Off-peak Background Modeling in PFE

[John Donovan]

This is a very subtle example (I probably should have chosen something more egregious!), but it serves to illustrate the steps in adjusting one's off-peak positions based on the background model fit to a wavescan (we all run wavescans before we acquire any data, right?).

This is a complex mineral with a lot of REEs, so here is a scan of Pr in the sample at default precision (6 secs per point using the continuous (ROM) scanning mode:



and here is a slightly "zoomed in" view:



Note that there are previous off peak positions specified in magenta and new (current) selections in green. Why? Because in the zoomed in view we can see that in the case of the low off-peak position (left side), the off-peak position was not quite at the lowest measured continuum intensity, while the high off-peak position was just a little too close to the tail of the Ce La1 peak.

So both off-peak positions where moved slightly outwards.   But what about the background fit?

So next click the Model Background button as seen here and you will see the fit to the default (linear) model:



Not bad, but this spectrometer range is near the low side and therefore the background tends to be more exponential. We can see the default bgd fit intensity here:



So now lets try an exponential fit as seen here:



Note the lower background intensity value and also the ability to adjust the "tension" of the exponential fit. 

The new off-peak positions are automatically saved, but if we want to save changes to the background fit type we can simply select the samples to assign it to specific samples at the bottom of the dialog (note All Samples option and Select button) using the mouse and then click the Assign Background Model To Selected Samples button as seen here:



Remember, if you also include the currently displayed wavescan sample in this background fit assignment, the software will automatically load that assignment the next time you click the Model Backgrounds button.

Or you can modify the fits from the Elements/Cations dialog as seen here:

[John Donovan]

Here's a better (more obvious) example of the benefit of background modeling looking at N Ka in AlN.

Here we see the wavescan sample is plotted from the Plot! window and the deafult high and low off-peak positions (in magneta) were moved to their current positions (in green), and when the Model Background button is clicked, the default linear fit is displayed:



Note the interpolated background intensity circled in red:



It is clear that the linear fit is not sufficient to model the background at such a low sin theta position, so the exponential background fit is selected as shown here:



Again note the improved (lower) interpolated background intensity circled in red:

[Ben Buse]

Hi,

I'm not sure this is the right place to post, I have a question how best to do a slope background with single background measurement.

I'm trying to do a slope background using a single background point - Sr in Plagioclase - upper background only, as lower background on Si Kb satelite.

I would like to measure the upper background for twice as long (i.e. same as peak time) - and not measure lower background. And apply a slope

Is this possible? What should I be doing? I've tried the following

Upper background only, still measures lower background, but calculates from upper background only.

Average background - I can set the upper and lower background positions to the upper background position - thereby doubling the time on the upper background and not measuring the lower background position. This works great if the background is assumed to be flat. However when I come to model the background - and assign a slope - it can't because upper and background positions are in the same place.

Maybe it's time I started using multi-point background

Thanks

Ben

[BenjaminWade]

Hi Ben
That's interesting, I have used single point with slope a little in the past and didn't realise it was still wasting time analysing the other background. I have since changed these to MP now though as I figured rather than assuming certain slope from a single wavescan I did, why don't I just measure it with MP and model the real slope, and made do with the increased time taken with MP...

Just my two cents, a comment on MP though if you are doing all points only on one peak side, do more than just a couple of bkgd measurements and space them apart a reasonable amount. If you only do a couple you would probably only be putting a linear fit through them (you have the option of linear/exp/poly), and depending on how much scatter is on those two background measurements, and the intensity of the peak measurement, the slope on that line could vary and run the danger of it fitting a background intensity higher than peak measurement. What would be really cool is if we could get the option to refine the equation of the linear/exp/poly fit through MP backgrounds...

Also another comment, if putting on Hi side to measure Sr, depending on how far you are going out there is a large Si Ka1 peak to the Hi side as well, as illustrated in attached figure from a run measuring Sr in plag.

Cheers

[Ben Buse]

Hi Ben,

Thanks, I'll do some experimenting - to see if I can get the real background shape with multipoint. What I also just realised is I could change low background count time to 0.01 sec.

[Probeman]

Hi Ben and Ben,
The high/low only and high/low only slope background models were intended to be utilized *after* the normal off-peak data was acquired. Particularly in cases where the user subsequently discovers that one of the off-peak positions is interfered with by a secondary emission line.

The idea being that one could change from the normal two points acquired to utilizing only *one* of the off-peak positions acquired in an effort to "salvage" the data. That is to say these off-peak fitting models are not intended for acquisition, but for data re-processing.

And yes, as you mentioned if you really only want to acquire data on one side of the peak you can move both off-peaks to the same side (hence the so called "same side off peaks" term in the PFE documentation), and then use the average option.

Or as you also mentioned, just use an interpolation method and space the off-peaks on the same side but far enough apart to prevent over extrapolation. This latter method is the classic "same side off-peak" method.  I've used it for example with U Ma when one is near the Ar absoprtion edge as in this nice example:

http://smf.probesoftware.com/index.php?topic=42.msg3915#msg3915

For acquisition of only one off-peak position Ben suggestions to set one background to a short integration time is a good solution.  To reduce the spectro movement time just put it near the other off-peak background but be sure you don't utilize it, unless you are using the average option.
john

PS Here's a nice discussion of using multi-points backgrounds where there are lots of interferences:

http://smf.probesoftware.com/index.php?topic=701.msg4283#msg4283

[Probeman]

Here's an example of a "nasty" background (for O Ka) in a Cr alloy I ran last night.  Not only is the background quite exponential but it is also interfered with by Cr L lines (and most likely interfering with the oxygen Ka peak as well).



I'm thinking I will go ahead and try using the MAN background model since I have lots of pure metal standards to fit.  This is done by fitting the on-peak intensity data from the already acquired off-peak acquired standards, and using the Use Off-Peak Elements for MAN fits menu as described here:

http://smf.probesoftware.com/index.php?topic=4.msg189#msg189

Then at least I won't have to worry about off-peak interferences!

[John Donovan]

Just wanted to post a quick tutorial on adjusting ones off-peak background models and provide a few pointers.  Note, typically we mostly worry about our off-peak backgrounds when performing minor/trace element analyses, but if the off-peak interferences are large enough, they can actually affect our major element totals!

Of course there are as many different ways to make such adjustments as there are microanalysts, so again, this is just one example.  In this case I analyzed for Si ka, Pb Ma and Fe ka in using SiO2, Fe2SiO4 and PbSiO3 as primary standards. Then I acquired wavescans on SiO2, HfSiO4 and PbSiO3 as seen here for Fe Ka:



No peaks interfering with the off-peak positions for Fe in these matrices, so we can just stick with the default two point linear model. What about Si Ka:



Looks OK, but what happens when we zoom in a little and check our background fit using the Model Backgrounds button?



As we can see the two point linear model isn't working very well because the background between the Si ka and a secondary peak (from Hf) doesn't quite reach the true background level. Remember about Bragg crystal polygonization effects during manufacturing which greatly extends the tails of our Bragg crystals?

Time to go to the high or low side background and apply a slope (SEM Geologist's favorite method!) as seen here using a Slope Low fit:



Now that our background fit is better we can apply these background model changes to all samples using the Select button and the Assign Background Model To Selected Samples:



Remember, one cannot change the off-peak background positions for samples that already contain data, but they will be applied to new samples. However, one *can* change the background fit model for all samples regardless of whether they contain intensity data or not!

[Probeman]

Here's a cute new feature in Probe for EPMA from the Model backgrounds dialog.

When plotting multiple wavescans and opening the Model Backgrounds dialog, one can now select which wavescan to apply the background model to as seen here:



Selecting a different wavescan we get this:

[Probeman]

I was going to post this discussion on adjusting ones' off-peak background positions in the topic on consensus k-ratios here:

https://smf.probesoftware.com/index.php?topic=1442.msg13908#msg13908

but maybe it's better here.

Although the background correction is not usually critically important when performing quant on major element concentrations, it is always worth doing right.  I have been very dismayed by how often I am seeing many labs just using the "default" off-peak background positions when acquiring off-peak samples.  I think this is part of the problem we are seeing from different labs responding consistently different results from each other...

Probeman tip: always acquire wavescan samples on the compositions you will be analyzing for and adjust the background positions before starting your quant:

https://smf.probesoftware.com/index.php?topic=68.msg11595#msg11595

https://smf.probesoftware.com/index.php?topic=592.msg4888#msg4888

And if you need to widen your wavescan ranges to find the "true" backgrounds on either side of your analytical peak, just do it!  In PFE, it's this adjustment which is accessed from the Acquire! window by clicking the Peak/Scan Options button:



Then clicking the element row of interest here:



OK, here's an example from the consensus k-ratio topic linked at the beginning of this post. Once your spectrometers are peaked up and PHAs adjusted you'll want to check your off-peak positions by performing wavescans, not only on your unknowns but also on your standards (if they haven't been already determined).  Let's start with Mg Ka looking at just the unknown MgAl2O4 sample:



Notice that the default background positions are not adequate at all. Of course one can adjust the default positions that are utilized:

https://smf.probesoftware.com/index.php?topic=444.0

but even so, always acquire a wavescan to check.  In fact, one should ideally acquire wavescans not only on your sample, but also on your standard materials as seen here:



At first glance that might look OK, but of course we'll want to zoom in a bit as seen here:



Now we can see that things are more interesting and in fact maybe we need to move the high off-peak positions a bit more to the right!  Here we have for Al Ka zoomed in:



and more interestingly, for O Ka, which has a non-linear background:



And that's just three materials containing three or fewer elements in each material!  Those who are interested in more on background fit models can see here and other topics on this forum:

https://smf.probesoftware.com/index.php?topic=68.0

But the point is, whatever software you use (Probe, JEOL or Cameca) one must take care in selecting their off-peak measurement positions for accurate quantitative work:

https://smf.probesoftware.com/index.php?topic=1442.msg13896#msg13896

Yes, especially with regards to minor and trace elements, but even for major elements such as oxygen where the P/B isn't quite as good and/or the spectrometer position is near the low end of the spectrometer range and the background goes non-linear