Evaluation of Correction Procedures for Quantitative Light Element Epma

The performance of correction models used in light element EPMA is assessed, with a special emphasis on the energy dependence of the absorption correction factor. Recently Love and Scott /I/ have assessed the performance of the established correction procedures and some new models used in electron probe microanalysis ( E p U ) . They concluded that all methods are approximately equivalent in accuracy when applied to a wide range of medium to high Z microanalysis data (430 measurements), giving the root mean square (m) error in the calculated concentration of the order of 5.4-7.3 %. Serious improvements are required, however, in light element studies. The conventional ZAP procedures give for light element systems unacceptably large RMS errors (9-14%) in contrast with the correction model proposed recently by Love and Scott /2/ which appears to give reasonable results, with an W error of 5.6 %. It should be noted, however, that so far 1) only a limited set of experimental data, say for oxides /3/, has been used to estimate the efficiencies of some particular correction methods, 2) the "optimised" mass absorption coefficients have been used to minimise errors in the concentration obtained by a given method, 3) microanalysis data for a wide range of electron energies E have been included into the estimated RMS errors whereas the performance of dlfferent models used in light element EPU is known to be sensitive to Eo (ork/c? ) values. In the present work microprobe data available for light element systems containing 0, C and B are considered in order to estimate the efficiency of different correction procedures, Since for light element systems the absorption correction factor &(and its dependence upon E,) is dominant over atomic number effects, we shall compare here the perfornance of different absorption corrections, using in all cases Love-Cox-Scott (LCS) expression /4/ for the stopping power factor Ps and the tables of R values /5/ to compute the backscatter factor F6 . The mean ionisation potential J and the surface ionisation parameterf(0) are obtained with the use of ~erger-Seltzerls and LCS1s formula respectively. Adopted values of mass absorption coefficients are those of Henke and Ebisu /6/, the backscattering coefficients are taken from tables /7/. The LCS approach to the calculation of the stopping power factor based upon a modified form of the Bethe-Ashkin equation for electron eneegy loss permits to avoid dE/dx becoming zero when ~=~/1.166. This would be the case, for example, in analysis for light element 1 from the binary system 1-2 if c E,S J, (J, /J, )Y1.166, Article published online by EDP Sciences and available at http://dx.doi.org/10.1051/jphyscol:1984202 C2-10 JOURNAL DE PHYSIQUE where J, , J, are the ionisation potentials of the constituent elements, C is the light element mass concentration, and E,the critical excitation energy. Such situation occurs, in particular, in analysis of Be in Cu-Be alloys (c,< 67 %), in the Fe-B system (C,< 21 %) etc. In the present paper we estimate the performaace of the following correction models: the simplified Philibert approximation (\P(o)=o) with I)%= 4.5 and n = 1.65 (Heinrich), 2)%= 2.39, n = 1.5 (Dwncunb-Shields); 3) the Philibert rigorous model with = 2.54, n = 1.5 (Duncumb-Eelford), 4) the Ruste-Zeller (RZ) absorption correction /8/ in which and n are Z de endent and h is energy dependent; 5) the ICS approximation /9/ with <= 9.5, n = 2; 6) the Love-Scott (LS) absorption correction based upon Bishop's rectangle approximation to 'f(p2); 7) the thin film (TF) model by Duncumb and Melford which comblnes atomlc number and absorption corrections. The expressions for h recommended for the above corrections are used in the calculations. In addition to the microanalysis data for binary and ternary oxides /3/ we take into consideration the oxygen, carbon and boron analyses /10,11/ carried out on the ARL instrument with @= 52.$, the results of Ruste and Gantois /12/ obtained on the MS-46 microprobe withe = 18Oand some other data. The whole collation comprises 272 measurements on 49 systems at 6 different voltages (5-30 kV). The measured intensity ratios, k, except those from /3/ were taken from the figures in the papers /10,11/; errors arising from reading the curves vary from 2-3 % to 8 %, while the statistical counting error will be-1.5 %. Some of the measurements /10,11/ obtained at low (5 kV) voltage and the results for Pb0 /3/ were excluded from the assessment as unreliable due to large RMS errors ( > 15 %70) in the estimated concentration. Typical overall correction factors, P, calculated by dif f erent methods f or moderately ( )( > 20000,~/$~ > 3) and heavily (f < 20000, $/x, < 3) absorbing systems are plotted wn Fig. 1,2 as a function of probe energy E,. For comparison, the experimental F(= c/cook ) values are also given in the figures. The RMS errors in the estimated concentration (Table, next page) were evaluated a) for all the systems in the energy range 5-30 keV and, separately b) for low (% c 20000) and high absorbing systems 0(>20000). The RMS data are also subdivided into two droups accordingly to the probe voltage (5-15 kV, 20-30 kV) and separately for low and high absorbers. DSU F F