FakultΓ€t fΓΌr Chemie
Infrared Spectrometric Purity Control of Chemical Substances using R
Fayaz Kondagulaa and Karl Moltb
1 Introduction
A typical Infrared Spectrum contains about 4000 data points and computer soft-ware is needed for processing the spectral data. Instrument manufacturers nor-mally provide proprietary software for this purpose. But this is often limited toa restricted number of common applications. To enable a more flexible and uni-versal numerical and statistical evaluation of spectral data, we have developedmethods to directly read the spectra into R. If the spectrum of its pure referenceis known, there are two ways for determining the purity of a potentially contam-inated substance. The first way is to calculate the correlation coefficient betweenthe spectrum of the substance whose purity has to be controlled and the corre-sponding reference spectrum and the second one is using difference spectroscopy.
2 Spectral Purity by spectral comparison with linear regression
2.1 Simulated example (Fig.1)
A simple way of calculating the spectral purity of a substance is regressing itsspectrum on the spectrum of the corresponding pure reference substance [2] us-ing the R function lm(substance~reference). The resulting correlation co-efficient is used as a measure for the spectral purity:
SPR1 = 100ΓR (1)
The correlation coefficient however has the disadvantage that with increasingimpurity it only slowly moves away from one. This βconvexβ behaviour is shownin Fig. 2(a).
600 800 1000 1200 1400
0.0
0.5
1.0
1.5
Wavenumbers
A
A
C
B
Bβ²β²
Figure 1: Simulated example of a spectral impurity. A is the spectrum of the pure component (av-erage from 9 spectra replicates) and B the spectrum of the contaminating substance. B is addedto the pure component A stepwise from 0.001 - 0.01 concentration units (spectra Bβ²). C showsthe spectra of the corresponding mixtures, on which some noise (amplitude 0.001) was superim-posed.
Compared to R the z coefficient due to Fisher [3] reacts more sensitive to im-purities. The z-transformation was performed using the R function fisherzincluded in the package survcomp:
z =1
2Γ ln
1 + R
1βR(2)
Fig. 2(b) demonstrates that z shows a βconcaveβ behavior which is favorable forpurity control because even small impurities have a large effect.
βββββββββ βββββββββ βββββββββ
βββββββββ
βββββββββ
βββββββββ
βββββββββ
βββββββββ
βββββββββ
βββββββββ
βββββββββ
βββββββββ
0.000 0.002 0.004 0.006 0.008 0.010
0.99
9970
0.99
9980
0.99
9990
1.00
0000
Impurity
R
(a) Dependency of the correlation coeffi-cent (R) from the degree of contamination.
βββββββββ ββ
βββββββ
ββ
βββββββ
βββββββ
ββ
βββ
ββββββ
βββββββββ
βββββββββ
βββββββββ
βββββββββ
βββββββββ
βββββββββ
βββββββββ
0.000 0.002 0.004 0.006 0.008 0.010
0.80
0.85
0.90
0.95
1.00
Impurity
zΒ΄
(b) zβ² values calculated by transformingthe R values in the left figure.
Figure 2: Results obtained by regressing the spectra of the mixtures C in Fig. 1 on the referencespectrum A. The impurity concentration belonging to zc resp. zcβ² corresponds to the limit ofdetection (LOD) resp. the limit of capture (LOC).
Repetitively measuring the spectra of the pure reference substance and thencomparing the single spectra with the mean spectrum allows the calculation ofa one-sided confidence interval for z. Its lower limit is as follows:
zc = zΜ β tsz
with t = Ξ¦β1t (1 β Ξ±) and Ξ± = 0.01. zc determines the limit of detection (LOD) and
zcβ² = zΜ β 2tsz the limit of capture (LOC). A normalization of z is then performedaFayaz Kondagula,M.Sc., UniversitΓ€t Duisburg-Essen, Fachgebiet Instrumentelle Analyik, 47048 Duisburg, [email protected]. Dr. Karl Molt, UniversitΓ€t Duisburg-Essen, Fachgebiet Instrumentelle Analyik, 47048 Duisburg, [email protected]
as followszβ² =
z
zc(3)
The corresponding spectral purity parameter is calculated as
SPR2 = 100Γ zβ² (4)
3 Spectral Purity based on Difference Spectroscopy
Another way of determining the spectral purity is by calculating the differencebetween the sample spectrum (C) and the reference spectrum (A):
D = C β f Γ A (5)
f should be chosen in such a way that an optimal compensation of the referencespectrum is achieved. A special algorithm was developed for determining the op-timal difference factor fopt. This works by calculating a whole series of differencespectra with decreasing difference factors and determining the integral of the pos-itive part of the difference spectra (Ο =
β«D+). From this the so called βDifference
Operating Characteristicβ (DOC) is calculated by applying a lagged difference ofsecond order on Ο using the R function diff(Chi,differences=2). Fig 3(a)shows a plot of the DOC against the factor f . The optimal factor is where theDOC exceeds a certain threshold set by the user.
β1.0 β0.5 0.0 0.5 1.0 1.5 2.0
0.00
000.
0005
0.00
100.
0015
0.00
20
PANMEAN Vs PP_6B3
Shift = 0 /Sensitivity =6.5 /Factor = 0.968Factor
Diff
2
(a) Difference Operating Characteristic Plot
3200 3100 3000 2900 2800
4060
8010
0
PANMEAN vs PP_6B3
Shift = 0 /Sensitivity =6.5 /Factor = 0.968x
dif
(b) Difference Spectrum
Figure 3: Purity control based on Difference spectroscopy
Ο and fopt are used as spectral purity parameters based on following equations:
SPR3 = 100Γ fopt (6)
SPR4 = 100Γ (1ββ« Ξ½Μ2
Ξ½Μ1D+β« Ξ½Μ2
Ξ½Μ1A
) (7)
4 Purity control of Palatinol N
With the example of a certain plasticizer (Palatinol N) it is shown that a con-tamination originating from another plasticizer (Palatinol 911P) can be detecteddown to about a concentration of 0.1 g/100g of this impurity by Diamond ATRSpectroscopy and down to 0.9 g/100g with Near Infrared Spectroscopy.
ββββββββββββββββββ
βββββββββ βββββββββ
βββββββββ
βββββββββ
βββββββββ
ββββββ
βββ
βββββββββ
βββ
ββββββ
βββββββββ
0 2 4 6 8 10
99.8
699
.92
99.9
8
Conc [g/100g]
SP
R_1
ββββββ
β
β
β
βββ
ββββ
ββ
βββββββββ
βββββββββ
ββββββββββββββββββ
ββββββββββββββββββ βββββββββ βββββββββ βββββββββ
0 2 4 6 8 10
6080
100
Conc [g/100g]
SP
R_2
LOD=0.1 g/100gLOC=0.3 g/100g
βββββββββ βββββββββ βββ
ββββββ
βββββββββ
ββ
ββ
β
ββββ βββ
ββββββ β
βββββ
β
β
β ββββββββ
β
ββ
ββββββ
β
βββ
ββββββ
βββ
ββββββ
0 2 4 6 8 10
9395
9799
C [g/100g]
SP
R_3
βββββββββ
ββββββ
βββ
βββββββββ
βββββββββ
ββββ
βββββ
ββββββ
βββ
β
β
β
βββ
β
β
β ββββββββ
β
ββββββββ
β
βββββββββ
ββ
β
ββββββ
0 2 4 6 8 10
9496
9810
0
C [g/100g]
SP
R_4
(a) Spectral Purity based on Correlation coefficient and Dif-ference Spectroscopy (MIR)
βββββββββ βββββββββ ββββββββββββββββββ
βββββββββ
βββββββββ
βββββββββ
βββββββββ
βββββββββ
βββββββββ
βββββββββ
0 2 4 6 8 10
99.9
8099
.990
100.
000
C [g/100g]
SP
R_1
β
β
β
ββ
β
β
ββ
βββββββββ
βββββββββ
βββββββββ
ββββββββββββββββββ
ββββββββββββββββββ βββββββββ βββββββββ βββββββββ
0 2 4 6 8 10
7090
110
C [g/100g]
SP
R_2
LOD=0.9 g/100gLOC=2.5 g/100g
βββ
βββ
βββ βββββββββ
βββββββββ β
ββββββββ
βββββββββ
βββ
ββββββ
ββββββ
ββββββββββββ
βββββββββ
βββ
βββ
βββ
βββ
βββ
βββ
0 2 4 6 8 10
97.0
98.0
99.0
100.
0
C [g/100g]
SP
R_3
ββββββ
β
β
β βββββββββ
ββββββββββββββββββ
βββββββββ
βββββββββ
βββββββββ
βββββββββ
βββββββββ
βββββββββ
βββββββββ
0 2 4 6 8 10
97.0
98.0
99.0
100.
0
C [g/100g]
SP
R_4
(b) Spectral Purity based on Correlation coefficient and Dif-ference Spectroscopy (NIR)
Figure 4: Dependency of Spectral Purity grades of Palatinol N as function of the concentration ofan impurity (Palatinol 911P)
5 Summary
The purity of chemical substances can be calculated either by difference spec-troscopy or spectral comparison via the correlation coefficient. The necessarystatistical computations can easily be performed with R.
References
[1] Karl Molt und Alfred Schlachter, VDI-Berichte Nr. 1959, 3-26, 2006[2] Horst Weitkamp und Dieter Wortig, Mikrochimica Acta 1983 II, 31-57[3] Jacob Cohen, Statistical Power Analysis for the Behavioral Sciences, Second
Edition, S. 110 f., Lawrence Erlbaum Associates, Inc. , Hove and London (1988)