The future of atomic absorption spectrometry: a continuum source with a charge coupled array detector† Plenary Lecture James M. Harnly US Department of Agriculture, Agriculture Research Service, Beltsville Human Nutrition Research Center, Food Composition Laboratory, Building 161, BARC-East, Beltsville, MD 20705, USA Received 29th September 1998, Accepted 17th November 1998 Continuum source atomic absorption spectrometry (CS-AAS) has made impressive progress in the last 5 years thanks to the availability of high resolution e� chelle spectrometers and solid state array detectors.With these new spectrometers and detectors, the capabilities of CS-AAS exceed those of conventional, line source-AAS (LS-AAS). For CS-AAS, absorbances are more accurate (corrected for stray radiation and non-specific broadband background absorption and integrated with respect to height in the furnace), detection limits average a factor of 2 lower, calibration ranges are a factor of 1000 greater, multi-wavelength data are available for correction of spectral interferences, sensitivity is a powerful quality assurance measure since it is independent of all instrument parameters except atomization temperature and, of course, multi-element detection is possible. The future appears bright for CS-AAS.Whereas, previously, CS-AAS was striving for parity with LS-AAS, it is now reasonable to state that it is CS-AAS which is setting the standard. Substitution of a continuum source for HCLs, without Introduction changing the rest of the instrument, is not a reasonable Continuum source atomic absorption spectrometry (CS-AAS) approach.The instability of the most intense continuum has long appealed to the spectroscopic community because of sources, xenon arc lamps, gives noisy baselines and poor the potential for simultaneous multi-element AAS determi- detection limits. Medium resolution monochromators, that are nations, a shortcoming of conventional line source AAS ideal for isolating HCL emission lines, provide a spectral (LS-AAS). Unfortunately, owing to limitations of the source, bandwidth that is too large for use with a continuum source.CS-AAS has failed for many years to compete with LS-AAS The large spectral bandwidth results in poor sensitivity and with respect to detection limits. Today, with the high radiation specificity, non-linear calibration curves and greater susceptithroughput and spectral resolution of e� chelle spectrometers, bility to broadband background interferences. In addition, the the multi-wavelength detection capability, low read noise and intensity of most continuum sources decreases dramatically high quantum eYciency of charge coupled detectors (CCDs) below 280 nm.Consequently, the use of a continuum source and the high speed data acquisition capabilities of modern for AAS requires the redesign of the whole instrument. computers, CS-AAS has surpassed the analytical capabilities As shown in Table 1, a variety of instrumental designs have of LS-AAS.These advantages will be explained in detail in been explored for CS-AAS. The three main challenges were this review. Although it takes time for the implementation of to obtain sensitivities, detection limits and calibration ranges new concepts, especially when it requires a considerable capital comparable to LS-AAS. Sensitivity was initially enhanced by investment, CS-AAS appears ripe for development.It seems the use of multipass absorption cells.3,4 It soon became obvilikely that if AAS was being developed today for the first ous, however, that the best approach to recovering the lost time, it would be with a continuum source. sensitivity was the use of a high resolution spectrometer AAS, as first described by Walsh1 and Alkemade and (interferometers and e� chelles)7–14 to provide the narrow analyt- Milatz,2 derived its success from the use of hollow cathode ical bandwidth previously supplied by the HCL.Obtaining lamps (HCLs) as the radiation source. These lamps, with their detection limits equal to those for LS-AAS required minimizing stable and narrow emission lines, oVered high analyte speci- the flicker noise of the xenon arc lamp. This was accomplished ficity, excellent detection limits (for the time) and linear using wavelength modulation with phase sensitive deteccalibration curves (2.5–3.0 orders of magnitude of concen- tion.5–7,10–14 Wavelength modulation has been implemented tration).Unfortunately, these lamps are not suitable as multi- using quartz refractor plates,5,6,10,11 oscillating e� talons7 and, element sources. Multi-element HCLs are restricted to compat- ultimately, array detectors.12–14 The linearity of the calibration ible elements and are always less intense than the single- curves is determined primarily by the spectral bandwidth of element lamps. Combinations of single- and multi-element the analytical measurement.Thus, the e� chelles and interfer- HCLs suVer from loss of intensity due to the necessary beam ometers have provided the best linearity,8,10 although the splitters/combiners. Consequently, although AAS is a powerful oscillating e� talon produced an unusual, multi-humped calianalytical tool, it has always remained a single-element bration curve owing to the narrowness of the spectral bandtechnique. pass. It should be noted that all the instruments described in Table 1, with one exception, are single-element designs.The only functional, multi-element instrument was that described †Presented at the Ninth Biennial National Atomic Spectroscopy Symposium (BNASS), Bath, UK, July 8–10, 1998. by Harnly et al.11 in 1979. J. Anal. At. Spectrom., 1999, 14, 137–146 137Table 1 Research on continuum source AAS Year Researchers Instrument Design Reference 1966 Fassel, Mossotti, Grossman Long cell paths 3 and Knisely 1967 McGee and Winefordner Long cell paths 4 1968 Snelleman Wavelength modulation 5 1972 Elser and Winefordner Double modulation (wavelength 6 modulation and optical chopping) 1972 Nitis, Svoboda and Fabry–Perot interferometer with 7 Winefordner oscillating etalon 1973 Veillon and Merchant Fabry–Perot interferometer 8 1974 Keliher and Wholes E� chelle monochromator 9 1976 Zander, O’Haver and E� chelle monochromator and 10 Keliher wavelength modulation 1979 Harnly, O’Haver, Golden E� chelle polychromator, wavelength 11 and Wolf modulation and computerized data acquisition (SIMAAC) 1993 Harnly E� chelle monochromator and linear 12 photodiode array detection 1997 Harnly, Smith, Wichems, E� chelle polychromator and segmented, 13 Ivaldi, Lundberg and Radziuk linear charge coupled array detectors 1998 Harnly, Fields and Schuetz E� chelle monochromator and two- 14 dimensional, thinned, back-illuminated charge coupled array detector Wavelength modulation is the heart of CS-AAS.Intensities data. A frame consists of all the data obtained from an array following an exposure for a predetermined period of time. A on and oV the analytical line are ratioed to correct for broad band shifts in the spectral intensity, whether caused by the frame may consist of data from all the pixels or select groups of pixels (sub-arrays) covering specific wavelength regions of lamp flicker or non-specific background absorption. Wavelength modulation is successful, however, only if it is interest.The frame rate of an array detector can be viewed as equivalent to the modulation frequency of the mechanical performed at a frequency greater than that of the intensity fluctuation. With a xenon arc lamp, a modulation frequency modulation devices. Since each frame consists of intensities measured over the whole wavelength region of interest, high of 50–60 Hz is needed to minimize the source fluctuation noise.15 The development of electrothermal atomization did frame rates are not necessary for accurate correction of intensity fluctuation.High frame rates, however, are still not significantly influence the data acquisition rate for CS-AAS. The 60 Hz data rate dictated by the flicker noise of necessary to characterize accurately the rapid, transient signals which result from furnace atomization. Intensities mp was suitable for furnace atomization.16 Prior to 1993, wavelength modulation was implemented converted to absorbance and integration must take place in the absorbance domain to ensure linearity with respect to mechanically and the resultant signal was detected using a lock-in amplifier.10,11 The mass of the moving mechanical concentration.Frame rates of at least 60 Hz are required for furnace atomization and even higher rates are preferred for device (the torque motor on which the quartz refractor plate was mounted) limited the modulation frequency.Frequencies the newer furnaces with heating rates of greater than 1000 °C s-1. above 60 Hz were not possible without sacrificing the signalto- noise ratio (S/N). Detection with lock-in amplifiers pro- The current state-of-the-art for CS-AAS is embodied in the latest instrument listed in Table 1, developed by Harnly et al.14 vided signals (I0-I ) that were proportional to absorbance at low concentrations. With respect to data processing, any of This design, which will be described in detail, is a singleelement instrument. Its characteristics, detailed in Table 2, the earlier instruments in Table 1 could have been adapted to multi-element detection using parallel analog circuits.illustrate the potential capabilities of CS-AAS using the latest solid state detector technology. The low pixel read noise results Coupling of a computer to CS-AAS in 197911 permitted the use of an exotic modulation waveform and high-speed data in photon shot noise being the dominant noise source. As a consequence, the best S/N can be achieved with the highest acquisition for all elements.SIMAAC acquired 20 intensities per modulation cycle for 56 cycles per second for 16 elements, resolution. The use of a high-resolution e� chelle spectrometer restricts the spectral width of each pixel, allowing high sensi- a total data acquisition frequency of 17.9 kHz. The combination of the unusual waveform and high data rate resulted in intensity measurements at discrete wavelengths.The 20 Table 2 Spectrometer characteristics intensity measurements acquired during a single sweep across E�chellea Dectectorb the absorption profile were used to compute a family of absorbances of varying sensitivity. This approach allowed 750 mm focal length Split frame transfer array detection of 16 elements, improvements in the S/N at low 63° 26¾ blaze angle 80×80 pixels concentrations and extension of the calibration curve linearity 46×96 mm grating 18×18 mm pixels to 4–5 orders of magnitude. 79 grooves mm-1 2.5 mm channel stop In the 1990s, use of solid state array detectors permitted 25 mm entrance slitwidth Thinned 500 mm entrance slit height Back-illuminated simultaneous multi-wavelength detection and eliminated the Spectral bandwidth: UV antireflection coating need for mechanical modulation.12–14 Multi-wavelength detec- 2.1 pm at 200 nm 100 000 e- charge capacity tion provides the equivalent of infinitely fast wavelength 7.0 pm at 580 nm 70 Hz frame rate modulation (on- and oV-line intensities are measured simul- Luminosity at 200 nm: 25 e- read noise taneously) provided that all the pixels have a common exposure 0.013 mm2/nm 50% quantum eYciency at 200 nm time, i.e., the exposure start and stop times are the same.The aRef. 19. bRef. 20. array detectors compute absorbances from a single ‘frame’ of 138 J. Anal. At. Spectrom., 1999, 14, 137–146practice, the data from the camera in Table 2 had 80 columns and 80 rows of pixels.The conversion of the intensities in a single frame to absorbance is relatively simple when a computer is available to do the data processing. In Fig. 1, linear arrays of pixels running parallel to the wavelength axis will be called rows and linear arrays perpendicular to the wavelength axis will be called columns. Simplistically, the intensities at the top and bottom of each column (i.e., intensities between orders) are subtracted from the intensities in the middle of the column to correct for stray light.Intensities at the ends of each row are used to compute absorbances for each pixel in the middle of the row. Absorbances from pixels in individual columns are summed to provide height-integrated absorbances. The heightintegrated absorbances are summed to provide a wavelengthintegrated absorbance. Finally, the height and wavelength inte- Fig. 1 Simulated data showing a short wavelength section of one of grated absorbances from each frame are summed to provide the dispersed orders of the e� chelle spectrometer with a slit height of 500 mm and an absorption profile with an absorbance of 0.5 and a a time integrated absorbance.full width at half-height of 2 pixels. Pixels in black above and below Mathematically, the height, wavelength and time integrated the order are used to correct for stray light. Pixels in black on the absorbance, AInt, can be expressed as order to each side of the absorption profile are used to correct to broadband background absorption interferences. AInt=tf .frame lp . row hp . column log(I0,r/Ii,r) (1) where tf, lp and hp are normalization constants (tf is the time tivity and permitting the acquisition of detailed spectral inforbetween frame reads in seconds, lp is the spectral pixel width mation for the surrounding wavelength region. The instability in picometers and hp is the pixel height in micrometers), I0,r is of the continuum source and non-specific background absorpthe reference intensity for row r obtained from the average of tion are eliminated by simultaneous multiwavelength detection pixel intensities at the ends of the row and Ii,r is intensity of with the array detector.In addition, the high throughput of pixel i in row r. The reference pixels must be predetermined the e� chelle and the high quantum eYciency of the array by the analyst. The S/N is quadratically dependent on the detector provide increased intensity in the far UV region number of reference pixels; hence the more used the better, (below 240 nm).The vertical dimension of the array (when although there are diminishing returns as the number grows coupled with an appropriate optical arrangement) can provide large. The normalized time, wavelength and height integrated absorbance information as a function of height in the furnace absorbance will have units of mm pm s.The mass necessary to and can measure stray light intensities between orders. provide an integrated absorbance of 0.0044 pm s (integrated The instrument described in Table 2 provides analytical with respect to wavelength and time) was previously defined information that is unique to the field of atomic spectrometry. as the intrinsic mass. The mass necessary to provide an A simulation of a typical frame of data is shown in Fig. 1. absorbance of 0.0044 mm pm s (with the addition of height From these data, it is possible to compute absorbances that integration) has not yet been defined.are stray light corrected, background corrected and integrated Fig. 2 and 3 present results obtained for the atomization of with respect to wavelength, height in the furnace and time. In 250 pg of Cu (324.7 nm). The intensities shown in Fig. 2 were addition, the data are suitable for resolving overlapping specobtained by averaging the data for four frames near the time tral interferences, although this area is largely unexplored for of the peak maximum.These data were obtained with an absorption spectrometry. In theory, after normalization, the entrance slit height of 100 mm (compared with the 500 mm computed absorbances are independent of the characteristics height used in Fig. 1 for the simulated data) so the order width of the continuum source, monochromator and solid state is much smaller. Fig. 3 provides the computed absorbances detector.The analytical sensitivity will depend only on the physical characteristics of the furnace (diameter and length) and the accuracy and repeatability of the temperature program. These analytical characteristics now make L’vov’s concept of absolute analysis17,18 a realizable possibility. Computing absorbance The simulated data in Fig. 1 shows a band of elevated intensities through the frame parallel to the wavelength axis. This band is a single order of the e� chelle viewed over a short wavelength interval, typically less than 1 nm.In Fig.1, intensities above and below the band (between orders) are also seen. Stray radiation and oVsets in the electronic circuitry determine the magnitude of the intensities between orders. The height of the order above the between order intensity is determined by the source intensity and the order being viewed and fluctuates with time. The width of the band (on the order axis) will depend on the entrance slit height. The dip in the middle of the band is the analyte absorption and the depth will vary with analyte concentration.Additional dips in the band may be seen if non-analyte absorption, or spectral line Fig. 2 E� chelle order containing absorption profile for 250 pg of Cu interferences, occur. The simulated data in Fig. 1 are con- (324.7 nm) obtained with an entrance slit 25 mm wide and 100 mm high. Large vertical oVset is electronic and not stray light. structed with only 22 columns and 54 rows of pixels.In J. Anal. At. Spectrom., 1999, 14, 137–146 139model and the fact that diVerent atomization temperatures were used, the agreement between the intrinsic mass for LS-AAS and CS-AAS is very reasonable. The lack of dependence of CS-AAS on the source and detection parameters places new emphasis on the performance of the furnace. Absorbances for any CS-AAS instrument can be directly compared and absorbance will have a fixed relationship to concentration.The only variable that can cause a change in sensitivity is the furnace. The number of atoms in the light path is dependent on the temperature, the furnace length and matrix interferences. A two-step furnace24 will allow better control of two of these variables; temperature and matrix interferences. Since a two-step furnace does not require a rapid temperature ramp, accurate and reproducible temperatures should be easier to achieve. Volatilization of the sample into a constant temperature should alleviate many of the chemical interferences arising from the sample matrix.Only the furnace length would have to be standardized to obtain a Fig. 3 Absorbance map for the same data as shown in Fig. 2 (250 pg uniform response between diVerent instruments. of Cu). Absorbance was computed using the intensities from the first and last column (perpendicular to wavelength axis) as reference Noise characteristics intensities. There was no stray light correction for these data.An instrument composed of a continuum source and an array detector can be expected to have three major noise sources: for each pixel in Fig. 2. For the purpose of illustration, the fluctuation and shot noise from the xenon arc lamp and read absorbances in Fig. 3 were computed without stray light noise from the array detector. correction and using the average of the first and last pixel in Fluctuation noise is inherently eliminated using an array each row as the reference intensity.Stray light correction was detector provided that the pixels have a common exposure omitted so that the absorbances for the pixels between orders time. More specifically, this means the exposure start and stop (i.e., at the top and bottom of each column) could be computed. times for each row of pixels (parallel to the wavelength axis) With stray light correction, pixels between orders will have must be the same. If an LPDA is read sequentially, some of intensities randomly distributed around zero.As a result, the the fluctuation noise will be retained.12 A typical exposure logarithm of the ratio for many of these values will be consists of an integration period followed by a sequential read undefined and the rest will have a large variance. In routine period. Each pixel, although integrating for the same length practice, absorbances are only computed for pixels on the of time, will not be exposed to the same radiation (since the order, i.e., pixels whose intensity will be significantly greater starting and stopping times are not the same) and can be than zero after stray light correction.subject to diVerent fluctuations of the source. The longer the In theory, the height, wavelength and time integrated integration period, with respect to the read period, the more absorbance is independent of the characteristics of the source, the fluctuation noise will be reduced. Blocking exposure of the disperser and detector.Absorbances computed as described array during the read period will eliminate fluctuation noise12 above are corrected for stray light and broadband background by establishing a common starting and stopping time for each absorption and are constant for a constant number of absorbpixel for the integration interval. ing atoms in the absorption cell (furnace). The source intensity Frame transfer devices, which shift the data from the image and fluctuation, the monochromators throughput and spectral array to unexposed storage arrays prior to reading, will also bandwidth and the detectors pixel count, pixel size and quaneliminate fluctuation noise, provided the shifts are perpendicu- tum eYciency will not alter the computed absorbance.The lar to the wavelength axis.13,14 The split frame transfer camera noise component will increase as the intensity decreases, but used in this study shifts 40 rows upward and 40 rows downward the mean absorbance will not change.In addition, wavelength from the image arrays to the storage arrays. Data from the and time integration also mean that the integrated absorbance storage arrays are then read systematically during the next is not dependent on the spectral distribution of the absorption integration period. Since the shifts are perpendicular to the coeYcient (i.e., changes in width due to collisional broadening wavelength axis, the integration period for each row is diVerent at higher pressures or changes in distribution due to diVerences but the exposure period for each pixel in a row is common.in the isotopic composition) and is not dependent on the Thus, absorbances computed for each row or absorbances appearance function of the analyte (provided that the loss computed from intensities summed vertically (binned) will function is constant). eliminate fluctuation noise. In general, shifts perpendicular to Table 3 lists the intrinsic masses determined for a variety of the wavelength axis allow the elimination of fluctuation noise CS-AAS instruments over a 10 year interval.The spectral while shifts parallel to the wavelength axis will retain some of width of a pixel (at 196 nm) varies from 1.9 to 100 pm. For the fluctuation noise. some elements, the data in this table should show better Minimizing the read noise of a pixel is important for agreement. This table, however, represents data put together obtaining the best possible detection limits with the highest in retrospect and was not part of a systematic study.The resolution. It was shown, using propagation of errors,12 that biggest source of variation in intrinsic mass is the furnace the absorbance noise, sA, for the read noise limited case was atomization. In each case, an HGA 500 graphite furnace atomizer was used but the atomization temperatures between sA= 0.43sreadÓn+1 I (2) experiments varied by as much as 500 °C.The estimated intrinsic mass for LS-AAS is also listed. The estimated intrinsic mass was obtained by correcting the characteristic mass for where sread is the read noise for a pixel, n is the number of pixels needed to cover the absorption profile (and also the the linewidth ratio of the HCL and the absorption profile using a simple triangle model and assuming absorption over number of reference pixels) and I is the intensity read from each pixel. Eqn. (2) is very useful because it is directly pro- the entire absorption profile.Considering the simplicity of the 140 J. Anal. At. Spectrom., 1999, 14, 137–146Table 3 Intrinsic massa LS-AAS mi for CS-AAS E� chelled LPDA E� chellee LPDA H-20f E� chelleg SCD Wavelength/ (25 mm)h (50 mm)h LPDA (25 mm)h Element nm m0 b mi c (50 mm)h As 193.7 17 9.2 9.7 9.9 – 10.5 Se 1986.0 30 10 11 10 – 12 Zn 213.9 0.40 0.50 1.4 0.17 Pb 217.0 5.0 1.9 1.9 1.8 2.1 3.0 Sb 217.6 38 15 – – – 10 Sn 224.6 20 10 9.2 9.4 – – Cd 228.8 0.35 0.14 0.27 0.19 0.23 0.29 Ni 232.0 13 5.5 8.0 8.4 6.4 – Co 240.7 6.0 2.4 2.3 2.3 – – Fe 248.3 5.0 1.9 1.6 1.8 – – Tl 276.8 10 4.0 – – – 11 Mn 279.5 2.0 0.61 0.83 0.88 – 1.4 Pb 283.3 11 4.2 3.9 3.8 – 6.7 Cu 324.7 4.0 1.0 1.00 1.1 – – Ag 328.1 – – 0.4 – – – Spectral width (pm) 1.9 3.8 100 6.1 of Pixel at 196 nm aAll values determined with HGA-500 graphite furnace atomizer (Perkin-Elmer, Norwalk, CT, USA).bCharacteristic mass.21 cComputed from m0 as described in ref. 22. dSpectraspan III e� chelle (Spectrametrics) with 256 pixel LPDA and 25 mm pixel width.22 eSpectraspan III e� chelle (Spectrametrics) with 128 pixel LPDA and 50 mm pixel width.22 fH-20 monochromator with 128 pixel LPDA and 50 mm pixel width.23 gOptima e� chelle (Perkin-Elmer) with 256 pixel LPDA and 25 mm pixel width.13 hPhysical width of pixel. Table 4 CS-AAS detection limits portional to the detection limit. The previous section demonstrated that the normalized, integrated absorbance was Element Wave- HGA-500 furnace THGA furnace independent of the source and detector characteristics.length/ Consequently, the detection limit is directly proportional to nm LSa CS-LPDAb CS-SCDc LSd CS-DEMONe the absorbance noise. As 193.7 20 28 12 6 4 Eqn. (2) shows that the absorbance noise will grow smaller Se 196.0 30 50 16 9 13 as the intensity increases or the read noise decreases. This Zn 213.9 1 2 0.1 0.4 0.2 equation also indicates that the absorbance noise will decrease Pb 217.0 10 6 4 4 – with increasing spectral bandwidth.For example, if the Sb 217.6 15 – 8 4 2.5 entrance slit width is doubled, the intensity will double, the Bi 223.1 6 – 5 – – number of pixels necessary to cover the profile will double Sn 224.6 20 26 – 10 – Cd 228.8 0.4 0.4 0.07 0.1 0.1 and sA will decrease by Ó2. The LPDA used previously with Ni 232.0 10 11 – 8 – CS-AAS12 had a read noise of about 3000 e-. As a result, Be 234.9 1 – – 0.1 0.08 read noise was dominant at all intensity levels.The best Co 240.6 2 4 – 4 – detection limits (Table 4) were obtained with the largest Fe 248.3 2 2 – 0.8 0.6 entrance slit width of the e� chelle, 500 mm. At the time, these Si 251.6 40 – – 15 6 detection limits were the best ever achieved for CS-AAS, but Tl 276.8 10 – 1 9 3 Mn 279.5 21 0.5 0.2 0.6 0.3 they precluded operation in the high-resolution mode. Pb 283.3 5 0.9 0.4 4 1 The photon shot noise limited case, the ideal case, is achieved Al 309.3 4 – – 3 0.8 if the fluctuation noise is eliminated and the read noise is low.Mo 313.3 4 – – 1 2 A high quality CCD will typically have a read noise of less Cu 324.7 1.0 0.6 – 4 1 than 25 e-. Consequently, all but the lowest intensities Ag 328.1 0.5 – – 0.4 0.2 (<625 e-) will be shot noise limited. The absorbance noise Cr 357.9 1 – – 0.4 0.8 for the shot noise limited case is aModel 5000 (Perkin-Elmer).21 bCS with linear photodiode array (LPDA) detector.22 cCS with segmented charge coupled array detector (SCD) of Optima (Perkin-Elmer).13 dSIMAA 6000 (Perkin-Elmer).25 sA= 0.43ÓIÓn+1 I = 0.43Ón+1 ÓI (3) eCS with double e� chelle monochromator (DEMON).26 since sI=ÓI.This equation shows that, for the shot noise limited case, the absorbance noise is independent of the spectral bandwidth. Consider again the case where the entrance slit more intense sources, with little success. More signifcant improvements have been made in increasing the luminosity of width is doubled.As stated previously, both n and I will also double. With eqn. (2), sA is reduced by Ó2. With eqn. (3), the spectrometer and quantum eYciency of the detectors. Table 2 shows that the two-dimensional array is thinned and however, sA remains the same. Consequently, the use of a narrow slit width to maximize spectral resolution does not back illuminated. The result is a quantum eYciency of 50% at 200 nm. degrade the detection limit.Eqn. (2) shows that if I, the level of radiation striking each Table 4 presents detection limits for some of the most recent CS-AAS instruments. A comparison of eqns. (2) and (3) pixel, increases without opening the entrance slit width, the absorbance noise will decrease. Three means of increasing I shows why better detection limits are achieved in the far UV with a CCD array. For the LPDA described above (read noise are to increase the source output, increase the transmission eYciency of the spectrometer and increase the detection about 3000 e-), the detected intensity must be 9×106 e- in order for the shot noise to equal the read noise.For the same eYciency. Numerous attempts have been made to develop J. Anal. At. Spectrom., 1999, 14, 137–146 141detector, elements lying between 190 and 230 nm had intensit- matrix, it seems likely that they arose from molecular bands of NO. The non-analyte absorption around the As and Se ies ranging from 3.8×105 to 3.2×106 e- for a 20 ms integration interval.Thus, the predicted absorbance noise for the lines could be both spectrally and temporally resolved. The non-analyte absorption disappeared when the elements were shot noise limited case ranged from 4.5 (at 193.7 nm) to 1.6 (at 232.0 nm) times lower. The results in Table 4 show that determined in a Pd(NO3)2 chemical modifier. Selenium (196.0 nm) was the only element of those examined for which the measured detection limits for the segmented CCD detector (read noise 15 e-) ranged from equivalent to a factor of 3.1 a Pd line was observed.The only consequence of the nonanalyte absorption in the nitric acid matrix is to emphasize times lower. The improved detection limits of the CS-double e� chelle the necessity for choosing the oV-line reference pixels with care. Obviously, this same problem is of much greater concern monochromator (CS-DEMON) instrument22 can be attributed to better imaging of the lamp on the entrance slit of the for oV-line background correction for emission spectroscopy because the operating temperatures are much higher and the spectrometer.CS-DEMON used a small, xenon short-arc lamp with a well defined bright spot at the tip of the cathode. This spectra are much more complex. Schuetz26 has reported much more complex structured bright spot was easier to image on the entrance slit than that of lamps employing the reflector as an integral part of the background spectra around Cd (228.8 nm), Pb (217.0 nm) and Se (196.0 nm) when using ammonium dihydrogenphosphate lamp housing.The bright spot provided higher UV intensities and slightly better detection limits. and magnesium nitrate as a chemical modifier. An obvious solution is to not use this particular modifier, but the more general question is still of interest: can interference from Spectral resolution complex line structures be corrected? In his thesis, Schuetz used spectral subtraction with some success.This approach, The combination of high S/N with high resolution makes CS-AAS an ideal instrument for examining spectral line inter- however, requires matching the sample matrix qualitatively and quantitatively. Initial results suggest that AA spectra are ferences. LS-AAS only has emission intensity over the narrow spectral width of the HCL emission line. It is impossible to be suYciently uncongested to allow the more empirical approach of reconstructing the spectra with modeled peaks.28 At this sure what absorption lines lie just outside the emission line width.Absorption interferences that fall within the mono- time, more study is necessary. The major point, however, is that there is suYcient information available to allow correction chromator’s spectral bandwidth (for background correction with a secondary, continuum source) or are shifted within the for spectral line interferences. The only question is how much eVort is justified in making the correction an of emission line bandwidth by the magnetic field of Zeeman AAS, can only be inferred.With CS-AAS, however, it is now automation that can be achieved. possible to inspect the spectral regions around the absorption profile in a manner analogous to that used for emission Spatial resolution spectroscopy. Fig. 4 shows the spectra in the region of the Se (196.0 nm) The columns of pixels perpendicular to the wavelength axis record intensities as a function of the order height.The order absorption line as a function of time.13 In this case, Pd(NO3)2 was used as a chemical modifier so the mass of Pd (5 g) was width, determined by the height of the entrance slit, is 500 mm. With an appropriate optical arrangement,17 the order width 2500 times that of Se (2 ng). It can be seen that after 4 s of atomization the Pd (peak B) was still present, whereas the Se reflects intensities as a function of height in the furnace.In eVect, the array detector (with pixels 20 mm high) divides the (peak A) was relatively short lived. Vacuum wavelength tables from the National Institute of Standards and Technology27 radiation transmitted through the 6 mm high furnace into 25 individual sections. Separate absorbances can be computed show the presence of a weak Pd line (196.011 nm) just 15 pm from the Se line (196.026 nm). The two peaks are fully resolved for each section and summed to provide a furnace-height integrated absorbance that is linear with respect to with a spectral bandwidth of 3 pm per pixel.Similar data for As, Bi, Cd, Mn, Pb, Sb, Se, Tl and Zn in concentration. Gilmutdinov and co-workers29 have shown that there is a a 5% v/v nitric acid matrix showed some non-analyte absorption around the As and Se lines but none around the wave- non-uniform distribution of source intensity and analyte atoms in the furnace. The lateral distribution of atoms tends to be lengths of the other elements.13 The source of the non-analyte absorption was not identified although, with a nitric acid symmetrical and relatively homogeneous at any height compared with the vertical distribution which is asymmetric and non-homogeneous owing to the dosing hole at the top of the furnace. The non-homogeneous distribution of atoms produces non-uniform transmitted intensities that, with integration in the intensity domain, can result in a non-linear response with respect to the analyte concentration.Gilmutdinov and co-workers speculated that, with complex sample matrices, the absorbances computed for LS-AAS are significantly diVerent for standards and samples owing to the non-homogeneity of the spatial distribution of the analyte atoms. McNally and Holcombe30 demonstrated a non-uniform height distribution for Cu atoms in a pure standard. With the furnace height divided into nine subsections, they measured absorbances at the bottom and the top of 0.39 and 0.30, respectively, at the time of the peak maximum.This diVerence of 0.09 is 23% of the signal at the bottom of the furnace at the time of the peak maximum. In Fig. 5, the diVerence in working range arise from the finite width of the HCL emission line (the HCL line is not monochromatic, just 3–5 times narrower than the absorption profile at atmospheric pressure) and stray light. With CS-AAS, there is no theoretical limit to the calibration range, only the practical limits imposed by the size of the array, the increasing possibility of spectral interferences and the ability to clean the furnace between atomizations.The shapes of the calibration curves were predicted theoretically in the original work by Mitchell and Zymanski.32 At low concentrations, absorbance increases linearly with increase in concentration. This relationship provides linear plots with a slope of 1.0 when the logarithm of absorbance is plotted versus the logarithm of concentration.As the concentration increases, the absorption at the peak center reaches a maximum determined by the stray light. At this point, the calibration curves for LS-AAS will reach a plateau. With a continuum source, however, absorption in the wings of the profile can be measured. At any wavelength in the wings, Fig. 5 Vertical resolution of absorbance corresponding to absorbance absorbance will increase linearly with increase in concenas a function of height in the furnace.Pixel 5 is approximately the tration. However, the wings broaden as a function of the bottom of the furnace and pixel 25 is the top of the furnace. square root of the concentration, so the wavelength integrated Absorbance was measured over 200 ms intervals centered at (#) 2.2, absorbance will also increase with the square root of the (%) 2.4 and (6) 2.6 s (the peak maximum) for 500 pg of Cu concentration. This relationship at high concentrations pro- (324.7 nm).vides a linear plot with a slope of 0.5 on a log–log plot, as shown in Fig. 6. it is only 12%. This lower percentage may be due to the larger The inflection point, the point where the calibration curve mass of Cu used in Fig. 5 (500 pg) as opposed to that of makes the transition from a slope of 1.0 to a slope of 0.5, is McNally and Holcombe (200 pg). determined by the ‘a-value’ and the hyperfine splitting of the The data in Fig. 5 demonstrate the suitability of the absorption profile.The a-value is the ratio of the collisional two-dimensional CCD array for measuring intensities and width to the Doppler width and determines the width of the computing absorbance as a function of height in the furnace. absorption profile. The a-value will vary between elements and With the detector described in Table 2, it is now possible to for each element as a function of temperature. Hyperfine measure routinely the inhomogeneity of the analyte distrisplitting is determined by the coupling of the electron trans- bution in the furnace.Reading the full array (80×80 pixels), itions and determines the number of components of the however, requires a read rate 80 times faster than that required absorption profile. Elements with low a-values and fewer if the columns were binned or if a linear array of 80 pixels components will have deeper, narrower profiles and elements with an 8051 aspect were used. Further studies will determine with a larger a-values and more components will have shal- whether the increased linearity of the analytical data justifies lower, broader profiles.The former elements will reach the the increased demand for data acquisition speed. stray light limit sooner and have inflection points at lower concentrations. Calibration curves In Fig. 6, absorbance is plotted versus the normalized concentration, i.e., the concentration of the standards divided Fig. 6 shows calibration curves for Ag (328.1 nm), Cd by the intrinsic concentration (the concentration necessary to (228.8 nm) and Pb (283.3 nm)31 that consist of two linear give an integrated absorbance of 0.0044). At lower concen- regions and cover 5–6 orders of magnitude of concentration. trations, each normalized concentration will have the same A major detraction for LS-AAS has always been the relatively absorbance even though the shapes of the absorption profiles short linear region of the calibration curves, from 2.5 to 3 are diVerent.In Fig. 6, the inflection point for Cd occurs at a orders of magnitude of concentration. The limits for the linear lower concentration than that of Ag and Pb. This indicates that the resultant absorption profile for Cd is narrower and deeper than those of Ag or Pb. The calibration points for all three elements in Fig. 6 can be fitted with a single calibration curve shape.33 Half of a hyperbola was rotated and the parameters were adjusted to provide a curve with a slopes of 1.0 and 0.5 in the linear regions. This shape was fit to all three data sets (solid lines in Fig. 6) by simply oVsetting the x and y axis coordinates. Two calibration standards (one in each linear region) are suYcient to determine a unique calibration curve, but four standards provide better reproducibility for the curves. Obtaining more information from ETAAS In light of the recent interest in metal speciation, it is ironic that researchers in the field of AAS have spent many years trying to optimize furnace parameters in order to force diVerent Fig. 6 Calibration curves for (6) Ag (328.1 nm), (%) Cd (228.8 nm) species into uniform behavior, i.e., volatilization of all metal and (1) Pb (283.3 nm). Absorbance integrated with respect to species as a single analytical peak, regardless the chemical wavelength and time is plotted versus normalized concentration (concentration divided by the intrinsic concentration).environment. It seems reasonable that use of less than J. Anal. At. Spectrom., 1999, 14, 137–146 143‘optimum’ parameters would oVer more information about the metal species. It is also true that analytical peaks acquired using ‘optimum’ parameters oVer information that is not currently being used. Alternative approaches to furnace design, temperature programs and data processing can provide much more information about the sample species and how they compare with known standards.The peak shape contains a significant amount of additional information about the analyte with respect to condensed and gas phase interactions. Researchers have known for years that the leading edge of the analytical peak can be used to determine the activation energy of the metal, an indication of the chemical bond(s) that had to be broken to allow volatilization.34 Harnly35 showed that the times for the appearance, peak maximum and decay to predetermined values could be used to predict interferences.It was later shown that the phase angles of the Fourier transform of the analytical peak were more sensitive and precise indicators of diVerences between Fig. 8 Thermal separation of heme Fe and inorganic Fe (248.3 nm) in a two-step furnace. the behavior of the standards and the samples.36 Most recently, it has been shown that principle component analysis is the best means of incorporating data from all the phase angles into a comprehensive evaluation of the signal behavior.37 are not necessarily suYcient to induce atomization. The solution to this problem is to use a two-step furnace.With this Fig. 7 shows the principal components analysis of the Fourier transform phase angles for the determination of Pb in furnace, the cup temperature can be slowly ramped to maximize thermal separation and the furnace can be held at a 5% HNO3 (Group A), in 5% HCl (Group B) and in various concentrations of NaCl in both HNO3 and HCl (Groups C temperature suYcient to atomize any of the compounds.Since the temperature of the furnace is held constant, diVusion from and D, respectively). The arrows show that there was a systematic shift in position with respect to the standard the furnace will be constant and the time integrated absorbance will be independent of the introduction rate from the cup. The concentrations in the HNO3 and with the NaCl content in both matrices. The standards in 5% HCl (Group B) showed constant diVusion rate will also allow one set of standards to be used for all compounds (peaks).no systematic shift in position with respect to concentration. The recoveries for 2 ng of Pb in 0.1, 0.2, 0.5 and 1.0% NaCl Fig. 8 shows the determination of heme Fe (Fe2+ in protoporphyrin IX) and inorganic Fe. The heme Fe content in 5% HNO3 (by mass) were 90, 80, 72 and 64%, respectively. The recoveries in similar concentrations of NaCl in 5% HCl of foods is of importance to nutritionists because it is absorbed by the body at about three times the rate of inorganic Fe.were 80, 74, 72 and 69%, respectively. Unfortunately, there are not enough repeat determinations to assess the precision Heme Fe sublimes at about 250 °C whereas inorganic Fe volatilizes around 1800 °C. It can be seen that the peaks are of the measurements. Certainly, however, the data suggest the diagnostic potential of the peak shape data. nicely resolved. The two step furnace provides an ideal means of quickly determining the two Fe concentrations.At this A second approach to acquiring more information about the analyte is to slow the temperature ramp of the furnace to stage, more work is needed to determine the best method of lysing the sample matrix so that the heme Fe is not trapped allow the detection of species with diVerent volatilities. Unfortunately, temperatures suYcient to volatilize the analyte and thermally degraded before it can sublime.The temporal resolution capabilities of the two-step furnace are appropriate for metals that have high volatilities or will not thermally degrade. Several applications suggest themselves, most notably the determination of elemental species of environmental interest such as the various organic forms of As, Hg, Sn and Tl. Future of multi-element AAS The instrument described in Table 2 and throughout this paper has unique analytical capabilities. As a single-element instrument, it surpasses LS-AAS with respect to absorbance accuracy, detection limits, calibration, spectral and spatial information and the portability of the figures of merit.The general analytical features of this instrument, when used with a graphite furnace atomizer, are summarized in Table 5. Such an instrument, at the listed cost, would be competitive in the market place. The real advantage of CS-AAS, however, is multi-element determinations. Fig. 7 Principal component analysis of phase angle arrays for varying Construction of a multi-element instrument, incorporating masses of Pb (283.3 nm) in matrices of 5% HNO3, 5% HCl and various concentrations of NaCl.(A) 0.2–4.0 ng of Pb in 5% HNO3; the characteristics listed in Table 2, would require the developdirection of arrow indicates direction of increasing Pb mass. (B) ment of a new detector. Although any one of several existing 0.2–4.0 ng of Pb in 5% HCl; there was no correlation between the e� chelle spectrometers would be suitable, a solid state detector principal component scores and the Pb mass.(C) 2.0 ng of Pb in 5% would have to be developed consisting of approximately 45 HNO3 and 0.1–1.0% NaCl (by mass); arrow indicates the direction two-dimensional CCDs. In theory, one possibility is a mono- of increasing NaCl concentration. (D) 2.0 ng of Pb in 5% HCl and lithic detector in which the two-dimensional CCDs are embed- 0.1–1.0% NaCl (by mass); arrow indicates the direction of increasing NaCl concentration.ded. Such a detector would be analogous to the segmented 144 J. Anal. At. Spectrom., 1999, 14, 137–146Table 5 Summary of characteristics of single-element CS-AAS instruments as to how this principle is applied. These diVerdescribed in this paper ences may be advantageous or disadvantageous for CS-AAS. The flexibility of the Iris and Vista in selecting analytical Feature Evaluation wavelengths is a distinct advantage for CS-AAS.The fixed wavelength position of the arrays of the Optima results in Detection limits 0.1–10 pg Calibration range 105–106 omission of about one third of the resonance wavelengths routinely used for AAS. With the Iris and Vista, it is only Interferences— necessary to program data acquisition for a diVerent suite of Stray light Severe—corrected with oV-order measurements Chemical Moderate—correctability depends on furnace sub-arrays. design The low pixel read noise of the Optima and Vista is preferred Spectral—broad Severe—corrected with oV-line measurements over the higher levels of the Iris.Whereas multiple non- Spectral—line Rare—correctability varies with overlap destructive reads of the CID can eVectively reduce the read severity noise for AES applications, the rapid, transient signals of the Mass tolerance >100% m/m Resolution 2 pm at 200 nm furnace will not permit suYcient multiple reads with CS-AAS. Cost ~$35 000 Consequently, the eVective read noise of the Iris will always be greater than that of the Optima or Vista.Of course, the read noise of the CID is still suYciently low that only a few elements in the UV would be rendered read noise limited. CCD detector described by Barnard et al.,38 except that each The Iris is the only instrument with the inherent ability to detector would have two dimensions. Since far fewer wavemeasure intensities between orders for correction for say lengths are used for AAS (compared with AES), only one light.In the single-element mode, a double monochromator array would be necessary for each element routinely run by can be used to reduce the stray light arising from the CS. In ETAAS, approximately 45. Such a detector, however, would the multi-element mode of operation, this is not possible and be expensive. the far stray light (from flaws in the grating) and stray light The most obvious means of reducing the complexity of the from overlapping adjacent orders can be very high.In the detector is to sacrifice the vertical dimension for height resomulti- element mode, the Iris is the only instrument that oVers lution of the furnace. If this is done, then a series of linear stray light correction capabilities. arrays can be used. The next logical question is whether any The high quantum eYciency of the Optima and Vista arises of the solid state detector–e� chelle spectrometer instruments from the linear nature of the CCD arrays.Without overlying that have been developed for AES (Table 6) are suitable for control lines, quantum eYciencies as high as 50% at 200 nm multi-element CS-AAS. While such instruments would not be can be achieved without the use of a fluorescent coating. The ideal for AAS, the lack of development costs is appealing. two-dimensional nature of the CID requires a grid of control Table 6 presents the characteristics of the three major e�chelle lines and reduces the quantum eYciency to about 34% using spectrometer–array detector instruments which have been a fluorescent coating.developed for ICP-AES.38–41 Each incorporates the same The high luminosity (Ll) of the Optima predicts better fundamental principle; a sub-array of pixels is used for each detection limits than the other two instruments. Higher lumin- wavelength of interest. This principle is also valid for CS-AAS. There is, however, a great deal of diVerence between the osity means greater intensity striking each pixel and, from Table 6 Comparison of commercially available atomic emission array detectors Optimaa Irisb Vistac Spectrometer characteristics— Type E� chelle E� chelle E�chelle Focal length/mm 504 384 400 Blaze angle (°) 63.4 19.5 44.8 Grating area/mm×mm 80×160 40×40 50×70 RLDd (200 nm)/nm mm-1 0.100 0.735 0.252 Entrance slit height/mm 250 58 50 Entrance slit width/mm: Normal 62 58 25 High resolution 31 32 – Spectral bandwidth/pm: Normal 6.2 43 6.3 High resolution 3.1 23 – Lle (200 nm)/mm2 nm-1 0.028 0.00038 0.0015 Detector characteristics— Type of array CCD CID CCD Segmented-linear Two dimensional Diagonal-linear Number of arrays 224 1 140 Array size(s) 1×20 to 1×80 512×512 1×750 to 1×1521 Number of pixels 6336 262 144 70 908 Pixel size/mm 12.5×170 (UV) to 12.5×80 (VIS) 28×28 12.5×105 (UV) to 25×45 (VIS) Read noise (e-) 15 260 <10 Quantum eYciency 56 34 50 (200 nm) (%) Sub-array processing Inherent Yes Yes Binning capability Not necessary Yes Yes Wavelength coverage/nm 167–782 170–800 165–785 aPerkin-Elmer.38,39 bThermo Jarrell Ash (Franklin, MA, USA).40 cVarian, Optical Spectroscopy Instruments (Mulgrave, Australia).41 d Reciprocal linear dispersion.eLuminosity per unit wavelength as defined in ref. 39. J. Anal. At. Spectrom., 1999, 14, 137–146 145P. L. Lundberg and B. Radziuk, J. Anal. At. Spectrom., 1997, eqn. (3), less absorbance noise. Since the wavelength and time 12, 617.integrated absorbance is constant, lower absorbance noise 14 J. M. Harnly, R. E. Fields and M. Schuetz, J. Anal. At. Spectrom., specifies better S/Ns and lower detection limits if the same in the press. furnace is used. 15 R. L. Cochran and G. M. Hieftje, Anal. Chem., 1977, 49, 2040. The Optima and Vista oVer seven times better resolution 16 R. E. Sturgeon and C. L. Chakrabarti, Prog. Anal. At. Spectrosc., 1978, 1, 5. than the Iris in the ‘normal’ mode of operation.The resolution 17 B. L’vov, Spectrochim. Acta, Part B, 1978, 33, 153. of both the Optima and the Iris can be improved by a 18 A. Kh. Gilmutdinov and J. M. Harnly, Spectrochim. Acta, Part B, factor of approximately two in the ‘high’ resolution mode. As 1998, 50, 1003. discussed earlier, the S/N is independent of the spectral 19 Spectraspan IIIB Emission Spectrometer Operators Manual, Part bandwidth. The bandwidth is an important factor, however, Number 1506029, Spectrametrics, 1977.in resolving spectral line interferences. 20 Specifications for FastOneTM, PixelVision, Beaverton, OR, 1997. 21 W. Slavin, Graphite Furnace: a Source Book, Part Number Although no specifications are presented in Table 6 for the 0993–8139, Perkin-Elmer, Norwalk, CT, 1984. speeds of operation, the general characteristics of each instru- 22 C. M. M. Smith and J. M. Harnly, Spectrochim. Acta, Part B, ment allow the determination of at least 10 sub-arrays every 1994, 49, 387. 17 ms (60 Hz). The number of elements determined may 23 J. M. Harnly, Spectrochim. Acta, Part B, 1993, 48, 909. require compromises with respect to integration times, shift 24 W. Frech and S. Jonsson, Spectrochim. Acta, Part B, 1982, 37, rates and frequency of analog-to-digital converter operation 1021. 25 SIMAA 6000 Atomic Absorption Spectrometer, Part Number which will alter the S/N. Exact values for detection limits are B050–6158, Perkin-Elmer, Norwalk, CT, 1989. diYcult to predict. 26 M. Schuetz, PhD Thesis, Technical University of Berlin, 1997. It can be seen that each of the commercially available 27 NIST Circular 488, Section 4, Ultraviolet Multiplet Table (Z=1 e� chelle–solid state detector instruments oVers diVerent capa- to 64), and Section 5, Ultraviolet Multiplet Table (Z=72 to 88), bilities. None is ideal. The cost of development and construc- National Institute of Standards and Technology, Gaithersburg, tion of the ‘ideal’ detector would be expensive. In the current MD, 1978. 28 J. M. Harnly, unreported results, 1998. economic climate, where development of technology for future 29 A. Kh. Gilmutdinov, B. Radziuk, M. Sperling and B. Welz, products is sacrificed for quarterly profits, the cost of the Spectrochim. Acta, Part B, 1996, 51, 1023. detector may be suYcient to block the development of multi- 30 J. McNally and J. A. Holcombe, Anal. Chem., 1987, 59, 1105. element CS-AAS, regardless of the enhanced analytical 31 D. N. Wichems, R. E. Fields and J. M. Harnly, J. Anal. At. capabilities. Spectrom., 1998, 11, 1277. 32 A. C. G. Mitchell andM.W. Zymansky, Resonance Radiation and Excited Atoms, Cambridge University Press, Cambridge, 1961. References 33 J. M. Harnly, C. M. M. Smith and B. Radziuk, Spectrochim. Acta, Part B, 1996, 51, 1055. 1 A. Walsh, Spectrochim. Acta, Part B, 1955, 7, 108. 34 R. E. Sturgeon, C. L. Chakrabarti and C. H. Langford, Anal. 2 C. T. J. Alkemade and J. M. W. Milatz, J. Opt. Soc. Am., 1955, Chem., 1976, 48, 1792. 45, 583. 35 J. M. Harnly, J. Anal. At. Spectrom., 1988, 3, 43. 3 V. A. Fassel, V. G. Mossotti, W. E. Grossman and R. N. Knisely, 36 J. M. Harnly, J. Anal. At. Spectrom., 1988, 3, 485. Spectrochim. Acta, Part B, 1966, 22, 347. 37 J. M. Harnly, Paper presented at the Pittsburgh Conference and 4 McGee and J. D. Winefordner, Anal. Chem., 1967, 37, 429. Exhibition, New York, March 5–9, 1990, Paper 572. 5 W. Snelleman, Spectrochim. Acta, Part B, 1968, 23, 403. 38 T. W. Barnard, M. J. Crockett, J. C. Ivaldi, P. L. Lundberg, 6 R. C. Elser and J. D. Winefordner, Anal. Chem., 1972, 44, 698. D. A. Yates, P. A. Levine and D. J. Sauer, Anal. Chem., 1993, 7 J. G. Nitis, V. Svoboda and J. D. Winefordner, Spectrochim. Acta, 65, 1231. Part B, 1972, 27, 345. 39 T. W. Barnard, M. J. Crockett, J. C. Ivaldi and P. L. Lundberg, 8 C. Veillon and P. Merchant, Appl. Spectrosc., 1973, 27, 361. Anal. Chem., 1993, 65, 1225. 9 P. N. Keliher and C. C. Wohlers, Anal. Chem., 1974, 46, 682. 40 M. J. Pilon, M. B. Denton, R. G. Schleicher, P. M. Moran and 10 A. T. Zander, T. C. O’Haver and P. Keliher, Anal. Chem., 1976, S. B. Smith, Jr., Appl. Spectrosc., 1990, 44, 1613. 48arian Optical Spectroscopy Instruments, Sugarland, TX, 11 J. M. Harnly, T. C. O’Haver, B. Golden and W. R. Wolf, Anal. personal communication, 1998. Chem., 1979, 51, 2007. 12 J. M. Harnly, J. Anal. At. Spectrom., 1993, 8, 317. 13 J. M. Harnly, C. M. M. Smith, D. N. Wichems, J. C. Ivaldi, Paper 8/07586G 146 J. Anal. At. Spectrom., 1999, 14, 137–146