How Accurate Are Modern Fundamental Parameter
Transcription
How Accurate Are Modern Fundamental Parameter
THE RIGAKU JOURNAL VOL. 21 / NO. 2 / 2004, 17–25 HOW ACCURATE ARE MODERN FUNDAMENTAL PARAMETER METHODS? MICHAEL MANTLER AND NAOKI KAWAHARA* Institute of Solid State Physics, Vienna University of Technology, Vienna, Austria * Rigaku Industrial Corporation, Osaka, Japan Fundamental parameter methods comprise a wide variety of applications and mathematical approaches. The question of achievable accuracy requires therefore a rather detailed view, and answers will vary with respect to the range of the analyte elements and the analyte lines, the employed mathematical models (such as conventional fp-equations, Monte-Carlo approaches, or theoretical influence coefficients), the type of specimen (bulk material, layered material, fused samples, etc.), and the instrumentation including the geometrical setup, excitation source, and detection system. Reference materials, their eventual dispensability (standardless methods) and—if provided—their optimum composition require careful attention. Finally, the reliability of the tabulated values of fundamental parameters themselves is a key element of such discussions. 1. Basic Considerations As with any other method, the fp-approach can take advantage of reference materials, and an analysis employing a standard of similar composition as the unknown material will give better results than by using a very different standard. Theoretically, the best possible result will therefore be expected if the compositions of unknown and standard (as well as the method of their preparation) are identical. In this case the obtained count data are defined only by the repeatability of the measurement, which is the closeness of agreement between independent results obtained with the same method on identical test material, under the same conditions. It is measured by the repeatability standard deviation, a purely statistical quantity. Hence, it can be made as small as desired by repetitions and long counting times within the general limitation of systematic deviations introduced by long-term drifts. Therefore, in principle, fundamental parameter methods (and other methods as well) are not inherently limited in accuracy. The achievable accuracy in an actual experiment is ultimately limited by repeatability issues, i.e., statistical errors, and the availability and quality of reference materials, i.e., the closeness of their assumed compositional data (“best possible estimates”) with (never observable) true values. The discussed hypothetical case of “analyzing” an unknown by comparison with a standard of identical composition is of course kind Vol. 21 No. 2 2004 of extreme, particularly in view of the main advantage of fp-methods which is that similar standards are not necessarily required. The other extreme is to use no reference materials at all (truly standard-less method*1). Then all parameters in the employed mathematical model must be known at an absolute scale (rather than in comparison to a standard). The obtained accuracy depends on all statistical and systematical errors of the fundamental parameters, the systematical errors introduced by deficiencies of the mathematical models, the numerical and statistical computational errors, and the statistical errors of the experiment. In an attempt to explore the limitations of such standard-less analysis by state-of-the art instrumentation and fp-methods, experiments have been carried out by Beckhoff et al. [1] and Mantler et al. [2] at fully calibrated synchrotron beam-lines of the PTB*2 laboratory at Bessy-II*3 by using certified pure element standards as *1 In many commercial products the term Standard-less Method is used as well, but with the understanding of employing pre-calibrated instruments, where data from reference materials are stored in the factory and remain valid over extended periods of time due to the exceptional longterm stability of modern instruments. In the context of this paper, [truly] standard-less is used for methods which are based on absolute counts rather than count rate ratios, and which use no reference materials for calibration. *2 Physikalisch-Technische Bundesanstalt (National Metrology Institute of Germany), Berlin, Germany. *3 Berlin electron storage ring company for synchrotron radiation, Berlin, Germany. 17 objects for testing. Measured and computed absolute photon counts (not count-rate ratios) are in agreements of less than 4% relative for thin layers of 5 nm–100nm of Cu and Ni. This matches the estimated uncertainties in the complex absolute calibration of all beam-line components and proves the validity of the basic fpapproach. Such purely standard-less methods are highly useful for development work, for example studies of the complex indirect excitation effects in the regime of light elements or XRF of L- and Mlines. All practical methods are, however, based on count rate ratios and some kind of reference materials including pure element reference standards. This allows working with “relative intensities”, where several factors (including some with less well known values) in the equations of standard and unknown cancel. 2. The “Ideal” Fundamental Parameter Approach The simplest model for the computation of the number of photons observed from a sample with known type and composition and under known experimental conditions is based upon absorption of primary and fluorescent beam within the sample, and the excitation of analyte line i by absorption of a primary photon of energy E (see Fig. 1). The following equations for the number of observed photons, ni, are valid for a thin compound layer (see the Appendix for a definition of the symbols): S 1 Ω ⋅κ i ⋅ ci ⋅ i ⋅ pi ⋅ ω i Si 4 ⋅ π ⋅ sin ψ 1 ni ∫ E0 ⋅ Ei N 0 (E ) ⋅ τ i (E ) ⋅ (1e T ⋅µi*⋅ρ ) ⋅ dE µi* (1) where L µi* ∑c j j 1 and c i L µ j (E ) µ j (i ) ⋅ c j ⋅ µ ji* sin ψ 2 j 1 sin ψ 1 ∑ ˆ mi m i M Mˆ L and Mˆ ∑ mˆ j j 1 Replacing the concentrations by mass-thicknesses, m̂i , eliminates also the thickness, T : S 1 Ω ˆi ⋅ i ⋅κ i ⋅ m ⋅ pi ⋅ ω i Si 4 ⋅ π ⋅ sin ψ 1 ni ∫ E0 ⋅ Ei 18 ⋅ N 0 (E ) ⋅ τ i (E ) ⋅ (1e ∑ j mˆ j ⋅µ ji* ) ⋅ dE ˆ mj ⋅µ ji* ∑ j (2) Fig. 1. Basic geometry of fundamental parameter approach. For bulk materials one obtains after division of numerator and denominator by m̂i : S 1 Ω ⋅κ i ⋅ i ⋅ pi ⋅ ω i Si 4 ⋅ π ⋅ sin ψ 1 ni ∫ E0 ⋅ Ei N 0 (E ) ⋅ τ i (E ) ⋅dE ˆj m ⋅ µ ji* j m ˆi ∑ (3) The count rate of a layer is a function of all mass-thicknesses constituting the layer, while for the bulk material it is only a function of mass ratios and independent on the total mass, M, (or thickness). If each of the L elements is measured by a single analyte line, equation (2) delivers (by an iterative algorithm) the L unknown mass-thicknesses m̂i . This is equivalent to L1 concentrations (with the L’th concentration obtained from cL1 Í cj ) and the thickness, T. In the case of bulk materials, however, there are only L1 unknown ratios m̂j /m̂i (or L concentrations after complementation by cL1 Í cj ), which makes the system of equations over-determined by 1. Unless the redundant equation is used for the analysis of an unmeasured element, the overall accuracy may be thereby improved. An undesirable side-effect may be that least-squares methods, as inherently used for the solution of an over-determined system of equations, have the general tendency of minimizing the sum of errorsquares by best-matching in first place the parameters with large numerical values, i.e., the high concentrations, at cost of low concentrations and traces. The Rigaku Journal The accuracy of the predicted counts, ni , depends upon • The general validity of the model (e.g., contributions by indirect excitation, errors introduced by the simplified geometry, etc.). • The accuracy by which the spectral distribution, N0(E ), of the primary photons is known. • The accuracy of the fundamental parameters. The simple model above includes only direct excitation in a homogeneous specimen. Indirect excitation is often seen synonymous with secondary and tertiary excitation and treated as a correction term, but there exist a few effects which can in specific situations by far outweigh direct excitation. The most important contributions are: • Secondary excitation (up to 25–30% of primary photons for bulk materials, much less for thin layers). • Tertiary excitation (up to 5%). • Coster–Kronig transitions (up to 30% in L-subshells, up to 70% in M-shells, due to modifications of the electronic configuration). • Cascading electron transfers during relaxation, whereby K-ionization initiates L-line (or M-line) emission within the same atom. The contribution varies considerably. It can be a large multiplicative factor in case of efficient [monochromatic] excitation by energies above the K-edge. • Excitation by photo-electrons and Auger-electrons (can be a multiplicative factor). Special sample types such as multi-layered materials, inhomogeneous samples, non-ideal geometries, rough surfaces, etc. must be dealt with by further extensions to this model (or treated by alternative methods, such as MonteCarlo computations). 3. The “Real” Situation: Divergent Beams In most modern instruments the X-ray tube is situated rather close to the specimen (Fig. 2). Therefore, the beam from the X-ray tube hits the specimen under rather widely divergent angular directions. Three effects must be taken into account: • As a consequence of the lateral incidence of the beam, the spot with the highest intensity on the specimen is not at the center but closer to the spot of minimum distance to the target (following roughly the “distancesquared-law”). This can easily be made visible by placing a photographic film (wrapped into black paper) into the specimen holder, as shown in Fig. 2. The uneven illumination by the primary Vol. 21 No. 2 2004 Fig. 2. Geometry of a crystal spectrometer. The sample is irradiated by a divergent beam under an angle of about 45° by the X-ray tube. The photographic image shows the intensity-distribution across the specimen. beam may also be an issue with imaging techniques, where a small spot of the specimen is selected by a mask and the specimen scanned. Only if the specimen is moved (as in modern Rigaku Systems) rather than the mask, a constant illumination by the primarybeam is maintained. • The spectral distribution and intensity of the partial beams in each direction vary due to the varying angle of emergence from the target and transmission through the Beryllium window. Radiation in the peripheral area is slightly hardened up. Mainly low energies and therefore very light elements or analyte lines from outer shells are affected. • Assuming a single “average” angle of incidence is only a rough approximation for a divergent polychromatic beam. While the deviation due to the geometrical divergence can be 19 Fig. 3. Computed dependence of count rates and count rate ratios of incident and observation angle. [Note: The count-rate ratios R_CrLa as a function of observation angle (at left/bottom) are practically constant at a value of 1.0. The count-rates n_NiKa , n_FeLa , n_CrLa , and n_NiLa as a function of observation angle are all within a narrow band (at right/bottom)]. easily computed, it is difficult to estimate the effect of spectral intensity variations, as this is a function of tube geometry, target material, and operating voltage. Computed data for stainless steel samples with varying angles are shown in Fig. 3. Building count-rate ratios is not necessarily helpful for all analyte lines (but certainly is in some cases). For an incident-beam divergence of 45°10° a relative error (without accounting for variations of the primary photon flux) in the 0.1% range for K-lines and up to 1% for low-energetic L-lines must be expected, for absolute photon counts as well as for count rate ratios. • While the observed fluorescent beam is quite parallel in crystal spectrometers, this is not the case of energy-dispersive detectors. Besides the geometrical error which can be derived from the fp-equation(s), the various angular directions can cause different detector response properties in EDXRS. 4. The Accuracy of Fundamental Parameters Fundamental parameters include 20 • Atomic interaction coefficients which describe the probabilities for photo-electric effect, coherent scattering, and incoherent scattering, • Absorption edge jump ratios, which are defined as the ratio of photo-absorption above and below the absorption edge and are used to compute the probability that a photon is absorbed in a specific energy level (edge), • The transition probability, which describes the probability of a specific electron transfer into the vacancy created by photo-absorption, and • The fluorescent yield, which is the probability that the relaxation is associated with the emission of a photon rather than of an Augerelectron. Additional parameters are required to describe interaction in the regime of light elements and low energies, such as • Probabilities of Coster–Kronig transitions, whereby electrons move into vacancies of a sub-shell from an other sub-shell of an atom, such as L1 ←L2 or L2 ←L3 or L1 ←L3, or correspondingly in the regime of the M-shell, • Various specific Auger-probabilities (see below), and • Cross sections for elastic and inelastic scatterThe Rigaku Journal ing of electrons by atoms, for the computation of indirect excitation effects by electrons. In addition, several other parameters are required, such as the spectral distribution of tube photons, the detector response function (“detection efficiency”), and the geometry of the instrument. For tube spectra, experimental data [3] as well as algorithms for their computation [4] are available. The match between theoretical and experimental data may be impaired by the age of a tube due to the contamination of target and window by tungsten and copper, and by deterioration of the target flatness. This affects mainly absolute photon counts and cancels in most cases effectively by building count rate ratios. As far as the detector response function can be treated as a purely energy dependent “counting efficiency”, it cancels by building count rate ratios. However, dead-time problems, and—with energy dispersive detectors— the complex behavior of the detector response function may require special corrections with unpredictable errors in case of neglecting these effects. The accuracy by which the atomic parameters are known depends in general on atomic number and energy (there are a few elements within the “well known-range”, such as radioactive materials and artificial elements, for which individual data are nevertheless unavailable or unreliable). It should be kept in mind that many coefficients have never been measured and their tabulated numerical values are derived by interpolation from neighboring atoms or energies, or by theoretical methods. Estimates of the accuracy of fundamental parameters by Elam [5], who has published a freely downloadable collection of fundamental parameters (Elam et al. [6]), are: A general observation in the field of interac- Cross sections, fluorescent yields Other parameters Below 1–3 keV Above 1–3 keV 10–50% A few % 50%? Frequently unknown 10–20% Mass Absorption Coefficients of Zinc 1.E+04 Line: XCOM v3.1( M.J. Berger. S.M. Seltzer, 1999 Dots: McMaster et.al. 1.E+03 1.E+02 1.E+01 0 5 10 15 20 25 30 keV m Zinc: Differences between Tables 5% 4% 3% 2% %Difference 1% 0% -1% 0 5 10 15 20 25 30 keV Fig. 4. Interaction coefficients for Zinc as a function of energy. Comparison of data by McMaster et al. and XCOM. scale. The overall effect on count rate ratios is typically in the range of up to 0.1%. 5. The Advantage of Building Count-rate Ratios The equations for directly excited photons of element i in a compound sample, ni , and for the pure element, Ni , show several common factors which cancel completely by building ratios (indicated by solid lines). The same effects would be achieved by using a compound reference material. In addition, the spectral distribution function, N0(λ ), and the photo-absorption coefficient, τ i (E ), appear in the integral of both equations and the effect of an error will at least cancel partially. This is particularly true if the error in these functions is closely a constant factor. tion coefficients is that the experimental values obtained by various authors differ mainly in the near-absorption edge region and are often in good agreement otherwise, as shown in Fig. 4 with data by McMaster et al. [7] and XCOM [8]. The differences are not visible in the logarithmic Vol. 21 No. 2 2004 21 ni ,direct Ω S 1 ⋅κ i ⋅ i ⋅ pi ⋅ ω i Si 4 ⋅ π ⋅ sin ψ 1 ∫ E0 ⋅ Ei Ni N 0′ (E ) ⋅ τ i (E ) ⋅ dE µi* Ω S 1 ⋅κ i ⋅ i ⋅ pi ⋅ ω i 4 ⋅ π ⋅ sin ψ 1 Si ∫ E0 ⋅ Ei N 0 (E ) ⋅ τ i (E ) ⋅ dE µii* ni, sec( j →i ) ⋅c i ⋅ (a) (b) Ω 1 S 1 ⋅ ⋅κ i ⋅ i ⋅ pi ⋅ ω i 2 4 ⋅ π ⋅ sin ψ 1 Si Si 1 ⋅ pj ⋅ ω j ⋅c j ⋅ Si ∫ E0 Ej N 0 (E ) ⋅ τ i (j ) ⋅ τ j (E ) µ j* sin ψ 1 µ(E ) ⋅ ln 1 ⋅ µ(j ) ⋅ sin ψ 1 µ(E ) sin ψ 2 µ(i ) ⋅ ln 1 ( j ) ⋅ sin ψ 2 µ(i ) µ ⋅ dE In the equations for secondary excitation the situation is less favorable, because all factors describing the excitation probability of the primary fluorescent photon, j, remain in effect (in the marked box). Their error propagates directly into the secondary excitation term and affects the total accuracy in proportion to the contribution of secondary excitation. If fluorescent yield, absorption edge jump, and transition probability are each affected by a statistical error of 1%, the error propagation law gives a total error in (relative) secondary excitation of also about 1%. Assuming that the contribution is nsecondary / ntotal10%, the overall result has then an error of 0.1%. 6. Specific Effects at Light Elements Analysis of light elements differs from that of medium- and high-Z elements by the fact that • Their fluorescent yield is very low (in the order of 103 for carbon and 104 for beryllium), and therefore the observed photon count rates are extremely low. • Excitation by tube photons is inefficient because there are no photons in the vicinity of the absorption edges of light elements. • Absorption of the fluorescent photons by the matrix is high. • Chemical state may affect the observed photon count rates. 22 Fig. 5. (a) Exitation of light elements by photoelectrons. (b) Contribution of Photo-electrons and Auger-electrons to the observed carbon Ka -counts in percents of the total photon count (computations by Pavlinksy et al. [11]). Singular data-point: Own computations for Fe–C, for a range of different geometrical set-ups. Excitation by Photo- and Auger-electrons When a photon is absorbed by an atom in an inner shell, in practically all cases a photo-electron is emitted which carries the energy-difference between the photon and the electron’s binding energy. Because most tube-photons have energies much higher than the binding-energy of a light element, the energy of the photoelectron may be rather high. The example in Fig. 5 shows the situation for EPhoton20 keV, EBinding0.28 keV, EElectronEPhotonEBinding19.72 keV. The energies of such electrons are in the same order of magnitude as those used for example in electron probe microanalysis (EPMA), and photo-electrons are therefore very well capable of exciting one or more other (light) elements. Other than in EPMA, electrons are emitted anywhere in the sample where photons are absorbed rather than injected from an external source (Fig. 6). In a similar way, Auger-electrons contribute to indirect excitation. The implementation of indirect excitation efThe Rigaku Journal 10 -8 10 -9 Measurement Calculation: without photoelectron excitation, without cascade effect with photoelectron excitation, without cascade effect with photoelectron excitation, with cascade effect [a] NB: parameters for the ionization crosssection of Bethe: b=0.209, c=1.053 [c] [b] 10 -8 10 -9 Calculated Cr-L Intensity Measured Cr-L Intensity Fig. 6. Left: Electrons in electron probe microanalysis (source is outside the specimen). Right: Photoelectrons emitted from some point within the specimen. Any point within the specimen, where tube-photons are absorbed, is such a source-point of electrons. -10 10 0 1000 2000 3000 4000 5000 6000 7000 8000 Incident Photon Energy / eV Fig. 7. Circles: Photon counts of CrLa , as a function of monochromatic excitation energy. [a] Effect of excitation by photo-electrons. [b] Jump at Cr–K edge due to additional vacancies in the K-shell, which are filled by K ←L3 transitions, freeing L3 for L3 ←M5 transitions (La 1). [c] Contributions of Coster–Kronig transitions. After Kawahara et al. [10]. fects by electrons in fundamental parameter software is difficult. While Monte-Carlo methods appear to be most suitable for computation of electron interactions, Kawahara [9, 10] has shown a successful solution for bulk materials and thin films which is based on conventional sets of equations (Fig. 7). Because of the fact that the electrons can originate from any atom in the matrix, those will abound whose host atoms absorbs high energy photons best and are contained in high concentration. In steel this is naturally iron. Since the size of micro-structural entities (grains) in iron-based materials is often in the Vol. 21 No. 2 2004 same order of magnitude as the range of electrons and carbon-photons, the microstructure of the material (and therefore the manufacturing process and specimen preparation) have an essential influence. Such situations constitute still a challenge in light element analysis and require careful selection of reference materials. Estimates by Pavlinsky et al. [11] and own computations show that in cases with small concentrations of a light analyte element (e.g., 1% C) in a heavy matrix up to 85% of the observed intensity may originate from indirect excitation by electrons. 23 7. Effects in the L- and M-Region Coster–Kronig Transitions and Auger-effect Coster–Kronig transitions and Auger-effect describe essentially the same type of de-excitation of an atom. The starting point is an ionized atom, as it exists for example after an initial absorption of a photon in an inner shell, after emission of a photo-electron. The mechanism of ionization is, however, of no importance. In XRF the subsequent relaxation is favorably a cascade of electron transitions from outer shells (e.g., K ←[Ka 1] ←L3 followed by L3 ← [La 1] ←M5 and possibly further transitions from outer shells). The competitive process is an energy transfer to an electron, which is then emitted, leaving the atom in a double ionized state. For example, when an electron transition K ←L3 takes place, the energy (equivalent to Ka 1) can be transferred to an electron, for example in M1, which is thereby emitted with an energy of EKa 1EM1. The electron is identified as a KLMAuger-electron (in detail: KL3M1-Auger electron). The Auger-process should be considered a single quantum mechanical relaxation event, which is not bound to the quantum mechanical selection rules for photons. In particular, there exists no “intermediate” state with a Ka 1 photon and hence no “internal absorption”. In this particular example, three different shells are involved: K, L, and M. If initial ionization and subsequent electron transition refer to the same principal shell (e.g., a LLM-type effect), it is called a Coster–Kronig transition. If everything takes place within the same principal shell, it is called a super-Coster–Kronig transition. These effects have the consequence of rearranging the distribution of electrons (in effect, of the vacancies) within the electronic shell of the atom and thereby changing the observed line intensities, which are proportional to the number of pertinent vacancies. For example, the emission of a La 1 photon (L3 ←M5) requires a vacancy in L3. This can be created in various ways: • Direct ionization of the L3 shell (e.g., by a tube-photon). • Direct ionization of the L2-shell, followed by a L2L3 Coster–Kronig transition. • Direct ionization of the L1-shell, followed by a L1L3 Coster–Kronig transition or a double transition L1L2L2L3. • Direct ionization of the K-shell, followed by a K ←[Ka 1] ←L3 transition. • Direct ionization of the K-shell, followed by a 24 K ←[Ka 2] ←L2 transition, followed by a L2L3 Coster–Kronig transition. • KL1X or KL2X Auger-effects with subsequent Coster–Kronig-transitions. Note that Auger-effect leaves the atom in a doubly ionized state. The change of energy levels may have a measurable effect. • KL3X Auger-effects, where X can be any outer shell for which EXEKEL3. These effects can be rather large. For example in many practical applications the ionization of the K-edge may be much more efficient than of the L3-edge and the cascading relaxation, amplified by Coster–Kronig transitions, may change the observed intensity considerably. Some selected numerical values have been given by J. Kawai [12] in detailed discussion of such effects (also related to the influence of chemical states) in a previous issue of the Rigaku Journal. A rather complete set of Coster–Kronig probabilities is contained in the compilation of fundamental parameters by Elam et al. A general discussion including a variety of numerical values of (detailed) Auger-probabilities is given by Bambynek et al. [13]. 8. Summary The current error-limit of absolute (truly standard-less) XRF by fundamental parameter methods under favorable circumstances (fully calibrated synchrotron beam-lines, medium-Z elements) is 3–4% relative. By using relative intensities with conventional instrumentation accuracies of a few 0.1 wgt% relative can be achieved with K-analyte lines in case of well prepared, homogeneous specimens. Main sources of errors are—besides experimental errors—inaccurate values of fundamental parameters including spectral distributions of tube spectra as well as neglect of accounting for real geometries and indirect excitation effects, particularly when light elements or L- and M-lines are involved. Symbols and Terminology i, j, k,..., n : Subscripts for elements (or compounds) in any given sample/specimen. Subscript i usually designates the analyte and j the remaining elements (or compounds) in the specimen (referred to as matrix elements or compounds. The subscripts are also used to indicate the (analyte or otherwise referred to) line of that element, the energy of that line, or the absorpThe Rigaku Journal tion edge belonging to that line. Atomic number Energy of a photon Highest energy in the tube-spectrum Ej : Energy of the absorption edge for analyte line j (element j) m iE : Mass absorption coefficient for element i for a given energy E m i*: Total effective mass absorption coefficient for element i, m i*m im i Ni : Count-rate from analyte line in pure element standard ni : Count-rate from analyte line in compound specimen N0(E ) · dE : Incident photons within interval (E, EdE ), per cm2 and second (flux) ci : Concentration (weight fraction) of element i t : Mass photo-absorption coefficient (subscripts as for m ) ˆ : Indicates items “per unit area”, such as m̂ m/A (mass/area) or T : Thickness (of thin layers) w : Fluorescent yield pi : Transition probability into a vacant shell in context of analyte line i S : Absorption edge jump ratio S(Photo-absorption coefficient at high energy side of edge)/(Photoabsorption coefficient at low energy side of edge) Z: E: E0 : P: Vol. 21 No. 2 ω ⋅p ⋅ 2004 Literature [1] [2] [3] [4] [5] [6] [7] [8] [9] [10] [11] [12] [13] B. Beckhoff, M. Mantler, M. Kolbe and G. Ulm, European Conference on X-ray Spectrometry, Alghero, 2004. M. Mantler, B. Beckhoff, M. Kolbe and G. Ulm, 53th Denver X-ray Conference. Steamboat Springs, USA, 2004. T. Arai, T. Shoji and K. Omote, Advances in X-Ray Analysis, 29, 413–426 (1986). See also http: //www.cstl.nist.gov/nist839/839.01/xrffpbiblio.html#x5 P. A. Pella, L. Feng and J. A. Small, X-Ray Spectrometry, 20, 109–110 (1991) and H. Ebel, X-Ray Spectrometry, 28, (1999) and K. Araki, A. Yoshioka, R. Shimizu, T. Nagatomi, S. Takahashi, Y. Nihei, Surface and Interface Analysis, 33, (2002) 376. W. T. Elam, Workshop “Quantitative Analysis”, Denver X-ray Conference, Denver, 2001. W. T. Elam, B. D. Ravel and J. R. Sieber, Radiation Physics and Chemistry 63(2), 121–128 (2002). W. H. McMaster, N. Kerr Del Grande, J. H. Mallett and J. H. Hubbell, Lawrence Livermore National Laboratory Report. UCRL-50174 Section II Revision I (1969). M. J. Berger, J. H. Hubbell, S. M. Seltzer, J. S. Coursey and D. S. Zucker, (1999). XCOM: Photon Cross Section Database (version 1.2). National Institute of Standards and Technology, Gaithersburg, MD, 1999. N. Kawahara, T. Yamada, B. Beckhoff, G. Ulm, R. Herbst and M. Mantler, 51st Denver X-ray Conference, Colorado Springs, U.S.A, 2002. N. Kawahara, T. Shoji, T. Yamada, Y. Kataoka, B. Beckhoff , G. Ulm and M. Mantler, 50th Denver X-ray Conference, Steamboat Springs, U.S.A, 2001. C. G. Pavlinsky and A. Yu. Dukhanin, XRS, 23, 221 (1994) and M. Mantler, Adv. X-ray analysis, 36, 1992, pp. 27–33. J. Kawai, The Rigaku Journal, 18, 1 (2001). W. Bambynek, B. Crasemann, R. W. Fink, H.-U. Fruend, H. Mark, C. D. Swift, R. E. Price and P. Venugopala Rao, Rev. Mod. Phys., 44, 716 (1972). S 1 S 25