Problems with the Analysis
Quantitative analysis is always beset with many difficulties and it is often difficult to pinpoint the cause of "bad" analyses. What is a bad analysis? This is usually based upon one or two tests: is the weight percent elemental (or oxide) totals very close to 100%? And if the material should have a certain stoichiometry, does it?
More often than not, the effect is the sum of more than one problem. Several things need to be checked. First of all, start by examining the standards.
Are the standards really "good" standards (both on their own, and for your particular suite of unknowns). Have their compositions been determined by a reliable analytical method? Traditionally, major element concentrations of standards have been determined using a bulk chemical technique, commonly "wet chemistry". The greatest problem is with natural minerals and glasses, which may be somewhat heterogeneous, and may have a sample "split" that is of a slightly different composition compared to the "official" composition. It is in situations like this, the use of secondary standards can really be helpful in determining what is going on. If the primary and secondary standards do not agree with each other, there is a problem: which is correct?
Traditionally, analysts have run analyses of several standards, using another as the standard for a particular element, and then repeat with different standards, to see which gives the "most correct" values the most times. The Probe for EPMA "Evaluate" application takes this one step further and plots up all standards for one element against each other (stated composition vs. ZAF corrected intensity), giving a clear indication of which standards are "good" (consistent with each other) and which are not (i.e., off the 1:1 line).
Also, are the concentrations entered in the Standard.mdb database correct and without typographical errors? And were counts acquired on the actual intended standard? It is easy to get lost at 300-400 magnification when using a standard mount which contains many standards (EDS helps here).
Some other things to consider :
1. Is the operating voltage correctly specified? Is the correct x-ray line tuned for each element? Check the on-peak position offsets from the Peak/Scan Option dialog and see if they are reasonable. The program will usually type a warning if the actual and calculated peak positions are very different. Be sure that the spectrometer is not tuned on a nearby line of another element if using multi-element standards.
2. Be sure that no "bad" data points are in the standards samples used for the quantitative calibration. The best way to check for this is to analyze each standard and examine the results to look for points with obviously bad or low totals (epoxy, bad surface polish, bad carbon coat, etc.). If a bad point is found, one can disable it. Remember, one can always enable data later on. A disabled point is simply not used in calculating the analytical calculations but is still present. A good rule of thumb is to only disable points that have low totals since generally most of the problems mentioned above will result in fewer x-ray counts. Avoid disabling points just to get better agreement between the primary and secondary standards. Points that have high totals should not be arbitrarily disabled. It may be necessary to look for other problems such as points with low totals in the primary standards.
3. Look for interferences on the analyzed elements. One easy way to do this is to use the Interferences menu item in the Standard program. One can also perform a wavelength scan and display possible interfering peak markers. If the element causing the interference is present in significant concentrations in your unknowns, and is not being analyzed for, it may be necessary to add the interfering element to the run by creating a new sample with the interfering element as an analyzed element. Be sure that the proper standards are available to use for the interference correction.
4. It would be useful to examine the precise choices for x-ray peaks; did the software/firmware routine choose the best peak position? Sometimes the algorithm doesn't. You want to use the center of the peak quasi-plateau where either any small spectrometer drift or peak/shape effects will be minimized. Probe for EPMA has a "post-scan" confirmation option where you can verify this. This issue would be for the main elements in a multi-element phase, and not that important for minor or trace elements where the determination of the background is the critical endeavor.
5. If none of the above suggestions seem to help, try acquiring the standards again. Probe for EPMA uses an automatic standard drift corrections which can make a significant difference in situations where one or more of your analytical channels is drifting. Note that since the program will perform an automatic drift correction not only on the standards, but also the interference standards and the MAN background standards, it might be also be necessary to run additional sets of those standards or MAN standards.
6. In the case of trace or minor elements, also check to see that none of the off-peak positions are interfered with by another peak. This can cause a reduction in the on-peak counts, sometimes enough to result in a negative k-ratio. Always run at least one wavelength scan on a sample, using the same count time as your quantitative analyses, and if a peak is seen interfering with the off-peak marker, use the Low and/or High buttons in the Graph Data window to select a new off-peak position that is not interfered with.
10. Detector electronics: Did you explicitly check/set the bias, gain, baseline? If operating in differential mode, did you verify the window width? Are you aware that high count rates on a standard will give you artificially low counts, as the pulse shifts to the left, potentially outside the lower window? To prevent this occurrence always check your PHA setting by running the Graph PHA Distribution button in the PHA setting dialog.
11. If you have high totals, have
you considered the possibility of secondary fluorescence, if you are measuring
small phases within a matrix or near another phase? Normally this is a small
effect on the total but when measuring trace or minor elements it can be a
source for large relative errors.
In addition, to solve problems with trace analysis, it is always useful to run at least one "blank" that is similar to your unknown sample, at least in terms of average atomic number, if not actual chemistry. For example, when analyzing trace elements in quartz, try to run an unknown analysis on a "pure" synthetic quartz sample, one that is known to contain known concentrations of the trace elements. The point is to try to determine how closely one can measure "zero" concentrations.