interference, we will over estimate the on-peak continuum using off-peak measurements whether they are applied in the off-peak correction normally, or in the MAN fit as we are doing here.

All good reasons to avoid off-peak backgrounds!   

Now the good news is that if we are using these MAN fits to background correct a major (and even minor) element intensity, an error of a few hundred PPM accuracy error will often be less than our on-peak intensity precision. Also for quant x-ray mapping, where we usually won't have enough precision per pixel for these sorts of background inaccuracies to matter much.

And by the way, these off-peak measurement precision errors? Well they become *accuracy* errors when using the MAN method!  Why? Because we normally repeat our off-peak measurement for each data point, but the MAN background gets calibrated only once or twice in a run!  So given the same average Z, the same background intensity will be estimated! And remember, the average Z precision is controlled by the *major* elements, not the continuum statistics! That's part of the reason the MAN method has higher precision than off-peak measurements. Read the 2016 Amer. Min. paper for details on this subtle issue.

And just to demonstrate that even in the case of a large concentration of the element of interest being present, there's sometimes no significant observable off-peak accuracy effects:



This standard 260 is pure synthetic Al2O3 measured at the Al Ka peak position. And while probeman did select the off-peak background positions carefully, it is still pretty good considering the intensity of the Al ka peak.  At least the background in this vicinity is not curved as it is in SiO2!

So please update your PFE and try it out. One helpful hint, you can "use" your MAN intensities from older probe runs by creating a new MDB file, then making a new sample, clicking the Load File Setup button and saying "Yes" to load standard intensities. It won't import your unknowns, but it will allow to to try out this new feature on some existing standard data.

Anyway, I think this new feature should get your MAN fits going, at least until you get some more pure oxide and other synthetic standards.

[orlandin]

John, this is amazing, thank you! I can't wait to give this a try tomorrow morning. What a cool way to check if your off-peak placements are appropriate, whether you intend to further use them for unknowns or not. Thank you so much for taking the time to work on this when really, neither one of us should have been worrying about this on a Saturday. Even if I do get a good MAN set going here soon, I think I will derive great satisfaction in getting my more complicated standards to fall along the MAN curve with the rest due to careful placement of off-peaks.

[John Donovan]

Ok, so I didn't want to leave everyone thinking that the use of these (interpolated) off-peak measurements for use in the MAN fits is not entirely worthwhile.

It's just that using off-peak intensities for the background correction (whether they are utilized in normal off-peak background corrections or in MAN background corrections), is fraught with a number of issues because, they *are* interpolations. With off-peak background methods, we are *not* measuring the background intensity *at* the actual emission line position, we are interpolating to it.

For one thing these off-peak (interpolated) MAN fits are worthwhile if one doesn't have a nice set of pure simple synthetic oxides. Which is the situation that Phil Orlandini in Austin is in (Phil: please show us some of your interpolated off-peak MAN fits when you get some!). Probeman in Oregon is lucky because he managed to obtain a number of pure synthetic oxides, specifically MgO, Al2O3, SiO2, TiO2, MnO and NiO early in his career and has added these simple oxides to all his standard mounts.

Of course one can also utilize pure metals, but for background correction of common oxides and silicates their average Z values will be roughly between 10 and 18 (SiO2 and Fe2SiO4). So while one can utilize Si metal, and maybe Ti, etc on up for one's MAN calibrations, the Z of Si starts at 14, so not ideal for the lower average Z silicates such as forsterite (Mg2SiO4).

That said, there is one other really nice use for these off-peak (interpolated) MAN fits and that is when one has undeclared concentrations of elements in ones (usually natural) standard materials (as was the case for Ben Hanson at Corning). Let's examine the case here which is a normal (on-peak) MAN fit for Ni Ka showing that the natural chromite standard has an undeclared trace concentration of around 600 to 700 PPM of Ni:



Now, of course we can simply remove that standard from the fit as is usually done as seen here:



But of course now we can also utilize the (interpolated) off-peak MAN fit as seen here:



Note that the above plot also documents the use of the (interpolated) off-peak MAN fit.  The idea with all this stuff being that we can acquire our standards using off-peak measurements, use these off-peak standards to fit our MAN curves as described here:

https://probesoftware.com/smf/index.php?topic=987.msg8889#msg8889

and in other posts in this topic, and then utilize MAN acquisitions on our unknown samples to save time and improve precision for point analyses and quantitative x-ray maps. And depending on the samples, maybe even improve the accuracy of our unknowns by avoiding possible off-peak interferences.   

[orlandin]

So, I finally figured out how to get administrator access on my own lab computers and have done some quick experimentation with this new feature. I absolutely love it, and this will really enhance the quality of analyses coming out of this lab. I've attached the results of applying this to an apatite analysis setup that I inherited from another operator in 2014 using File Setup importing. The client said that this worked, they wanted exactly what this guy did that time, and I didn't have time to reinvent the wheel then - but I would love to now!

There aren't any particular notes, but it seems like they really wanted to use MAN from the standard setup but ended up defaulting to off-peak for everything. I used the exact same MAN-spirit setup when I ran these analyses, so this seemed like just the place to start out. These were collected at 10 nA, 10 micron beam, 10-pt TDI. Error bars are 3-sigma. As you can see, maybe I should have started with a simpler project if I wanted to be able to explain everything like John did! I look forward to thinking hard about all of this (maybe using the EPMA Method Development Tool to start off with), but for the time being I will just leave the 'Assign MAN' results here to illustrate maybe an endmember of applying this method:



(edited to use the cool inline image feature, but be warned that it is 2500x1548 pixels!)

[John Donovan]

As many of you know, the mean atomic number (MAN) background correction method is based on Kramer's Law which states that (all other things being equal), the continuum intensity is a function of the average atomic number of the material.

https://epmalab.uoregon.edu/publ/Improved%20MAN%20(Jour.%20Micros.%20Microa.,%201996).pdf

https://epmalab.uoregon.edu/publ/A%20new%20EPMA%20method%20for%20fast%20trace%20element%20analysis%20in%20simple%20matrices.pdf

Generally this MAN calibration curve is obtained by measuring continuum intensities in standards (of known composition and hence of known average atomic number), at the element emission line position, that *do not* contain the element of interest. For geological work on silicates and oxides this ideally means a selection of pure synthetic oxides, e.g., MgO, SiO2, Al2O3, TiO2, MnO, NiO and/or other pure synthetic oxides or silicates (though it should be mentioned that the MAN method can be applied to any material assuming one has standards with a similar range of average atomic numbers).

Unfortunately some EPMA standard collections consist mainly of natural materials which often contain traces of many elements, thus "interfering" with obtaining a "clean" continuum measurement at the element of interest peak position. To avoid these "contamination" issues when calibrating one's MAN curves, Ben Hanson proposed a modification of the MAN method that would utilize off-peak measurements on standards, which could then use these interpolated continuum intensities for the MAN calibration curve, which would then be applied to unknown samples.  The advantages being faster data acquisitions on unknown samples with improved precision.

Subsequently Phil Orlandini proposed utilizing even standards which contain the element of interest to allow even more standards to be utilized in the MAN calibration curve. See M&M abstract attached below which Phil presented at the virtual M&M earlier this month.

More recently Anette von der Handt inquired as to whether we could also utilize not only traditional off-peak measurements on standards for these MAN calibration curves, but also use the multi-point background (MPB) background method:

https://search.proquest.com/openview/208992e799d48f657d3508a6ef11ca1c

I hadn't considered this option before, but I thought, well why not?  After we examined the code, it appeared to us that this might already be taken into account since an MPB measurement is really just multiple off-peak measurements! So I ran a quick and dirty simulation in Probe for EPMA, acquiring an element using the MPB method on a number of standards, as seen here for Cr Ka in chromium metal:



I then checked the menu shown here to enable the program to search for MAN standards even if they were acquired using off-peak (or MPB) methods:



Then after opening up the Assign MAN Fits menu dialog we see the following results for Cr Ka:



Oh, right, the chromium peak is slightly interfered by the V Kb line in the vanadium metal standard, even in simulation mode! But after deselecting the vanadium standard we update the fit and now obtain a much better MAN calibration curve for Cr Ka:



So we now go to the Analyze window and analyze one of our standards (just for fun!) and obtain the following results:



Note that Cr is acquired (and still quantified) using the MPB method, while Mo is acquired (and quantified) using the traditional off-peak (linear fit) method. Next we go back to the Analysis Options menu and check the menu for analyzing off-peak elements as though they were acquired using the MAN method as seen here:



Now there was one tiny change we did have to make to the code which was a labeling issue when this Use MAN Correction for Off-peak Elements menu is utilized and one opens the Display Multi-Point Backgrounds dialog from the Run menu as seen here:



Because previously, it would indicate that the element was acquired as a "shared" MPB. Which is (yet!) another option for using multiple off-peak elements on the same spectrometer and Bragg crystal to create a pseudo MPB sample as described here:

https://probesoftware.com/smf/index.php?topic=9.msg9043#msg9043

Anywho, now we go again to the Analyze window and analyze our standard again, and now the results are based on the MAN calibration curves which utilized interpolated off-peak measurements (MPB for Cr and traditional off-peak for Mo):



OK, now data 2 points isn't enough to draw any conclusions with, but it is interesting that the variances for both Cr and Mo are smaller when utilizing the MAN method (as one would expect)!

Does this sound complicated?  Actually it's quite easy once you acquire data on a few standards and take a few moment to go through the menus, but do feel free to ask questions and we'd be happy to help.   

The neat thing about these interpolated off-peak options is that one can analyze the same data using both off-peak and MAN background methods and compare the results. Try it out and let us know what you find.

[Rom]

Greetings, could you precise what model the system use when linked 2 or more off-peak points for develop interpolated on-peak intensity for MAN background curve.
Is this only linear model or we can choose different (exponential, polynomial) models as well?

[John Donovan]

The off-peak background fit utilized for the interpolated MAN option depends on the background acquisition mode, and also the background fitting model that was selected by the user for each MAN standard sample.

This could mean that if the MAN standard samples were acquired using normal two point off-peaks, any of the off-peak models could be selected subsequent to acquisition. For example, linear, exponential, slope, polynomial, etc.

If the background acquisition mode was multi-point background (MPB), then again, the MPB fit model can be selected subsequent to acquisition as either linear, exponential or polynomial.

The MAN calibration curve fit itself can be linear or 2nd order polynomial. Attention should be given to the "Z fraction" Zbar averaging method when analyzing moderate to high Z materials.

[Rom]

Thank you John,
1. Could you give some recommendations about "Attention should be given to the "Z fraction" Zbar averaging method when analyzing moderate to high Z materials". What is the average range of high Z materials.
2. How I can increase my knowledge about reasons to think about Z when I try to use MAN-method of background correction? 

[John Donovan]

1. Could you give some recommendations about "Attention should be given to the "Z fraction" Zbar averaging method when analyzing moderate to high Z materials". What is the average range of high Z materials.

Well most silicates and oxides are in the range of Zbar of 10 to 20.  Materials with average Z values higher than that would be moderate Zbar and high Zbar materials would even higher than that (higher than 30 or 40?). Check the Standard application which can list your standard materials by average Z from the Standard | List All Standard Names and Average Z menu.

2. How I can increase my knowledge about reasons to think about Z when I try to use MAN-method of background correction?

Here is a general discussion of the various background acquisition/correction methods in PFE:

https://probesoftware.com/smf/index.php?topic=1378.0

I should also note that ideally one should utilize pure synthetic materials for use as MAN standards. This in fact is one of the reasons we should work with Will Nachlas on obtaining, characterizing and distributing (globally) high purity synthetic oxides and silicates as discussed here:

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

In fact there are 5 reasons for these high purity synthetic materials:

• For use as primary standards.
• For use as secondary standards (which is why Will is working on obtaining at least two synthetic materials with the same element).
• For use as interference standards (high purity is important for the quantitative interference correction).
• For use as MAN standards (high purity is essential for creating accurate background curves, though the interpolated off-peak MAN method is one alternative, though it still has all the defects of the traditional off-peak methods).
• For use as blank standards, for the determination of a zero concentration in a given matrix, e.g., synthetic ZrSiO4.

Here are some other topics to read:

https://probesoftware.com/smf/index.php?topic=1438.0

https://probesoftware.com/smf/index.php?topic=4.msg9953#msg9953