Reliability in the analysis of rocks and minerals - Analytical Chemistry


Reliability in the analysis of rocks and minerals - Analytical Chemistry...

1 downloads 117 Views 6MB Size

Reliability in the Analysis of Penland

A.

Dane

by

right)

and

(cover

photos

Smithsonian

Sydney Abbey Geological Survey of Canada 601 Booth St. Ottawa. Ontario, Canada K1A 0E8

Rocks and Minerals We are all familiar with the many spectacular advances in analytical methodology in recent years. We have

striking improvements in selec-

seen

tivity, sensitivity, precision, accuracy,

and speed. The analysis of rocks for geological studies has been no exception in these developments, but the matter of overall reliability is another question. What is so special about rocks?, you may ask. One might as well ask: What is a rock? From a chemical point of view, a rock may be regarded as a nat-

urally occurring, essentially inorganic material, a heterogeneous mixture of solid phases, called minerals, with highly varied chemical, physical and mechanical properties. A rock may contain any or all of the elements in the periodic table. Its major and minor chemical constituents normally include Si, Ti, Al, Fe(III), Fe(II), Mn, Mg, Ca, Na, K, H, C and P, generally expressed as oxides, and possibly F and S as well. Other constituents are considered “trace elements.” How does a rock differ from an ore? After all, an ore may be considered as merely a rock containing economically significant concentrations of a useful constituent. However, the real difference lies in the purpose of the analysis. An ore is generally analyzed for its economically recoverable constituents and possibly also for others that may have effects on processing. Perhaps most important of all is the fact that ores are normally analyzed by those who have a financial stake in the qual-

Table

I.

Standard Deviation of Silica Results

Year

1931

Glass

1951

Granite Tonal ite

1963

1970 1972 * *

0.09 after eliminating

0003-2700/81 /0351-528A$01.00/0 American Chemical Society

©1981

5

34 14 9

30 36 35

Syenile Granodiorite

1974

*Sld,

results

Feldspar Granite

1972

%, absolute,

NO. Of

Sample type

reported

*

Rocks, as mentioned earlier, must be analyzed for just about everything. There are very few laboratories, if any, that are competent to determine all the constituents of a rock that may be of interest to an earth scientist. The classical giants of rock analysis, such as W. F. Hillebrand and H. S. Washington, were able to report extraordinarily reproducible results, from which it was assumed, perhaps questionably, that their results were very accurate as well. There was a general belief that careful work by an experienced analyst could produce results comparable to those of Hillebrand or Washington. Then the roof fell in. In 1951, there appeared the first detailed compilation of data on the two so-called standard rocks, G-l and W-l (7). All concerned were shocked by the highly discordant values reported for each constituent by the 30-odd analysts involved. For example, results reported for silica in G-l ranged from 71.05 to 72.74%—enough to set both Hillebrand and Washington spinning in their graves. The agonized debates that followed are a matter of record. Many and varied were the explanations, excuses, and downright alibis put forward. Inhomogeneity in the samples was blamed by many as a cause of the disparate results, but no one really tried to show just how badly the mineral constituents of the samples would have to be segregated to explain such

ity of the results.

one

result

Ref.

0.28“ 0,37 0.26 0.10 0.18

10 1

3 11

12

1.06

13

0.46

14

large discrepancies.

Continued study and discussion regarding G-l and W-l revealed a number of shortcomings in the procedures followed in their preparation, distribution, and analysis. With the benefit of that experience, the U.S. Geological Survey embarked on the production of six new samples—G-2, GSP-1, AGV-1, BCR-1, PCC-1 and DTS-1. At about the same time, several geological groups in various other countries initiated similar programs. Many laboratories throughout the world contributed analytical data for many constituents of all of those programs, using a wide variety of methods. With the benefit of experience with G-l and W-l, one might have expected that the later programs would have produced more consistent results. Unfortunately, that did not happen. As shown in Table I, not only was there no improvement, but some distressing additional facts were revealed. There was, in effect, no significant improvement in the overall precision of silica determination over a 43-year interval (1931-1974) and, worse still, the disparities seemed to increase with the number of results involved! Shortly after publication of the first compilation of data on the second set of reference rocks from the U.S. Geological Survey, we attempted to arrive at “best values” for many of the constituents of those samples (2). The purpose of the exercise was to provide “standards” for atomic absorption spectrometry. Although some progress was made toward that goal, the resulting studies served to reveal much more about the reliability of rock analysis—or, perhaps, the unreli-

ability. It became clear early in the game that application of rigorous statistical interpretation of the raw data would be of highly questionable value be-

the data were so grossly imbalanced in terms of the number of results obtained per constituent, the degree of replication per reported result, the number of constituents reported per sample—and in just about any other parameter one can imagine. To cause

ANALYTICAL CHEMISTRY, VOL. 53, NO. 4, APRIL 1981



529 A

The Acknowledged Symbol of Excellence Since 1959

HPLC Solvents from Burdick & Jackson Purified to the exacting requirements of HPLC, gas chromatography, fluorescence and spectrophotometric analysis. Acetone Acetonitrile Benzene Butanol-1 Butanol-2 n-Butyl Acetate Butyl Chloride Carbon Tetrachloride Chlorobenzene

Chloroform Cyclohexane Cyclo pentane o-Dichlorobenzene Diethyl Carbonate Dimethyl Acetamide Dimethyl Formamide Dimethyl Sulfoxide Dioxane 2-Ethoxyethanol Ethyl Acetate Ethyl Ether Ethylene Dichloride Heptane Hexadecane Hexane Isobutyl Alcohol Iso-hexanes Methanol 2-Methoxyethanol 2-Methoxyethyl Acetate Methyl t-Butyl Ether Methyl Ethyl Ketone Methyl Isoamyl Ketone Methyl Isobutyl Ketone Methyl n-Propyl Ketone Methylene Chloride N-Methylpyrrolidone Nonane Pentane Petroleum Ether beta-Phenethylamine Propanol-1 Propanol-2 Propylene Carbonate Pyridine Tetrahydrofuran Toluene Trichloroethylene Trichlorotrifluoroethane 2,2,4-Trimethylpentane Water for HPLC o-Xylene

Call or write for more information.

BURDICK & JACKSON LABORATORIES, INC. 1953 South Harvey Street Muskegon. Michigan U S A. 49442 (616) 726-3171

Table II. Some Results for Arsenic (ppm) Sample

1

Sample 2

»

3

Analyst

c

82

94

16.2

16.4

0.67

F

12

14

2

G

22

24

12

H

14

13

4

J

20

22

1.4

K

17.9

0.48 0.6 0.03

L

14

19.8 19.6 4.7

18

21

17.8

106

M N P

coin a phrase, we were faced with “heterogeneous data.” To illustrate these points, let me tell you a few horror stories. All of them involve comparatively respected names, working in renowned institutions. Case A. Prof. A, a world-famous geochemist, decided, for some reason, to determine potassium in a group of proposed reference samples of rocks

by means of a radiometric counting technique. His results averaged about 20% lower than the consensus of other reported results, which were based on a variety of analytical techniques. Was there no flame photometer available for Prof. A to cross-check his results? Case B. Prof. B, another worldfamous geochemist, offered to provide trace-element data on another group of reference samples. His results were far removed from all others reported by other contributors, for about half of the elements reported. As for “method used,” he merely described it as “our regular spectrographic method, as used by our analyst, Mrs. X” (more about that below). Case C. This one is really sad. Table II shows results for arsenic in three proposed reference rocks. Please note the truly sad case of Analyst C. I find it particularly distressing because I had visited C’s laboratory a few years earlier and had come away with a favorable impression of his competence. Incidentally, subsequent studies revealed that the true arsenic content of sample 3 is probably close to those reported by F, L, and M. Several other reported results are therefore questionable, but none as questionable as C’s. C reported similarly absurdly high results for B, Ge, Li, Nb, Pb, Sb, Sr, Y, and Zr. It must be mentioned that after C reported his results, the coordinator of the program thanked him heartily for providing more data than anyone else (some of his data—on elements other than those shown—being fairly good). However, C was provided with median

CIRCLE 29 ON READER SERVICE CARD

530 A

Sample

ANALYTICAL CHEMISTRY, VOL. 53, NO. 4, APRIL 1981

values of results reported by others for the same elements, from which he could easily see that something was wrong. He made no attempt to rectify the situation. Case D. In a sense, this one is even sadder than Case C. The laboratory involved is in a major governmental research center in a highly developed country. Let us look at its results for cerium (in Table III). The two sets of results from D will be explained in a moment. Please note that D’s results

lower than nearly all others on Samples 1 and 2, but close to most of the others on Sample 3, whose cerium content is comparatively low. Our old friend Prof. B checks in here with a high result on Sample 2 and an apparent “pass” on Samples 1 and 3. Perhaps his bid on Sample 2 should have been registered as “double,” but we are not playing bridge. D reported similarly low results for Dy, Eu, La, and Tm. As happened in the preceding case, the coordinator of the program on Samples 1, 2, and 3 provided D with the median values of results from the are

other participating laboratories. This happened after D had reported the first set of results. Some months later, a second report came from D. It stated that the apparatus used had been dismantled, cleaned, and recalibrated, and the analysis repeated. Since no procedural change was mentioned, it must be assumed that exactly the same procedure was followed. Not surprisingly, the second set of D’s results was much the same as the first. The method in this case involved preconcentration by ion exchange and final determination by a highly precise technique. Reference to the original literature on the method and on earlier applications clearly indicated that the method was fundamentally sound. The only problem, apparently, was that it had never before been applied to samples with such high rare-earth contents! It then appeared to be a sim-

Table III. Some Results for Cerium (ppm) Sample

300 160

237 130 214 150

215 237 110

1

Sample 2

2550 3500 2400 2309 1140 2640

960 2250 2274 2400

Sample 3 Analyst

AM B

AC

25

S

26

D