1304
Anal. Chem. 1982,5 4 , 1304-1308
lining of the chromatographic outlet and the mass spectrometric inlet tubing should reduce this unwanted additional dilution by the make-up gas. Table IV shows also that the base line noise at low flow rate is generally somewhat higher, especially in the case of the open tubular column. This result is probably caused by a not sufficient depression of the air diffusion into the ion source of the mass spectrometer by the scavenge flow of helium. A decrease of the noise level of one-third to one-tenth of the present values seems to be possible if the interface is designed in such a way that the air diffusion into the ion source is avoided more effectively. ACKNOWLEDGMENT The assistance of B. Goose in writing the paper is appreciated.
(7) McFadden, William H. J. Chromatogr. Scl. 1980, 18, 97. (8) Kenndler, Ernst; Schmid, Erich R. ”Instrumentation for High-Performance Liquid Chromatography”; Huber, J. F. K., Ed.; Eisevler: Amsterdam, Oxford, New York, 1978; Vol. 13, pp 163-177. (9) Ten Noever de Brauw, M. C. J. Chromatogr. 1979, 165, 207-233. (10) J . Chromafogr. SCl. 1979, 77, 1-112; special issue “Gas Chromatography-Mass Spectrometry”. (11) Huber, J. F. K. 2.Anal. Chem. 1975, 277, 341. (12) Huber, J. F. K. Chlmk Suppl. 1970, 24. (13) Huber, J. F. K.; Hylsman, J. A. R. J. Anal. Chim. Acta 1967, 38, 305. (14) Huber, J. F. K.; Hulsman, J. A. R . J.; Meijers, C. A. M. J. Chromatogr. 1971, 6 2 , 79. (15) van Deemter, J. J.; Zuiderweg, F. J.; Kllnkenberg, A. Chem. Eng. Sci. 1956. 5 , 271. (16) Golay, M. I n “Gas Chromatography, 1956”; Desty, D. H., Ed., Butterworth: London, 1958; p 36. (17) Huber, J. F. K.; Lauer, H. H.; Poppe, H. J. Chromatogr. 1975, 112, 377. (18) Giddings, Calvin; Seager, Spencer L.; Stucki, Larry R.; Stewart, Georg H. Anal. Chem. WSO, 3 2 , 867-870. (19) Sternberg, J. C.; Poulson, R. E. Anal. Chem. 1964, 3 6 , 58-63.
RECEIVED for review October 20, 1981. Resubmitted April 5, 1982. Accepted April 5, 1982. The paper was presented at
LITERATURE CITED Holmes, J. C.; Morell, F. A. Appl. Spectrosc. 1957, 1 7 , 86. Gohlke, R. S . Anal. Chem. 1959. 31, 535. Henneberg, D. 2.Anal. Chem. 1961, 783, 12. Watson, J. T.; Biemann, K. Anal. Chem. 1964, 36, 1135. Rhyhage, R. Anal. Chem. 1964. 3 6 , 759. Grotch, S. L. Anal. Chem. 1970, 42, 1214.
the 6th International Symposium “Advances and Application of Chromatography in Industry”, Bratislava, Sept 16-19,1980. We wish to express our gratitude to the SEA Foundation for financial support of the project.
Identification of Crude Oils by Selective Chemical Ionization Mass Spectrometry’ P. Burke and K. R. Jennings” Department of Chemistry and Molecular Sciences, University of Warwick, Coventry C V 4 7AL, England
R. P. Morgan” and C. A. Gilchrlst Shell Research Limited, Thornton Research Centre, P.O.
Box 1,
Chester, England
OH- chemlcal lonlratlon has been used to Ionize selectively a serles of crude olis. The results show that this method can provlde positive ldentlflcation of the geographlcai area of orlgln of an oil and in the majority of cases can ldentlfy the individual flelds.
One of the problems of electron impact mass spectrometry is that a gross mixture of 100-200 compounds can yield as many as 5000 significant (intensity >0.1% of the largest peak) ionic species. Thus, much time must be spent on interpretation when a total analysis is required. Use of a soft ionization process such as field desorption (FD) (1)or chemical ionization (CI) (2) can reduce the number of ionic species formed, by reducing the amount of energy imparted to the ions during the ionization process. Recently, further simplification of spectra has been demonstrated under CI conditions by using special reagent gases which selectively ionize only certain types of molecules. By use of this latter technique, a method has been developed that should provide a fast and efficient method for the identification of crude oils. Analysis of Crude Oils. Considerable effort has been devoted to the characterization and identification of crude oils by environmental protection agencies and the petroleum industry. Unfortunately no single technique exists which is T h i s paper is dedicated to t h e m e m o r y o f Geoff Steel who tragically d i e d March 29, 1980, and who i n i t i a t e d and inspired this work. 0003-2700/82/0354-1304$01.25/0
sufficient for crude oil identification and hence a number of complementary methods have to be used. These methods primarily include (i) low resolution gas chromatography ( 3 4 , (ii) neutron activation (7), (iii) X-ray fluorescence spectroscopy, (iv) sulfur-print gas chromatography (8), (v) infrared spectroscopy in the 2-15 jtm range (9, IO), (vi) 13C nuclear magnetic resonance spectroscopy (II), (vii) gas chromatography selective ion monitoring (12-14). What is required is a simple and rapid method of crude oil identification which can be used to identify a spilled oil from a short list of possible candidates. The various processes involved in weathering deplete the classes of hydrocarbons present in oils at different rates. Of the four main weathering processes that of biological degradation is known, for example, to deplete n-alkanes at a faster rate than branched alkanes (15). The different physical properties of various hydrocarbon classes (boiling points, aqueous solubilities, etc.) are reflected in the differential rates of depletion by the other main weathering processes of evaporation (the most rapid and largest bulk effect) (16),solution, and photolysis. What is desirable, therefore, in addition to the above mentioned rapidity and ease of data analysis, is a technique that observes as few compound classes as possible, in order to reduce the interference of competing weathering processes, whilst retaining sufficient diagnostic power for crude oil identification. OH- Chemical Ionization (CI). An ionization technique which would yield molecular ion profiles will have potentially 0 1982 American Chemical Society
ANALYTICAL CHEMISTRY, VOL. 54, NO. 8, JULY 1982
1305
SEPTUM INLET
VA R1ABLE
2~a!o OIL/HEXANE
D MAGNET
Flgure 1. Representation of apparatus used to measure the OH- chemlcal ionlzatlon mass spectra of the crude oils.
greater diagnostic power than any mass spectral technique in which contributions are made to peaks from higher or lower mass, e.g., fragment ions in 70 eV electron impact spectra and addition ions in positive chemical ionization spectra. Such a technique is less susceptible therefore to differential rates of weathering and will have greater reproducibility. The excess energy involved in OH- proton transfer reactions can be considered to ble partitioned mainly in the new bond formed in the neutral (HO-H)
OH- + M-H e M-
+ HzO
so little fragmentation of the resulting anion is seen. For all components which undlergo this type of reaction the desired molecular weight ion profile is produced. Due to the relative strengths of the gas phase acidities, the thermodynamic center of the above reaction lies to the left when MH is benzene, a cycloalkane, or an alkane. For the alkylated aromatics and heteroatomic compounds the thermodynamic center lies to the right, so that only these compounds will be ionized by reaction with OH- in a negative chemical ionization source. Therefore straight, branched, or naphthenic ring alkanes do not contribute to the mass spectrum of the bulk mixture and this allows a further simplification of the effects of weathering since those compounds which do contribute to the spectrum have more uniform weathering characteristics. A similar analysis with an alternative soft ionization technique could be achieved only by means of a time-consuming chromatographic separation. The OH- plasma Waf3 originally produced by a mixture of nitrous oxide and methane. Production of OH- occurs by the process of dissociative resonance capture
N20 + e-
-
0-.
+ Nz
followed by a fast H atom abstraction
0- + R-H
R. + OH-
In the case of methane R = CH,. However, problems can occur in using this mixture, and in many cases the spectra are complicated by peaks ascribed to [M - H + NzO - HzO]- as well as those due to [M - HI-. As a result, we have chosen to use n-hexane as the hydrocarbon, R-H (equation above), because it does not yield the extraneous high mass adduct ions (18). EXPERIMENTAL SECTION All the experiments were carried out at the University of Warwick, using the apparatus dhown in Figure 1. The apparatus is adapted from the work of Sieck (19,201. A 1O-wL sample of a 3% v/v solution of the crude oil in hexane is introduced into the glass vessel which is at 225 "C. The vapor formed by the solution is allowed into the ion source through a capillary to prevent discharges of the mass spectrometer's high voltage to the glass vessel surrounds. N20is introduced into the source through a separate inlet where it is allowed to mix with the oil/hexane vapor for 30 s. The gaseous mixture is then bombarded with electrons to form OH- ions which in turn ionize the aromatic fraction of the crude oil. The gases are introduced such that the ratio of N20to hexane is 1:1, yielding a total pressure in the source of 0.4 mbar. A mass spectrum cm then be obtained with an AEI 1073 mass spectrometer. A typical spectrum over the lower end of the mass range is shown in Figure 2. Once a spectrum has been obtained, the crude oil/hexane vapor is pumped out through valve (A) and the system flushed with pure hexane. The time taken for the entire operation is approximately 5 min. Operating conditions are as follows: temperature of glass line, 230 "C; source temperature, 260 "C; electron energy, 320 eV; accelerating potential, 1 kV; source pressure, 0.4 mbar; scan speed, 100 s/decade; mass resolution, 1000 (at 10% valley). ANALYSIS OF RESULTS The results of measuring the OH- CI mass spectra of 17 different crude oils have been analyzed by using a "nearest neighbor" method. Each spectrum is represented by 30 mass peaks in order to reduce the computing time of the statistical
1306
ANALYTICAL CHEMISTRY, VOL. 54, NO. 8, JULY 1982
Table 11. Mismatch of the Repeat Runs of D, with the Spectrum Used in the Reference Data Set and the Other Spectra in the Reference Set Illustrating the Reproducibility of the Technique crude oil Dl (av)
D, D, D, Dl Dl *
I
*
A,
I
A2
A, A,
c3
Figure 2. OH- chemical ionization mass spectrum from a typical crude oil, over the mass range m l z 100-270.
Table I. Peaks Used as a Fingerprint of the OHCrude Oil Mass Spectra 181 195 209 23 7 251
183 197 211 225 253
185 199 213 227 241 255
187
189
201
203 217 231 245 259
215 229 243 257
247 249
calculations. These peaks are marked in Figure 2 and are listed in Table I. Each peak is made up from contributions from (M - 1) ions given by various isomeric compounds from one main class. For example, the C,H2,4 peaks ([M - HI is m / z 91,105, ...) are alkyl benzenes and the CnH2n-12([M - HIis m / z 141, 155, ...) are alkyl naphthalenes. These are the dominant peaks below m/z 180. However, initial studies have shown that rapid evaporation of hydrocarbons containing up to approximately 14 carbon atoms takes place, making these peaks highly unsuitable for a fingerprinting technique of weathered oils. The selection of the 30 peaks was based on the inclusion of the CnH2n-12and the CnH2,+ ions for each carbon number (C14 and above), and the homologous series ions which have intermediate m / z ratios, i.e., CflH2,-lO and C,H2,+ Peaks above m/z 259 were not chosen for two reasons. Firstly the intensity of such peaks compared to ions in the range 181-259 gives them a very small contribution to the measure of similarity, the Euclidean distance metric, which can be dominated by a few large differences. Secondly at the highest operating temperature of the reservoir (225 “C), some of the components contributing to ion intensities around m / z 300 and above may be insufficiently volatilized introducing large variations of peak size with very small variations of reservoir temperature. Each spectrum was normalized so that the total ion current of the spectrum was equal to 100 arbitrary units. Initially, comparison of a spectrum with the reference spectra was made by using the following approach. The algorithm used is represented by the equation 30
F = C (I(U), - I(R)rJ2 n=l
where I(U), and I(R), represent the intensities of peak n in the unknown and reference spectrum, respectively, and F represents the degree of fit (or mismatch) between the two spectra (21). Thus if the spectra U and R were identical, F
(5) (4) (3) (2) (1)
mismatch crude oil 0.0 0.9 2.0 2.3 2.3 2.6 5.5 7.1 7.2 7.4 8.6
B3 A3 A, A,
c,
B2
B4
B, c2 ’6
Bl
mismatch 10.5 11.2 11.9 12.1 12.5 13.9 14.2 14.3 14.7 15.8 16.3
would be zero, while the larger the differences the larger would be the values of F (up to a value of 141.41). In the reduction of each spectrum to 30 peaks, each oil is effectively represented as a point in a 30-dimensional plot. This approach provides a method of obtaining a measure of the distance between one point (the unknown) and each of the other points in this 30-dimensional space. However, this method has recently been improved by weighting each peak’s contribution to a classification metric according to its reliability. These weights are calculated by finding the standard deviation of each peak over a series of replicate runs of one oil, spread over a period of months. The weighting factor, which is the inverse of the standard deviation, is a measure of the reproducibility. Using these weighting factors, one obtains a greater spacing between dissimilar oils in the 30-dimensional space. As a result the “resolution” of the technique has been improved.
DISCUSSION OF RESULTS The crude oils used to evaluate the technique were taken from four geographical areas which cover the main oil-producing countries. They are represented by Al to As, B1 to B,, C1 and C2, and D1. The letters signify the different geographical areas while the subscripts represent the different fields within those areas. The oils were chosen to cover as wide a range in both type and geographical origin as was possible from the available crude oils, The first experiments carried out were to test the reproducibility of the experimental technique. This is not a test of the reproducibility of the method. A test of the value of the method for typing crude oils would require examination of the oil from different wells from the same field obtained over a long period of time. Such a test would yield a measure of the variation within a given oil field and whether this method can distinguish between oil from wells within a given field. Test of Reproducibility. The reproducibility of the technique was investigated by comparing the differences observed in spectra obtained at monthly intervals of the same crude oil with the differences observed between different crude oils. The results for oil D1are shown in Table 11. The spectrum resulting from the average of the five repeat runs is compared with the five repeat runs individually and with the rest of the oils in the reference library. It can be seen that the differences between the repeat runs are considerably less than the differences between D1 and the other crudes. As a result, any major differences between spectra must reflect differences in the chemical composition of the oil, as only relatively minor differences will arise from the use of the experimental techniques described above. These smaller intraclass differences may be used to calculate a 95% confidence limit (one-tailed) so that a short list of closest possible candidates may be drawn up. In Table 11, use of the t distribution with four degrees of freedom (for the five rep-
ANALYTICAL CHEMISTRY, VOL. 54, NO. 8, JULY 1982
Table 111. Top Ten Mismatches of Unknown I (C,) with the Reference Library of Spectra crude oil
mismatch
crude oil
mismatch
1.0
BS
5.7 6.1 6.6 6.7 7.4
4.2 5.3 5.5 5.6
B, A,
Table VI. Top Ten Mismatches of Unknown IV (A,) with the Rest of the Oils in the Reference Library of Spectra crude oil
mismatch
crude oil
mismatch
Aa
0.9 1.0 1.3 1.5 1.8
A3
2.2 2.8 3.8 4.0 4.3
A1 A5 A8
A4
Table IV. Top Ten Mismatches of Unknown I1 (D,) with the Reference Library of Spectra crude oil
mismatch
crude oil
mismatch
Dl
0.6 3.0 4.1 4.1 4.4
A4
5.2 6.3 7.8 8.4 8.6
A6
A2 A1 A5
B3
Cl Ba B4
Table V. Top Ten Mismatches of Unknown I11 (B,) with the Reference Library of Spectra crude oil
mismatch
crude oil
B4
1.0 2.3 3.1 3.8 4.1
B6
Bl B, B3
B2
Ca A4
AS A2
mismatch 4.5 5.0 5.6 6.1 6.5
A, A6
Ba
D,
Table VII. Top Ten Mismatches of an Oil (V) from Outside the Reference Library of Spectra crude oil
mismatch
crude oil
mismatch
A, Aa A, A, B,
4.2 4.6 5.5 5.8 5.8
A8 BS
5.9 6.8 7.1 7.6 7.6
Dl A3
Table VIII. Top Ten Mismatches of B3 and a Contaminated Sample of B, with the Reference Library of Spectra crude oil B3 B, a B2 B6
licate runs) gives a 95% confidence limit of 2.8. Here, no other reference oil spectrum Falls within this limit-the technique has sufficient diagnostic power to differentiate between D1 and all other library oils. In several cases reference oils fall inside the confidence limit of an oil from the same region. This suggests a common geological origin of the oils in question, discrimination betweein which is likely to be much more difficult. Test Matching of “TJnknowns”. The results of matching the partial mass spectrum of the unknown “I”, which was in fact C2,with the reference spectra are shown in Table 111. The mismatch between the unknown and Cz is 1.0. This value lies well within the instrumental reproducibility of the spectra (vide supra) and so the unknown could be positively identified as Cz if the number of‘ possibilities were limited to the 17 crudes used in this work. Note that C2 is clearly distinguishable from C1 crude, also from the same geographical area. Unknown “II“ (Dl). The results of matching unknown “11”, which was D,, with the mass spectra in the reference set are shown in Table IV. Again the best mismatch is with D, and the figure lies within the value expected for variations within the experimental techniique. Therefore, it would appear that it is possible to positively identify both of these first two unknowns from the rest of the reference spectra. Unknown “111”(B4). The results for the match of unknown I11 (B4) with the reference spectra are shown in Table V. In this case I11 can almost be positively identified as B4, although B1 cannot be completelly ruled out as it lies on the limits of instrumental reproducibility. Note how all the oils from geographical area B have grouped together a t the top of the mismatch table. This property should enable oils from outside the reference library to be assigned to their correct geographical are (see the section Test Matching of a Crude Oil (V) Outside Reference Library). Unknown “1V” (Al). The results of matching unknown IV (Al) with the reference apectra are shown in Table VI. In this case positive identification of the unknown cannot be made as a large number of oils from this geographical region fell within the reproducibility limit. However, the correct oil
1307
A, Bl A2 B4
AI BS
mismatch with B, 0.0 5.2 2.7 3.9 6.0 3.1 8.4 4.5 8.2 7.0
mismatch with B, (contaminated) 5.2 0.0 7.2 7.3 7.4 8.4 8.9 8.9 9.4 12.2
Contaminated.
(A,) is predicted as one of the two most likely fits for the unknown. Test Matching of a Crude Oil (V) Outside Reference Library. In order to examine how the method would react to an oil from outside the reference data set, we ran a sample of crude oil from another field and matched that against the reference spectra. The results are shown in Table VII. In this case, no positive mismatch (a score less than the reproducibility limit) is made, and so the possibility that this oil comes from outside the reference set is clearly suggested. In addition, as the four best fits to this oil are with crude oils from the geographical area A, a clear indication that the oil is a crude from that area is presented. Analysis of a Contaminated Crude Oil. All the previous “unknown” samples had been taken from shipments of oil of the same period as the reference oils, but during the course of this project a sample of a B3 crude was acquired which came from a much later shipment than the reference oil. It was suspected to have been contaminated during shipment with small amounts of another crude oil. The mismatch of this oil and the reference B3 oil with the rest of the reference spectra is shown in Table VIII. The scores of 0.0 in Table VI11 are perfect matches, as would be expected from comparing a spectrum with itself. It is clear, from the fact that the smallest mismatch with the contaminated oil is B3, that the bulk of the oil is B3. Nevertheless, the mismatch of 5.2 (vide supra) shows that the oil has been contaminated. However, if the contaminant was another oil from the reference set, it would be expected that the contaminated oil would score a smaller mismatch to the contaminant than the reference B3 oil. It can be seen from Table VI11 that this does not occur.
ANALYTICAL CHEMISTRY, VOL. 54, NO. 8, JULY 1982
1308
P ? I Y C I P A L COMPONENT ANAL"SIS,EXCL.UDING
M/E 217
REGIONAL. GROUP A N A L Y S I S
3.5 DD
3.3
B
2,s
0
3
+
2.0
CL
:.5
3
+
il 4 5.0 U
0.5 0.0 -0.5 -1.0
-1.5
-2.0
c
FACTOR ONE (DATA FROM
SCODFILE/MINUSFl
Flgure 3. Plot of the first and second variates in which each crude oil is represented by a point ( x , y ) (see text).
METHODS OF GRAPHICAL REPRESENTATION It would be advantageous to obtain plots of the distribution of the 30 dimensional spectra as an aid to interpretation and as an assessment of the technique. Various methods have been discussed (22, 23) but the two methods chosen here are principal component analysis and discriminant function analysis. P r i n o i n - l Pnmnnnan) A n a l v a i c l d P P A \
T n PPA
nnw
variates are created which are linear combinations of the original (30) measurements but which contain the maximum amount of total variance possible in the first variate, the greatest amount of the remaining total variance in the second variate, and so on. A plot of the first and second variates will therefore contain far greater variance than is possible in a plot of two of the original variables (peak intensities). Figure 3 shows such a plot of the first two factors. For these analyses all the peaks were autoscaled, i.e.
where z = new autoscaled score for peak j , oil i, x i j = normalized intensity of peak j , oil i, %j = mean of peak j , aj = standard deviation for peak j . This was carried out to ensure that each peak would have an equivalent contribution before the PCA was brought into operation. No weighting was introduced. The plot contains 75% of the total variance. Factor one, which separates out regions A, B, and C from each other is dominated by contributions from the CnH2n..14series and covers the lighter mass end of the 30 peaks. Some of the oils in region A are pairs of runs from the same field, which accounts for the overlapping of spectral points. The second factor is dominated by peaks from the high mass end of the range and separates out the heavy oils from region C. With a more comprehensive choice of oils and a weighting scheme, a better overall spread and contribution to factors should be achieved. Discriminant Analysis. The second visual display technique involves the pregrouping of data into, in the case of this oil identification system, the different regional groups. Discriminant analysis separates the specified regions into
L
8
8
-7
-6
-5
8
-4
c ,
8
8
-3
-2
-I
CANON. V A R I A B L E 1
,
0
1
I
1
2
#
#
3
(DATA FROM
I
4
,
5
,
6
7
BIGFILE)
Flgure 4. Plot illustrating regional group analysis of the crude oils and indlcatlng the reliabilii of the chosen peaks In differentiating between oils of different regions.
clusters, using the minimum number of peaks; these peaks are the best at interclass distinction and may be more heavily weighted when an indication of the geographical origin of an unknown sample is required. Two new variables are created which are combinations of the peaks chosen to distinguish between classes; a scatter plot of these is shown in Figure 4. As expected, no mismatches occurred, which confirms that the selected peaks can be used to discriminate between samples of different chemical composition derived from different geographical regions. LITERATURE CITED (1) Beckey. H. D. "Principles of Fleld Ionization and Fleld Desorption"; Pergamon Press: Oxford, 1977. (2) FleM, F. H. I n "MTP International Review of Sclence: Physical Chemistry, Series One"; Maccoli, A., Ed.; Butterworths: London, 1973; Vol. 5. (3) Ehrhardt, M.; Biumer, M. Envlron. Pollut. 1972, 3 , 179. (4) Rasmussen, D. V. Anal. Chem. 1976, 48, 1562. (5) Cole, R. D. Nature (London) 1971, 233, 546. (6) Adlard, E. R. In "Chromatography In the Petroleum Field", Gould, T. H., Ed.; Marcel Dekker: New York, 1976. (7) Dnewer, D. L.; Kowalski, B. R.; Shatzki, T. F. Anal. Chem. 1975. 47, 1573. (8) Adlard, E. R. J. Inst. Pet. 1972, 58, 63. (9) Brown, C. W.; Lynch, P. F.; Ahmadjian, M. Appl. Spectrosc. Rev. 1975, 9 , 223. (IO) Mattson, J. S.; Mattson, C. S.; Spencer, M. J.; Starks, S. A. Anal. Chem. 1977, 49, 297. (11) Shooley, J. A.; Budde, W. L. Anal. Chem. 1976, 48, 1458. (12) Gallegos, E. J. Anal. Chem. 1975, 47, 1524. (13) Albaiges, J.; Albrecht, P. Int. J. Anal. Chem. 1079, 6 , 171. (14) Koons, C. B.; Rodgers, M. A.; Mercer, J. N.; Flory, D. A.; Rubensteln, A. E.; Lichtenstein, H. A. I n "Pattern Recognition: Oil Spill Identificatlon" (Conf Proceedings); Coronado, CA, 1977; p 151. (15) Nagara, S.; Konds, G. API Pub/. 1076, 4284, 617. Bentz, A. P. "Pattern Recognltlon: Oil Spill Identification" (Conf Pro(16) ceedings); Coronado, CA, 1977; p 21. (17) Smlt, A. L. C.; Fleld, F. H. J . Am. Chem. SOC. 1977, 99, 6471. (18) Sieck, L. W.; Burke, P.; Jennings, K. R. Anal. Chem. 1979, 57, 2232. (19) Sieck, L. W. Anal. Chem. 1079, 51, 128. (20) Sieck, L. W. Chem. Brit. 1980, 16, 38. (21) Sokal, R. R. Syst. Zool. 1961, 10, 70. (22) Kowalskl, B. R.; Bender, C. F. J. Am. Chem. SOC. 1972, 9 4 , 5623. (23) Kowalskl, B. R.; Bender, C. F. J. Am. Chem. SOC. 1973, 9 5 , 686.
RECEIVED for review March 27, 1981. Resubmitted January 25, 1982. Accepted March 31, 1982.