Available online at www.sciencedirect.

com
Journal of Chromatography A, 1182 (2008) 113118
Validation of a gas chromatography/thermal conductivity
detection method for the determination of the water content
of oxygenated solvents
W.K. OKeefe, F.T.T. Ng

, G.L. Rempel
The University of Waterloo, Department of Chemical Engineering, 200 University Avenue West,
Waterloo, Ont. N2L 3G1, Canada
Received 4 July 2007; received in revised form 6 December 2007; accepted 10 December 2007
Available online 23 December 2007
Abstract
A gas chromatographic (GC) method for the determination of the water content of acetone solutions containing methyl isobutyl ketone (MIBK)
and other by-products of MIBK synthesis was developed and validated. Linearity is demonstrated (R
2
=0.9999) over a broad range of water
concentrations in acetone ranging from 0.04 to 15.2% (w/w). The quantitation limit (QL) of this technique was found to be 0.346% (w/w) and
the detection limit (DL) was estimated to be 13 ppm. Validation of this analytical methods accuracy against a coulometric Karl Fischer titration
(KFT) method showed excellent agreement between these two techniques. The GC/TCD method is accurate to within a relative error of 3.49%
with respect to the KFT method for water concentration above the QL. However, the analysis of six samples below the QL showed an average
relative error of 10.96%. Consequently, the KFT method is recommended for the moisture analysis of samples for water concentrations below
0.35% (w/w). Excellent precision is realized for water concentration above the QL (RSD0.8%). Good precision (RSD=1.1%) was obtained
for the sample with the lowest water concentration investigated (0.04%, w/w). Repeatability, robustness and excellent specicity of this technique
are demonstrated.
2007 Elsevier B.V. All rights reserved.
Keywords: GC/TCD; Water analysis; MIBK; Acetone; Oxygenated solvents
1. Introduction
4-Methyl-2-pentanone, also known as methyl isobutyl ketone
(MIBK), is a valuable derivative of acetone, third in consump-
tion behind bisphenol and methyl methacrylate [1]. MIBK is
used primarily as an industrial solvent in the paints and coatings
industry and in metallurgical and solvent extractions as well as
a precursor in the production of specialty chemicals. The cur-
rent global demand for MIBK is estimated to be in excess of
300 10
6
kg/year. Increasingly stringent environmental regula-
tion and rising energy costs provide a substantial incentive for
the development of novel catalysts and highly efcient catalytic
reactor technologies for MIBK synthesis. Consequently, MIBK
synthesis remains an area of signicant research activity.

Corresponding author. Tel.: +1 519 888 4567x33979; fax: +1 519 746 4979.
E-mail address: fttng@cape.uwaterloo.ca (F.T.T. Ng).
The synthesis of MIBK involves three major reaction steps.
First acetone is dimerized to produce diacetone alcohol (DAA).
Second, DAA is dehydrated to produce mesityl oxide (MO),
an ,-unsaturated ketone, its isomer isomesityl oxide (IMO)
and water. Lastly, the C C bond of mesityl oxide is selec-
tively hydrogenated to produce MIBK. Water is co-produced
with mesityl oxide in an equimolar amount and consequently at
least one mole of water is produced per mole of MIBK synthe-
sized from acetone. In addition, water is usually present in the
acetone feedstock and is also produced from numerous possible
undesirable competing reactions which produce higher molec-
ular weight species including phorone, mesitylene, diisobutyl
ketone (DIBK), diisopropyl ether and 2,6-dimethyl-4-hydroxy-
2-heptene via condensation reactions.
The accurate quantication of water is of particular impor-
tance in the study of MIBK synthesis. The presence of water
is known to have a signicant effect on the reaction kinetics
due to its strong adsorption on acid sites on the catalyst. It is
also well known that water will inhibit mesityl oxide forma-
0021-9673/$ see front matter 2007 Elsevier B.V. All rights reserved.
doi:10.1016/j.chroma.2007.12.044
114 W.K. OKeefe et al. / J. Chromatogr. A 1182 (2008) 113118
tion and can lead to catalyst deactivation. Unfortunately, water
cannot be detected by the GC/ame ionization detection (FID)
technique typically used to quantify the organic species. Con-
sequently, many researchers proceed under the assumption that
the water concentration in the bulk may be accurately inferred
from the material balance. However, such an approach does
not account for the residual water in the solvent and reactor
system. Moreover, water may be produced by numerous unde-
sirable condensation reactions that may collectively lead to a
signicant contribution to the total water content. To complete
the mass balance, the reactants involved in these condensation
reactions must be identied and accurately measured. Uncer-
tainty in their quantication will introduce uncertainty in the
estimate of the water concentration. Therefore, a direct mea-
surement of water concentration is preferred over an inferential
method. Here, the validation of a GC technique utilizing ther-
mal conductivity detection (TCD) for the direct quantication
of water in acetone solvent is reported.
2. Experimental
2.1. GC/TCD method and apparatus
An Agilent Technologies 6890Ngas chromatograph (Agilent
Technologies Canada, Missassagua, Canada) equipped with a
Hayesep P packed column constructed of stainless steel tubing
(8 ft. 0.125 O.D.) connected to an EPCpurge packed inlet. 1/8
in. stainless steel tubing has a wall thickness of 0.028 in. and
a nominal internal diameter of 0.069 in. A TCD system on the
outlet was used to analyze the water content of acetone solu-
tions containing various organic species produced from MIBK
synthesis. The carrier gas ow rate was controlled with a state
of the art Agilent Technologies electronic pneumatic control
(EPC) module, which according to the manufacturer is accu-
rate to within 5% and repeatable to within 0.35%. The
GC system was equipped with an Agilent Technologies 7863
Series autoinjector. The liquid samples were contained in GC
autosampler vials sealed with septum caps. It was determined
that a 10 L Agilent Technologies syringe (P/N 51813361)
with a non-standard PTFE tip plunger gives excellent perfor-
mance and is recommended for the GC/TCD method described
herein. The Agilent Technologies Chemstation software (Rev.
A.09.03 [1417]) was used to control the operation of the GC
systemas well as for data acquisitionandthe analysis of the chro-
matograms. The parameters for the GC/TCD method of water
analysis (Table 1) are based on a method developed previously
in our group by Podrebarac [2].
2.2. GC calibration standards and control samples
External calibration standards were used to carry out the
moisture analysis via GC/TCD. A 4-L volume of HPLC-grade
acetone (SigmaAldrich Canada, Oakville, Canada, 99.98%)
was dried with 32.6 g of 5A molecular sieves (SigmaAldrich
Canada). The sieves were activated in air at 350

C for 3 h,
then cooled to ambient temperature under vacuum in a vacuum
ask and subsequently transferred directly into the 4-L bottle
Table 1
GC/TCD parameters for the analysis of water content of an organic mixture in
acetone solvent
Carrier gas Helium
Carrier ow rate (mL/min) 20
Inlet temperature (

C) 250
Detector temperature (

C) 250
Reference ow (mL/min) 20
Make-up ow (mL/min) 3
Injection volume (L) 1
Oven prole
Initial temperature (

C) 140
Hold at initial temperature (

C) for 2 min 140

Ramp to (

C) at 20

C/min 220
of acetone and retained there during its storage and use. Five
amber bottles (ca. 250 mL) were dried in an oven at about 80

C
overnight then sealed with screw caps and allowed to cool to
ambient temperature. The bottle of dried acetone was refrig-
erated in an explosion-proof refrigerator at temperatures below
20

C. The empty bottles were weighed precisely to 0.005 g.

Predetermined amounts of dried acetone ranging fromabout 140
to 165 g were added to the bottles and then the bottles were
sealed and re-weighed. Pre-determined amounts of high purity
water (Millipore) were added dropwise using Pasteur pipettes
to the 5 bottles of acetone to give water mass fractions ranging
from 0.04 to 15.2% (w/w) water. The bottles were sealed and
re-weighed. The water mass fraction of the ve calibration sam-
ples was calculated from the recorded masses. The calibration
standards were stored in an explosion proof refrigerator at less
than 20

C when not in use. In order to validate this analyti-

cal method, 9 stock solutions of water in acetone were prepared
in a similar fashion as described above with the exception that
smaller glass vials (ca. 20 mL) were used instead of amber bot-
tles. Three liquid samples were extracted from each of the 9
stock solutions using Pasteur pipettes and transferred to 27 GC
autosampler vials. The Pasteur pipettes and autosampler vials
were also dried overnight in an oven at about 80

C prior to use.
These 27 control samples were sealed in the autosampler vials
with septum caps and used to validate the analytical method.
2.3. Method validation
The GC/TCD method was validated following protocols
and guidelines outlined by the US Food and Drug Adminis-
tration (FDA) for the validation of analytical methods [36].
Specically, the GC/TCD method was evaluated for accuracy,
precision, intermediate precision, robustness, specicity and
linearity to demonstrate that the technique meets the FDA stan-
dards. The quantitation limit (QL) and detection limit (DL) are
reported as required. The accuracy of the technique was deter-
mined by direct comparison against a coulometric Karl Fisher
titration (KFT) method, which is known to be highly accurate.
The moisture content of the 27 control samples were accurately
measured via the coulometric KFT method using a Mitsubishi
CA-06 106 moisture meter with sample injections of 50 or
100 L depending on the nominal concentration of the control
W.K. OKeefe et al. / J. Chromatogr. A 1182 (2008) 113118 115
Fig. 1. A typical chromatogram of the water analysis via GC/TCD of a mixture
containing products of the MIBK synthesis in acetone solvent.
sample. The amounts of samples injected were weighed pre-
cisely to 0.00005 g on a Mettler AE 100 balance. The water
content of the 27 control samples was subsequently measured by
GC/TCDfor comparison. The precision of the GC/TCDmethod
was ascertained by 10 successive analysis for each of 3 samples
which span the range of the GC/TCD calibration curve includ-
ing the high and low calibration standard and one of the control
standards of intermediate composition. Intermediate precision
is demonstrated from the GC/TCD analyses of 15 of the control
samples on a separate Agilent Technologies 6890N GC sys-
tem using the same Hayesep P column. The robustness of the
GC/TCDtechnique was evaluated by the repeated analysis of 15
of the control samples on the principal GC instrument used in
this work after over 48 h had expired and by directly manipulat-
ing GC parameters to see the effect on the resultant quantitative
analysis. The specicity is demonstrated fromthe analysis of the
chromatogrampresentedherein. Linearity, range, the QLandDL
were determined from the statistical analyses of the calibration
curves.
3. Results
3.1. Specicity
Fig. 1 shows a typical chromatogramfor a sample containing
water, MIBK, mesityl oxide and other products from the syn-
thesis of MIBK from acetone, which is the solvent. The 1 L of
sample injected to the GC had 0.0178 mg of water and a water
concentration of 2.26% (w/w). The chromatogram in Fig. 1
demonstrates excellent specicity with water eluting early in
the analysis corresponding to the sharp peak observed around
1.28 min, followed several minutes later by acetone at around
4.5 min and other organic species which elute in broad convo-
luted peaks much later. The minor peak at 0.47 min corresponds
to air. Fig. 1 also demonstrates that the water peak is base-
line resolved meaning that the signal is at the baseline prior
to elution of the water peak and returns to the baseline prior
to the elution of the next analyte. The Chemstation software
conrmed this observation by reporting the peak code as BB,
meaning baseline to baseline, for the water peak in the chro-
matogramreport. Since the organic peaks are generally found to
be severely convoluted, broad and lack symmetry, this technique
should not be used to quantify the organic species. However, the
water concentration may be calculated from the water peak area
with excellent accuracy by correlation to external GC standards
of known moisture content in HPLC-grade acetone.
3.2. Linearity
The ve calibration standards described in Section 2.2 were
analyzed by GC/TCD. The resultant calibration curve showed
excellent linearity with R
2
=0.9999 over 5 observations span-
ning a water content of 0.0415.2% (w/w). This exceeds the
FDA requirement of R
2
0.999 over a minimum of 5 observa-
tions [3,4]. The calibrationcurve, whichrelatedthe mass fraction
of water (%, w/w) on the ordinate to the GC/TCD peak area
(25 Vs) on the abscissa, had a slope of 0.004221 and an inter-
cept of 0.004517. The standard error of estimate was 0.0548%
(w/w) and the relative standard deviation (RSD) was 0.79%.
3.3. Quantitation limit
The QLis dened as the lowest concentration of an analyte in
a sample that can be can be estimated with acceptable accuracy
and precision. The QL can be calculated from the slope (S) of
the calibration curve and the standard deviation of responses ()
as outlined in Eq. (1) [4]. The standard deviation of responses
is simply the standard deviation of the y-intercepts for a series
of calibration curves [4]. Ten calibration curves were arbitrarily
selected and the standard deviation of their y intercepts were
calculated giving =0.034152 wt.%. Using this value for and
the slope of the calibration curve (S) reported in Section 3.2,
Eq. (1) gives QL=809.07 (25 Vs). Substituting this value
of QL with the dimensions of peak area into the calibration
curve reported in Section 3.2 gives QL=0.346% (w/w).
QL =
10
S
=
10 0.034152 (%, w/w)
0.0004221 ((%, w/w)/Vs)
= 809.07 Vs = 0.346 (%, w/w) (1)
3.4. Detection limit
The DL of an analytical method is dened as the lowest con-
centration of a particular analyte in a solution for which the
analyte can be identied with 99% condence, without nec-
essarily quantifying the amount of analyte. The GC detection
limit was estimated using Eq. (2) from the assessment of the
magnitude of the background noise and the slope of the calibra-
tion curve reported in Section 3.2 [4,7]. in Eq. (2) represents
the standard deviation of the analytical background response
and can be estimated from the standard deviation of multiple
estimates of the magnitude of the background noise [4,7]. The
magnitude of the background noise was calculated using the
Agilent Chemstation software noise determination feature and
represents 6 times the standard error of a regression line through
the baseline signal through the specied interval. The baseline
116 W.K. OKeefe et al. / J. Chromatogr. A 1182 (2008) 113118
noise for the retention time interval from0.7 to 1.1 min (i.e. after
the elution of the minor peak observed at around 0.47 min and
before the elution of the water peak at 1.3 min) was analyzed for
the chromatograms of the 5 calibration standards for the cali-
bration curve reported in Section 3.2. The average background
noise was 0.0850 [25 V] and the standard deviation of these
5 baseline noise measurements was 0.009445 [25 V]. Sub-
stitution of this value as an estimate of the standard deviation
of the analytical background response [25 Vs] into Eq.
(2) and using the value of the slope (S =2368.89 ([25 Vs/%,
w/w]) obtained from a plot of the calibration data from Fig. 2
with (%, w/w) on the abscissa and peak area [25 Vs] on the
ordinate, gives a detection limit of 13 ppm.
DL =
3.3
S
=
3.3 0.009445
2368.89
= 0.00013157 10
5
(%, w/w) = 13 ppm (2)
3.5. Accuracy
The accuracy of the GC/TCD method was validated against
a coulometric KFT method, which is considered to give the
accepted true value. 27 control samples were analyzed by each
method. One obvious outlier was observed and rejected from
the analysis. The analyses of the remaining 26 control sam-
ples presented in Fig. 2 show an excellent agreement between
the two techniques. The average relative error of the GC/TCD
method with respect to the KFT technique was 3.49% for sam-
ples above the QLwith water concentrations ranging from0.7 to
14.8% (w/w). However, some disagreement between these two
techniques is evident at very low moisture concentrations. Six
measurements near the lowest concentration (ca. 0.27%, w/w),
below the QL for the GC/TCD method, gave an average relative
error of 10.96% with respect to the KFT analyses. For higher
moisture concentration, excellent agreement is observed. The
Fig. 2. Validation of the GC/TCD method against a coulometric Karl Fischer
titration method (26 sample analyses by each technique).
Table 2
Precision of the GC/TCD analysis method
Analysis no. Peak area [Vs]
Control 3 Std. 1 Std. 5
1 18,497 219 37,306
2 18,582 218 37,450
3 18,524 224 37,418
4 18,506 222 37,591
5 18,402 224 37,524
6 18,295 223 37,313
7 18,232 219 37,017
8 18,318 217 37,184
9 18,181 220 37,848
10 18,202 221 37,371
Average 18,374 221 37,402
SD 147.35 2.45 227.71
RSD (%) 0.80 1.11 0.61
95% Condence
bounds on average
(Vs) 91.3 1.5 141.1
(%) 0.50 0.69 0.38
GC/TCD measurements of samples with water concentrations
3% (w/w) and higher were accurate to within a relative error of
2.12% with respect tot the KFT method.
3.6. Precision
The precision of the GC/TCD technique was ascertained by
10 successive analyses for each of three samples spanning the
range of the calibration curve. The results are illustrated in
Table 2. The GC/TCD technique exhibits excellent precision.
The samples at the intermediate and high concentration were
highly repeatable with an RSD well below 1%, which exceeds
the FDAguideline which stipulates an RSDless than or equal to
1% over 10 injections [3]. Calibration standard 1, which had a
moisture content of 0.04 wt.%, well belowthe QL of this analyt-
ical method reported in Section 3.3, showed a reasonably good
repeatability with an RSD of 1.1%.
3.7. Intermediate precision
In order to assess the reproducibility of the GC/TCDmethod,
the 15 control samples were analyzed for moisture content
via GC/TCD using a separate GC, denoted GC-2 in Fig. 3,
using the same Hayesep P column. The results were compared
to those obtained previously on the principal GC used in this
work, denoted GC-1 in Fig. 3. The results in Fig. 3 demon-
strate that the analytical technique is highly reproducible. As
expected, there was some disagreement between these two GC
analyses for data below the QL. The average relative error for
the analyses at the lowest concentration (ca. 0.3%, w/w) was
15.1% for GC-2 with respect to GC-1. However the two instru-
ments gave excellent agreement for the data above the QL with
an average relative error of 0.85% for GC-2 with respect to
GC-1.
W.K. OKeefe et al. / J. Chromatogr. A 1182 (2008) 113118 117
Fig. 3. Reproducibility of the GC/TCD method on a separate GC system.
3.8. Robustness
The robustness of the GC/TCD method of moisture analysis
was demonstrated by the repeated analysis of 15 control samples
after a period of over 48 h had expired. The GC instrument was
calibrated on both days. The results in Table 3 show very good
agreement for the 12 measurements above the QL. Therefore it
can be concluded that a delay time prior to analysis, even on
the order of days, will not signicantly affect the results. How-
ever, the 3 measurements below the QL did show inconsistent
results.
Two other tests were carried out to assess the robustness of the
analytical method. The purpose of these tests was to determine
the repeatability of the method when an error () was deliber-
ately introduced into a GCparameter, specically the carrier gas
ow rate and oven temperature. The GC system was calibrated
and a baseline analysis was obtained following the established
GC method. After which, the GC system was not recalibrated
with the corresponding error introduced into the GC parameter
prior to these experiments, otherwise the recalibration would
account for the perturbation introduced to the GC parameter. A
sample from stock solution 3 had its moisture content measured
three additional times with the temperature of the initial isother-
Table 3
Repeated GC/TCD analyses of samples after 3 days
Water analysis (%, w/w)
14 January 17 January Change (%)
0.29 0.34 19.0
0.29 0.34 17.9
0.29 0.35 19.1
3.36 3.46 3.02
3.36 3.42 1.76
3.37 3.42 1.37
7.98 8.22 3.05
8.07 8.15 1.07
8.07 8.10 0.30
12.39 12.55 1.28
12.43 12.53 0.79
12.56 12.51 0.36
14.63 14.94 2.11
14.83 14.93 0.65
14.90 14.84 0.39
mal plateau in the oven prole reduced to 135 from140

C. Note
that the water peak elutes during this initial isothermal condi-
tion. The sample was also analyzed for moisture content three
times with the carrier gas ow rate reduced by 10% from 20 to
18 mL/min. The results in Table 4 show that the change in tem-
perature of the rst isothermal plateau in the GC oven prole
did not have a signicant effect on the resulting estimate of the
water concentration. However, the change in carrier ow rate
did signicantly affect the result. It can be concluded that the
GC/TCD method is relatively insensitive to uctuations in the
oven temperature. However, it is important to ensure that the
carrier gas ow is carefully controlled.
3.9. Range
The range of the analytical method is bounded on the lowend
of concentrationbythe QL, whichbydenitionis the lowest con-
centration that can be measured with acceptable accuracy, and
bounded on the high end of concentration by the concentration
of the calibration standard having the highest water concentra-
tion, since extrapolation is not recommended. Thus, the range of
the GC/TCD method demonstrated in this study is from 0.346
to 15.2%(w/w). Linearity, precision and repeatability have been
demonstrated over this range.
Table 4
Evaluation of robustness by direct manipulation of GC parameters
Description Peak area (25 Vs) Water (%, w/w) Average (%, w/w) Change (%)
Benchmark, control sample 3
with no changes to GC/TCD
method
1,7844 8.038
17,903 8.064 8.053
17,890 8.058
Lower initial isotherm from
140 to 135

C
17,834 8.033
17,924 8.073 8.056 +0.034
17,898 8.062
Reduce carrier gas ow rate
by 10% from 20 to 18
(mL/min)
19,467 8.759
19,557 8.798 8.804 +9.31
19,678 8.853
118 W.K. OKeefe et al. / J. Chromatogr. A 1182 (2008) 113118
4. Conclusions
A GC/TCD method for the determination of water concen-
tration in acetone solutions was developed and validated. The
GC/TCDmethod is highly accurate, to within 3.49%and repeat-
able with an RSD less than 1% based on data obtained for water
concentrations ranging fromabout 0.79 to 15.2%(w/w). The QL
for this analytical method was estimated to be 0.346% (w/w).
The Karl Fischer titration method is recommended for routine
moisture analysis of samples below this threshold value. It is
reasonable to assume that this GC/TCD technique for water
analysis may be applied to other systems of oxygenated organic
solvents.
Acknowledgements
The authors thank professor R. Legge for the use of his Karl
Fischer Titrator and Joe Clifford and Amanda Warboys for their
assistance in the operation of the instrument. Funding for this
project was provided by the Natural Sciences and Engineer-
ing Research Council (NSERC) of Canada. Financial support to
W.K. OKeefe in the form of an Ontario Graduate Scholarship
by the Government of Ontario, Ministry of Training, Colleges
and Students is gratefully acknowledged.
References
[1] D. Muthusamy, I. Fisher, in: J.I. Kroschwitz (Ed.), Methyl Isobutyl Ketone.
The Kirk-Othmer Encyclopedia of Chemical Technology, vol. 16, fth ed.,
2004, p. 329.
[2] G.G. Podrebarac, Development and modelling of a new catalytic distillation
process, Ph.D. Thesis, The University of Waterloo, Waterloo, Canada, 1996.
[3] Center for Drug Evaluation and Research (CDER), US Food and Drug
Administration (FDA), Reviewer Guidance: Validation of Chromatographic
Methods, 1994; www.fda.gov/CDER.
[4] Center for Drug Evaluation and Research (CDER), US Food and Drug
Administration (FDA), Guidance for Industry, Q2B Validation of Ana-
lytical Methods Protocol: Methodology, 1996; http://www.fda.gov./CDER/
guidance/index.htm.
[5] Validation of Analytical Procedures: Methodology USDepartment of Health
and Human Services, Food and Drug Administration Center for Veterinary
Medicine Guidance for Industry, 1999.
[6] R. Brown, M. Caphart, P. Faustino, R. Frankewich, J. Gibbs, E. Leutzinger,
G. Lunn, L. Ng, R. Rajagopalan, Y. Chiu, E. Sheinin, LCGC 19
(2001) 74.
[7] V. Gambaro, C. Benvenuti, L. De Ferrari, L. DellAcqua, F. Fare, Farmaco
58 (2003) 947.