A General Method for Calculating the Detection

Limit in Chemical Analysis

Robert Gabriels
Chemistry Department, Experimental Station for Ornamental Plants,
Caritasstraat, 17, B-9230-Melle, Belgium

Most methods of quantitative chemical analysis use standard The proposed method for calculating the detection limit D
series and calibration curves in order to convert the apparatus is based on the relation:
response into the actual sample concentration.
Around the blank however, the problem of sensitivity =
*/-?_ (t\ + ti) (1)
arises: this comes down to the ability to discern a true dif- 1

ference between very small amounts of a substance. In a

where = true difference that is desired to detect with a
more specialized sense, the detection limit is defined as the
smallest quantity or concentration of a substance, that can be
probability P
= number of replications
detected at a chosen probability level. =
true standard error per unit, in practice the
In this connection statistical methods are frequently used
standard deviation s
to come to a mathematical expression of the sensitivity of the
ti =
significant value of t in the test of significance
applied method. In spectro-chemistry, this problem is being
(with care to distinguish between one- and
regarded as of such importance that in 1964 a conference took two-tailed tests), ti corresponds to a signifi-
place in Exeter, England, solely devoted to limitations of de- cance level ai
tection (7).
h = /-value in connection with the desired probability
As appears from the literature, many authors (7-9) calculate
P. The significance level a2 of /2 is given by
the detection limit as a function of the standard deviation on
the relation: 2 2 (1 P). = —

the blank value, e.g., x times the standard deviation (70). But
when this value is being considered as the difference between It should be emphasized that formula 1, based on previous
the blank and the smallest detectable quantity, the variability statistical work (13, 14), is approximate in nature but quite
of both has to be taken into account (77,12). A i-test can satisfactory for most practical purposes down to 10. =

then be carried out on the difference between the mean of the
blank and the mean of a sample with a very low content of the
substance.
The aim of this study was to develop a method for calcu-
lating the detection limit covering any chosen concentration
range. Especially in routine multi-element analysis, one ele-
ment is found in the concentration range of e.g., 0-2 %, while
another is only likely to be detected using a 0-2 ppm standard
series, depending on the element level in the sample. It is
evident that the latter case requires the highest sensitivity
conditions with regard to analytical equipment. Here, and in
trace analysis in general, it is not unusual to find a detection
limit which is not lower than the one calculated for a standard
series with a larger range, e.g., 0 to 20 ppm, because of the
photomultiplier dark current or other background elevating
phenomena which generally result in lower precision and de-
creased signal-to-noise ratios.
(1) “Limitations of Detection in Spectro-chemical Analysis,”
Hilger and Watts Ltd., London, 1964.
(2) P. W. J. M. Boumans and F. J. M. J. Maessen, Z. Anal. Chem.,
220, 241 (1966).
(3) H. Delaffolie, Mikrochim. Acta, 1, 304 (1955).
(4) K. Doerffel, “Reinststoffprobleme, II Reinststoffanalytik,”
Akademie Verlag, Berlin, 1966.
The actual calculation of the detection limit is as follows:
if we have, for a blank of a certain substance or element to be
analyzed, n replications with a mean Xh we can calculate the
variance and the standard deviation s, the latter representing an
estimate of the true standard error per unit. A sample with
a very low content, also n times analyzed, results in a mean j?2
and a standard deviation which we assume to be equal to the
same s. The sample will contain a detectable quantity when
Xi- }
which has probability P of being detected by means of a /-test
with significance level ai. Now it is also evident that
it- ^ Xi +
Hence, the minimal detectable quantity D is given by
D =
Xi + (2)
Because we are only interested in detecting differences in
one direction with regard to the blank (one-tailed test), the
value of h has to be read corresponding to twice the signifi-
cance probability. Here, the test was done at the 5% sig-
nificance level, so that the value of / in the standard tables is
to be read at 0.10 or 10%.
If one is satisfied with a 90% chance (P 0.90) of detecting =

(5) R. Gerbatch and H. Scholtze, Rev. Yena, (2) 89 (1963). D and by combining Equations 1 and 2
(6) H. Kaiser, Z. Anal. Chem., 209, 1 (1965).
(7) Ibid., 216, 80 (1966).
(8) O. G. Koch and G. A. Koch-Dedic, “Handbuch der Spuren- D =
Xl + í
" (f0.10; n—1 4" /0.20; J!_])
analyse,” Springer Verlag, Berlin, 1964. n
(9) . H. Ruessmann and K. R. Brooks, Acta Chim. Acad. Sci.
Hung., 42, 1 (1964).
(10) ASTM “Methods for Emission Spectrochemical Analysis,” (13) W. G. Cochran and G. M. Cox, “Experimental Designs,” 2nd
4th Ed., p 227, Philadelphia, Pa., 1964. ed., J. Wiley and Sons, Inc., New York, 1957, p 19.
(11) L. A. Currie, Anal. Chem., 40, 586 (1968). (14) J. Neyman and B. Tokarska, J. Amer. Statist. Ass., 31, 318
(12) F. Scharf, Experientia, 20 (8), 470 (1964). (1936).

ANALYTICAL CHEMISTRY, VOL. 42, NO. 12, OCTOBER 1970 · 1439

 freedom ( — 1). For a specific test result, multiplying the
_
Table I. Numerical Values of corresponding value of this factor by the standard deviation
V2jn (in.io; n-i + to.20; n-i) in Function of -
1 of the blank determines the fraction of scale units to be added
Degrees of Degrees of to the mean blank, in order to obtain D.
freedom y/2jn (/o.io; 7.-1 freedom y/2¡n (Zo.io; n- Since the value increases when the degrees of freedom
n -

1 + Í0.20 ; n-l) n -
1 + io.20; n-l) become small, it is customary, where justified, to pool the
1 9.39 21 0.92 estimates sft1 and s*22 by calculating
2

2 3.92 22 0.90
3 2.82 23 0.88 , S*12 + S*22
0.86 Sx
-

2.32
—

4 24 *
2
5 2.02 25 0.84
6 1.81 26 0.82
7 1.65 27 0.81 This increases the degrees of freedom from ( — 1) to 2(n —

1),
8 1.54 28 0.79 and the formula becomes
9 1.44 29 0.78
10 1.36 30 0.76 D =
Xl + Si /2 [/o.lO; 2(n—1> + .20; 2<n-l)]
11 1.29 35 0.71
12 1.23 40 0.66 When planning experiments it is sometimes useful also to
13 1.18 45 0.62
know the number of replications required to obtain a detection
14 1.13 50 0.59
15 1.09 55 0.56 limit which is, for example, defined as 1 X s, an ASTM rec-
16 1.06 60 0.54 ommendation (10). From Table I it is easily seen that in this
17 1.02 70 0.50 case 19 replications are needed.
18 0.99 80 0.46
19 0.97 90 0.44
20 0.94 100 0.42 ACKNOWLEDGMENT

I would like to thank W. G. Cochran, Harvard University,

and A. Rotti, University of Ghent, Belgium, for stimulating
Here D is expressed as an analytical signal (mV, absor- discussions of the material in this study.
bance...) that can be transformed (calibration curve) into con-
centration values. Table I shows the numerical values of Received for review March 12, 1970. Accepted June 23,
y/2jn (to.io; n-i + to.2o; n-i) in function of the degrees of 1970.

Gas Chromatographic Determination of Nitrated Derivatives of

Glycerine in Aged Double-Base Propellants
J. M. Tro well
Hercules Incorporated, Bacchus Works, Magna, Utah

The studv of the mechanism of aging and expected useful

life of all types of nitroxy based solid propellants involves the Table I. Gas Chromatography Conditions
measurement of the various propellant stabilizers [nitrogen Dual columns
oxides (NOz) scavengers] under various storage conditions. 6 ft X Vs-inch tubing

The methods normally used for analyzing aged propellant Packed with 2.5% OV-17 and 2.5% QF-1 on 60-80 mesh
Gas Chrom Q
samples for residual stabilizers and other components are a Column temp, 70-230 °C
combination of infrared and thin-layer chromatography as Column temp, program rate, 10 °C/min
reported by Macke (1) or a gas chromatography method pre- Detector temp, 250 °C
viously reported by the author (2). Injection port temp, 70 °C
Helium flow, 65 cc/min
A recent attempt to apply the gas chromatography method
Hydrogen flow, 28 cc/min
to accelerated aged samples in which essentially all of the Air flow, 500 cc/min
stabilizers had been nitrated, was not successful because of Sample size, 0.004 cc
interference of 1,3-dinitroglycerine with the measurement of
low concentrations of resorcinol. Therefore, a new gas chro-
matography method for the analysis of propellant components
was investigated. investigation. The instrument was modified by placing an
injection port inside the oven compartment to allow true on-
EXPERIMENTAL column injection. Detailed instrument parameters are
Apparatus. A commercial gas chromatograph, F&M summarized in Table I.
5754B with dual flame ionization detectors was used in this Column packings were prepared using the Applied Science
Laboratories Hi-EfF Fluidizer. The columns were packed
(1) G. F. Macke, J. Chromatogr., 38, 47 (1968). under vacuum with vigorous vibration. The ends of the
(2) J. M. Tro well and M. C. Philpot, Anal. Chem., 41, 166 (1969). column were then plugged with silanized glass wool and

1440 · ANALYTICAL CHEMISTRY, VOL. 42, NO. 12, OCTOBER 1970