The depth of analysis not only depends on the matrix, but also on the elements you are looking for. So which elements are important to you?

The most important elements are Fe, Si, Al, Mn, P, Mg and Ca. The first three elements are the most abundant.

Two other aspects come into play here. 

1. Are you analyzing the rocks as whole rock samples or are they prepared in anyway (powdered, pressed etc.) 
2. What types of rocks are these? 

The natural heterogenities of the samples will come into play if you are analyzing whole rock samples. If they were fine grained basalt, no problem, but if they are coarse grained granite or pegmatites then you will have to be careful in interpreting the individual analyses as the chemistry will change a lot depending on what is illuminated by the X-ray beam.

I'm analyzing powdered rock samples, which are Banded Iron Formation. It's a fine grained rock used as iron ore by steel industry. All the samples are crushed under 1mm.

What accuracy do you want from the lower atomic number elements ? If you're looking for very accurate figures (a vague description I know, but without values it's difficult to be accurate myself) then analyzing the rocks as loose coarse powders may not be the best sample preparation approach. You may well need to grind the samples to something like a 50um powder to avoid particle size effects such as shadowing. One thing you will not avoid will be mineralogical effects unless the standards you use have come from the same mineralogical source as your samples, unless you're OK about about significant errors.

The elements concentration vary like this:

• Fe: from 70% to 35%
• SiO2: from 75% to 0.5%
• Al2O3: from 20% to 0.5%
• P: from 0.3% to 0.02%
• Mn: from 20% to 0.02%

The ideal accuracy for these elements should be like this:

• Fe, SiO2, Al2O3: 0.5%
• P: 0.05%
• Mn: 0.25%

The standards I used for calibration comes from the same geological source.

The simple answer to your question is that "it depends". However, it isn't too much trouble to calculate the absorption coefficient for each line that you're measuring in a hypothetical sample with the average concentration that you see. From that and the angle of the detector, you can calculate the "infinite thickness" depth for each radiation. That will give you the depth that a given characteristic radiation can escape from, independent of the irradiation.

I've seen the sample prep stuff you found the link to, but I can't comment on them one way or the other because they're made by a competitor. You'll need to evaluate them yourself.  In one of your earlier posts. you mention that the grain size is about 1mm. If you want accurate light element performance, you'll need to go a lot smaller than that to avoid surface effects caused by large particles. Ideally you should have a particle size of about 50 microns.

X-rays, whether they come from a HH-, ED- or WD-XRF will all do the same; penetrate the sample down to "a certain" depth (depends on the anode material of your X-ray tube and the applied kV). What is important is what comes out of the sample (into the detector) and this will depend on the energy (keV) of the elements you are interested in (and some other parameters like sample preparation, density, etc.). The lightest element you seem to be interested in is Al (1,5keV) and the heaviest one Fe (6,4keV). This means that for whatever the sample form (preparation) is you will get a signal for Fe which comes from deeper than Al does. In the original rock we will probably talk about maximum 5 - 20µm for Al and 100 - 200µm for Fe (calculated by Lambert-Beer's law). So (sorry Stephen...) but 50µm won't be fine enough. The penetration depths won't be much more as Fe is quite a heavy mineral. Using this as a base you have to understand that sample preparation will be THE defining parameter for the accuracy of your analysis. I don't believe you can ever do a good enough analysis on the original rock! Better you lick the sample and estimate the concentrations 🙂 

In general, we make for iron ores fused beads and apply a correction called "the loss ignited alphas". But that's generally applied with WD-XRF. Only this method will give you the (absolute?) accuracies you are looking at! As you are a handheld user I guess you wanted to have a fast and cheap solution, right? I don't like to disappoint people, but I would say you probably didn't make the best choice (unless you can live with a worse accuracy than you defined).

Just adding to the great dialogue on this forum to date:

This website has a lot of information on X-ray Interactions with Matter http://henke.lbl.gov/optical_constants
This can give you some theoretical values for transmissions through various solids http://henke.lbl.gov/optical_constants/filter2.html

But obviously, as pointed out above, density and sample preparation are going to have the largest and most significant effects on both the precision and accuracy of your results with pXRF.

In saying that, we have done plenty of successful trial work on iron ores around the world with pXRF on well prepared samples. Feel free to drop me a line if you wish to discuss

50um has worked fine in cements, rocks, etc for me in the past, but if you think spending longer to grind the sample finer would be beneficial Pol, then I'll take a look at it and let you know. You've also got to bear in mind that when grinding to a particular particle size, a significant number of particles will have been ground finer.

I fully agree when you make pressed pellets out of the ground samples (to minimize grain size effects) and let the samples spin during measurement! What I mentioned before was particularly towards the combination HH-XRF, loose powder preparation and accuracy of lighter elements.

Great discussion. I am particularly impressed with Pol's unique Taste and Analyze recommendation. If your analyzing powders, presumably in a sample cell and using a tripod mount, then I would use a fine powder and just fill the sample cup up with about a cm of material since I am guessing with BIF samples you are not sample limited. You will then be certainly infinitely thick for your elements of interest, + most others you might pick up in the spectrum when you are analyzing such as Sr, Rb, Zr, that might be of interest. If the standards you measured are already certified for additional elements then if you are working with a good software package you should be able to add these additional elements to your calibration without needing to reanalyze the standards or samples. In effect more data for no more analysis. I used to take this approach all the time in my old application lab.

There have been many good comments about the important factors that impact data quality, sample homogeneity, particle size, matrix effects, sample preparation and the quality of the calibration standards, but as analysts we are all striving for the best quality data to answer questions in an imperfect world. Understanding the trade off between data quality and the data quality required to answer a particular question is key. Framing the question well that you are trying to answer with analytical data is really important, otherwise you run the risk of collecting a lot of very accurate data that might not answer the question you are posing. I thinks its always good to have the larger prospective about what you are trying to achieve with the data you want to collect.