A TOOL FOR PREDICTING DETECTION LIMITS AND ERRORS IN EPMA

摘要

Detection limits and errors in electron probe microanalysis (EPMA) can be accurately predicted using a small dataset and the phi-rho-z routines - here we use phi-rho-z routines implemented in CalcZAF (freeware produced by Probe Software [1]). An accurate prediction aids the choice of spectrometers, instrument conditions (kV, beam current) and count times. The tool (implemented in Excel) uses a small reference dataset which consists of net intensities for each element and mean atomic number background regressions. The data can be easily extracted from previous analyses to include different accelerating voltages and different spectrometer/crystal arrangements. For each condition (element, kV and spectrometer/crystal) a net intensity for a known composition (i.e., standard) is required. Net intensity for a compound standard is converted into intensity for a pure element standard using the matrix correction factor (ZAFCOR) calculated from phi-rho-z routines in CalcZAF (Table 1). Background intensities are stored as a second-order polynomial regression of background height and mean atomic number, where background height is adjusted using the absorption correction factor (ABSCOR) to remove the effect of continuum absorption. This follows the background method of Donovan et al. [2], and can be automated in the Probe for EPMA software [3] or calculated individually in CalcZAF. The data can then be interrogated (Fig. 1) to calculate detection limits and errors for the sample of interest. The pure element net intensity is converted into the sample net intensity using the element ZAFCOR calculated for the sample. The background intensity is determined from the calculated atomic number of the sample and corrected for continuum absorption using element ABSCOR calculated for the sample. Once the net intensity and background intensity are known the detection limit and error can be readily calculated using the formulas of Scott and Love [4]. If we consider the example shown in Fig. 1 of Ca at 20 kV in olivine (SH11), data is present for spectrometers 3 and 5 (spectrometer 1 is missing background intensity). It can be seen in the orange pane that spectrometer 5 gives lower errors and detection limits, and the effect of summing the two spectrometers is also calculated. The effect of increasing beam current and count time can be invested in the lower two green panels - here we calculate the errors and detection limits at 200 nA and 120 second count times. Both error and detection limit are reduced by an order of magnitude. The calculated values give a good approximation of measured values at the same condition (Table 2).