JOURNAL OF ANALYTICAL ATOMIC SPECTROMETRY, FEBRUARY 1987, VOL. 2 67 Silicate Rock Analysis by Energy-dispersive X-ray Fluorescence Using a Cobalt Anode X-ray Tube Part 2.” Practical Application and Routine Performance in the Determination of Chromium, Vanadium and Barium Philip J. Potts, Peter C. Webb, John S. Watson and David W. Wright Department of Earth Sciences, Open University, Walton Hall, Milton Keynes, Buckinghamshire MK7 6AA, UK This paper investigates the practical determination of the elements Cr, V and Ba by energy dispersive X-ray fluorescence using the excitation conditions dedcribed in Part 1. Particular attention was paid to the selection of a reliable set of geochemical reference materials to form a coherent calibration set. Discrepancies caused by the overlap of calcium - aluminium and calcium - silicon sum peaks on Ba Land Cr K lines in those samples containing more than 10% CaO were eliminated by reducing spectrum data accumulation rates.New data are presented for the elements Cr, V and Ba (and Ti and Mn) in nine reference materials recently distributed by the Geological Survey of Japan and these results show satisfactory agreement with other published values. Keywords: Silicate rock analysis; energy-dispersive X-ray fluorescence; cobalt anode X-ray tube; chromium, vanadium and barium determination The elements chromium, vanadium and barium are important in many geochemical studies, for which determinations are required down to concentrations of a few p.p.m. Several techniques are capable of achieving this goal, including wavelength-dispersive X-ray fluorescence (WD-XRF).However, this method is not without some analytical difficul- ties. In particular, although conventional wavelength-disper- sive X-ray spectrometers have high nominal resolution, several serious spectrum overlap interferences affect the elements of interest, e.g., Ti KP on V Ka, V KP on Cr Ka and Ti Ka on Ba La. Indeed the latter interference is relatively severe in silicate rock analysis applications, causing some workers to prefer the less intense Ba Lp line for analysis.’ Recent work in this laboratory2J has established that energy-dispersive X-ray fluorescence (ED-XFW) is as effec- tive as the wavelength-dispersive technique in determining routinely a wide range of major and trace elements in silicate rocks. Detection limits, particularly for the heavier trace elements (Rb, Sr, Y, Zr, Nb, Pb and Th), were found to be equivalent to those encountered in routine WD-XRF schemes of analysis.3 However, analytical data for the trace elements Cr, V and Ba were unsatisfactory and not reported.The reasons for this were because (i) K lines of these elements were not adequately excited using the general purpose silver tube then available and (ii) the resolution response of an energy-dispersive detector compares unfavourably with that of a wavelength-dispersive spectrometer in this region of the X-ray spectrum. Data presented in Part 1 4 show that the difficulties in determining Cr, V and Ba by ED-XRF analysis are largely overcome by exciting samples with a cobalt anode X-ray tube. Not only are the trace elements of interest excited efficiently by the characteristic tube lines, but in combination with a 12.5-pm iron primary beam filter, the excitation of these elements is selectively enhanced relative to that of iron.As the abundance of iron in silicate rocks is normally several orders of magnitude higher than that of the trace elements in question, this selective excitation has beneficial analytical consequences as discussed previously.4 In this paper, the practical application of selective cobalt * For Part 1 of this series see reference 4. tube excitation is examined in the ED-XRF analysis of silicate rocks. In particular, the choice of reference materials and the reliability of the resultant calibration data set is examined critically.The quality of refined calibrations is assessed through determinations of the elements Cr, V and Ba (and additional data for Ti and Mn) in nine new silicate reference materials recently distributed by the Geological Survey of Japan. Instrumental Procedures Analytical data were obtained from an energy-dispersive X-ray fluorescence spectrometer (Link Systems MECA 10- 44) fitted with a cobalt anode side-window X-ray tube. Samples were excited at 20 kV, 0.1 mA using a 12.5-pm iron primary beam filter. The nominal resolution of the Si(Li) detector was 165 eV at 5.9 keV and the X-ray tube was operated in the pulsed mode to permit data acquisition rates of 10000 counts s-1 at only 3&35% dead time. Spectra were counted for 800 live seconds and peak areas were quantified using a proprietary filtered least-squares deconvolution pro- gram.5 This deconvolution program has been shown to calculate the areas of overlapping spectrum lines reliably in ED-XRF applications,2 and is an important factor that contributed to the success of the present study.Proceedings for deriving the X-ray profiles used in deconvolution and other experimental details have been described earlier.2 The instrument was calibrated using the wide range of international reference materials listed in Table 1, prepared as pressed powder pellets. Calibration equations were obtained for individual elements by linear regression of X-ray count data against apparent fluorescence concentrations6 calculated using the “usable” compositions of Abbey.7 In routine analyses, apparent concentrations were derived from the calibration equation.“True” concentrations were then calcu- lated from an iterative mass attenuation correction procedure using data for the major elements determined from photo- peaks in the same spectra that were recorded for the measurement of Cr, V and Ba count data. The mass attenuation correction procedure utilised the coefficients of Leroux and Thinh8 and was adapted to account for structural water as estimated from the loss on ignition of a separate aliquot of each sample.68 JOURNAL OF ANALYTICAL ATOMIC SPECTROMETRY, FEBRUARY 1987, VOL. 2 Results and Discussion The success of any calibration based on international refer- ence materials depends on the reliability of “usable” values taken from the literature.As discussed by Abbey,9 geochem- ical reference material compositions cannot be adopted uncritically when choosing a calibration data set. Of particular concern in this work were the data used for the calibration of chromium. Most silicate reference materials contain less than 400 p.p.m. of chromium. However, three widely used reference samples contain significantly higher concentrations, namely: peridotite USGS PCC-1 (2800 p.p.m.), dunite NIM-D (2900 p.p.m.) and pyroxenite NIM-P (24000 p.p.m.). Preliminary examination of the ED-XRF calibration data set indicated that significant discrepancies could affect the fit of the calibration line at low concentrations if these exceptionally high values were included. This effect is demonstrated in Fig.1 , which plots the ED-XRF calibration data for ten reference materials selected arbitrarily to cover the range 0-400 p.p.m. of chromium together with data for PCC-1. The complete data set is plotted in Fig. l(a). In Fig. l(b) and (c), a linear regression for these data is plotted over the range 0-0.6 (apparent fluorescence values as YO oxide). All ten data points were included in the linear regression plotted in Fig. l(b). The datum for PCC-1 was excluded from the regression plotted in Fig. l(c). Examination of these graphs shows that bias is introduced into the regression line plotted in Fig. l(b) characterised in this instance by a small but significant positive intercept with the intensity axis. This bias is virtually eliminated when PCC-1 is excluded from the data set.This phenomenon could be caused by a non-linear response from the instrumentation at high chromium count rates. Alterna- tively, rounding errors (+ 50 p.p.m.) in the quoted composi- tion could also contribute to this discrepancy. These results thus highlight the unreasonably high weighting that may be placed on a single datum by a linear least-squares regression procedure, applied to a calibration data set which includes one point that is significantly isolated from the main body of data. To avoid such errors, Thompson10 recommended that the simple linear regression procedure was likely to give satisfac- tory fit only if ten or more data points were used and if these data were more or less uniformly spread over the entire concentration range from zero upwards.This latter criterion is not upheld when PCC-1 is included in the calibration set. This sample was, therefore, excluded from the calibration used in this work. By similar reasoning, the reference samples Mica-Mg (4000 p.p.m. Ba) and NIM-S (2400 p.p.m. Ba) were not included in the barium calibration. Table 1. Cr, V and Ba compositions of reference materials (p.p.m.). Data in italics were used in the final calibration. All data are abstracted from Abbey.’ Data marked with ? are “less reliable” values; the remainder are “usable” values Reference material AGV-1 . . . . . . BCR-1 . . . . . . G-2 . . . . . . . . GSP-1 . . . . . . PCC-1 . . . . . . BHVO-1.. . . . . QLO-1 . . . . . . RGM-1 . . . . . . SDC-1 . . . . . . sm-1 . . . . . . BIR-1 . . . . .. DNC-1 . . . . . . w-2 . . . . . . BCS-375 . . . . . . BCS-376 . . . . . . G A . . . . . . . . GH . . . . . . . . BR . . . . . . . . Mica-Fe . . . . . . Mica-Mg.. . . . . USGS (USA)- BCS ( U K j CRPG (France)- ANRT (France)- DR-N . . . . . . FK-N . . . . . . GS-N . . . . . . AN-G . . . . . . BE-N . . . . . . MA-N . . . . . . GSJ (Japan)- NIM (South Africa)- JB-1 . . . . . . JG-1 . . . . . . NIM-D . . . . . . NIM-G . . . . . . NIM-L . . . . . . NIM-N . . . . . . NIM-P . . . . . . NIM-S . . . . SY-2 . . . . . . SY-3 . . . . . . CCRMP (Canada*)- Cr V Ba 10 15 8 12 2800 300 4.2? 4? 66? 4? 370? 270? 92 125 420 36 54 29 320? 61 14? I05? 31 0 150 260 - 1200 680 1900 1300 135 1400 800 650 560 120 175 4? 6. l? . . . . . . . . . . . . . . . . . . . . . . . . . . 25 - 90? 450? .. . . 12 6 380 90 100 38 5? 240 135? 90? 850 22 1050 145 4000 . . . . . . . . 230 - 62? 70 240 4.6? 390 21 O? 1400 34 1050 42 42 55 50 360 3? 3? * . . . . . . . . . . . 400 53 21 0 24 490 460 . . . . lo? 120? 450 100 2400 46? 2900 12 lo? 30? 24000 12 40 81 220 230 10 2? . . . . . . . . . . 12 10 52 51 460 430 . . . . x104 125 v) U 5 100 8 . > C .z 75 +I - 50 25 0 xi04 12 8 4 x104 ( b ) PCC-1 included 12 - 8 - 4 - 0 I (c) PCC-1 excluded I (a) all data PCC-1 0 z I 1 0.2 0.4 0.6 0 0.02 0.04 0.06 0 0.02 0.04 0.06 Apparent fluorescence values, O/O oxide Fig. 1. ED-XRF calibration lines for chromium. (a) Original data set (X-ray count intensity versus apparent fluorescence concentration in % oxide) for ten reference materials having Cr com ositions between 0 and 600 p.p.m. (AGV-la, JG-1, SDC-1, GS-N, BCS-375, JB-1, BE-N, BHVO-1, BR, DR-N) and PCC-1 (2800 p.p,m.C$. (b) Linear regression for data set including PCC-1. (c) Linear regression for data set excluding PCC-1. The lower part of the calibration line only is plotted in Figs. (b) and (c)JOURNAL OF ANALYTICAL ATOMIC SPECTROMETRY, FEBRUARY 1987, VOL. 2 69 In order to select reference materials suitable for setting up a reliable calibration, as wide a range of silicate reference samples as possible were analysed, the resulting data being examined carefully for internal consistency. In addition to the potential calibration bias arising from samples with unusually high elemental compositions, as discussed above, it was found that fitting errors were also large if all other values were included uncritically in the calibration.Data were, therefore, examined carefully to identify outliers both in calibration data plots, (that is X-ray counts plotted against apparent fluores- cence values) and in plots of expected versus analysed compositions. After exclusion of samples that either plotted as outliers or had unusually high concentrations, the reference values accepted for the final calibration data set are listed in Table 1. The self-consistency of the resultant calibration is demonstrated in Figs. 2 (Cr), 3 (V) and 4 (Ba). In some instances the values discarded from the calibration data set correlated with spectral overlap interferences caused by sum peaks. These are considered further below. In other instances, discrepancies could not be correlated with any identifiable source of systematic bias.BIR-1 (13.33% CaO), W-2 (10.89y0 CaO), AN-G (15.92% CaO), DNC-1(11.52% CaO) andNIM-N (11.50% CaO). The magnitude of this effect was particularly exaggerated when the abundance of coexisting Cr or Ba was low, as may be judged by the data for these samples identified separately in Figs. 2 and 4. To minimise the influence of these sum-peak interferences in analytical results reported in this work, excitation con- ditions were modified from those in the original proposal.4 Because of design limitations in our instrumentation, it was not possible simply to reduce the tube current below the recommended value of 0.1 mA. As an alternative, it was found effective to double the thickness of the iron primary beam filter to 25 pm to reduce significantly the rate of data accumulation and so eliminate sum-peak interferences.Com- pensation for the expected reduction in analytical precision was achieved by increasing spectrum count times from 800 to 1200 s. In consequence, detection limits for Cr, V and Ba are expected to be similar to those originally reported.4 Sum-peak Interferences The analysis conditions derived in the previous paper4 were selected for materials having relatively low calcium contents. However, examination of ED-XW data for a wide range of silicate reference materials showed that unacceptably high count rates were observed for samples containing more than about 10% CaO. This major element is efficiently excited under the conditions optimised for trace chromium determi- nations and sum peaks arising from coincident detection of calcium Kar and both the aluminium and silicon K lines then become significant.These sum peaks were not taken into account by the spectrum deconvolution software used here and caused significant spectrum interferences on Ba Lp2 and Cr Ka as follows: Ba Lp2 (5.16 keV) is overlapped by Ca Kar + A1 Ka (5.18 keV); and Cr Ka (5.41 keV) interfered with by Ca Kar + Si Ka (5.43 keV). No comparable interference was detected on the vanadium K lines. The effect of these sum-peak interferences is to increase the apparent intensity of the trace element line causing systematically high determinations of Cr and/or Ba in 400 E 300 c- al C 2 6 200 c 8 % -0 z P - 100 I 1 I I 100 200 300 400 Expected Cr content, p.p.rn.Fig. 2. Analysed versus expected content of _chromium in silicate reference materials between 0 and 400 p.p.m. 0, Values accepted in the final calibration; 0, values discarded from final calibration; and + , values subject to significant sum peak interferences (see text) 400 300 2 c al c 8 > 200 U v) z m - 2 100 100 200 300 400 Expected V content, p.p.m. Fig. 3. Analysed versus expected content of vanadium in silicate reference materials between 0 and 400 p.p.m. See Fig. 2 for symbol identification 1500 E 2 w- al c = 1000 8 m P U rn > m 4 500 - 1 I I 500 1000 1500 Expected Ba content, p.p.m. Fig. 4. Analysed versus expected content of barium in sificatc reference materials between 0 and 2000 p.p.m. See Fig. 2 for symbol identification70 JOURNAL OF ANALYTICAL ATOMIC SPECTROMETRY, FEBRUARY 1987, VOL.2 Table 2. Analytical results for reference materials from the Geological Survey of Japan. Analyses carried out on two powder pellets (1 and 2); x, mean determination; CV, coefficient of variation of the four analyses listed for each element = 100 x standard deviatiodx. All data listed in p.p.m. JA-la JB-la Element 1 2 R CV,% 1 2 x CV,% Cr V Ba Ti Mn Cr V Ba Ti Mn Cr V Ba Ti Mn Cr V Ba Ti Mn Cr V Ba Ti Mn 6 5 107 117 402 434 5080 5024 1162 1149 5 5.5 10.5 6 117 114.8 4.5 118 399 414.8 4.1 424 5099 5073 0.66 5090 1162 1157 0.55 1154 JB-2 366 371 198 201 660 592 7823 7879 1119 1129 372 369.2 0.76 368 204 203.5 2.6 21 1 579 635.5 9.7 711 7906 7857 0.54 7821 1132 1124 0.69 1116 JB-3 1 27 22 546 546 295 339 7047 6992 1752 1732 2 x CV,% 30* 26.5 12.5 27 551* 544.0 1.4 533 410* 315.8 25.3 219 6984* 7013 0.43 7029 1753* 1742.5 0.66 1733 JG-la 1 50 48 349 364 405 343 7455 7445 1348 1346 2 x CV,% 45 * 51.0 13.6 61 373 * 354.5 5.1 332 855*,f 355.3 12.6 318 7249 * 7382 1.3 7380 1352* 1345.3 0.54 1335 JGb-1 1 18 16 24 28 482 437 1593 1603 437 490 2 x CV,% 26 22.5 29.4 30 19 21.8 24.4 16 453 461.0 4.3 472 1597 1598.5 0.28 1601 432 446.5 6.6 427 1 65 71 614 613 c95 190 9587 9588 1553 1549 2 x CV,% 67 66.3 5.7 62 616 616.3 2.6 622 c94 169 9618 9575 0.5 9507 1559 1551.5 0.69 1545 - - JP-1 JR-1 1 2 B CV,% 1 2 x CV,% 2722 2745 18 13 < 12 < 12 141 143 987 983 3001 2868.8 5.5 3007 17 15.5 15.4 14 <12 <12 - < 12 142 142.0 0.57 142 994 990.5 0.68 998 JR-2 1 <3 (3 <4 <4 <23 <23 491 498 790 794 2 x CV,% - - <3 5 <4 <4 - <4 <23 45 502 493.0 1.9 48 1 789 791.0 0.27 791 - - 12 11 8 7 53 48 737 740 710 717 - - <3 5 <4 <4 45 48.0 7.4 46 728 734.0 0.75 73 1 699 707.3 1.1 703 - - * These analyses were for pellet 1.7 Determination not included in the calculation of the mean.JOURNAL OF ANALYTICAL ATOMIC SPECTROMETRY, FEBRUARY 1987, VOL. 2 71 Table 3. Comparison between analytical results and published values for the new GSJ reference materials (p.p.rn.). Values in parentheses may be subject to additional uncertainty in extrapolating the calibration (see text) Ti V Reference material Ja-landesite . . . . JB-labasalt . . . . JB-2basalt . . . . JB-3basalt . . . . JG-la granodiorite .. JGb-lgabbro . . . . JP-lperidotite . . JR-lrhyolite . . . . JR-2rhyolite . . . . This work . . 5073 . . 7857 . . 7013 . . 7382' . . 1559 . . 9575 . . 142 . . 734 . . 493 Bower And011 et al. 13 5220 5100 k 120 8030 - 7130 7070 f 120 8690 - 1560 - 9710 - - - 600 660+60 540 - Cr This work 115 204 544 354 22 616 16 <4 - Ando" 103 21 1 540 24 - - - - - Mn Bower et al. 13 111 f 6 590f40 - - - - - 7 + 2 - Ja-1 andesite . . . . JB-labasalt . . . . JB-2basalt . . . . JB-3 basalt , . . . JG-la granodiorite . . JGb-lgabbro . . . . JP-lperidotite . . JR-lrhyolite . . . . JR-2rhyolite . . . . This work . . 5.5 . . 369 . . 27 . . 51 . . 22 . . 66 . . (2869) - . . - . . Ando 6 405 28 53 - Bower et al. 13 7 2 2 28 f 2 - - - - - 4 f 2 - Potts and Rogers12 10 430 31 62 20 62 (2942) 4 6.7 This work 1157 1124 1742 1345 446 1552 990 707 79 1 Bower Ando" et al.13 1160 1180 f 20 1240 - 1550 1780 f 40 1240 - 490 - 1320 - - - 770 770 f 15 850 - Ba Ja-landesite . . . . JB-labasalt . . . . JB-2 basalt . . . . JB-3basalt . . . . JG-lagranodiorite . . JGb-lgabbro . . . . JP-lperidotite . . JR-1 rhyolite . . . . JR-2rhyolite . . . . This work . . 415 , . 636 . . 316 . . 355 . . 461 . . <12 . . 48 - . . - . . And011 307 490 208 462 - - - 40 - Bower et al. 13 370 k 40 290 f 40 - - - - - 64 k 40 - Potts and Rogers12 303 509 246 238 484 - - 107 105 Analysis of Reference Materials from the Geological Survey of Japan The value of the proposed scheme was tested by analysing, as unknowns, nine reference materials recently distributed by the Geological Survey of Japan.These sampl'es were not included in the calibration data set although granite JG-la and basalt JB-la appear to be re-issues of earlier samples, JG-1 and JB-1 , which were used as calibration reference materials. Duplicate samples of pressed pellets of each reference material were each analysed twice using the reduced excita- tion conditions outlined above. The results for each determi- nation are listed in Table 2 together with the mean values and coefficients of variation. Data for the elements Ti and Mn are included in this table as these are also efficiently excited under the conditions adopted here. The precision of the results listed in Table 2 is considered to be satisfactory for all elements, but with some reservations about the barium data, an element for which interferences in ED spectra are expected to be most severe.In Table 3, average results are compared with the proposed values of Ando" and analysed data of both Potts and Rogers12 (neutron activation data for Cr) and Bower et ul. 13 (data averaged from XRF, neutron activation and ICP atomic emission determinations). Data presented here gener- ally lie within the spread of results reported by other workers, thus justifying the use of this ED-XRF method for the routine determination of Cr, V and Ba in silicate rocks. In view of the uncertainties described earlier in analysing reference materials of unusually high chromium content, some reserva- tion must be expressed in the reliability of extrapolated data for JP-1 (2869 p.p.m. Cr average) presented in Table 2.Conclusion Energy-dispersive X-ray fluorescence analysis using cobalt anode X-ray tube excitation may be employed successfully to determine the geochemically important trace elements Cr, V and Ba in silicate rocks. The accuracy and reliability of results depends on a critical evaluation of reference material values incorporated in the calibration data set. It was necessary to remove data for reference samples containing unusually high concentrations of chromium and barium from the calibration set to prevent bias at low concentrations in the corresponding calibration line. If samples containing more than cu. 10% CaO are to be analysed, care must be taken to avoid excessively high data accumulation rates and so prevent sum-peak interferences on the elements Cr and Ba. In the analysis of nine reference materials distributed by the Geological Survey of Japan, results lie within the spread of data reported . previously.72 JOURNAL OF ANALYTICAL ATOMIC SPECTROMETRY, FEBRUARY 1987, VOL, 2 The authors gratefully acknowledge Link Systems Ltd. for loan of the cobalt X-ray tube used in this work and John Taylor for assistance in preparing this manuscript. 7. Abbey, S., Geol. Surv. Can. Pap., 83-15, 1983, 114 pp. 8. Leroux, J., and Thinh, T. P., “Revised Tables of X-ray Mass Attenuation Coefficients,” Corporation Scientifique Claisse, Quebec, 1977,46 pp. 1. 2. 3. 4. 5. 6. References Nomsh, K., and Chapell, B. W., in Zussman, J., Editor, “Physical Methods in Determinative Mineralogy,” Academic Press, London, 1977, pp. 201-272. Potts, P. J., Webb, P. C., and Watson, J. S., X-Ray Spectrom., 1984, 13, 2. Potts, P. J., Webb, P. C., and Watson, J. S. ,Analyst, 1985,110, 507. Potts, P. J., Webb, P. C., and Watson, J. S., J. Anal. At. Spectrom., 1986, 1, 467. Statham, P. J., Anal. Chem., 1977, 49,2149. Nornsh, K . , and Hutton, J. T., Geochim. Cosmochim. Acta, 1969,33,431. 9. Abbey, S., Anal. Chem., 1981, 53,529A. 10. Thompson, M., Analyst, 1982, 107, 1169. 11. Ando, A., in Govindaraju, K., Editor, Geostand. Newsl. Spec. Issue, 1984, 8 , Appendix I. 12. Potts, P. J., and Rogers, N. W., Geostand. Newsl., 1986, 10, 121. 13. Bower, N. W., Gladney, E. S., Hagan, R. C., Trujillo, P. E., and Warren, R. G., Geostand. Newsl., 1985,9, 199. Note-Reference 4 is to Part 1 of this series. Paper J6l49 Received July 2nd, 1986 Accepted October 15th, 1986