Laser Ablation Inductively Coupled Plasma Mass Spectrometric Transient Signal Data Acquisition and Analyte Concentration Calculation* Analytical Atomic Spectrometry HENRY P . LONGERICH SIMON E. JACKSON AND DETLEF GUNTHER Department of Earth Sciences and Centre for Earth Resources Research Memorial University of Newfoundland St. John's Newfoundland Canada A1 B 3XB Laser ablation inductively coupled plasma mass spectrometry (LA-ICP-MS) produces complex time-dependent signals. These require significantly different treatment both during data acquisition and reduction from the more steady-state signals produced by solution sample introduction. This paper discusses in detail data acquisition and reduction considerations in LA-ICP-MS analysis. Optimum data acquisition parameters are suggested.Equations are derived for the calculation of sample concentrations and LOD when time-resolved data acquisition is employed sensitivity calibration is obtained from reference materials with known analyte concentrations and naturally occurring internal standards are used to correct for the multiplicative correction factors of drift matrix effects and the amount of material ablated and transported to the ICP. Keywords Laser ablation inductively coupled plasma mass spectrometry; transient signal data acquisition; transient signal analyte concentration calculations; limit of detection Laser ablation inductively coupled plasma mass spectrometry (LA-ICP-MS) is an extremely powerful analytical technique for spatially resolved in situ analysis of solids for a wide variety of trace elements.The often complex rapidly changing transi- ent signals produced can provide not only quantitative concen- tration data but other information about the sample (i.e. the presence of inclusions and chemical zoning within an ablation volume) and can permit meaningful analyses of contained inclusions both solid' and liquid. A wide variety of different data acquisition conditions have been reported for LA analy- sis.2-6 The conditions cited in a number of papers together with the default settings which are used in the software of some ICP-MS instrumentation suggest that data acquisition parameters for transient signal data acquisition have not been considered in detail by some users and manufacturers. The purpose of the present paper is to discuss data acqui- sition and reduction considerations in LA-ICP-MS analysis and to provide guidance on optimum data acquisition param- eters and data reduction algorithms.Many of the consider- ations are applicable to other transient sample data acquisition protocols. DATA ACQUISITION To achieve the maximum benefit from LA-ICP-MS especially when heterogeneous or multi-phase samples are being analysed * Presented at the 1996 Winter Conference on Plasma Spectrochemistry Fort Lauderdale FL USA January 8-1 3 1996. it is very important to acquire and visually examine the signals as a function of time. Without real-time data acquisition and display important information is lost and is unavailable to the analyst (Fig. 1). Further the ability to observe the signals while the ablation process is taking place is important as it allows the operator to make changes to the laser operating parameters (e.g.laser power and focus) during an analysis. To obtain background intensity a 'gas background' signal (i.e. the signal when the sample is not being ablated) can be obtained during the initial part of the acquisition before 30 60 Time/s Fig. 1 Intensity uersus time for LA of (a) a synthetic garnet and glass sample and (b) a fluid inclusion in quartz. The intensity is on a linear scale and has been normalized to the range. In (a) the A1 and Y are contained in the garnet. During the ablation the underlying glass was penetrated as signified by the quickly rising signal of Ba. In processing the signal from the garnet integration of the analytes would end when the Ba signal increased sharply.In (b) opening of a fluid inclusion is indicated by rising intensities of Bi and T1 (dissolved in the fluid) following ablation through the host quartz (shown by the Si signal) Journal of Analytical Atomic Spectrometry September 1996 Vol. 11 (899-904) 899ablation is started. Selected signal intervals are then integrated and net count rates are determined for each element from background count rates and gross analyte ablation count rates measured subsequently in time. While a very few specialized ICP-MS instruments are now available which use simultaneous detection most instruments use quadrupole mass selection and are operated in a fast peak hopping mode in which intensities for various m/z values are detected sequentially.Laser ablation signals being noisy even when compared with other transient signals require a data acquisition sequence that very rapidly jumps between m/z values in order to approach simultaneous detection. Instrument manufacturers use different terminology for the various data acquisition parameters that define the sequence which has no doubt added confusion among operators attempting to keep the syntactic definitions in mind. In this laboratory there are both SCIEX and Fisons/VG quadrupole-based ICP-MS instrumentations but the terminology of the SCIEX software will be used in the present paper where applicable. ‘Number of elements’ ‘quadrupole settling time’ ‘dwell time’ ‘sweeps per reading’ ‘readings per replicate’ ‘number of replicates’ and ‘points per spectral peak’ are all parameters (not independent) that define a data acquisition procedure. A quadrupole mass spectrometer is a sequential device in which a sweep is defined as one cycle of the mass spectrometer in which data are acquired on each of the selected m/z values.Readings and replicates are available in the SCIEX but not the VG software. ‘Readings’ are the mean of a number of sweeps. The mean of a set of ‘readings’ is a ‘replicate’. Quadrupole Settling Time Quadrupoles are very fast peak hopping mass selectors. However they require a finite time to settle which imposes a limitation on scanning speed. Quadrupole settling time is the time for the detector signal to stabilize (to some arbitrary level) following a discontinuous change (peak hop) in the selected m/z value.The quadrupole settling time can be software controlled (and might or might not be adjustable by the operator) or can be hardware controlled. Further the time could or could not be different depending upon the magnitude of the m/z jump. The largest required settling time is expected when the change of m/z is the largest. If intensity data are acquired sequentially in order of increasing m/z the ‘fly back’ a term borrowed from television technology when the large jump is made from the highest to the lowest selected m/z (e.g. U to Li) requires the greatest settling time. The effect of choosing a settling time which is too short will be exacerbated when the jump is from or over a very intense peak to a small spectral peak and will result in a detected signal which is biased high.Conversely when the two peaks are of the same magnitude and close together the error will be much smaller. When short dwell times are used the settling error will be larger than when using longer dwell times. The settling time for quadrupole mass spectrometers is between 0.1 to 10 ms. If the time is set at too low a value it could be necessary to insert an additional ‘dummy’ mass in the sequence from which data are acquired but are not used. It was found necessary to do this in several data acquisition protocols. However where selectable a quadrupole settling time of 1.5 ms was found to be adequate for most but not all applications. It is the short settling time of quadrupole mass analysers that makes them useful for transient signal data acquisition.Instruments that use magnetic mass analysers are by compari- son slower to settle following an m/z jump. The time for magnetic sector instruments to settle is variable and depends upon the magnitude of the m/z jump and the m/z values involved since magnetic sectors do not have linear relationship of mass to magnet current. Although recent developments have produced significant improvements in switching speed in mag- netic instruments they are at the present time less suitable for transient sampling applications than quadrupole-based instruments. Dwell Time Knowledge of the quadrupole settling time is required in order to make an appropriate selection of the dwell time the integrating time on each selected m/z. Often the software allows dwell time to be independently selectable for different m/z values but for simplicity only analyses in which a constant dwell time is used for all m/z values will be described.In some data acquisition software the dwell time might not be treated as an independent variable but instead could be a dependent variable which is set indirectly by adjusting the time per slice (sweep) until the desired dwell time results. Selecting this parameter requires a compromise between a long dwell time which maximizes counting efficiency (the time spent acquiring data relative to system overhead) and an infinitely short dwell time that allows the system to approach in a mathematical sense the ideal of simultaneous data acquisition. A reasonable compromise is to select a dwell time that is approximately six times the quadrupole settling time (e.g.84 ms dwell time for a 1.5 ms quadrupole settling time). Then approximately 15% of the total time is lost to overhead while approximately 85% of the total time is used for data acqui- sition. Dwell times longer than 10ms result in small but insignificant increases in counting efficiency and fewer mass sweeps in the total acquisition period resulting in reduced ability to characterize fast changing transient signals precisely. Owing to the presence of noise from the mains frequencies (60 Hz in North America and 50 Hz in many other countries) dwell times which are an integer multiple of the mains fre- quency are recommended since noise-power spectral measure- ments made on ICPs often show noise that is related to power line frequencies.With 60 Hz power 60 Hz harmonics will be present along with 120Hz the frequency of rectified 60Hz power. When three-phase power sources are used (common with older vacuum tube ICP power supplies) 180 Hz will also be present. Since half a cycle is 83ms one cycle (1 60s) is 16$ms dwell times of 84 16 or 50ms are recommended. In countries where the mains are 50 Hz (one cycle is 20 ms) slightly longer dwell times of 10 20 or 40 are recommended. In most of the present work 50ms were used for solution nebulization data acquisition and ca. 8.3 ms for laser sampling data acquisition. A typical silicate mineral element analysis element menu consists of 25 elements two major element internal standards (e.g. Ca and Ti) fifth-row elements (Sr Y Zr and Nb) sixth- row elements including all the lanthanides (Ba La Ce Pr Nd Sm Eu Gd Tb Dy Ho Er Tm Yb Lu Hf and Ta) and the actinide elements (Th and U).More specialized analyses can determine fewer elements but a set of 25 elements is very typical. Using the suggested compromise dwell time of 8; ms ( tdwell) with a quadrupole settling time of 1.5 ms gives 85% counting efficiency i.e. 85% of the time is used in acquiring data and 15% of the time consumed in overhead. Measuring one point per spectral peak (see below) the time per sweeps (isweep) is given by tsweep = lZelernents( tdwell -k tquadsettling) where tqudsettling is the quadrupole settling time and nelernents is the number of elements. With a sequence of 25 elements a dwell time of 84 ms and 1.5 ms quadrupole settling time tsweep = 246 ms.A typical analysis consisting of approximately 1 min of background measurement followed by approximately a 1 min measurement of sample ablation requires approxi- 900 Journal of Analytical Atomic Spectrometry September 1996 Vol. 11mately 480 sweeps and gives an integrated acquisition time for each element of 2 s. Sweeps per Reading Some data acquisition software allow the averaging of a number of sweeps into a ‘reading’. If the instrument data acquisition software does not allow this it can be accomplished off-line by averaging (typically 3-10) sweeps. Averaging of the signals is helpful to reduce the noise and allow the user to distinguish small analyte signals from the background (note that one detected ion in 8.3 ms equates to a count rate of 120 counts s-l which is large compared with instrument back- grounds that are often less than 10 counts s-’ for quadrupole instruments).Combining sweeps also reduces the size of the resulting data files which can be very large (e.g. 25 elements x 480 sweeps per analysis). However excessive aver- aging (the limit being when all of the data are integrated into a single replicate) can result in the loss of important information contained in fleeting signals or spikes related to real and interesting small inclusions of different phases. The choice of the number of sweeps per reading is thus a compromise between noise reduction (large number of sweeps per replicate) and loss of information from real fine structure in the signals.Experience has shown that integration should not reduce the replicate time to less than approximately 1 s (three sweeps per reading). The number of readings per repli- cate is always set to one and the number of replicates is selected by the operator for the appropriate total acquisition time. Points per Spectral Peak Most ICP-MS data acquisition software allow the user to select the minimum step size along the m/z axis of a spectrum. This step size could be for example 0.1 u (10 points per m/z value) or 0.05 u (20 points per m/z value). Users often report the use of three or more points per peak where a point at the peak centre is acquired along with one or more on the high and one or more on the low m/z side. Some users select a large number of points per peak (e.g.seven or nine) the equivalent of a total integration. Consideration of the consequences of the selection of the number of points per peak clearly indicates that one is the optimum number of points per peak because one point per peak results in (1) the maximum signal (2) the minimum change of intensity with any change in calibration of m/z (3) the maximum counting efficiency (fraction of total time used for data acquisition) when the total sweep time is fixed and (4) the best abundance sensitivity. (1) Signal. That one point per peak results in the maximum signal is intuitively obvious since using more than one point per peak results in the averaging of a measurement at the peak centre where intensity is the largest with measurements made off the peak centre where intensities are smaller.(2) Mass calibration. That measurements made using one point per peak are less sensitive to a change or drift in the calibration of m/z is not intuitively obvious and the expectation that using three or more points per peak will minimize the effects of drift in m/z calibration is presumably the reason that users often select more than one point per peak. The reader can confirm that it is better to use only one point per peak by acquiring signal intensities over a spectral peak and calculating a moving average of a fixed number of points across the peak. As the set of data points selected is moved from the peak centre the mean changes by a larger amount as the number of points per peak is increased. (3) Counting eficiency.When more than one point per peak is acquired several quadrupole settling times per peak are used. Assuming a constant quadrupole settling time (1.5 ms) is used for all m/z changes and a dwell time of 85ms dwell one point per peak and a 25 element analysis a counting effici- ency of 85% is obtained. If three points per peak are chosen and the same time per sweep is maintained [246ms=25 x (8$ + lS)] the dwell time would have to be reduced to 1.78 ms and the efficiency would be lowered to 54%; this results in a reduced total integration time per element of from 2.0 to 1.3 s. (4) Abundance sensitivity. In inorganic quadrupole MS abundance sensitivity is the most important measure of mass spectrometer resolution. When adjusting m/z calibration and resolution the operator will probably use as a measure of resolution the peak width at some fraction of the maximum intensity (e.g.1 m/z unit at 10% of the maximum peak intensity). However abundance sensitivity which is a measure of the interference from adjacent peaks is a more important measure of resolution that measures the interference owing to tailing of the signal of an adjacent ion on the m/z peak being determined. The abundance sensitivity is the ratio of the intensity at rn/z m t 1 or m- 1 relative to the intensity at m/z m when there is a large ion signal at m/z m and there are no ions of rn/z rn + 1 or m - 1 present. It is intuitively obvious that abundance sensitivity will be degraded if off-peak points which lie closer to adjacent peaks are included in the integration.Data Acquisition and Sampling Protocol Once the data acquisition parameters have been selected a sampling sequence (i.e. samples calibration materials and reference standards) must be established. While most data acquisition parameters will be similar regardless of application the sample sequence will vary significantly depending upon application. Summarized below is the data acquisition and sampling protocol used in this laboratory which specializes in high spatial resolution analysis of geological minerals in petrographic sections. Typical data acquisition parameters These are as follows dwell time = 85 ms; quadrupole settling time = 1.5 ms; number of m/z sweeps for total analysis = 480 where 240 are used for background and 240 are used for ablation; sweeps per reading = 3; number of elements = 25 (typical but can vary from 2 to 50); points per peak = 1; and total analysis time = ca.120 s. Data acquisition (a) Acquire approximately 1 min of data with the laser beam blocked to obtain the background. Note that this is an instrumental background as opposed to a reagent blank. More time can be used but the uncertainty in the background does not improve greatly with increased time since the uncertainty is inversely proportional to the square root of the integration time. When very low instrument backgrounds are encountered it is possible that zero ions can be detected giving a count rate of zero with accompanying analytically unrealistic esti- mated detection limit of zero. In this situation one can be added to the total number of background ions to produce a higher non-zero estimate of the background.It is rec- ommended that the laser actually be on and firing with the beam blocked during background acquisition time. This is because in the past electrical pick up by the ion detector of electrical noise from the laser firing circuits has been observed. This also helps to stabilize the laser output. (b) Ablation is commenced (the beam path unblocked) and data are acquired for up to about 1 min using a pulse energy and repetition rate predetermined to produce the required Journal of Analytical Atomic Spectrometry September 1996 Vol. 11 901ablation intensity (i.e. pit size signal). The time can be reduced when boring very shallow pits or when the ablation penetrates through the sample into the support as indicated either visually or by the signal for one or more diagnostic elements in the sample the epoxy or the underlying glass support.Petrographic selections are now prepared in this laboratory using epoxy spiked with Bi an element which occurs at very low concentrations in most samples so that the concentration of Bi is 200 pgg-' in the solid epoxy. This is achieved by premixing bismuth 2-ethylhexanoate (Alfa Aesar Johnson Matthey Ward Hill MA USA) with epoxy resin (Buehler epoxide resin 20-8 130-032 Tech-Met Ontario Canada). When mounting the rock section to a glass slide the resin is mixed as usual four parts resin with one part hardener (Buehler epoxide hardener 20-8 132-032). Acquiring data Data are acquired in runs of up to 20 analyses.Each run starts and ends with two data acquisitions on a calibration material. The 20 analysis limit ensures that calibration is performed on no more than about an hourly basis in order to monitor and correct for the drift of the inter-element sensitivities (ie. analyte-internal standard) with time. Reference materials are also analysed at least twice during a run to monitor accuracy. Selection and Integration of Background and Analyte Data In the analysis of the data it is first necessary to select the data intervals which are to be integrated. Data Selection Choose the time intervals (replicates) over which to integrate (the mean is more convenient to calculate noting that an integration is the mean count rate multiplied by time) the background and the ablation count rates (Fig.1). The first background interval is not used and 2-3 replicates immediately before the initiation of the LA are not used. The first few (typically 1-10 depending upon the application) ablation repli- cates are generally rejected in order to avoid the effects of laser removal of surface contamination and to allow the signal to stabilize. Data integration is terminated at the point when the laser starts to sample the support an underlying mineral phase or an inclusion or other anomalous chemical heterogeneity within the sample or the signal falls significantly (< FZ 50%) either owing to termination of ablation or laser defocusing. While rejection of this tail results in loss of some signal it gives rise to higher signal-to-background noise for the inte- grated interval and consequently improved detection limits.7 Calculate mean background intensity Defining variables nb = the number of sweeps in the back- ground; and tdwell = the dwell time.For each element calculate the mean background intensity (counts s-'). Calculate mean gross analyte ablation intensity Defining variables n = the number of sweeps of gross analyte signal. For each element calculate the mean gross intensity (counts s-l). Apply background correction Subtract from the gross mean analyte intensity (counts s-') the mean background intensity (counts s-l). Since the inte- gration time of the background is not usually the same as the ablation integration time the mean count rates are used. Concentration Calculation In LA analysis calibration can be carried out using external calibration when samples are carefully matrix matched and the time and power of the ablation are reproduced precisely.However in the same manner that internal standards are used routinely to improve precision and accuracy in solution nebul- ization ICP-MS the use of 'naturally occurring' internal stan- dards in LA produces a more robust calibration. The use of internal standards corrects for the sources of bias which are multiplicative i.e. sensitivity drift matrix effects and differences in the volume (mass) of the sample ablated relative to the calibration material (ablation sensitivity correction). Biases that are additive (e.g. polyatomic ion interferences) are not corrected by the use of internal standards.Since the addition of internal standards is not easily and routinely carried out with solid samples 'naturally occurring' elements are used as internal standards. These are elements that are found in both the samples and calibration material and for which the con- centrations are known in both materials. The concentration of the internal standard can be obtained from an analysis using an alternative method or from the known elemental stoichiometry when crystalline materials are analysed. The concentration of the analyte element in the sample (CANSAM) is given by the count rate for the analyte (RAN,,,) in the sample divided by the normalized sensitivity (S) as follows The normalized sensitivity (S) is the sensitivity determined on a calibration standard (CAL) corrected for the volume (mass) of sample ablated.When using naturally occurring internal standards the sensitivity (counts s-' per unit of concentration) normalized to the mass of the sample (SAM) ablated in the determinations is where RANCAL is the count rate of the analyte in the calibration material; CANCAL is the concentration of the analyte in the calibration material; RISsAM is the count rate of the internal standard in the sample; RISCAL is the count rate of the internal standard in the calibration material; CIS, is the concentration of the internal standard in the sample; and CIscALI is the concentration of the internal standard in the calibration material. The sensitivity (S) is prone to amass dependent drift which can change with time. This is a particular problem in the analysis of geological samples where a light major element is generally used as an internal standard which could suffer from significant drift in sensitivity relative to much heavier trace element analytes.In the absence of more than one internal standard it is assumed that this relative drift occurs linearly with time. Consequently a correction to the sensitivity ratios of analyte and internal standard is applied using a linear interpolation (with time) between the calibration samples ana- lysed at the beginning and the end of the run. Limit of Detection The amount of material ablated in LA sampling is often significantly different for each analysis. Consequently LOD are different for each analysis and must be calculated for each individual acquisition.The first step in the calculation is to find the standard deviation of an individual sample count rates in the selected background interval using the well known equation for the calculation of the standard deviation of an individual determi- nation in a population (in commercial spread-sheet software 902 Journal of Analytical Atomic Spectrometry September 1996 Vol. 1 1this function is often referred to as the sample standard deviation as opposed to the population SD) Uindividual = Jz (x - n b - 1 where Xi are the individual measurements and nb is the number of determinations of the background accumulated. This standard deviation is a measure of the distribution of the individual determinations in the set of nb samples of the count rate.However what is required is not a measure of the distribution of the individual determinations of the back- ground but the distribution of the mean of nb background measurements. That is if the data acquisition was repeated a large number of times how would the mean background count rate be distributed or what would be the standard deviation of the mean. The standard deviation of the mean is given by the well known but less well understood equation Oindividual 6mean background = - In the calculation of the LOD it is necessary to estimate the variation of an ablation signal that would be found if the sample contained no analyte. It is assumed that the standard deviation of an individual determination from the signal derived from ablation of a sample with zero concentration is the same as it is for a gas background interval.Thus the standard deviation of an individual determination in the ablation region is also assumed to be dindividual. The standard deviation of the mean ablation signal is then Oindividual O'mean ablation = - A3 where na is the number of slices in the ablation time interval that are integrated. The net count rate for an analyte in any analysis is the difference between the gross count rate for the selected ablation interval and the background count rate The rules of propagation of error through the subtraction give gRnet2 = ORablatian' + ORbaskgroun where the o values are of the mean since it is the means that are subtracted and it is the standard deviations of the means which are of interest. Thus g.. . 2 individual Oindividua12 +- - Oindividua? (k + ;) na or 1 OR, = Oindividual 4; + If na=nb=n as is often the situation then the following simplification results Oindividual Users often omit the square root of two in LOD calculations which results in a lower estimate of the LOD. Consideration of the propagation of error through the subtraction described above gives a more realistic estimate of the LOD. Defining the LOD as three times the standard deviation of a sample that contains zero analyte then This converts the variation in the net intensity from count rates into appropriate concentration units. The sensitivity (S) is the sensitivity normalized to the amount of sample ablated and includes all multiplicative factors that are corrected by the use of the naturally occurring internal standard.Note that the standard deviation that is calculated is an estimate of the variation in a population. There is an uncer- tainty in the calculated standard deviation which follows a chi-squared distribution having the notable characteristics that negative values are not possible and that the distribution tails asymmetrically in the positive direction. When the standard deviations are compared the uncertainty in the standard deviations must be considered; consequently when one or even more experiments demonstrate a lower standard devi- ation than does another experiment it might not be possible to conclude that the difference is statistically significant. Only when a large number of estimates of the standard deviation are obtained do the estimates of the standard deviation approach the true standard deviation.CONCLUSIONS While most of the discussions in the present paper are amenable to experimental demonstration many of the conclusions can be more easily demonstrated using mathematical modelling and statistical considerations. For example experiments to demonstrate the effect of calibration drift in m/z are difficult to carry out since it is not easy to operate an instrument with a known drift. Furthermore experiments can be carried out using conditions that provide accurate and precise analysis over a short period of time but do not demonstrate what happens under conditions of severe instrumental drift or when unusual samples are analysed. From practical and theoretical considerations together with simple mathematical modelling the following acquisition par- ameters are recommended for data acquisition using LA sampling with quadrupole-based ICP-MS detection ( 1) ana- lyses should be performed using time-resolved data acquisition software; (2) a pre-ablation background of similar duration to the ablation should be acquired; (3) for instruments with a quadrupole settling time c a .1.5 ms a dwell time is rec- ommended that is about six times the quadrupole settling time and is a multiple of the mains frequency i.e. 8$ ms for 60 Hz mains or 10 ms for 50 Hz mains; (4) one point per peak should always be used; ( 5 ) pre-integration of small groups of sweep^^-^ provides some smoothing and reduces data file size while allowing fine detail to be recognized; and (6) for all real samples that are heterogeneous to some extent selection of signal intervals for integration should be performed only after careful inspection of time-resolved signals.To calculate sample concentrations and LOD the necessary equations are summarized below. The sensitivity ( S ) is calcu- lated when using naturally occurring internal standards from the signals obtained from a calibration material of accurately known concentrations. Sensitivity is normalized to the mass or volume of sample ablated in a determination. Correction for mass dependent drift should be applied using measurements on the calibration material made before and after the sample data are acquired. The concentration of an analyte in a sample is then given by The LODs in LA-ICP-MS which must be calculated for each Journal of Analytical Atomic S p e c t r o m e t r y September 1996 VoZ. 11 903analyte in each analysis are given by Ingo Horn acquired the garnet sample data shown as an example of time-resolved data. 1 S n 3cindiviciual Ji +- Detection limit = This work was carried out as part of the development of the ICP-MS facility at Memorial University of Newfoundland. Two ICP-MS instruments and the LA system were purchased with Natural Sciences and Engineering Research Council of Canada (NSERC) equipment grants and the system is operated with the assistance of an NSERC infrastructure grant and numerous NSERC research grants including one to H.P.L. A grant from Fisons Instruments with matching funding from NSERC under the Industrial Oriented Research programme is also acknowledged. The generous support of Memorial University to the operation of the facility is also acknowledged. REFERENCES Taylor R. P. Jackson S. E. Webster J. L. and Jones P. Geochim. Cosmochim. Acta submitted. Pearce N. J. G. Perkins W. T. and Fuge R. J. Anal. At. Spectrom. 1992 7 595. Garbe-Schoenberg C.-D. and McMurtry G. M. Fresenius’ J. Anal. Chem. 1994 350 264. Jarvis K. E. and Williams J. G. Chem. Geol. 1993 106 251. Durrant S. F. Fresenius’ J. Anal. Chem. 1994 349 768. Jackson S. E. Longerich H. P. Dunning G. R. and Fryer B. J. Can. Mineral. 1992 30 1049. Denoyer E. At. Spectrosc. 1992 13 93. Paper 51075071; Received November 16 199.5 Accepted February 27 1996 904 Journal of Analytical Atomic Spectrometry September 1996 Vol. 11