Matrix correction accuracy vs. matrix matched standard accuracy

[Probeman]

There are two ways in which we might approach the question of standard selection.

1. We might attempt to find a standard which is close in composition to our unknown material. This has the benefit of minimizing the matrix correction necessary for quantification. This method was the recommended approach 40 years ago when the physics of matrix corrections was still not well understood. However, today we find that our matrix corrections in EPMA (WDS and EDS) are on average around 2% or better.

The problem with using a matrix matched standard is that we might be depending on a standard which although is close to our unknown in composition, it might not be accurately characterized. Do we really know that our natural matrix matched standard composition is within 2% accuracy?

2. The other approach is to utilize a primary standard which is known to be highly accurate, so let's consider this latter situation a bit more: if we want to analyze say Mg in a pyroxene or garnet, we need to ask ourselves which is the source of greater error? A natural material known to be heterogeneous and containing multiple inclusions or perhaps we should obtain a synthetic high purity MgO or even better, as we saw in Will Nachlas' FIG-MAS presentation during M&M 2025, a synthetic high purity Mg2SiO4, and utilize our matrix corrections which are known to be within 2% accuracy or better?

Then there is a question of trace element accuracy: e.g., a TiO2 primary standard for measuring trace Ti in SiO2. TiO2 is of course which is of course quite dissimilar to our unknown, and therefore we depend on the accuracy of matrix correction physics to extrapolate from our TiO2 standard to our SiO2 matrix. The correction factor for Ti Ka at 20 keV is ~1.19. If we accept a ~2% accuracy on this extrapolation, we would expect a +/- 2 PPM accuracy at 100 PPM Ti concentration, or +/- 0.2 PPM at 10 PPM. This significantly smaller to our expected sensitivity for these measurements.

If on the other hand we utilized a primary standard which was a matrix matched standard, say an SiO2 with a "doped" concentration of Ti, we have to ask: to what accuracy do we actually know this dopant level? In other words, is the concentration 110 PPM or 90PPM, and is it even homogeneously distributed?

Far better is to extrapolate from our highly accurate TiO2 standard, and also utilize a blank synthetic SiO2 to determine the absolute accuracy of our ability to measure zero:

https://smf.probesoftware.com/index.php?topic=454.msg6694#msg6694

And please don't get me started on the statistical insanity of utilizing natural trace element matrix as a primary standard for a trace element:

https://smf.probesoftware.com/index.php?topic=610.msg11833#msg11833

In other words, we should judge the relative accuracy of our matrix corrections vs. the accuracy of the matrix matched standard. Yes, sometimes we will perhaps require a close matrix match to our unknown (highly absorbing matrices?), but we need to consider all sources of accuracy error, and make a scientifically informed choice.

See here for general considerations on major vs trace element analyses:

https://smf.probesoftware.com/index.php?topic=610.msg11752#msg11752

[Nicholas Ritchie]

The uncertainty associated with the natural similar standard should be included in the total uncertainty budget.  It is only when you compare the uncertainty associated with the matrix correction with the uncertainty associated with the natural similar standards can you make a well-reasoned choice between relying on a similar standard or conventional matrix correction.

[Probeman]

The uncertainty associated with the natural similar standard should be included in the total uncertainty budget.  It is only when you compare the uncertainty associated with the matrix correction with the uncertainty associated with the natural similar standards can you make a well-reasoned choice between relying on a similar standard or conventional matrix correction.

I completely agree, but unfortunately people seem to just assume the "published" values applies to the particular grain they standardized on.

And to make matters worse, when natural standards are concerned (e.g., Smithsonian standards), the published values are the average of the actual mineral matrix plus the inclusions. That is, the "uncertainty" (actually an accuracy error) due to the averaged inclusions are not accounted for by the wet chemistry.

Yet another reason to move to *global* high purity synthetic minerals as soon as possible when feasible.