Use of a Continuous Source in Flame FIuo rescence S pect rometry CLAUDE VEILLONf1 J. M. MANSFIELD, M. L. PARSONS, and J. D. WINEFORDNER Department of Chemistry, University of Florida, Gainesville, Flu.
b Low limits of detection for 1 3 elements were obtained b y atomic fluorescence flame spectrometry using a 150-watt xenon arc continuous source, a total-consumption atomizerburner, and a low resolution monochromator. Some properties of a new flame, argon, hydrogen, entrained air, and scattering of incident radiation by salt particles in the flame gases were also studied. Copper, silver, gold, bismuth, magnesium, zinc, cadmium, mercury, and thallium exhibited relatively intense atomic fluorescence in flames excited by a continuous source. The shape of the analytical curves of zinc, cadmium, and thallium were different from those obtained with line sources.
P
ERHAPS
Present address, SpectrochemicalAnal-
ysis Section, National Bureau of Stand-
ards, Washington, D. C. 204 *
PHASE-SENSITIVE
AMPLIFIER
I
REFEREWCE
MONOCHROMATOR
the most undesirable feature
of atomic fluorescence and atomic
absorption flame spectrometry has been that a separate line source is usually required for each element to be determined. This has been a serious drawback in attempting to develop the atomic fluorescence technique into a generally applicable analytical method, because the sources used for atomic fluorescence flame spectrometry must be more intense than those used for atomic absorption spectrometry, and practical intense line sources are as yet unavailable for many elements. Despite the drawbacks of using individual sources, atomic absorption spectrometry has been able to flourish by use of low intensity, hollow cathode discharge tubes, which can be easily manufactured for most elements. Past atomic fluorescence flame spectrometric studies have employed the Osram-type metal vapor lamps and RF-operated electrodeless discharge tubes. Recent studies in the authors’ laboratory have indicated that continuous sources can be successfully used for atomic fluorescence flame spectrometry. Continuous sources have recently been used in atomic absorption flame spectrometry (3). However, high-quality monochromators with very narrow spectral bandwidths are required due to 1
-
ANALYTICAL CHEMISTRY
HIGH
VOLTAOE
Figure 1 . Schematic diagram of the apparatus used to make flame fluorescence measurements (not to scale)
the very narrow absorption linewidth of most spectral lines. Because the detector does not “see” the source in flame fluorescence spectrometry, a much simpler and less expensive monochromator can be used. The monochromator is needed only to isolate the fluorescence radiation of interest. Thirteen elements were investigated using a 150-watt xenon arc lamp as the source of excitation. Limits of detection obtained for most of these are in the low p.p.m. range and lower. The authors believe that the results obtained with this source will greatly influence future efforts in the field of atomic fluorescence flame spectrometry. EXPERIMENTAL
Apparatus. The apparatus used to make flame fluorescence measurements is shown in Figure 1. The monochromator and optics were the same ones used by Mansfield, et al. (6). Light-trap tubes were added to an originally rectangular baffle-box in an attempt to reduce the level of scattered radiation reaching the entrance slit of the monochromator. All fluorescence studies were carried out using a total-consumption aspirator-burner (4020, Beckman Instruments, Fullerton, Calif.). Studies on the scattering of
radiation by salt particles were carried out with both the total-consumption aspirator-burner and a laboratory constructed heated-chamber aspiratorburner similar to that described by Dubbs (2). In the polarization studies, two polarizer assemblies (B54-6440, American Instrument Co., Inc., Silver Spring, Md.) were used. One was positioned in front of the slit of the monochromator and the other between the source and the mechanical chopper. The signal from the RCA type 1P28 multiplier-phototube (PM) was fed into a pre-amplifier (CR-4, Princeton Applied Research, Princeton, N. J.), The pre-amplifier was made necessary by the relatively low input impedance (approximately 50 kohms) of the phasesensitive amplifier (JB-4, Princeton Applied Research). The pre-amplifier provided additional gain and, by means of its adjustable frequency bandpass, increased the percentage of the peak input voltage to the phase-sensitive amplifier, which was due to the desired signal. The reference signal was obtained from the lamp-photocell arrangement shown in Figure 1. The lamp was a General Electric 82 miniature bulb, and the photocell was a Raytheon EM 1502 CdSe fastrswitching photcconductivity cell. A &volt automotive storage battery powered the lamp and the reference circuit. The chopper-
p'
IOOOk
Y
L c
3 a
METAL
CONCENTRATION,
P.P.Y.
Figure 2. Analytical curves for several elements using a continuous source
modulated lamp radiation modulated the photocell resistance, thus modulating the current through (and therefore the voltage across) the 3.3-kohm resistor, Because the same chopper was used for both the reference lamp and the source, the signal and reference were of the same frequency. The circuitry in the phase-sensitive amplifier is designed so that the output is independent of the reference signal magnitude over a fairly large range and is dependent only upon its frequency. The output of the phase-sensitive amplifier is differential, and the quiescent d.c. level is -8 volts with respect to ground. Therefore, to use a recorder having a single-ended input, the circuit shown between the amplifier and recorder was necessary. -1 -8volts bucking-voltage was derived from a 10-kohm potentiometer across eight RM-12-R mercury cells connected in series. The 10-kohm variable resistor across the recorder input was used to adjust the recorder sensitivity to the full-scale output voltage of the amplifier (5 volts). All data used for the analytical curves and limits of detection were recorded on a 10-mv. recorder (SR, E. H. Sargent and Co., Chicago, Ill.). -411 scattering spectra were recorded on an x-Y recorder (1620-814, American Instrument Co., Inc.), The source was a 150-watt xenon arc lamp (Engelhard, Hanovia, Yewark, K. J.) powered by a regulated d.c. supply (422-818, American Instrument Co., Silver Spring, >Id.). The chopper consisted of a circular aluminum disc having eleven 0.75-inch, equally-spaced holes (the space between the holes was also equal to 0.75 inch). The chopper was rotated by an 1800r.p.m. synchronous motor (KYC-23, Bodine Electric Co., Chicago, Ill.). Negative high voltage for the multiplier-phototube was furnished by a regulated power supply (412.1, Fluke Mfg. Co., Seattle, Wash.). Conditions. The phase-sensitive amplifier had no response to thermal emission from the flame, except for a n increase in the noise level due t o the resulting phototube current. For all of the analytical studies, the time-
l'',~O:l.
""""I
' METAL
" " " " IO
'
" " " " 100
CONCENTRATION,
'
"""d
1000
P.P.Y.
Figure 3. Analytical curves for several elements using a continuous source
constant of the phase-sensitive amplifier was 10 seconds. It was noted on several occasions that the signal-to-noise ratio could be improved by going to higher phototube voltages with an accompanying decrease in the monochromator slit width. This indicated that the phototube signal was increasing faster than the noise currents. Therefore, all measurements were made using an overall phototube voltage of - 1000 volts. Voltages higher than this mere not tried, to minimize possible phototube damage and fatigue effects. The' phase-sensitive amplifier was found to overload before the maximum current rating of the 1P28 multiplier phototube was reached. No attempt was made to optimize exhaustively the experimental conditions such as flame gas flow rates, region of flame viewed, monochromator slit width, etc. Argon-Hydrogen Flame. While investigating the fluorescence of magnesium, it was found that the fluorescence intensity could be increased by adding argon t o the oxygen supplied to the burner. However, the intensity increased as the proportion of argon in the argon-oxygen mixture increased (all the way up t o pure argon with only entrained air maintaining combustion). Under these conditions, the intensity was approximately 10-fold greater than with oxygen alone. Because the temperature of such a flame must be much lower than an oxy-hydrogen flame, the observed enhancement of the fluorescence with argon was probably due to decreased quenching of the excited atoms by collisions with molecular species present in the flame and decreased compound formation of the atom of interest with flame gas products, The argon-hydrogen flame was evaluated with many of the elements studied, and the results are presented with the discussions of each of these elements;
RESULTS AND DISCUSSION
The elements copper, silver, gold, bismuth, magnesium (in the hr-Hz flame), zinc, cadmium, mercury, and thallium exhibited relatively intense atomic fluorescence in flames exited by a continuous source. Analytical curves are shown in Figures 2 and 3. Analytical conditions used and limits of detection obtained are given in Table I. The limit of detection i? defined as that solution concentration resulting in a sgnal-to-noise ratio of 0.75 (6). -111 of the analytical curves in Figures 2 and 3 are terminated at a fluorescence inteniity equal to the r.m.s. noise level obtained when pure solvent is aspirated into the flame. The point? qhonn in the analytical curves are the averageq of at lea-t two determinations. In no case did ci single measurement differ by inore than 570 from the average value. The precision of measurement was similar to that obtained by llansfield, Winefordner, and Veillon ( 5 ) -Le., the relative standard deviation for atomic concentrations near the midpoint of the analytical curve of any element was less than 5.0%. All metals were introduced into the flame of interest as chlorides or nitrates. For a single determination a t least 1 ml. of sample solution was required. The light-trap tubes added to the baffle-box resulted in a decrease in incident radiation scattered from solid surfaces. However the majority of the incident radiation reaching the monochromator entrance slit was due to reflection and scattering from solution droplets in the flame of the totalconsumption burner. Lead. The atomic fluorescence of lead a t 2833 and 4058 A. was nearly equal in intensity. However, the lamp intensity a t 4058 A. mas considerably greater than a t 2833 d., resulting in a larger signal-to-noise VOL. 38, NO. 2, FEBRUARY 1966
205
b Figure 4. Scattering and spectral response curves' A
Typical recorded scattering curve. Experimental conditions: 1.5-mm. slit width, 950 volts on photomultiplier tube, 6.5 cm. above burner tip; H 2 flow, 5000 cc./minute; 0 2 flow, 3250 cc./minute; the heating tape on the heated chamber was adjusted such that the gases were at about 150' C. at the exit of the chamber B Spectral response curve of the experimental apparatus. Experimental conditions: 20-micron slit width, 800 volts on photomultiplier tube; light from the xenon lamp was reflected .into the monochromator b y means of an aluminum wire C Scattering curve corrected for the spectral response of the system aThe signal axis is relative; therefore, the heights of the curves cannot b e compared direttly
ratio. This indicated that these results might be improved considerably with a more powerful continuous source. Other lead lines exhibiting fluorescence were 2802 and 3683 A. Very weak fluorescence was also observed at 2614 and 2873 A. The fluorescence intensity of the 4058 A. lead line mas about two-fold greater in the Ar-H2 flame than in the 02-H2 flame. In Figure 3, analytical curves for lead are given for both flames. It should be noted that although the fluorescence intensity in the .2r-H2 flame is twice that in the 02-H2 flame, the noise level in the 02-H~flame is lower and consequently the limit of detectability is about the same in both flames. This lower noise level in the 02-H2 flame was presumably due t o the much higher flame temperature which reduced scattering due to more efficient solvent vaporization. Bismuth. -4tomic fluorescence was observed a t 3068 A. Because of the close proximity of this line to the intense OH-bands, the noise level was somewhat higher than was the case with lead. Increasing the H2/02
I 200
flow ratio a t a fixed 0 2 flow sharply increased the fluorescence signal, but the resulting increase in flame size necessitated going to a greater height in the flame to improve the signal-tonoise ratio. The same increase in the signal-to-noise ratio could be attained by reducing the H2/Oz flow ratio slightly and using the same burner height used for lead. The analytical curve for bismuth is shown in Figure 3, In the $r-H2 flame, the atomic fluorescence intensity increased. However, the resulting increase in scattering necessitated using smaller slit widths which essentially cancelled the increase. Copper, Silver, and Gold. The analytical curves obtained for the atomic fluorescence of copper (3248 A), silver (3281 A,), and gold (2676 A.) are shown in Figure 2. The poorer sensitivity of gold compared to the other two is partly a result of a lower source intensity a t the gold line.
Experimental Conditions" Used and Limits of Detection Obtained for 14 Determinations 0 2 H2 flow, flow, Limit of Wavelength, liters/ liters/ Slit width, detectability, Element A. minute minute mm. p.p.m.
cu A; Pb
3248 3281 2676 4058
Ph
4n5x
Bi
3068 2852 2139 2288 3776 4227" 40OOc 468OC
^ I
m Zn Cd T1 Ca
Ba
Ga Ni
~
_ I _ _
337OC
2.40 2.10 2.10 2.25b 2 . 2~. 5 2.25 2.25b 2.25 2.25 2.28b 2.28 2.25 2.25 2.25
,
7.10 7.10 7.10 7.20 7.20 6.70 6.20 7.20 7.20 7.20 6.20 5.90 7.20 7.20
0.50 0.50 0.50 0.30 0.30 0.35 0.40 2.8 1.8 0.28 0.30 0.25 0.30 0.33
0.35 0.08 3.5 7.5 10 2 0.2 0.6 0.08 0.55 1.5 7 20 13
a Vertical height from bottom of slit to tip of burner was 6.0 cm., except Pb in Ar-Kt flame (6.5 cm.). b Argon flow (approx.). c Due to incident light scattering from salt particles.
206
ANALYTICAL CHEMISTRY
400
500
600
700
W A V E L E N G T H , rnQ
Table I.
AE
300
Magnesium. While investigating the atomic fluorescence of magnesium a t 2852 -4.in the 02-H2 flame, some very peculiar results were obtained which indicated that the signal observed was not due to the atomic fluorescence of magnesium atoms alone, but was also probably due to scattering by some molecular aggregates in the flame. In the Ar-HZ flame, the fluorescence was due to magnesium atoms. No evidence of scattering by molecular aggregates or molecular fluorescence could be obtained. The r e m s . noise level in the Ar-H2 flame was about twofold greater than in the 02-H2 flame, resulting in a net signal-to-noise ratio increase of five-fold. The analytical curve obtained for magnesium in the Ar-H2 flame is shown in Figure 2. Zinc, Cadmium and Mercury. Zinc gave atomic fluorescence a t 2139 A. Although the xenon arc source was extremely weak in intensity a t this wavelength, the fluorescence was quite intense. The analytical curve is shown in Figure 3. Note the large monochromator slit width used for this measurement (see Table I). Considerable improvement would certainly result with a more intense continuous source. I n the Ar-H2 flame, the fluorescence intensity was about 30Q/, greater than in the O2-H2 flame. Cadmium exhibited intense atomic fluorescence a t 2288 A. The wide slit width (Table I) indicates the low source intensity a t this wavelength. In the Ar-H2 flame, the fluorescence intensity was about 20% greater than in the 02H2 flame. The analytical curve is shown in Figure 3. Mercury exhibited very weak atomic fluorescence a t 2537 A., being about 10fold less than that of lead in the O2-H2 flame. Even in the Ar-H2 flame, the fluorescence intensity was very low. These results were somewhat surprising because mercury has little tendency to
form stable compounds in the flame, and t'he gA value (1) of the 2537 A. line is reasonably high. Considering only the gA values and the source intensity, mercury should exhibit fluorescence comparable to that of lead. The low sensitivity is possibly due to a low quantum efficiency for the 2537 A. niercury radiation. Thallium. All of the ground state thallium lines, and all of the lines originating from the 7793 em.-' state and having large gA vadues (1) exhibited atomic fluorescence. The 3776 A. line was the most intense. In the ,4r-H2 flame, the fluorescence intensity was about five-fold greater than in the 02H 2 flame. V7hile aspirating reasonably high concent'rations of thallium at a burner height of 6.0 em. and increasing the Hz flow above 7.20 liters/minute (Ar flow = 2.25 litersjminute), a sharp increase in the noise level resulted. This was due to the relatively intense thernial emission of thallium becoming visible t o t,he phot'odetect'or because of the increase in flame height. The optimuni burner height-flow ratio combination was taken as that' n.hich allowed the fluorescence t o be measured just above the visible flame. This same effect was observed for all of the elements exhibiting appreciable thermal emission and/or having lines in regions of high flanie background intensity. The analytical curve obtained for thallium in the .ir-Hz flame is shown in Figure 3. Other Elements and Observations. Khile investigating magnesium, calcium, barium, gallium, indium, nickel, and inanganiise in the 02-Hz flame, the photodetector signal decreased as the H z j 0 2flow ratio increased. This was contrary to what would be expected if the signal Ivere due to atomic fluorescence. One would have expected those results, however, if the signal were due to scat,tering by molecular aggregates or due to fluorescence of molecules rather than atoms. It was also noted that variation in the monochromator wavelength setting about the characteristic wavelengths of the various line gave little change in the photodetector signal. By scanning the wavelength, the measured signal was found t o exhibit structure inuch like band emission. Also, several of the elements present gave signals a t wavelengths Lvhere there were no lines and no known band eniission spectra due to coinpounds of these elements. The phase-sensitive det'ector system, while being insensitive to thermal emission from the flame, did respond to scattered radiation. There was then the possibility that the signal observed was a result of scattering from compounds (molecule aggregates with a diameter less than the wavelength of light used for excit'ation) resulting from the salt
introduced into the flame or from association of the element in concern with one of the flame gas products-e.g., 0, OH, etc.-or a result of fluorescence of molecules in the flame gases. The signal was not a result of water droplet scattering because this was electronically compensated for by iiieans of a zerosuppress. The signal was also not due to niolecular fluorescence (except for a small molecular fluorescence signal due to OH radicals). The oscillator strengths, f's, of molecular transitions for gaseous molecules are quite smalle.g., f's = for most molecules except for OH which has an f = 10-3, whereas the oscillator strengths for most resonance lines used in atomic fluorescence studies are of the order of unity. Therefore, molecular fluorescence Lvould only be expected to be detectable for very large sample concentrations-e.g., solution concentrations around lo5 p.p.m. Finally, recorded plots of signal vs. Lvavelength for 30 elements introduced as various salts into the 02-H2 flame were made. Each of these plots was corrected for the variation in the spectral response of the photodetector, the spectral output of the xenon lamp, and the transmission of the nionochromator as a function of wavelength. The resulting curves show the same characteristics as a typical scattering curve--Le., the response varies as 1/x4 ( 4 ) . In Figure 4, a typical uncorrected scattering curve (Curve -4) made with the heated chamber-type aspirator-burner to eliminate reflection and scattering from water droplets is given. Curve B in Figure 4 represents the spectral response curve of the xenon source-monochromator-phototube combination, and Curve C is the scattering spectrum denoted by Curve A corrected for instrumental characteristics. The spectra due to radiation from salt particles are exactly the sanie for all elements studied except for line fluorescence, and so only one typical spectrum (Figure 4, Curve C) is given. The relative signal due to scattering depends not only upon the concentration of the species introduced into the flame gaser but also on the type of element and the compound introduced into the flame. Because scattered radiation is known ( 4 ) t o be polarized, esperiments were also carried out using polarizers to determine whether the scattered radiation could be greatly reduced when the polarizing and analyzing prisms (polarizers) were oriented with their electric vectors perpendicular. Under these conditions, the scattering signal was reduced 20-fold compared to the scattering signal when the polarizers were oriented with paralled electric vectors. Another indication that the observed phenomenon was scattered radiation was evident from the shapes of the
analytical curves. (See Figures 2 and 3 for several representative plots.) -411 analytical curves of scattered signal us. sample concentration had the same slope within experimental error. The slope of the log-log plots is approximately 45" for all elements and for various wavelengths between 3000 and 6000 d. which indicates a direct proportion between the scattered signal and the number of scattering particles per cmS3of flame gases. Limits of detection for several elements obtained from the scattering curves are given in Table I. The scattering by salt particles is most likely a result of molecular aggregates of the salt introduced into the flame or of compounds foiined from the metal and flame gas products, such as 0, OH, etc. The scattering signal also increased wit'h decreasing fuel-to-oxygen flow into the flame which would tend to indicale that the scattering species might be osides or hydroxides. The use of organic solvents also decreased the scattering signal appreciably. The scattered signal a t 4000 of a 500p.p.ni. aqueous solution of a calcium salt reduced by 3.3-fold when 507, by volunie of methanol was used and by five-fold when a 7570 by volume methanol solution was used. Xcetone gave similar results. Magnesium in a OP-H2 flame gave appreciable scattering especially as the H2/0g ratio was decreased, but niagnesiuni in t'he *Ir--H? flame gave no scattering. Elements which are known to exist predominantly as compounds in H1-02 flames usually gave large scattering signals. Solvent droplets result in a large reflection signal when using the totalconsumption aspirator-burner, particularly with aqueous solutions. However a heated-chamber aspirator-burner gave only a negligible reflection signal. This large background signal due t o reflection and scattering of incident radiation from water droplets can therefore be minimized by using the heatedchamber aspirator-burner. It can also be substantially reduced by using organic solvents with the total-consumption aspirator-burner. For example a solution of 20% by volume of methanol reduced t'he reflected signal a t 4000 -4.by 2.5-fold, and 80% methanol reduced the reflected signal by 20-fold. =icetone gave similar results. In addition to decreasing the background signals, organic solvents also decreased the total peak-to-peak noise level due to reflection and scattering. '
CONCLUSIONS
Quite good sensitivities of analysis were obtained for 13 elements by flame fluorescence spectrometry using a small xenon arc as the source of excitation. Interesting, and sometimes anomalous, VOL. 38, NO. 2, FEBRUARY 1966
207
behavior was found for several of the elements in the Ar-H2 flame. This type of flame has never been used before, and the results obtained indicate that it should be investigated further. Scattering curves for several elements were obtained. The scattering is probably due to molecular aggregates of compounds formed with flame gas products. One may have noted in Figures 2 and 3 that the analytical curves for a few of the elements show something other than a first-order dependence of fluorescence intensity on concentration. The reason for this disagreement with the theory (6) is not known a t this time. The shape of the analytical curves of Zn, Cd,
and T1 are different from those obtained previously (5) with line sources. The reason for this is not known a t this time, It is possible that the continuous source may be of considerably more utility to atomic fluorescence than to atomic absorption flame spectrometry. The apparently wide spectral regions over which scattering is observed may cause some elements to interfere in a practical analysis. Also, scattering might constitute an interference for elements exhibiting atomic fluorescence. LITERATURE CITED
(1) Corliss, C. H., Bozman, W. R., Natl. Bur. Std. ( U . S.) Monograph 53, July
20, 1962.
(2) . . Dubbs, C. A.. ANAL.CHEM.24. 1654
(1954). ‘ (3) Fassel, V. A., Mossotti, V. G., Grossman, W. E. L., Kniseley, R. N., Pittsburgh Conf. Anal. Chem. Appl. Spect., 1 Qfi.5.
(4) Jenkins, F. A,, White, H. E., “Fundamentals of 0 tics,” p. 460, McGrawHill, New Yorf, 1957. (5) Mansfield, J. M., Winefordner, J. D., Veillon, C., ANAL. CHEW 37, 1049 (196:). (6) Winefordner, J. D., Vickers, T. J., Ibzd., 36, 161 (1964).
RECEIVED for review September 15, 1965. Accepted November 24, 1965. Work supported by Air Force Grant AF-AFOSR1033-66 and a NASA traineeship (J.M.M.).
Preconcentration and Spectrographic Determination of Ultratrace Metallic Impurities in Potassium Chloride M. C. FARQUHAR, J. A. HILL, and M. M. ENGLISH Trona Research Laboratory, American Potash & Chemical Corp., Trona, Calif.
b Ultratrace quantities of certain metallic impurities in chemical grade potassium chloride are critical in electrolytic potassium hydroxide cells. These impurities are preconcentrated by carrier precipitation with 8-quinolinol, thionalide, and tannic acid and determined spectrographically, similar to the method first proposed by Mitchell and Scott. However, in the present work, the method is extended in sensitivity and scope. A total of 39 elements is determined a t low partsper-billion concentrations. CONCENTRATIONS of certain E:oluble metallic impurities in potassium chloride brine cause excessive generation of hydrogen in electrolytic potassium hydroxide cells, similar to the effect reported by Angel and Lunden ( 2 , 3) in sodium hydroxide cells. Cell efficiency is reduced, and in addition the possibility exists for generating explosive mixtures of hydrogen in chlorine. 4 s a result, stringent purity requirements are imposed on chemical grade potassium chloride for these impurities. Target levels of about 6-p.p.b. NO,8-p.p.b. V, 10-p.p.b. Cr, 30-p.p.b. Co, and 50-p.p.b. Ni have therefore been indicated for this product. To achieve such purity, sensitive and precise analyses are required for monitoring production and for specification analyses of the potassium chloride product. The advantages of emission spectrography are apparent for the required multielement analyses. However, even
208
ANALYTICAL CHEMISTRY
the high sensitivity of spectrography is not adequate for determinations a t the concentrations mentioned earlier. It is necessary to preconcentrate the trace impurities before they can be determined. Several investigators have enumerated the advantages of preconcentration followed by spectrographic determinations. The preconcentration methods have for the most part utilized separations based on ion exchange, solvent extraction, or carrier precipitation, For example, Brody, Faris, and Buchanan ( 4 ) separated 0.1-p.p.m. quantities of impurities from plutonium and uranium by anion exchange prior to determinations by the copper spark spectrographic method. Van Erkelens (5) and also Koch (7) used solvent extraction for preconcentration of 0.1p.p.m. amounts of trace metals from biological ashes and from pure aluminum. The extracts were evaporated to dryness for subsequent spectrographic determinations, Carrier precipitation with organic precipitants was employed by Pohl ( I S ) for preconcentrating 0.1p.p.m. amounts of metallic impurities from aluminum prior to spectrographic determinations. Perhaps the best known preconcentration method is that of Mitchell and Scott (9, 10) for spectrographic determinations of trace constituents. Aluminum was used as a carrier in the precipitation with 8-quinolinol, thionalide, and tannic acid. The ignited precipitate was spectrographed to provide accurate determinations of 11 elements in soils and plant residues.
The method has the advantage that the impurities from large samples (for example, 50 grams of KCl) can be preconcentrated into 15 to 30 mg. of ignited precipitate. Heggen and Strock ( 6 ) , and Silvey and Brennan ( 1 4 ) used the precipitation method with indium as a carrier to determine 17 elements. Mitchell and Scott, as well as subsequent workers, found the method to be limited in sensitivity by impurities in the reagents. Below about 0.1 p.p.m., samples could not be differentiated from the blanks with sufficient accuracy. The present method provides for extensive purification of the reagents, and the blanks are sufficiently low in the critical elements to permit accurate determinations a t the required concentrations. The effect of reagent purification is illustrated in Figure 1, which is a photograph showing the spectra of blanks which were precipitated with specially purified reagents compared with those made with reagent grade chemicals. In addition, the present work has expanded the scope of the method to a total of 39 elements which can be preconcentrated and determined (Table I). EXPERIMENTAL
Preparation of Reagents. It is important that all reagents be as free as possible from metallic impurities. Methods are given here for preparation of reagents of the purity demonstrated in Figure 1. Purified reagents are always stored in polyethylene, rather than in glass, with the advantages pointed out by Thiers (15).
