I think that there might be confusion in these results.

1. It seems that you are removing carbon from your results. Or not detecting it at all. (see part 4)

2. Calcite - CaCO3. Arguably, your result for CaO could be the calcite.

3. Quartz - SiO2.... you have this in your table

4. That's not Nb. That is Au from your coating. Your carbon peak is probably also being mislabled as Nb.

For good quantitative EDS, you should mount and polish your sample. That way you don't have any strange geometry effects altering your results. It would probably also be worth changing your coating to carbon (as thin as possible) as well, so that you don't have gold peaks all over your spectrum.

I think that there might be confusion in these results.

1. It seems that you are removing carbon from your results. Or not detecting it at all. (see part 4)

2. Calcite - CaCO3. Arguably, your result for CaO could be the calcite.

3. Quartz - SiO2.... you have this in your table

4. That's not Nb. That is Au from your coating. Your carbon peak is probably also being mislabled as Nb.

For good quantitative EDS, you should mount and polish your sample. That way you don't have any strange geometry effects altering your results. It would probably also be worth changing your coating to carbon (as thin as possible) as well, so that you don't have gold peaks all over your spectrum.

Your results are a mess. Matthew is very right.

If lines marked as Nb are really Au and C, then it looks like you have extremely thick coating. 

You need to repeat your measurements, coat with carbon, or if you cannot do carbon, coat again with Au, but with much (much!) thinner layer. Find lines if Au and carbon. Do not exclude carbon from quantification. Use standards (standardless analysis with a lot of C and O is mostly useless), and do not normalize to 100%. If you do not have standards, do not quantify at all, use qualitative results. 

 Dear Matthew and Vladimir

Thanks for your good explanations and contribution, and also for sharing your knowledge.

None

Check if you have the capability to coat with Al. Of course this would preclude identifying Al minerals, but if you want to find everything you may have to do more than one run with different coating materials.

As the other respondents have noted, the gold coating is so thick that it is obscuring the EDS spectrum. I don't think the element used for coating is particularly important. I mostly use palladium, but I have also used gold, and I can't tell any difference in the results in terms of image quality. But the coating thickness makes a big difference for EDS analysis. I generally try to have very thin coatings, and for geologic materials I can sometimes get good EDS spectra with no coating at all. Even if you are getting some arc tracks or other indications of specimen charging (e.g., the specimen appears to shift position as you scan) in the SEM image, it may still be possible to get a useful EDS spectrum. It depends on the beam voltage, the scan rate, the size of the field of view, the degree of electrical conductivity, and the specimen surface texture. But the best approach would simply be to greatly reduce the coating time (or emission current) for your sputter coating. For a smooth surface, the coating layer can be very thin. For example, I commonly coat rough-textured geological specimens for 90 seconds, but only 15 seconds for polished specimens. The amount of time you will use will of course depend on the sputter coater you are using.

If  you include carbon in the quantification, please keep in mind that the calculations for the carbon value will probably not be very reliable. This is a result of the difficulty of measuring the very low energy carbon X-rays, and the low "fluorescence yield" for this element. Our Edax Genesis EDS detector commonly calculates carbon percentages that are far in excess of the actual value,maybe two or three times too high. This is a major problem for measuring the other elements, because in the common standardless ("ZAF") quantification procedure, the values for all elements are normalized to total 100%.  If the % of carbon is overestimated, all of the other elements will be too low. I have never found a solution to this problem.If anyone has a suggestion, I'd like to hear it.

Finally, if you are using  computerized auto-labeling to identify the X-ray peaks, be careful about accepting the validity of the results. My experience has been that auto-labeling commonly gives incorrect identifications, e.g., the erroneous listing of Nb instead of Au in your spectra. You can check identifications using a table of characteristic X-ray energies for elements, One one-line version can be found at http://minerva.union.edu/labrakes/xrayenergytable.pdf

Best wishes,

George

I agree with George Mustoe and also with the observations by M.R. Field. I would strongly advise to polish the samples or, alternately, carry on the analysis on a non-tilted, flat, clean and smooth surface. The average C content of your samples should ensure conductivity when using a low beam dosage (e.i. low sample current). A very thin C coating could also help. However all the caveats about coating the samples and analytical results mentioned by George Mustoe should be taken into account.

I agree with Mathew; try to change coating to carbon, a thin layer, after a good polishing for a smooth and homogeneous surface; gold tends to get too too thick, in addition. 

Very good comments above. Just to add about the sample preparation method. If you have coals you will need to vacuum impregnate the sample with a low viscosity resin that is stable under an electron beam. Polishing will then need to be done very carefully. Best if you can find someone who has experience in polishing this type material to start with - or have fun trying! Yes, use carbon to coat. One problem you may encounter is the size of the particles being too small. Your image 4-1 shows a 2um particle - be aware of the size of the interaction volume; you will likely be including x-rays from elements outside your area of interest.

One reason your software may be giving you incorrect peak identifications is the number of counts in each channel, you maximum is only 500 counts. This is nowhere near enough in order to gain correct peak shapes for auto identification (humans can do better with few counts ;-) I don't know what detector you have but with a Si(Li) detector we used to work at 2,500 counts per second on the cobalt standard and count for 60 seconds for good quant work. These days with faster detectors it can be done in few seconds gathering 10s of thousands per second. So look at your data acquisition strategy.