American Mineralogist, Volume 77, pages453-457, 1992

LETTER

Analysis of oxygen with the microprobe: Applications to hydrated and minerals

W. P. N.qsH Department of Geology and Geophysics,University of Utah, Salr Lake City, Utah 84112-1183,U.S.A.

Ansrru.cr The development of layered synthetic dispersion elementsmakes it possible to analyze routinely for O with the . The precision for O in silicate glassis 0.60lo of the amount present,comparable to that for Si. Optimum results occur when standards similar in composition to unknowns are used and care is taken in standard and sample preparation. The procedure is particularly advantageousin the analysis of hydrous glass and minerals in which conventional microprobe analysesyield totals of less than 1000/o becauseO attached to H remains unmeasured.The direct determination of O provides a measnreof the quality of the analysis and yields an indirect estimate of the HrO content of the sample by comparison of measuredwith stoichiometric O concentrations.

INrnonucrroN Onsracr,rs ro rHE ANALysrs oF LrcHT ELEMENTS The analysis by electron microprobe of light elements The problems encounteredin the analysis of light ele- (Z < 9) such as O presentsnumerous problems resulting ments are of two varieties: those resulting from the phys- from the low cross sectionsfor ionization, high absorp- ics of X-ray generation and detection, and those associ- tion ofthe low-energy X-rays, and spectral interferences ated with preparation of the sample and standard. Lrght by higher order X-ray lines of heavier elements. O has elementshave very low cross sectionsfor ionization that rarely been analyzed directly despite its abundance, and result in a low yield of X-rays, and theselow-energy X-rays its concentration is traditionally calculated by stoichi- are readily absorbedin the sample. Mass absorption co- ometry with measuredcations. This method neglectsO efrcients are often large, and if there are significant dif- associatedwith cations not measured,notably H in HrO- ferences in composition between standards and un- bearing materials, and other light elements such as Li, knowns, the mass absorption coefficientsmust be known Be, B, C, and N. BecauseO that may be associatedwith with high accuracy. Moreover, although O X-rays may theseelements is not determined, analytical totals are less be generatedat considerabledepth (micrometers) in the than l00o/0.Uncertainties in valence may also result in sample, the emission volume is much shallower because inaccurate totals when O is calculated by stoichiometry. of high absorption. Thus, the O signal may be strongly Under these circumstances,it is difrcult to assessthe influenced by surficial and near-surfaceconditions. quality of an analysis, although the calculation of the In the past, the problem of low X-ray yield was exac- structural formula for a mineral may confirm whether or erbated by inefficient diffraction of characteristic X-rays not the analysisis acceptable.In the caseofnatural glass, to the detector. The situation has been significantly im- no such recourseexists, and low totals are generally ac- proved by the development of layered synthetic disper- cepted and attributed to the presenceof HrO or the loss sion elements.These X-ray reflectorsconsist of alternat- of Na caused by interaction of the electron beam with ing layers of heavy and light elementsthat act as diffracting the sample. crystals,with d values determined by the thicknessof the The recentdevelopment of layeredsynthetic dispersion light element layer. The advantageof these X-ray reflec- elements(LSDE), often referred to as multilayer diffract- tors in the analysis of light elementshas been described ing crystals, has made O and other light elements mea- by severalauthors (Nicolosi et al., 1986; Love and Scott, surable,even in minor concentrations.Although it is now 1987;Armstrong, 1988;Kawabe et al., 1988;Bastin and possible to analyze for O routinely in minerals, the de- Heijligers,l99l; McGee et al., l99l). Not only do these termination of O is particularly advantageousin the anal- reflectors provide greater peak intensities, they also sup- ysis of hydrous minerals and glassesbecause, first, the presshigher order interfering X-ray lines, such as AlKa(IIf analytical total provides a direct measure of the quality on the OKa peak and AlKa(IV) on the N peak, which are of the analysis, and, second, HrO contents can be esti- presentif a lead stearatediffracting crystal is used.Count- mated from the differencein measuredand stoichiomet- ing rates with the LSDE crystalsare sufficiently high that rically calculatedO contents. the statistical precision for O is similar to heavier ele- ooo3404x/92l0304-o453$02.00 453 454 NASH: ANALYSIS OF O IN MINERALS AND GLASS ments conventionally analyzed with the electron micro- used for the remaining elements.Kaersutite was used as probe. Minor shifts in position and shapeof the OKa line the O. Si. and Al standard for the analysisof hornblende' occur with LSDE, as describedby Armstrong (1988)and and muscovite was used as the O standard for illite. Sam- Bastinand Heijligers(1989). In the applicaiiondescribed ples and unknowns were coatedwith C simultaneouslyto here, the problems of mass absorption and shifts in peak assurean equivalent thicknessofthe C coat. During anal- position and shape are minimized by the traditional ysis, repetitive measurementswere made on synthetic method of using standardssimilar in composition to the AlrO. to determine if O were present in amounts greater unknowns. than stoichiometric becauseof adsorbedH'O on the sam- In order to obtain accurate and reproducible micro- ples; no excessO was detected. probe results, care must also be taken with sample prep- H concentrations in melt inclusions were determined aration. The standardsand unknowns must be well pol- by secondaryion mass spectroscopywith a CamecaIMS ished and clean, have the same thickness of C coating, 4f microprobe at the University of Edinburgh (cf. and have surfacesperpendicular to the electron beam. Hervig et al., 1989,for discussionof SIMS measurements These proceduresare particularly critical for O because of HrO in melt inclusions). HrO concentrations of glass the signal is derived from very near the surface, and C separateswere determined manometrically on the H ex- has a high massabsorption coefficientfor OKa radiation. traction line at the University of Utah. Goldsteinet al. (1991)describe the effectsofsurficial ox- idation of metallic samples.Although this is not a prob- Rnsur-rs lem for silicate glassesand minerals, it could be a serious The O contents of 18 mineral standards,determined problem in the analysis of low concentrations of O in using synthetic AlrO, for O calibration, are illustrated in metals, sulfides, etc. In addition, adsorption of H.O by Figure l. Nominal O contents were calculated assuming the samplescould be problematic and should be evalu- stoichiometry of the mineral standards.The correlation ated. Finally, errors of up to l0loper degreeof inclination between measured and nominal O contents is generally are produced by samplesthat are inclined to the primary good. However, hematite and magnetite have measured electron beam. O contentsthat are low by l0loabsolute; these are the two data points at lowest O contents in Figure 1. This dis- AN.q.LvttcAl- coNDITIoNS crepancyis attributed to reduced peak intensities due to Analyseswere performed on a CamecaSX-50 electron peak broadening associatedwith differential energeticsof microprobe equipped with four wavelength-dispersive the bonding of Fe oxides compared with that of alumina spectrometers. The X-ray reflector, manufactured by or silicates. This emphasizes the need to correct for Ovonic Synthetic Materials Company (OVOMX OV- changesin peak shape,if present (cf. Bastin and Heijli- 060a) is a 60-A WSi multilayer deposited on a single- gers, 1989), or, alternatively, to utilize standardssimilar crystal substrateof Si (100). O and F were measuredon to unknowns in composition. a spectrometer with this crystal. The remaining spec- Conventional analysesofhydrated silicic glassesyield trometers were equipped with TAP, PET, and LiF ana- low analytical totals, which are attributed to HrO and lyzing crystals.Gas-flow proportional countersusing P-10 alkali mobility under the electron beam. Table I presents gas (900/oAr, l0o/omethane) were used on all spectrom- data for four silicic .The first is the averageof 130 eters.Analyses were made at l5 keV acceleratingvoltage, analysesperformed over a period of 2 yt on MM3, the a beam current of 25 nA, and a beam diameter varying high-silica obsidian standard.The standard deviation for from 5 to 25 pm. O was measured at the beginning of O represents a reproducibility of 0.60/oof the amount eachanalysis; counting time was 20 s. A similar counting present, slightly better in this case than the error on Si time was used for standardization so that any effects of (0.80/0,relative). The conventional analysis in terms of C contamination in the samplechamber would be similar oxidesis shown for comparison. for the standard and the unknowns. Background inten- The second data set is for glass shards from the Lava sitieswere measuredon both sidesof the analytical peaks. Creek B tufl erupted from the region of Yellowstone Na- Concentrationswere calculatedfrom relative peak inten- tional Park 0.6 Ma. A common practice is to assumethat sities using the Qbz) algorithm of Pouchou and Pichoir the difference between the analytical total and 1000/ois ( 199l), utilizing massabsorption coefficientsfrom Henke causedby the presenceof unanalyzed HrO and to nor- et al. (1982). On the natural glass standard containing malize the data to 1000/0,producing an analysis of the 49.8 wto/oO, these conditions yielded a peak to back- "anhydrous" glass.The procedure has merit only if the ground ratio of 41, comparable to that for Al (50). analysis is of good quality, Na has not been lost during Synthetic AlrO, was used as a standard for reconnais- analysis,and there has not been an exchangeofHrO for sancemeasurements of O in a variety of mineral stan- other elementssuch as alkalis during hydration. The lat- dards. A natural obsidian (MM3) was used for analyses ter is best addressedisotopically (Cerling et al., 1985). of silicic glass.The O content of MM3 was calculatedby However, the two former concernsare resolved by anal- stoichiometry, with 4l cations analyzedby wet chemical, ysis for O, whereupon a good analysis results in an ana- X-ray fluorescence,instrumental neutron activation, and lytical total approaching 1000/0.A conventional analysis atomic absorption methods. O, Si, Al, and K were cali- of glass shards from this same tephra unit by Sarna- brated with this standard; natural mineral standards were Wojcicki et al. (1987) is presentedfor comparison.Al- NASH: ANALYSIS OF O IN MINERALS AND GLASS 455

Trale 1. Analysesof hydratednatural glasses

MM.3 LCB HH FC FC(L) Sample n:130 n:11 LCB(S)n : 12 n: 4 n: 17 Si 3s.89(0.32) 3445 33.2 34.64 Ti 0.08(0.02) 0.06 0.00 o.12 AI 6.59(0.07) 6.15 8.25 6.39 't.o2 Fe 0.51(0.04) 0.36 0.83 Mn 0.04(0.02) o.o2 0.08 0.03 Mg 0.04(0.01) 0.02 0.00 0.06 0.38(0.02) 0.36 0.21 0.40 Na 2.59(0.09) 2.59 2.30 1.73 K 4.33(0.05) 4.04 3.24 5.14

30 40 50 cl 0.07(0.01) 0.1s 0.13 0.03 F 0.14(0.03) 0.01 2.23 0.24 o 49.41(0.30) 50.90 49.9 49.50 Fig.t. comparison#Hilffil.LJtln theelectron mi- Total 100.07(0.42) 99.73 100.1 99.09 croprobe(EPMA) vs. norninalO contents(assuming stoichi- sio, 76I 73.7 72.2 71.1 74.1 73.5 ometry)in a varietyof mineralstandards. The analyticalerror Tio, 0.14 0.10 0.10 0.00 o.20 0.22 is lessthan the sizeofthe Al,o3 124 11.61 11.7 15.6 12.1 12.3 dataooint. FeO- 0.65 1.31 1.33 0.46 1.06 1.16 MnO 0.05 0.03 0.03 0.10 0.04 0.01 Mgo 0.06 0.03 o.02 0.00 0.10 0.14 though it is not possible a priori, quality CaO 0.53 0.50 0 51 0.29 0.55 0.62 to assessthe of Naro 3.49 3.49 3.37 3.09 2.33 2.1 the latter, comparison of the two analysesconfirms its K.o 5.21 4.87 4.86 3.91 6.20 6.04 quality. cl 0 07 0.17 0.13 0.03 0.02 F 014 0.01 2.23 o.24 0.2 The third data set is the averageof 12 analyseson a Sum 99.5 95.7 94.3 94.8 96.9 96.4 single silicate melt inclusion, 120 pm in diameter, in a -o-. 0.08 0.04 0.97 0.11 0.08 quartz phenocryst from rhyolitic pumice. The elemental Total 99.5 95.7 94.3 93.8 96.8 96.4 analltical total is 100.l0lo compared to the stoichiometric H,Ot 0.6 4.5 4.5 z-6 Total 100.1 100.2 98.3 99.4 oxide total of 93.80/0. The fourth data set is for melt inclusions, 30-50 pm in A/ofe:Numbers in Darenthesesare standard deviations.MM-3: Mineral Mountainsobsidian. LCB : Lava Creek B ash, PromontoryPoint, Utah. diameter, from quartz phenocrystsin the Fire Clay unit LCB(S): Lava Creek B ash; average of six localities;analysis recast as ofCarboniferous age (Lyons et al., in preparation). Con- originalEPMA data from Sarna-Wojcickiet al. (1987,Table 1). HH : melt : ventional analysesyielded, as expected,low analytical to- inclusionin quartz phenocryst, HoneycombHills rhyolite, Utah. FC av- erage of melt inclusionsin Fire Clay tonstein (Lyons et al., in preparation), tals. However, the question remained as to whether the Universityof Utah. FC(L): averageof melt inclusionsin FireClay tonstein totals were low due to HrO content or Na loss during (Lyons et al., in preparation),U.S. GeologicalSurvey. - Total Fe as FeO. analysis. Subsequentanalysis for O in our laboratory " -o: F,cr. yielded analytical totals approaching l00o/0,with only t H,O estimated from differencebetween measured and stoichiomet- slightly elevated Na concentrations, supporting the as- ric O. sumption that th€ major deficit in the oxide total is HrO.

Esrrl.lrroN oF H2O coNTENTS oF NATURALGLASS In hydrous materials, the concentration of O measured analysisin which O is measured.For materialswith mod- by electron microprobe exceeds the stoichiometrically erate to high Fe contents, the estimate requires a knowl- calculatedcontent becauseH, as well as He, Li, Be, B, C, edgeof the valencestate of Fe. The method is well suited and N, have not been measured.Assuming that the latter for rhyolitic glassesbecause oftheir low Fe content (Table elementsare in low concentration and that all other ele- l). ments presentand their valencestates have beenaccount- In Figure 2 estimatesof HrO contents of silicic glasses ed for properly, the difference between measured and determined by electron microprobe are compared to stoichiometric O can be attributed to the presenceof HrO measurementsof HrO content by manometry and sec- or OH. The abundanceof HrO determined by this meth- ondary ion in natural glasses.EPMA od is, ofcourse, an estimate becausethere has not been results agree well with manometry and less well with a direct determination of HrO or the H concentration, SIMS, perhapsbecause of the lower precision of the lat- and the method relies upon the accurate determination ter. For moderate H2O contents in glasses,the method of all other elementspresent. In this respect,the method yields values within about lolo absolute of the amount is similar to that of assumingthat the differencebetween present. HrO contents estimated for silicate glassesare the oxide total and 1000/ois the HrO content. Other than given in Table l. measuring O directly, it has the additional advantageof monitoring the quality of the analysis.The method does Hvnnous MTNERALS not have the precision of (e.g., The analysisof O is advantageousin hydrous minerals Newman et al., 1986) nor the desirability of direct mea- insofar as the analytical total servesas a measureof the surement of H, as in secondaryion mass spectrometry; quality of the analysis. Furthermore, if the valence state however it is simple, and the result emergesfrom any of the cations is known. H"O contents can be estimated. 456 NASH: ANALYSIS OF O IN MINERALS AND GLASS

AcxNowr-SDGMENTS

5 Financial support was provided by NSF grants EAR-86 I 8 10 I and EAR- 8720627.James Webster provided the SIMS analysesof H, and Thomas performed the f Gavigan assistedin EPMA analysesof melt inclusionsand t manometric water determinations. Paul Lyons provided the ancient (!) early draft of the manuscript' -2 melt inclusions. F.H. Brown reviewed an Subsequentreviews by JamesMcGee and an anonymousreviewer further T o improved the paper.

0 RnrnnnNcns crrnn

0r23456 Armstrong, J.T. (1988) Accurate quantitative analysisofoxygen and ni- I..rurcd H2O (fl%) trogen with a WSi multilayer crystal. In D.E. Newbury, Ed', Micro- beam analysis 1988,p. 301-304. San FranciscoPress, San Francisco' Fig. 2. H,O contentof hydratedglass estimated by EPMA Bastin, G.F., and Heijligers, H.J.M. (1989) QuantitatiYe EPMA of oxv- vs. HrO determinedby secondaryion massspectrometry (SIMS) gen. In P.I. Russell, Ed., Microbeam analysis 19E9,p. 207-210- San on melt inclusions(open squares) and manometryon obsidian Francisco Press, San Fiancisco. glassseparates (solid squares). The line is thelinear least sqrures - (1991) Quantitative electron probe rnicroanalysisofultralight el- frt to the data. The error bars on EPMA values are standard ements Ooron-oxygen). In K.F.J. Heinrich and D.E. Newbury, Eds., deviationsfor tenanalyses on eachsample. SIMS error estimates Electron probe quantitation, p. 145-161. Plenum Press,New York. (1985) are0.50/o; manometry uncertainties are less than the sizeof the Cerling, T.E., Brown, F.H., and Bowman, J.R. Low-temperature volcanic glass:Hydration, Na, K, '8O and Ar mobility. datapoints. alteration of Chemical Geology, 52, 281-293. coldstein, J.J., Choi, S.K., van Ioo, F.J.J., Heijligers, H'J.M., Dijkstra, J.M., and Bastin, G.F. (1991) The influenceof surfaceoxygen contam- An example of an analysis of a hornblende is given in ination of bulk EPMA of oxygen in ternary titanium-oxygen-com- p. 57-58. San Table 2. The analytical total by conventional microprobe pounds. In D.G. Howitt, Ed., Microbeam analysis 1991, Press,San Francisco. presentedin Table 2 is the anal- Francisco analysis is 97.5o/o. Also Henke, B.L., Lee, P., Tanaka, T.J., Shimabukuro, R.L., and Fujikawa, ysis of an ammonium-bearing, tobelitic illite. Without B.K. (1982) Low-energy X-ray interaction coefrcients: Photoabsorp- the analysisof O and N, the conventional analytical total tion, scattering, and reflection. Atomic Data and Nuclear Data Tables, is 91.50/0.The elemental analysis totals 99.40loand thus 27, t-144. H.R., and Kyle, P.R. (1989) Pre- provides a measure of confidence in the quality of the Hervig, R.L., Dunbar, N.W., Westrich, eruptive water content of rhyolitic magmas as determined by ion mi analysis,particularly the determination of N content. The croprobe analysisof melt inclusions in phenocrysts'Journal of Yolca- analytical conditions for the measurementof N are given nology and Geothermal Research,36,293-302. in Wilson et al. (1992). The estimatedHrO content of the Kawabe,K., Takagi, S., Saito, M., and Tagata,S. (198E)Layered synthetic probe tobelitic illite is 4.80/0.The determination of O contents microstructure dispersion elements for electron microanalysis of carbon,boron, and beryllium. In D.E. Newbury, Ed., Microbeam anal- in hydrous B-bearing minerals would also be advanta- ysis 1988, p.341-344. San FranciscoPress, San Francisco. geousbecause it would permit an assessmentof the an- Love, G., and Scott,V.D. (198?) Progressin the EPMA of light elements. alytical accuracy,a problem experiencedby McGee et al. Institute ofPhysics ConferenceSeries, 90, 349-352. (1991). McGee, J.J., Slack, J.F., and Herrington, C.R. (1991) Boron analysisby electron microprobe using MoBnC layered synthetic crystals. American Mineralogist, 76, 681-684. Newman, S., Stolper,E.M., and Epstein,S. (1986) Measurementof water TABLE2. Analysesof hornblendeand N-bearingillite in rhyolitic glasses:Calibration of an infrared spectroscopic technique. American Mineralogist, 7 l, 1527-154 I' HornblendeAS-10 llliteFM-g Nicolosi, J.A., Groven, J.P., Merlo, D., and Jenkins, R- (1986) Iayered n: 10 n:8 synthetic microstructures for long wavelength X-ray spectrometry' Op- 25, 964-969. Si 22.5 23.0 tical Engineering, Ti 0.81 0.06 Pouchou,Jean-Louis, and Pichoir, F. (1991) Quantitative analysisofho- AI 3.34 19.1 mogeneousor stratified microvolumes applying the model "PAP." In Fe 10.4 0.37 K.F.J. Heinrich and D.E. Newbury, Eds', Electron probe quantitation, Mn 0.45 p. 3 l-75. Plenum Press,New York. Mg 8.64 0.71 Sarna-Wojcicki,A.M., Morrison, S.D, Meyer, C.E', and Hillhouse, J.W. Ca 8.34 0.04 (1987) Correlation ofupper Cenozoictephra layersbetween sediments Na 0.73 0.12 ofthe westernUnited Statesand easternPacific Oceanand comparison K o.47 3.29 and magnetostratigraphic age data. Geological So- cl 0.08 with biostratigaphic Bulletin, 98,207-223. T 0.00 0.60 ciety of America N 1.55 Wilson, P.N., Parry, W.T, and Nash, W.P. (1992) Tobelitic and ammo- 43.9 50.6 nium illitic vein phyllosilicates from hydrothermally altered black shale, Total 99.8 99.4 southem Oquirrh Mountains, Utah. Clays and Clay Minerals, in press. : FM-9 : tobelitic illite Notei AS-10 hornblendetrom Alta Stock, Utah; Mem:scnrrr REcEIVEDDeceln{srn 30, l99l trom OquirrhMountains, Utah (Wilsonet al., 1992). MANUscRIpTACCEFTED FesnuARv 12, 1992