2504
Anal. Chem. 1984, 56,2504-2509
Data Analysis in Elemental Gas Chromatography Ermes Pella* and Lucian0 Bedoni Farmitalia Carlo Erba, Ricerca & Sviluppo Chimico, Via Imbonati 24, 20159 Milano, Italy
Bruno Colombo and Guido Giazzi Carlo Erba Strumentazione, Rodano, Milano, Italy
I n order to Improve the accuracy of elemental analysls based on combustion and gas chromatography, callbratlon equatlons are studled wlth regards to thelr usefulness and llmlts In practlce. The functlonal relatlonshlps between elemental components C, H, N, and S and TCD lntegratlon data of thelr combustion gases CO,, H,O, N,, and SO, are rellably described by simple llnear equatlons obtalned from experlmental data by means of llnear regresslon analysls. These linear callbratlon equatlons serve as a means of checklng the llns arky response range of the system and as a means of quantlfylng adsorptlon effects; they also show the Influence of analytlcal parameters and reagent quallty. Analytical functions glven by callbratlon equatlons provlde more accurate results and make It posslble to extend the useful operatlng range of the Instrument wlthout loslng accuracy.
the dependent variable, elemental content is the independent variable, and I is regressed to C to obtain the calibration equations. These equations, which can be established by sampling a few reference substances with elemental contents well distributed in the analytical range, and recording the relevant count integration data, usually show very high correlation coefficients and therefore satisfactorily fit experimental data. By extension of the study of calibration equations to ranges higher than that involved in the analytical procedure, the system's limits of linearity can be established; the frequent occurrence of a negative Y intercept in eq 1 serves to record and quantify unknown adsorption phenomena in the analytical system. Inversion of the calibration eq 1 provides the analytical function
C = -I & - B m
In modern elemental organic microanalysis based on coupling combustion methods with GC separation and TC detection of combustion gases, little attention has been paid so far to the calibration procedure, even though applicability, accuracy, time consumption, and costs of the analytical process depend on it. The customary calibration procedure (1-3)consists of repeatedly sampling known amounts of a reference substance, measuring TCD signals (corresponding to peak areas) of its combustion gases COz, H20,N2and SO2,calculating the ratio between element amount and the digital integration value, and then averaging ratios for the same element, thus obtaining its calibration factor or constant (4). However objections can be raised to this calibration method on several grounds: (a) it describes the elemental componentlintegration data relationship well only for the point given by the averaging procedure; (b) it does not provide a constant calibration factor for an extended analysis range (5);(c) it is based on the assumption that the analytical process is affected only by positive errors (blank); (d) it fails in the case of very low (6, 7) and high N, H,and S percentages; (e) it does not take into account the linearity limits of the analytical system; (f) it provides a highly precise calibration factor when only one reference substance is used but the precision is limited when cross-checking different reference substances (5). As an alternative to the calibration factor, one must seek a suitable calibration function, based on analysis of more than one standard and calculated with the aid of the least-squares procedure, to interpret correctly the functional relationship between digital data from integration of TC signals and known amounts of elemental components of the standard. For each elemental component a calibration equation can be found, a linear equation of the type I=mC&B
(1)
where I represents the count intensity and C the elemental content in micrograms. Clearly count intensity is taken as 0003-2700/84/0356-2504$01.50/0
m
which serves to calculate the amount of elements in the unknown samples more reliably and accurately than is obtainable with the calibration factor.
EXPERIMENTAL SECTION Apparatus. The instrument employed for studying response/amount of element relationships is the Erba elemental analyzer, Model 1106 C (2, 3, 7), or a modified version for determining C, H,N, and S as described by us (8). Both are basically a combination of a combustion train and a gas chromatograph. Samples are weighed into light tin containers and dropped from a sampler at preset times into a vertical heated quartz tube, through which a constant flow (30 mL/min) of pure helium is maintained. When the sample is introduced, the carrier is temporarily enriched with pure oxygen; flash combustion takes place primed by oxidation of the container. Quantitative combustion is completed by interaction with oxidative catalysts; oxygen removal together with reduction of SO3to SO2and of nitrogen oxides to nitrogen is then accomplished on heated copper. The mixture of combustion gases is carried by helium into the chromatographic column, packed with Porapak QS and heated at about 90 "C. The individual components are separated and eluted in the sequence N2, C02, H20, and SO2. The peaks are recorded with a 1 mV recorder and TCD output signals are directly evaluated by electronic integrators independently of the recorder. Typical analysis chromatograms printed by the SP 4270 integrator-calculator are given in Figure 1. Nitrogen is firstly eluted roughly at 2 min in about 40 s, carbon dioxide at 3 min in 70 s, water at 5.5 min in 120 s, and lastly sulfur dioxide at 9.5 min in about 120 9.
EstablishingCalibration Equations. Samples of different reference substances (see Table I), dry and pure at the minimal level of 99.8% are accurately weighed in amounts of 0.1-2 mg into tih containers and analytically processed. The determination yields a four-peak chromatogram if the sample contains C, H, N, and S, as shown in Figure 1A. More frequently the standard does not contain sulfur and the chromatogram is like Figure 1B. Before the run was started, the amounts in micrograms of elemental components are calculated from sample weight and theoretical percentages and, when possible, entered in the calculating device as shown in Table 11. These element amounts 0 1984 American Chemical Society
ANALYTICAL CHEMISTRY, VOL. 50, NO. 13, NOVEMBER 1984
Table I. Suitable Reference Substances
n-hexadecane cholesterol phenanthrene atropine valine diphenylguanidine hydantoin biuret isatin sulfapyridine methionine thiourea
%C
%H
84.87 83.87 94.34 70.56 51.26 73.90 36.00 23.30 65.30 52.99 40.25 15.78
15.13 11.99 5.66
8.01 9.46 6.20 4.03 4.89 3.43 4.45 7.43 5.30
%N
%S
4.84 11.96 19.89 27.99 40.77 9.52 16.86 9.39 36.80
12.86 21.49 42.12
represent the X values selected in advance, with the corresponding experimental integration data as Y values. This gives a series of paired integration data/element amounts evenly distributed in the range under study. Suitable reference substances, covering the different concentration ranges of element, are listed in Table I. For nitrogen amounts higher than 400 Fg/analysis, biuret is used as standard, for carbon amounts over 1000 Wg/analysis phenanthrene or cholesterol is used, for hydrogen over 100 Fg/ analysis cholesterol and n-hexadecane are used, and for sulfur over 400 pg/analysis thiourea is used. To obtain equations, regression analysis based on the leastsquares method is applied manually to the pairs of values for the same element, using a programmable pocket calculator such as Texas TI 59, or automatically with an integrator-calculator like SP 4270 which provides printed slopes and Y intercepts of four
(Table IIA) or three calibration equations (Table IIB), and also correlation coefficients (R). These coefficients are necessary to state the correctness of calculation and the quality of fitting; they are given, for example, in Table I11 near the corresponding equations. Their high values mean that regression is optimal and the calibration equations are reliable. The surprisingly high coefficient value can be explained by the extremely repetitive gas chromatographic conditions of the analytical system and its good total efficiency. From data in Table 11, two series of calibration equations can be found, as follows: YN* = 0 . 1 6 1 7 ~-~0.342 yNZ 0 . 1 6 3 4 ~-~0.141
+
ycoz = 0 . 3 5 9 4 ~ ~ 0.745
+
ycoz = 0 . 3 6 1 2 ~ ~ 0.328
The series on the left was obtained by means of the modified analytical channel to determine C, H, N, and S; the series on the right was obtained with the usual CHN channel. Calibration equations are therefore simple linear equations, showing an obviously different response level (different coefficients or slopes) and a constant (or Y intercept) negative for three gases, Nz,HzO,and SOz (loss of component) and positive for COz (blank). Calibration equations present different slope and Y intercept values when integrators from different firms are used, because of technical differences. It is useful to find a standardization criterion in order to compare calibration equations obtained using different integration devices. Our criterion consists of assuming that the slope in the carbon equation is equal to 1 as carbon
Table 11. Reports Relevant to Chromatograms of Figure 1 (A) Four Calibration Equations
file 1, method 5 , run 4, index 3, calib name
MCRG %'
retention time, min
area
BC
re1 retention time
1.75 2.35 4.8 9.47
248049 1832890 463967 178383
02 03 01 01
0.745
H S
155.28 488.04 40.98 118.44
totals
802.74
N
C
1 2.043 4.03
2723289
Coefficients of least-squares fit to a linear equation file entry
KA
1 2 3
o.oooot + 01 o.oooot + 01 o.oooot + 01 o.0ooot
4
KC
KB
+ 01
0.1617t 0.35941 0.11497 0.1519t
+ 04 + 04 + 05 + 04
-0.342t 0.7453t -0.760f -0.202t
+ 04 + 04 + 04 + 04
(B) Three Calibration Equations file 1, method 5, run 4, index 3, calib name
MCRG %"
retention time, min
area
BC
re1 retention time
N C H
162.3 603 50.59
2.27 2.98 5.57
264689 2181861 576065
01 01 01
0.762 1 1.869
totals
815.89
3022615
Coefficients of least-squares fit to a linear equation file entry 1 2 3 a
KA o.ooo0t + 01
o.oooot + 01 o.oooot + 01
2505
KB 0.1634f 0.3612t 0.1143t
KC
+ 04 + 04
+ 05
MCRG is micrograms. MCRG 5% is used because the subsequent calculation can give either kg or %.
-0.141t + 04 0.3284t 04 -0.213f 04
+ +
2508
ANALYTICAL CHEMISTRY, VOL. 56, NO. 13, NOVEMBER 1984
Table 111. Applicability Ranges for Calibration Equations component
NZ COZ
element
range, pg/analysis
nitrogen
0-(40, 60) (40, 60)-(400,500) (400, 500)-2000 0-(1200, 1400) >(1200, 1400) 0-(20, 30) (20, 30)-(80, 100) (80, 100)-300 0-(40, 60) (40, 60)-400 400-1000 *y = counts x IO-*.
carbon
HZO
hydrogen
SO2
sulfur
'z = (sample w t (pg) x theor
%)/loo.
3
J
u; Figure 1. Analysis chromatograms for C, H, N, and S (A) and C, H, and N analysis (6).
dioxide, according to our results, is the only adsorption-free combustion gas. The Y intercept is accordingly calculated and the calibration equations, recalculated taking into account not the data of Figure 1 but the ones statistically observed during a year's work and considering the actual reciprocal ratio of TC response among the components, are the following: =0.366~ 0.647 ~ yco2 = 1.000Xc
+ 1.750
(3)
(4) (5)
ysop 0 . 4 4 5 ~- ~1.748
(6)
These standard equations are independent of flow rate also and are therefore useful to compare systems working in different conditions. The slope ratio 0.366/1.OOO/3.306/0.445 can be taken as a criterion of good working. Substantial deviations from this ratio or anomalous Y intercept values usually indicate a chemical malfunction. Influence of Working Parameters on Calibration. Flow rate of carrier is the parameter which most influences the slope in calibration equations, causing it to change in inverse proportion. Thus since close control of flow rate is essential for keeping constant the calibration, we managed to maintain a rate of 30
calibration eqa$b
corr coeff
eq no.
0.999 64 0.999 95 0.999 76 0.999 956 0.999 987 0.999 937 0.999 993 0.999 58 0.999 82 0.999 76 0.999 94
3a
3 3b
4 4b
5a 5 5b 6a
6 6b
mL/min helium in establishing calibration equations. This value may be slightly altered by granulation, packing degree, and working temperature of every newly employed reagent and has to be restored before establishing an ultimate calibration equation. Concerning system geometry, there are two different arrangements for the combustion section in C, H, and N and C, H, N, and S determination, oxidative and reductive sections may be either separate or combined (7,8).The different system geometry does not affect calibration equations. The same analyzer can accept two automatic samplers, for loading a series of 23 samples or for continuous loading. With the former, minimal pressure changes are noted in the system on nitrogen elution and eq 3 is obtained; with the latter, the internal pressure undergoes an appreciable change on nitrogen elution, causing an increase in the slope in the nitrogen equation. The slopes in the equations for C, H, and S, however, do not vary as pressure equilibrium has been restored by then. Concerning reagents, there is a substantial difference between oxidative and reductive catalysts employed in the combustion train. The former, provided they are free from alkaline impurities, do not influence calibration equations. The latter e.g., metallic copper, does, and the Y intercept varies accordingly. Nitrogen is partially adsorbed on the copper layer depending on the copper quality, quantity, and working temperature. Adsorption decreases for temperatures over 700 OC. Copper from not properly reduced copper oxide can adsorb sulfur dioxide. Regenerated copper adsorbs not only nitrogen but also carbon dioxide causing the appearance of a negative Y intercept in eq 4. Tin employed for making containers is responsible for the carbon blank and, partially, for sulfur loss. Strong SnO, residue from spent containers can cause slight variations in both slopes and Y intercepts. A1 containers are responsible for a high negative Y intercept in the nitrogen calibration equation and decreasing slope value in the carbon equation. Oxygen donors added to the sample in C, H, N, and S determination (9) clearly add nitrogen, carbon, and water to the system, changing the Y intercept. Auxiliary oxygen, which makes flash combustion of the sample possible, introduces an additional small nitrogen blank into the system which varies according to the oxygen quality and quantity and cylinder. A 5-mL dosing loop for oxygen was used in this work. GC Conditions, Columns are made of copper for C, H, and N analyses, 6 X 4 mm, 250 cm, and of Teflon for C, H, N, and S analysis, 6 X 4 mm, 200 cm, heated at 90 OC, filled with pretreated Porapak QS, 80-100 mesh. conditions are selected in order to work with the least amount of stationary phase since Porapak is responsible for HzO and SOz adsorption as already reported (9) and thus for negative Y intercepts in eq 5 and 6. Nontreated Porapak causes greater SOz adsorption. Poor Nz/COz separation affects the precision of the corresponding calibration equations. Variations due to electrical parameters have not been considered because the constant voltage mode of bridge supplying (IO)is reported to provide the best linear range for TC response. Regarding the analog to digital conversion used in the system, the various electronic digital integrators, Le., DP 110 CEST, H.P. 3390, or S.P. 4270, offered equivalent precision and performance.
ANALYTICAL CHEMISTRY, VOL. 56, NO. 13, NOVEMBER 1984
Last, ultramicro and micro analytical balances made by different firms were used with equivalent results. Calculation, Mathematical inversion of calibration equations, as already cited, gives the analytical function 2 for the different elements; applied by using the integration values of unknown processed samples, this serves to calculate their element amounts. Simple mathematical division by the sample weight then provides the required percentages. Calibration equation inversion and percentage calculation can be done automatically the by SP 4270 or HEC Computer 960 CEST, after having entered the weights of unknown samples. In routine analytical practice, at the beginning of the analytical cycle, three runs with different reference substances, weighted in similar amount, gives a rational distribution of values along the X axis and calibration equations can be confidently established (Rbetter than 0.999). It is not advisable to use only one reference substance with different weighings. Analytical function is checked daily and, assuming that no substantial modification has been made to the analytical system such as flow rate regulation or replacement of reactors, it is usudy confirmed. Small variations are noticed in the Y intercept value as oxidative and reductive reactors age. Comparing analytical function 2 with the customary calculation based on calibration factors, the term B/m present in (2) if positive corresponds to the blank to be traditionally subtracted. However if it is negative as in most cases (calculation of N, H, and S percentages),it represents a loss of component which could never be considered by the traditional calculation based on calibration factors. RESULTS AND DISCUSSION Calibration equations are important not only for relating digital data to composition, therefore providing a reliable analytical function for practical purposes, but also for demonstrating the behavior of the analytical system. They also permit a critical evaluation of TC measurement of the four components N2, C02, H20, and SO2at their different possible concentrations in helium. This study was not therefore limited to clearing up the relationship between TC response and element amount in the usually narrow range employed by analysts. It extended to the low and very low range (sample weights less than 200 pg) and to the relatively high range (weights higher than 1mg) too. The results are expressed by means of calibration equations reported in Table 111, and since the eluted components each behave differently, they are discussed separately according to each element. Nitrogen Calibration Equation. Considering the first gas eluted, nitrogen, if we plot the integration values of its TC signal against the amounts of element, the resulting curve is practically a straight line from zero up to 400-500 Hg nitrogen per analysis. However two equations are reported in Table I11 for low and normal concentrations (eq 3a and 3) because adsorption phenomena are small when the component is diluted in the carrier and increase until they become constant, as predicted by the adsorption law. Consequently the Y intercept is slightly positive at low nitrogen concentrations but becomes negative at increasingly higher concentrations, meaning some nitrogen is lost in the system. The value B / m corresponds to nitrogen loss, already cited in literature (6,7, 1 0 , probably due to adsorption of N-oxides on copper. When a blank determination is made without sample, a positive value is always detected because of the contribution of auxiliary oxygen and, when taken into account, it is usually, incorrectly, subtracted in the calculation based on calibration factor. The calculation method based on analytical function 2 however takes into account nitrogen loss due to adsorption on copper and provides a mathematical correction of the adsorptive N loss. Conditions are different for low nitrogen concentrations where adsorption is reduced. It is not possible to establish a clean cut between low and normal concentration ranges because the effects of adsorption on copper are influenced by
2507
the amount of combustion gases accompanying nitrogen, particularly water, which can reduce N absorption when processing hydrogen-rich compounds. On examining the calibration equation in the range of concentrations higher than 500 pg/analysis (eq 3b), we note increases in slope and negative Y intercept. In other words the plot integrals vs. element (in micrograms) tend toward a positive deviation (Le., integrals are larger than a linear relationship predicts). This behavior can be explained as an effect of the nonlinear response of the TC detector. As a matter of fact 500 pg of nitrogen eluted from the system in 40 s by a helium flow of 30 mL/min gives an average nitrogen concentration in the detector of 0.5%, or a maximum of 1%; at these concentrations deviations of the TC response from theoretical linearity are possible (12)and molecular properties (mass, collision diameter and intermolecular forces) with different levels for nitrogen and helium enable one to predict (13)that the total response deviation will be positive, as found experimentally. This is what happens in eq 3b where the slope appears greater; because of the increased slope the Y intercept is higher, so the N loss looks deceptively high. Examining from a statistical-mathematical point of view of the relationship between integration values and micrograms of nitrogen in the entire range (0-2000 pg/analysis), we no longer obtain a simple linear equation but a binomial expression as follows:
This describes a slightly parabolic convex (14) curve. Carbon Calibration Equation. Considering the second eluted gas, carbon dioxide, more favorable linearity conditions can immediately be seen than with the other gases and linearity extends from 0 to about 1300 pg/analysis or better, together with a positive, steady Y intercept. Calibration eq 4 does not show any appreciable variation in the range indicated. The pairs integral valuelpg of element when the carbon amount was higher than 700 pg/analpis were obtained by processing nitrogen-free reference substances in order to avoid interference by improper N2/C02 separation. The Y intercept is usually positive and the calculated blank value B / m corresponds perfectly to the real blank measured by sampling tin containers without samples. The Y intercept tends to zero with blank-free containers. Negative Y intercepts have been detected only using catalysts containing alkaline impurities or regenerated copper, as already cited. Adsorption phenomena can thus be generally ruled out by C 0 2 elution. This is the reason why we assumed the slope of the carbon calibration equation equal to 1,aiming to compare equations from different laboratories. A small slope increase (0.5%) is shown in the range over 1300 pg/analysis (eq 4b), that is some counts more than theoretical, while the Y intercept consequently becomes negative simulating carbon loss. For % C, calculation using the calibration equation is the same as that based on calibration factor, provided a blank is taken into account in the latter case. Use of the calibration equation however is more rational, because it provides a precise blank measurement, avoiding additional runs to determine the blank, and an unfailing slope value over a large range. The extended linearity of the relationship C02integrals/pg of C may appear surprising in that 1000 pg of carbon gives an average C02 concentration of 6% in the TC detector. However, in the case of the binary mixture of C02/He in spite of the very different molecular properties there is good compensation of nonlinearizing effects thus allowing extended response linearity (13).
2508
ANALYTICAL CHEMISTRY, VOL. 56, NO. 13, NOVEMBER 1984
Table IV. Rssults Obtained Using Calibration Equationsd
tetramethylammonium tetraphenylborate' hydrastine Ag diethyldithiocarbamate mercurisuccinimidee N-(2-chloro-4-nitrophenyl)-N'-(4-iodophenyl)thioureae sulfanilamide picric acid 4-p-fluorophenyl-3-thiosemicarbazide diaminomaleic acid, dinitrile bis(tripheny1phosphine)copper trifluoromethanesulfonate'
X
3.56 3.65 5.46 7.06 9.69 16.27 18.34 9.97 51.83
~
%H
%C
%N
compound
R
3.58 3.66 5.43 7.10 9.67 16.31 18.41 9.99 51.88
b
S
0.07 85.49 0.10 65.78 0.08 23.44 0.05 24.22 0.05 36.01 0.04 41.84 0.08 31.44 0.05 49.53 0.10 44.45 60.28 93.71 91.24 9.32 9.36 63.98 16.02 16.08 50.34
C
85.45 65.85 23.48 24.19 36.06 41.80 31.50 49.46 44.38 60.28 93.78 91.20 63.94 50.41
X
0.06 0.06 0.10 0.08 0.09 0.08 0.07 0.05 0.07 0.05 0.09
f
8.20 5.52 3.93 2.03 2.09 4.68 1.31 2.87 3.73 4.10 6.29 0.90 6.04 4.23
% S
S
X
8.22 5.55 3.97 2.02 2.07 4.67 1.34 2.82 3.75 4.15 6.29 0.93 6.02 4.28
0.05 0.04 0.06 0.04 0.07 0.04 0.05 0.06 0.04 0.05 0.04
f
S
X
Z
25.03 25.13 0.10 7.39 7.46 0.07 18.62 18.67 0.08 10.16 10.12 0.06
4.35 4.33 0.04 naphthalene CRM 004 cokee 0.50 0.54 acetanilide at 90% (mixed with Si02,9:l w/w) sulfapyridine at 95% (mixed with SiOz, 191 w/w) 12.22 12.28 ax, theoretical %. *.F, average %. O S , standard deviation. dFiveto ten runs for each sample; 0.3-2 wt range. eReferencesubstances from EEC Community Bureau of Reference/Brussels.
Hydrogen Calibration Equation. Considering the correlation between H 2 0 integration values and H micrograms, a linear relationship is again found (eq 5a and 5), basically the same for low and middle concentrations, the only change being in the negative Y intercept owing to the gradually increasing adsorption. We emphasize again that, in spite of the positive blank of water brought into the system by auxiliary oxygen and reagents, the equation indicates a loss of water of about 1.5 pg. This water loss is nominal, because the real loss should be calculated as 1.5 pug + external blank. The previous calculation procedure based on calibration factor and subtraction of the blank appears even less correct on account of the presence of adsorption effects. The calibration equation for hydrogen, in comparison with others, displays limited linear behavior, from 20 to about 100 pg/analysis. However in the case of 100 pg of hydrogen, 900 pg of water are produced, able to give an average concentration of 2% in the TC detector; this is objectionable with regards to response linearity. As a matter of fact, in the range above 100 pglanalysis, the hydrogen calibration equation (eq 5b) shows positive slope deviations of about 1-2% (integration values larger than theoretical), thus demonstrating deviation from linearity just beginning a t 80-100 pglanalysis. Plotting H20 integrals against micrograms of hydrogen, the resulting curve shows positive deviations from linearity at concentrations higher than 100 pg/analysis giving a deceptively high Y intercept. However by taking into account the water loss and providing that eluted water falls inside the range examined, very accurate results are possible. This is particularly true in analyzing liquid fuels when the use of the hydrogen calibration equation established using standards like cholesterol or n-hexadecane is mandatory. Sulfur Calibration Equation. The relationship between SO2integration values and micrograms of sulfur is still represented by linear equations (Table 111) in which the Y intercept is always negative and higher than that of the other calibration equations. This means that adsorption effects always play a leading role in this determination. S loss owing to SO2 adsorption is small at low concentration (eq 6a) but is notable in the analytical range, with an upward trend if improper reagents are employed (impure tin and tungstic acid, improperly reduced copper and untreated stationary phase). S determination, notoriously difficult (9) because critical conditions are necessary to convert all sulfur present to sulfur(1V) oxide, shows its most delicate point in this loss of component which can be accounted for in the % S calculation only by using calibration equations. The percentage calculation based on calibration factor, tolerable when sulfur
amounts are similar, is no longer applicable when unknown sulfur amounts are sampled in the analytical system. When the range is over 400 pg/analysis, calibration eq 6b shows an increased slope, constant over 2000 pg/analysis, and a deceptively high Y intercept. The rise in slope is clearly caused by a change in the response linearity of the TC detector (400 pg/analysis means an average concentration of 0.5% SOz in He in the TC detector). This is visible as a slight flex in the curve like that shown by the nitrogen calibration curve (eq 3c). The same behavior of binary mixture S02/He has previously been described (15) whereas the plot of SO2integral values vs. S amounts was found linear over the entire range by others (13);however a different dilution of the component in the carrier can explain differing results. In any event, S calibration eq 6a and 6 clearly take into account the analytical phenomena of S determination by combustion-gas chromatography explaining the occurrence of a negative error reported (9, 16). Table IV shows examples of results using analytical functions based on calibration equations. These results are unusually precise and accurate and, compared to those obtained by calculation based on calibration factors but not indicated here, appear far superior in the case of low nitrogen and hydrogen percentages and more reliable especially as regards sulfur determination. In all three cases, analytical functions derived from calibration equations are able to account for the right component loss which, in the case of sulfur (3.9 pg from eq 6), is such as to prevent application of the method, if not considered. In Table IV, moreover, the different percentages offer satisfactory equivalence in the case of substances containing inorganic residue, as shown for acetanilide and sulfapyridine. Beyond the analytical results presented, for the range of increasing element contents &e,, nitrogen higher than 500 pglanalysis, carbon higher than 1300 pg/analysis, and hydrogen higher than 100 pglanalysis) it turns out that a second-order equation should be used to describe better the response/composition relationship, or else suitable dilution of the components in a helium flow higher than 30 mL/min should be selected. Last, one could object that this study was carried out using only one particular analyzer. This analyzer however is known to combust all types of samples (17-19) and therefore to convert quantitatively any reference substance into the corresponding combustion gases. Consequently the theoretical inferences of this study can be applied to any analytical system based on coupling combustion methods with GC separation and TC measurement of combustion gases, with possible variations in the linearity limits according to the elution mode
S
Anal. Chem. 1084, 56, 2509-2512
and the Y intercept as adsorption phenomena certainly differ. Registry No. Carbon, 7440-44-0; hydrogen, 1333-74-0;nitrogen, 7727-37-9; sulfur, 7704-34-9. LITERATURE CITED (1) Ehrenberger, F.; Qorbach, S. “Methoden der organlschen Elementarund Spurenanalyse”: Verlag Chemle: Welnhelm/Bergstr., 1973 p 85. (2) Belcher, R. “Instrumental Organlc Elemental Mlcroanalysls”; AcademICPress: London, New York, San Francisco, 1977; Chapter 111. (3) Klrsten, W. J. “Organlc Elemental Analysls”; Academic Press: New York, London 1983; p 40. (4) Ma, T. S.; RMner, R. C. “Modern Organic Elemental Analysis”; Marcel Dekker: New York and Basel, 1979; p 62. (5) Mazzeo, P.: Mazzeo-Farlna, A. Mlcrochem. J . 1983, 28, 137. (6) Ehrenberger, F.; Kelker, H.; Weber, 0 Z. Anal. Chem. 1986, 222, 260. (7) Pella, E.; Colombo, 8. Mlkrochlm. Acta 1973, 697.
(8) (9) (10) (11) (12) (13) (14) (15) (16) (17) (18) (19)
2500
Pella, E,; Colombo, B. Mlkmchlm. Acta 1978, 271. Klrsten, W. J. Anal. Chem. 1979, 57, 1173. Wells, G.;Slmon, R. J . Chromatogr. 1983, 256, 1. Slmon, W.; Clerc, 0. T. Helv. Chlm. Acta 1963, 46, 236. Patterson, P. L.; Gatten, R. A.; Kolar, Q.; Ontlveros, C. J . Chromatwr. Scl. 1982, 20, 27. Kekkebus, B. 6.; Barsky, M. H.;Rossl, R. T.; Jordan, J. J. Am. Chem. SOC. 1988, 68, 2398. Carr, P. W. Anal. Chem. 1980, 52, 1746. Barsky, M. H. Ph. D. Thesls The Pennsylvania State Unlverslty, 1961, Unlverslty Mlcrofllm, Inc., Ann Arbor, MI. Haran, Q., unpublished work, Nottlngham Unlverlsty, March 1982. Borda, P. P.; Legrdlns, P. Anal. Chem. 1880, 52, 1777. Froellch, P. N. Llmnol. Oceanogr. 1980, 25, 564. Glarzl, Q.; Colombo, 8. J . Coal Qual. 1982, 2 , 28.
RECEIVED for review April 16,1984. Accepted June 18,1984.
Preparation and Properties of Open Tubular Columns Coated with Tetra-n-butylammonium Tetrafluoroborate Subhash C. Dhanesar and Colin F. Poole* Department of Chemistry, Wayne State University, Detroit, Michigan 48202
Reasonably eff lclent ( 1500-2000 plates/m), long-llved open tubular columns were prepared by using the organic molten salt tetra-n-butylammonlum tetrafluoroborate as the stationary phase. To form a stable, homogeneous ttrrtknary-phase film, the Inner surface of each soda lime giam capillary column was first roughed by etching twice wlth ammonium blfluorlde at 350 O C . The column was then coated by the statk method wlth a solution of 10-25 mg/mL of the Organic molten salt In dlchloromethane. The optimum concentration of coating solution was 15 mg/mL; the silane-coupling reagent (3-glycldoxypropyl)trhnethoxysllaneIs useful for stabliizlng low loaded (7.5-15 mg/mL) fllms. Above the melting point of the salt, the efflclency of the columns declines with Increasing temperature. Presumably at higher temperature, the less vlscous phase Is able to flow and forms an inhomogeneous stationary-phase film. The columns have a useable operating temperature range of 170-240 O C . The utlllty of the columns Is demonstrated by the separation of substituted naphthalene derivatives, fatty acid methyl esters, and Aroclors.
Organic molten salts comprise a new class of stationary phases for gas chromatography (1-4).Above their melting points they form stable isotropic solvents. Unusually strong orientation and proton acceptor properties make them unique among selective stationary phases used for gas chromatography. One of these molten salts, tetra-n-butylammonium tetrafluoroborate, is particularly useful as a stationary phase. It has a wide operating temperature range, 170-290 “C, and can be easily coated onto diatomaceous supports to yield columns of high efficiency, 2500-3000 theoretical plates/m (4).
No previous report has dealt with the preparation of open tubular columns coated with organic molten salts. The purpose of this paper is to describe a method for preparing efficient, stable, and long-lived glass open tubular columns coated with tetra-n-butylammonium tetrafluoroborate. 0003-2700/84/0356-2509$01.50/0
EXPERIMENTAL SECTION Unless otherwise stated, all chemicals and solvents were general laboratory or analytical grade in the highest purity available. Aroclor 1254 and the fatty acid methyl ester mixture (MESRM-03)were standard reference samples from Alltech (Arlington, IL). Tetra-n-butylammonium tetzduoroborate was obtained from Aldrich Chemical Co. (Milwaukee, WI) and (3-glycidoxypropy1)trimethoxysilanefrom Petrach Systems Inc. (Bristol, PA). Glass capillary columns of various lengths and 0.25-mm i.d. were drawn from soda lime blanks (Kimble 46485, 2.0 mm i.d. X 6.5 mm 0.d.) by using a Shimadzu GDM-1B capillary column glass drawing machine (Shimadzu, Columbia, MD). Extra fast setting epoxy glue (catalog no. K-8778-00) was obtained from ColeParmer (Chicago, IL). Whisker-walled, soda lime glass capillary columns were prepared by modification of the methods described by Onuska et al. (5) and Peters et al. (6).The soda lime capillary columns were filled with a solution of ammonium bifluoride (5% w/v) in methanol and allowed to stand for 1 h. The columns were then emptied by nitrogen pressure at a velocity of ca. 2 cm/s, and residual solvent was removed with a nitrogen flow rate of about 8 mL/min overnight. The columns were sealed at both ends with a microtorch and placed in a temperature programmable oven. The oven was temperature programmed from room temperature to 350 OC at 10 OC/min and held there for 3 h. Afterwards, the columns were cooled to room temperature, the ends broken off, and the column flushed with 10-20 column volumes of methanol at ca. 2 cm/s. The columns were then dried overnight with a nitrogen flow rate of 8 mL/min. Column preparation was completed by reetching the column a second time following the sequence of operations just described. To increase the wettability of some columns cold silanization with a silane-coupling reagent waa used (7,8).After etching, the columns were filled with a solution of (3-glycidoxypropy1)trimethoxysilane in methanol (5% w/v) and allowed to stand for several hours. The first five or so coils of the column were then emptied by nitrogen pressure; the coating solution was then forced into the emptied end of the column. The columns were then statically coated as discussed below. Each etched column was filled by nitrogen pressure with a solution of tetra-n-butylammonium tetrafluoroborate (10-25 mg/mL) in dichloromethane. Each column was allowed to stand for several hours to allow trapped gas to dissolve in the coating 0 1984 Amerlcan Chemlcal Soclety