V O L U M E 23, NO. 10, O C T O B E R 1 9 5 1 (2) Center, E. J . , and Willard, H. H., IND.ENQ.CHEM.,ANAL.ED., 14, 287 (1942). (3) Gentry, C. H. R., and Sherrington, L. G., Analyst, 71, 432 (1946). (4) Hawes, C. C., Mining Technol., 9, No. 1794 (1945). (5) Heczko, T., Chem. Ztg., 58, 1032 (1934). (6) Hill, U. T., Inland Steel Corp., private communication. (7) Johnson, H. 0.. Weaver, J. R., and Lykken, L., ANAL.CHEM., 19, 481 (1947). (8) Kassner, J. L., and Ozier, XI. -4., J . Am. Ceram. SOC.,33, 250 (1950). (9) Lang, R., and Reifer, J., 2. anal. Chem., 93,162 (1933). (10) Latimer, K. hi., and Hildebrand, J. H., “Reference Book of Inorganic Chemistry,” p. 286, New York, Macmillan Co., 1940. ( 1 1 ) Lundell, G. E. F., and Hoffman, J. I., “Outlines of Methods of Chemical Analysis,” p. 46, New York, John Wiley & Sons, 1938.
1455 (12) Ibid., p. 55. (13) Lundell, G. E. F., and Knowles, H. B., J . Research Xatl. Bur. Standards, 3, 91 (1929). (14) Moeller, T., IND.ENG.CHEM.,ANAL.ED., 15, 346 (1943). (15) Parks, T. D., and Lykken, L., ANAL.CHEM.,20, 148 (1948). (16) United States Steel Corp., “Sampling and Analysis of Iron and Manganese Ores,” p. 63, Method 111, Pittsburgh, Carnegie Steel Co., Bureau of Instruction, 1926. (17) Wiberley, S. E., and Bassett, L. G., BNAL.CHEY., 21, 609 (1949). ENQ.CHEM.,ANAL. ED., (18) Willard, H. H., and Center, E. J., IND. 13,81 (1941). (19) Willard, H. H., and Greathouse, L. H., J . A m . Chem. Soc., 42, 2208 (1920). RECEIVED January 30, 1950. Presented before the Division of Analytical Chemistry a t the 119th Meeting of the AMERICAN CHEMICAL SOCIETY, Boston, Mass. Reaearoh sponsored b y a grant from the Reaearoh Fund of the University of Alabama.
Flame Spectrophotometry for Determination of Sodium, Potassium, and Lithium in Glass Application of Analysis of Variance to an Analytical Problem EDWARD J. BRODERICK AND PHYLLIS G. ZACK Transformer & Allied Products Laboratory, General Electric Co., Pittsjield, Mass. An accurate flame spectrophotometric method was developed for the determination of alkalies in glass and similar refractory materials in place of the arduous, time-consuming gravimetric methods and a system of experimentation was applied, using a statistical design and analysis of variance, in order to minimize the amount of experimentation needed when working with a large number of instrumental and interference variables. The results were equal to or better than the classical gravimetric methods. This method compares favorably in accuracy with gravimetric methods and requires only one fifth the time. The application of analysis of variance to an analytical problem has proved very satisfactory in this work, and should be very useful to the analytical chemist in similar problems.
this solution, the main interferences will be from the effect of the presence of alkali metals on each other. Using the flame photometer in conjunction with the Beckman Model DU spectrophotometer, certain physical conditions must be controlled to eliminate excessive variability. These include flame background, reproducibility of conditions in the atomizer assembly, stability of the flamr, which depends on gas, air, and oxygen pressure, cooling water temperature, and spectral band width. Some quantitative measure of the experimental error must also be made. T o obtain a fairly comprehensive picture of the effect of small changes in these various conditions, the experiments were designed in the form of Latin and Greco-Latin squares. These designs made it possible t o study thwe or four main effects at the same time, with a n absolute minimum of experimenta APPARATUS
T
HE emission characteristics of the alkali metals in a hot &me (9, 12, 13) and the application of this phenomenon in
flame spectrophotometry t o the determination of small amounts of these elements (1, 2, 5, 11, 16)are well known. However, in the application of this technique for the analysis of any material in which the three alkalies are present together, a large number of instrumental and interference variables immediately become apparent. T h e time and labor required t o control these variables can be minimized by using an experimental design, which follows the statistical principles developed by Fisher and his co-workers (6). Such a design has proved very satisfactory in the present problem, and should provide a useful tool for the analytical chemist in this and similar work. I n analyzing a borosilicate gIass-60% SiO,, 20% B203, 5% KazO, 5Yc Li,O, 2% K,O, and the remainder R20s-it seemed probable that the alkali metals couid be determined using a flame spectrophotometric technique in place of the arduous, timeconsuming classical methods (8, 14). In the preliminary steps of dissolving the sample, the bulk of material is removed, leaving the sodium, potassium, and lithium in solution. If flame photometric determinations are made on
A Beckman DU spectrophotometer with flame photometer attachment was used. The fuel su ply was bottled propane gas of ordinary cooking quality. Bottgd oxygen and bottled nitrogen, which was used wherever compressed air is called for, nere from a factory supply. A resistance box assembly w’m installed on the phototube housing to permit t h e quick interchange of resistors for emission and absorption work. This was installed late in the work to permit the use of the spectrophotometer for other laboratory purposes, without the necessity of effecting a manual exchange of resistors, with the consequent delays in obtaining equilibrium conditione within the housing. Reagents all conformed t o ACS s ecifications. Volumetric equipment was all of &weau of Standards certifica- tion. X&ol-Vit brand glassware was used for the storage of the standard solutions of the alkali metals. STANDARD SOLUTIONS
T h e reagent grade chloride salts of sodium, otaesium, and lithium were dried a t 110’ C . to constant weiggt. A suitable weight (for final concentration of 1 mg. per ml. as the oxide) of the dried salt was dissolved in a volumetric flask. From this stock solution, suitahle aliquots can be diluted for comparison standards.
1456
ANALYTICAL CHEMISTRY Table I.
Typical Experinlent I
111 30 Ib 2a
I1 2b 3s
la 2c 3b
-1 B C
IC
Symbol Representation Oxmen, Inches 45 I I1
o;
111
DJ
Gas, Cm. Air, Lb. ri 2.0 1 15 2 17.5 B 1 7 C 2.3 3 20 \?-are length 770 nip Slit width 0 5 mni. Sensitivity 0 ,1
Potassium Concentration, P.P.hI. a 10 b 25 c 38
EXPERIhIESTAL
The instrumental conditions for reproducibility and stability were considered first; then the proper flame conditions and interference effects were examined st,at.ietically. Flame Background. T h e spectrum of the flame was established by setting the instrument t o zero with the shutter closed, and then rebslancing the galvanometer circuit wit,h the shutter open. The intensity \vas obtained a t various slit widths, as shown in Figure 1. Over the 400- t o 700-mp range the flame background is less than 0.25y0with a 1-mni. slit width and much less with t h r smaller widths. I n a n analysis, the ratio of flame background to snmplc intrnsity should he held as small as possihle.
Flame Conditions. Because of the large number of possible settings of air, oxygen, and gas pressures which must be measured t o obtain the optimum emission intensity in the determination of each of the elements, these settings were made the variables in a Greco-Latin square design t o determine the best relationship of the conditions in a determination. Based on preliminary designs, which showed t,he ranges of greatest promise, a typical experiment was designed (:is shown in Table I ) t o cover a n extensive study of these ranges. Using this design, nine different settings of the gas, oxygen, and air pressures mere niade with the three potassium solutions. I t will be seen that, in the design e:tch level of each variable appears once, and t h a t there is no unnecessary repetition. The great advantage of this design, aside from the srnall nuniber of experiments necessary, is the amount of information that can be obtained from it statistically. Knowing t h a t the sum of squares [Z(r - ?)*I has a n additive property in relation t o the total variation within t h r squaw, we can break down the variation into it,s compon >nts-i.e., the variation in results caueed by changes in gas pressure, the variation caused by changes in oxygen pressure, etc. The total variation equals the sum of the individual variations, plus any remainder which is assigned to experimental error. From st,atistical tables, the importance of the individual variation?, or the probability of their heing dur t o chance alone, can be determined. Thr actu:tl values so obtained are shown in Table 11. The arithmetical calculations can be found in most trstbooks on statistical and experimental design and ci.itica1 values are taken from tables of statistics. The hypothesis is formed t h a t there is no variation due t o a given condition :rnd t,hix hypothesis is either :tcc?pted or rejected. Consider the F value for air in Table 11. (It is the ratio of mean square due to air over mean square due to experimental error &here the mean square is a n estimate of the variance, u * , ) The critical values are 3.56 and 6.05 for 2 and 18 degrees of free-
Table 11. 1.03 1.07 0.91 4.63 4.85 4 03
2 79 3 02 2 75
wwf L m G m IN mumcrnfi Figure 1. 1.
10-mm. slit.
2.
3 4i
Flame Intensity 0.5-mm. slit.
3.
0.2-mm. slit
Flame Stability. Plotting flame intensity in per cent transmittance (the inst,rument scale readings, which are used as the unite of intensity throughout this work) against time, equilibrium was obtained in 60 seconds and held for 4 t o 5 minutes. Four condit,ions must b r oht:iined before consistent readings ~ can 1 1 made. The sample beaker niui;t he filled to a predetermined level. T h e atomizer and preheating chamber must be saturated with a strong (100 to 200 p.p.m.) solution of the element to be determined, and then flushed with distilled water until the flame background remains constant. This was found to be the quickest way of obtaining equilibrium conditions, and was required only at the beginning of a series of measurements. The atomizer was flushed, and cleared back as described in the literature (fO)! at frequent intervals to ensure uniform conditions at the atomizer nozzle. It was suggested t o the authors that a porous thimble over the atomizer nozzle would eliminate the obstruction by microscopic particles, yhich a t times produce results t h a t can only be attributed t o their presence. Cooling water flow must be controlled so that the effluent is just below the boiling point.
Factors Concn. Oxygen Gas Air Experimental
Experimental Values
Readings, % Transinittance 2.89 3.14 2.69 1 33 1 32 1 14 3 45 3 78
Sum of Squares 44.2618 0.2035 0.6555 2.9057 1.2037
Degrees of Freedoni 2
2 2 2 18
Mean Square 22.1309 0,1018 0.3278 1.4528 0 0669
4.90 5.15 4.19
2.27 2.10 1 85 1.03 1.22 1.17
Calcd.
Critica F
F
330.81 1.52 -1.89 21.71
Fa.01 = 6.0; Fo.06 = 3 . 5 6
Sum of squares = z(Z: - 512 = ~ ( z ? ) Degrees of freedom = one less than nuiiiber of levels of each variable sum of squares hIe:n square = degrees of freedom I." = mean square of condition mean square, experimental
Table 111. so.
I B C D
E €
G H I
iYax0, P.P.?d. _____-
I I1
I11 I
I1 111 I I1 111
10
25 50 10
25 50 10 25
50
Solutions L-sed KrO, P.P.M. a
a
Ylb b c c c
10
10
10 25 25
25 50 50 50
LixO, P.P.11: 1
2 3
2
10 25
50
2.5
3
1
50 10
3 1 2
50 10 25
1457
V O L U M E 2 3 , NO, 10, O C T O B E R 1 9 5 1 Table IV.
Determination of Lithium in Presence of Sodium and Potassium
further attention in actual dctcrminations, t o ascertain n-hcther the errors so introduced xverc more serious than the errore n-hich arise from variations in physical conditions.
Design Sndiiini
.-
2 -
11
1.i:
].is
1.i)
I)
E
1;
r
I.i3
I,il
1.h