Influence of reagent purity on the ion chromatographic determination
of bromate in water using 3,3'-dimethoxybenzidine as a
prochromophore for photometric detectionf
Edward T. Urbansky* and Stephanie K. Brown
United States Environmental Protection Agency, Office of Research and Development,
National Risk Management Research Laboratory, Water Supply and Water Resources
Division, 26 West Martin Luther King Drive, Cincinnati, OH 45268, USA.
E-mail: urbansky.edward@epa.gov; Fax: +1 513 569 7658; Tel: +1 513 569 7655
Received 8th August 2000, Accepted 28th September 2000
First published as an Advance Article on the web 1st November 2000
Variable availability of the purified dihydrochloride salt of 3,3'-dimethoxybenzidine (DMB; orfAo-dianisidine)
led us to investigate the effects of reagent purity on the analytical results obtained when this reagent is used in
the photometric determination of the disinfection byproduct bromate. After analyte ions are separated by ion
chromatography, a solution of DMB (post-column reagent) is added to the eluate and the DMB is oxidized,
thereby producing a chromophore detected by its absorbance. Although some commercial products of
undefined grade performed well, others did not. Variability was also observed between lots of purified material.
Sensitivity at low concentrations (<5 ug L~' BrOs") varied by a factor of up to 10. In some cases, the lower
limit of detection for photometric detection was greater than that obtained using conductivity detection, as high
as 5-7 ug L"1 BrO3~. An impurity or several impurities are suspected to be responsible for deviations from
linearity at low analyte concentrations. This investigation underscores the need for ensuring reagent purity in
environmental analyses. Ideally, chemical manufacturers will meet the needs of analytical chemists who test
potable water and begin producing a high grade material in sufficient quantities to meet monitoring
requirements. The establishment of third-party standards for a spectrophotometric grade of DMB-2HC1 would
be helpful in ensuring that a variety of manufacturers could supply products of uniformly high quality that
would be suitable for the measurement of bromate in public drinking water supplies.
Aim of investigation
Benzidine dyes have a long history of use in spectrophotometric
determinations of oxidizing agents.1'2 Of these, alkoxy-
substituted benzidines, especially 3,3'-dimethoxybenzidine
(DMB; o-dianisidine), have often been used because of their
favorable properties.3 Benzidine dyes act as prochromophores;
the actual chromophores are produced by a redox reaction
between the analyte and the benzidine dye. In the past, this
approach has been applied to the quantitation of higher
valence metal oxyanions, e.g., chromate4'5 and vanadate,5 as
well as non-metal oxidants, e.g., N-bromosuccinimide,6
hydrogen peroxide,7 and periodic acid.8
Most recently, DMB has been used for the spectrophoto-
metric determination of the disinfection byproduct bromate
(BrO3~) in bottled waters9 and drinking water in public
systems.10 In order to further reduce the lower limit of
detection, EPA's Method 317J has been developed to include
the addition of an acidified solution of DMB to the eluate
stream of the ion chromatograph after the conductivity
detector; this has been referred to as post-column reaction
by some investigators.11 While passing through a heated
reaction coil, the DMB is oxidized to form the chromophore.
tThis is the work of United States government employees engaged in
their official duties. As such it is in the public domain and exempt from
copyright. © US government.
{Information on EPA's Method 317 may be obtained from the
Technical Support Center, Office of Ground Water and Drinking
Water, Environmental Protection Agency, Cincinnati, OH 45268,
USA.
The eluate stream passes through a UV-visible absorbance
photometric detector, and the bromate concentration is
determined essentially using the Beer-Lambert law. Calibra-
tion curves are not linear; however, we do not believe that
deviations from the Beer-Lambert law are responsible for the
apparent concavity of the curves. Part of this investigation
stems from a curiosity about the cause of the nonlinear
calibration curves.
Over the past year, experience has suggested that the supply
of purified 3,3'-dimethoxybenzidine dihydrochloride is vari-
able, apparently due to difficulties associated with refinement.
However, products sold as lower grades have been widely
available. As a consequence, we set out to determine the impact
of using reagents sold as lower grades on the analytical results
to decide whether these might serve as satisfactory substitutes
for the unavailable purified reagent. At present, manufacturer
stocks of the material appear to be adequate to meet research
needs, but that does not negate the original rationale for
undertaking this study. Moreover, we hoped that we might
discern an explanation for the nonlinear calibration curves
(vide supra).
Experimental
3,3'-Dimethoxybenzidine solutions
The prochromophore was purchased from a number of
manufacturers as the dihydrochloride salt, DMB-2HC1:
Acros (through Fisher Scientific, Pittsburgh, PA, USA),
Aldrich (Milwaukee, WI, USA), Avocado/Johnson Matthey
(Ward Hill, MA, USA), Sigma (St. Louis, MO, USA) and
DOI: 10.1039/b006498j
This journal is (
J. Environ. Monit., 2000, 2, 571-575
) The Royal Society of Chemistry 2000
571
-------
Spectrum (."New Brunswick, "NJ, XJSA). These products ranged
in color and consistency from white crystals to a yellow powder
to pink/grayish clumps. Some appeared to have inhomogeneity
in color. The manufacturers' reported assays were not
confirmed independently. DMB solutions were prepared to
contain ~ 1.2 M HNO3 and ~ 0.042 M KBr as prescribed in
Method 317.0. Except for the DMB salt, the other commercial
reagents were identical in brand and grade to those used
previously.10 Millipore (Milford, MA,.USA) Milli-Q water
(p>18 Mflcm) was used to prepare all solutions. Although
Sigma sells a purified grade, no authoritative organization has
published reagent specifications for this material. Catalogs
report the following assays: Acros, 99%; Aldrich, technical;
Aesar, 99%; Spectrum, unspecified; Sigma, purified.
The material from Aldrich would not dissolve in the
methanol without the addition of a few drops of HNO3(aq).
The mixture prepared from the Acros product contained a
large amount of finely divided suspended matter and was
therefore filtered through Millipore 0.22 urn GS nitrocellulose
membrane niters. In the case of the acidification, it is
questionable whether that represents a significant procedural
deviation. On the other hand, filtering the sample is probably
significant—not necessarily in and of itself, but because it
indicates observable amounts. of impurities. Otherwise, the
preparations were unremarkable.
Bromate standards
Bromate standards were prepared from a commercial
product purchased from Spex (Metuchen, NJ, USA). For
individual injections, the commercial standard was diluted
into Milli-Q water using a dispensing pipettor. Additional
standards (1.00 gL"1 bromate) were prepared from ACS
reagent grades of NaBrO3 and KBrO3 and kept refrigerated.
These were serially diluted and used to verify the concentra-
tion in the commercial standard. The stock solutions were
used within 30 days of their preparation. Dilutions were
made immediately prior to chromatographic analysis. The
following bromate concentrations were used to assess
method performance: 0, 0.50, 1.0, 2.0, 5.0, 10.0, 30.0, 50.0,
60.0, 70.0, 80.0, 90.0, 100 ug IT1. The same stock standard
was used for all of the data reported herein.
Ion chromatography
A Dionex (Sunnyvale, CA, USA) DX500 ion chromato-
graph equipped with a 4 mm AG9HC guard and an AS9HC
separation column was used. The eluent was 9.0 mM
Na2CO3 solution. CD20 conductivity and AD20 absorbance
detectors were used. The DMB solution was added via a
pneumatic system, and the post-addition equipment (mixer,
500 uL heated reaction coil) were as previously
described.10'11 PeakNet software was used for instrument
operation, data acquisition, and data analysis. The con-
ductivity values (peak areas) for the lOOugL"1 standards
were used as a check of instrument performance. These
represent identical injections since the prochromophore is
added downstream of the conductivity detector. The
variation in peak area between the maximum (35300) and
minimum (28700) was 23%; average, 31100; estimated
standard deviation of the mean, 1100; relative standard
deviation, 8.2%.
Results and discussion
Sensitivity and linearity of the calibration curve
As demonstrated in Fig. la, all of the reagents produce linear
calibration curves for [BrO3~]>20 ug L""1. In order to produce
the chromophore, the bromate is reacted with bromide under
acidic conditions to rapidly comproportionate to dibromine,
0.0 1.0 2.0 3.0 4.0 5.0
Bromate concentration/fig L~1
Fig. 1 Response to bromate for 3,3'-dimethoxybenzidine (DMB) using
absorbance detection. Legend for brands/lots: Sigma 10K5082 (D),
Spectrum (•), Sigma 95H016 (O); Acros (•), Aldrich (A), Avocado/
Johnson Matthey (T). (a) High concentration, (b) Near the limit of
detection.
Br2(aq), [eqn. (1)]. By analogy with other chemical systems,7 it
is believed that dibromine reacts via a radical mechanism to
give the chromophore [eqn. (2)].
5Br-+BrO3~-t-6H+-»3Br2-t-3H2O (1)
,OCH3
NH2 .- chromophore (2)
The main objective was to ascertain whether other grades of
DMB could be used in this assay, and detailed investigations
into the deviations from linearity were not pursued. However, a
plausible explanation will be offered, and the discussion is
geared towards explaining why it is reasonable rather than a
rigorous confirmation. The behavior exhibited in Fig. Ib is
consistent with a reactive impurity that can react with Br2 faster
than DMB can. In other words, the rate constant for eqn. (3)
§Drinking water regulations generally write contaminant concentra-
tions in mass-based units, e.g., ugL"1. However, molarities are
required for limiting-reagent calculations. In addition, equilibrium
constants and rate constants are nearly always expressed using
molarities.
572 J. Environ. Monit., 2000, 2, 571-575
-------
rmlst be many times larger than the rate constant for eqn. (2)
because the impurity is perhaps < 1% w/w.
Br2+impurity -» non-absorbing products
(3)
The kinetics of the redox reaction between oxidizing bromine
species and DMB have been investigated; however, the original
work used bromate and bromide at roughly stoichiometrically
equivalent concentrations.12 Consequently, the authors con-
cluded that HOBr was primarily responsible for oxidizing the
DMB. In Method 317.0, there are considerable excesses of
hydrogen ion and bromide ion. A bromate concentration§ of
100 ugL"1 (0.79 uM) would consume 3.9 uM Br~, which is
negligible relative to the post-mixing [Br~]0=
0.015 M. Consequently, HOBr is rapidly converted to Br2.
Based on rate and equilibrium constants reported by Eigen and
Kustin [eqns. (4) and (5)],13 we calculate [BrJ/
[HOBr]=1.38xl06 at [Brj=0.015M and [H+] = 0.42M
(post-mixing concentrations).' Moreover, the pseudo-first-
order rate constant for dibromine formation from the
comproportionation of hypobromous acid and hydrobromic
acid should be 9.9 x 107 s~' at these reactant concentrations.
Br2+H2O ;=± HOBr+Br~ + H+, #=4.46 x 1Q-9 (4)
d[Br2]/df=(1.6x 1010 M~2 s-')[Br-][H+][HOBr] (5)
Given the favorable conditions for the formation of dibromine
in our reaction system, we believe that the primary oxidizing
agent attacking the DMB is most likely Br2 and not HOBr.
However, HOBr probably carries some fraction of the reaction.
At this bromide concentration, a negligible concentration of
dibromine is also tied up as the presumably unreactive
tribromide ion, Br3~.
Although there may be several reactive impurities, we will
assume that the competition for Br2 can be represented by only
eqns. (2) and (3). The kinetics of concurrent parallel reactions
are quite complex,14 and resolving them is impossible in this
case because neither the identities nor the initial concentrations
of the impurities are known. Consequently, combining all
impurities as a net reaction must be recognized as a practical
necessity.
At higher bromate concentrations, we can conclude that
eqn. (3) progresses to completion, or at least equilibrium, since
all of the calibration curves in Fig. la eventually exhibit
linearity. Thus, the linear portion of the calibration curve can
be assumed to result entirely from eqn. (2). If we extrapolate
back to the ordinate from the linear region, we can estimate the
amount of bromate effectively consumed by the impurity.
Thus, the apparent loss of bromate, A[BrO3~], due to the
impurity may be determined from the ^-intercept of the linear
region of the calibration curve. Using this approach for the
Avocado product, we find that up to 10 ug L BrO3~ can be
consumed by the impurity whereas the Spectrum had no loss.
If a highly reactive impurity is in fact present, a possible
curative strategy can be speculated: the initial addition of
10 ug L~l (0.39 uequiv L~') BrO3~ to the DMB solution. The
added oxidant should cancel out the apparent loss of bromate
recovery by reacting away all of the impurity. Alternatively,
adding 0.39 uequiv L"1 of any other suitably facile oxidant
(e.g., Cr2O72", H2O2, Ce4+, VO43")H might also be successful.
Even if a small excess of the oxidant were added, the only effect
would be to raise the baseline. Of course, this may be
accompanied by increases in noise or difficulty in peak
integration and would require further investigation. This line
of research was not pursued because high purity reagents were
^Because the chromogenic reaction is based on an oxidation-
reduction, the consumption of oxidant must be expressed in terms of
normality. It cannot be expressed in molarity since the stoichoimetry is
known for each of the oxidants.
available for actual analyses and deviating from the method as
written was viewed as undesirable. Such a strategy could
perhaps be employed if purified DMB-2HC1 cannot be
obtained and the performance is shown to be satisfactory.
Because the sensitivity of the method (slope of the
calibration curve) at low analyte concentrations is affected
by the identities and concentrations of the impurities,
individual lots of DMB-2HC1 should be tested to determine
the lower limit of detection (LOD) rigorously. From a
methodological standpoint, such rigor is not required for
Method 317.0 since a calibration check standard is used to
assess performance. For the less pure reagents, LODs cannot
be calculated using an approach based on the linear dynamic
range. After all, the dynamic range is decidedly nonlinear for at
least two of the curves in Fig. Ib. EPA's method detection
limit15 (MDL) requires that the property being measured and
the instrumental response be directly proportional. This
restriction may be relaxed somewhat when either the domain
and range of calibration region are sufficiently small that the
curve may be approximated by a linear function or when the
DMB is sufficiently pure. DMB purity appears to be
responsible for the linearity of Sigma lot 95H016 (O) and
Spectrum (•). Because Sigma lot 95H016 was used for the
previous study,10 the MDL calculation offered there was in fact
valid. While there are methods for approximating LODs for
nonlinear response,16 we will not address them here. Clearly,
the product purchased from Spectrum gave the most linear
response (R2=0.9998; slope =3700 ±17; j-intercept =
-2400 + 2300). It also gave the highest sensitivity. Therefore,
it is concluded to be the purest of the products we tested.
Furthermore, the behavior exhibited by this product indicates
that a highly reproducible extent of reaction is possible at 60 °C
as opposed to insufficient or variable reaction time.
In the case of Sigma lot 10K5082 and the Aldrich product,
the calibration curves can be satisfactorily fitted to an equation
of the form y^wf+bx (y = area, x=concentration) near the
detection limit, i.e., [BrO3~]<5 ugL"1 (Fig. Ib). At first
glance, the linear regression parameters suggest that linear
fits are acceptable, but the higher concentration data bias the
unweighted fits. The plots in Fig. 1 are decidedly nonlinear
when accounting for the whole concentration domain. For
Sigma 10K5082, unweighted linear regression gives
y=(3480±50)x with R2=0.991, and unweighted quadratic
regression gives y = (7.3 + 0.8)x2+(2830 ±80)x with
R2=0.9997. For Aldrich, unweighted linear regression gives
j>=(3150+40)* with R2=0.998, and unweighted quadratic
regression gives 7 = (4.8±0.5)x2+(2710 + 50)x with
R2 = 0.9998. Although the regression coefficients suggest that
the quadratic fit is better, the peak areas predicted for 0.5 and
1.0 ug L""1 bromate, are too high (by factors of ~3 and ~2,
respectively) for both products. It is not possible to generate a
simple function that fits the entire data set for either of these
products. Nevertheless, regions of the curve may be reasonably
fitted with simple functions so that the entire curve could be
defined using a piecewise function. Of course, piecewise
functions are undesirable for calibration curves since they
are conventionally defined for subsets of the domain rather
than the range. This does not necessarily preclude their use,
however, because samples may be screened to estimate analyte
concentrations or all samples may be known to have analyte
concentrations within a region of the curve that can be fitted by
a moderately simple function.
At a bromate concentration of about l.OugL"1, the
absorbance for these two DMB solutions was about half
that expected based on the high concentration linear fits. It is
important to point out that tailing can sometimes account for
apparent loss of the absorbance signal, especially if peak height
is used. To prevent this, each chromatogram was evaluated to
assess the performance of the automatic integrator, and peak
areas were always used (not peak height).
J. Environ. Monit., 2000, 2, 571-575 573
-------
Neither the material from Avocado nor that from Acros was
capable of attaining the level of sensitivity required for the
determination of bromate in most potable water supplies.
Chromatograms using the Acros product were particularly
noisy, making integration of the peaks extremely difficult and
designation of the baseline somewhat arbitrary. Presumably,
the noise in the absorbance traces is the result of scattering by
particulates that passed through the filter membrane. As noted
in the Experimental, a considerable amount of undissolved
material was present in the DMB reagent prepared from the
Acros product, and so it was filtered before use. The existence
of filterable material suggests the original commercial product
contained significant impurities. Consequently, the final
concentration of DMB was most likely lower than what
would be expected based on the tared mass of DMB-2HC1.
This may be responsible for the reduced slope relative to the
other products. In fact, it suggests the material contains about
50% w/w DMB-2HC1. For these products, the slopes of the
calibration lines at higher bromate concentrations
(> 10 ug L"1) are 50-70% of the slopes for the other products.
That notwithstanding, the calibration graphs for these two
products remain more nearly linear over the whole concentra-
tion domain than the Sigma 10K5082 or Aldrich ones. Rather
than the highly reactive impurity suggested above for the Sigma
10K5082 or Aldrich, this behavior suggests an inert impurity
whose presence simply lowers the concentration of the
prochromophore. The end result must be a kinetic consequence
rather than a limiting reagent problem. Even if the commercial
products were demonstrated to contain only 50% of the active
principle, there would still be at least a 100-fold excess of DMB.
There are a number of ways of synthesizing DMB, and we
did not contact manufacturers to see what scheme(s) they were
using. As a result, impurities may vary from vendor to vendor
and from lot to lot. Therefore, new lots from any of the
manufacturers whose products we used might outperform the
lots we tested; conversely, new lots might also underperform.
The presence of impurities is not unique to DMB, and has been
reported for KI used in the determination of halogen
oxyanions.17'18 Given this considerable variability, an estab-
lishment of third-party criteria for a spectrophotometric grade
of 3,3'-dimethoxybenzidine dihydrochloride would be bene-
ficial, for example, an American Chemical Society reagent
grade specification. Preferably, any written standards would
require a minimal level of performance in this assay because the
sensitivity to impurities is quite high. It is worth mentioning
that investigations on the determination of metal oxyanions •
involved measuring concentrations > 1 mg L~' and therefore
would not have experienced this phenomenon to any
observable extent.
The fact that two of the products produced virtually linear
calibration curves, combined with the fact that all of the
reagents produce linear curves at high concentration, suggests
that deviations from the Beer-Lambert law are not responsible
for the nonlinearity at lower bromate concentrations. It is well-
known that deviations from the Beer-Lambert law occur at
higher and not lower chromophore concentrations.19
Instrument performance
In our experience, some parts of the system (e.g., tubing,
mixers) are prone to clogging or restriction. Presumably, this is
effected by precipitation of salts of the protonated DMB or
possibly impurities either initially present or which form after
dissolution. If the flow of the DMB solution is reduced, the net
impact on signal is of course the same as that from an impure
or degraded batch of DMB. Sometimes the signal is simply
attenuated, but other times it exhibits quadratic behavior.
Several DMB solutions were discarded that were probably fine
because of an initial assumption that the equipment was
behaving properly and the prochromophore solution was
faulty. After trying several DMB solutions, other explanations
were sought for reduced performance. Perhaps the instrument
and other apparatus should have been examined first. This
raises an important issue, specifically, that it is necessary to
continually verify the flow of the DMB solution. This can be
done routinely by measuring total ehiate flow, assuming that
the eluent pumps are functioning properly. The DMB solution
is dispensed using a pneumatic system rather than a pump that
maintains a constant volume per time flow; accordingly,
restrictions anywhere result in reduced flow. In addition, any
coating on the wall of the reaction coil reduces the available
volume and thus reduces the time the reactants are heated,
again promoting reduced recovery.
Precipitation and staining have been observed in the reaction
coil, the mixer, and the tubing. Attempts to clean these items
using a combination of solvents, mineral acids, and bases were
of questionable utility. Somewhat surprisingly, we found that
equipment could appear to be visually clean, but still have poor
performance. In particular, the heated reaction coil—even
though it is constructed of PTFE-—appears to be resistant to
cleaning and must periodically be replaced. When no other
explanation could be found, replacing the tubing and reaction
coil restored satisfactory performance. Thus, it seems that some
changes in the apparatus are not detectable by gross
examination alone. Microscopic or more detailed inspections
of the surfaces were not pursued. This phenomenon may have
been worsened by the use of a range of DMB products, some of
which appear to have contained relatively significant amounts
of impurities of unknown solubilities.
After some initial work that showed the magnitude of these
instrumental problems and several part replacements, all of the
analyses were repeated. During that time and afterwards, the
performance of the ion chromatograph system was reconfirmed
by measuring flow rates and rerunning calibration standards
using the prochromophore solutions prepared from Spectrum
DMB-2HC1. Furthermore, we took the precaution of con-
tinually cleaning out the DMB delivery system and flushing
large volumes of eluent through the column. This seemed to
prevent the problems discussed earlier. We suspect that
impurities in lower grades of DMB-2HC1 may be responsible
for some of the problems; thus, additional caution is warranted
in terms of protecting the equipment from degradation should
these materials be used.
Conclusions
Purity of the prochromophore is of utmost importance in
obtaining satisfactory sensitivity. Impurities and insoluble
matter reduce precision and accuracy and can raise LODs to
unacceptably high levels for photometry, making conductivity
more reliable. Impurities are certainly present in some grades of
DMB. These appear to be at least partly responsible for
deviations from linearity. These effects may change in some
drinking water matrices and require further investigation. It
may be possible in some circumstances to react away
impurities, but not if strict adherence to the EPA method is
required. A stable supply of pure DMB-2HC1 must be assured
for this method to be used effectively in regulatory compliance
monitoring. Ideally, criteria for grading this reagent will be set
by some independent authoritative body, such as an analytical
reagent committee.
Acknowledgements
We acknowledge helpful comments from Daniel P. Hautman
of EPA's Office of Water Technical Support Center. Mention
of specific brand names or manufacturers should not be
construed as endorsement or recommendation for use by the
US government.
574 J. Environ. Monit., 2000, 2, 571-575
-------
References H
1 E. Bishop, in Indicators, ed. E. Bishop, Pergamon, Oxford, 1972,
pp.568-588.
2 S. J. Lyle, Talanta, 1959, 2, 293, and references cited therein.
3 D. I. Rees and W. I. Stephen, Talanta, 1959,2, 361, and references
cited therein. 12
4 K. L. Cheng and B. L. Goydish, Chem. Anal., 1963, 52, 73, and
references cited therein. 13
5 M. Ariel and J. Manka, Anal. Chim. Acta, 1961, 25, 248. 14
6 H. Sanke Gowda, R. Shakunthala and U. Subrahmanya, Analyst,
1979, 104, 865, and references cited therein. 15
7 M. K. Beklemishev, A. L. Kapanadze, N. V. Bakhilina and
I. F. Dolmanova, Talanta, 2000, 51, 389. 16
8 M. Guernet, Bull. Soc. Chim. France, 1964, 478. 17
9 C. R. Warner, D. H. Daniels, F. L. Joe, Jr. and G. W. Diachenko, 18
FoodAddit. Contam., 1996, 13, 633. 19
10 H. P. Wagner, B. V. Pepich, D. P. Hautman and D. J. Munch,
/. Chromatogr.. A, 1999, 850, 119.
H. P. Wagner, B. V. Pepich, D. P. Hautman and D. J. Munch,
Method 317.0. Determination of inorganic oxyhalide disinfection
byproducts in drinking water using ion chromatography with the
addition of a postcolumn reagent for trace bromate analysis,
Environmental Protection Agency, Cincinnati, OH, Revision 1.0.,
September 1999.
S. B. Jonnalagadda, N. M. Munkombwe, P. Hensman and
f. Mushinga, Int. J. Chem. Kinet., 1991, 23, 113.
M. Eigen and K. Kustin, J. Am. Chem. Soc., 1962, 84, 1355.
J. H. Espenson, Chemical Kinetics and Reaction Mechanisms,
McGraw-Hill, New York, NY, 1981, p. 57.
J. A. Glaser, D. L. Foerst, G. D. McKee, S. A. Quave and
W. L. Budde, Environ. Sci. Technol, 1981, 15, 1426.
L. M. Schwartz, Anal. Chem., 1977, 49, 2062.
Y. Bichsel and U. von Gunten, Anal. Chem., 1999, 71, 34.
E. Salhi and U. von Gunten, Water Res., 1999, 33, 3239.
H. H. Willard, L. L. Merritt, Jr., J. A. Dean and F. A. Settle, Jr.,
Instrumental Methods of Analysis, Wadsworth, Belmont, CA,
1988, 7th edn., pp. 162-163.
J. Environ. Monit., 2000, 2, 571-575 575
------- |