DIODE-ARRAY DETECTORS IN FLOW-~N~ECT~~N

ANALYSIS

MIXTURE RESOLUTION BY MULTI-WAVELENGTH ANALYSIS

M. BLANCO*,J. GENE, H. X~URRLAGA, S. MA~POCEIand J. RIBA

Departamcntode Q&mica, Divisiitn de Qunnica Analiti~ Fact&ad de Ciencias, Universidad Aut~no~
de Barcelona, 88193 Belfaterra (Barcelona), Spain

(Received 30 June 1986. Revised 21 May 1987. Accepted 27 June 1987)

Summary-The application of diode-array spectrophotometers ta multi-component analysis by flow-

injection analysis is reported. Two different aspects are considered: (a) the obtainment of a reproducible
spectrum cor~s~nd~ng to the maximum of the ??IA peak, and ($1 the rna~~ti~ treatment involved
in determining the mixture of components. The spectrum is recorded by two different procedures
according to the type of injection system used. The mixtures are resolved with the aid of three different treat-
ments: (a) a linear equation system, (b) a multi-component analysis program (Hewlett-Packard) and (c)
a graphical method (multi-wavelength linear regression analysis), All three procedures have been applied
to the determination of Fe(H) and Fe(III) with a mixture of l,lO-pbenanthroline and sulphosalicyhc acid.

Flow-in~~i5n anaiysis (FIA) is an analytical tech-
nique of well-known potential, and its application to
simultaneous determination of several analytes in the
same sample is a subject of growing interest. One 5f
the ways to increase its potential in the resolution
of complex mixtures is the use of image detectors.
Diode-array ~e~trophot5meters are image detectors
capable ofcompletely recording an ultraviolet-visible
spectrum in about 0.1 sec. Despite their increasing
use as HPLC detectors and their potential advan-
tages,* they have only very recently begun to be used
in FIA. For example, Vithanage and Dasgupta3 have
applied diode-array detectors (DADS) to the study of
proton-iigand and metal-&and equilibria occurring
in the FIA-peak gradient zone. Valcarcel et al. have
reported the simultaneous determination of Fe(U)
and Cu(I1) with a mixture of specific chromogenic
reapnts,4 the metal concentrations being calculated
by means of a linear equation system. The use of
DADs results in faster analyses thanks t5 the possi-
bility of making several simultaneous measurements
at various wavelengths. Although there are a number
of procedures available for optimization of the wave-
lengths to be used, 5-8 the resolution of mixtures on
the basis of a few linear equations is i~~~~~~ble to
systems having extensive spectral overlap.
Derivative spectra are known to be more selective
than absorption spectra9 (zeroth-derivative). This in-
creased selectivity has been exploited in the deter-
mination of a given component in a sample by
making measurements at a single wavelength. Diode-
array detect5rs considerably facilitate the obtainment
of derivative spectra. The application of a mdti-
*To whom all correspondence should he addressed.
component analysis program to these spectra
has allowed the resolution of mixtures of analytes
with strongly overlapped spectra and the cancellation
of matrix effects.10-2 Derivative spectra have
recently been applied to the resolution of a
quatemary mixture by use of a combined FIA-DAD
system.3
The aim of this work was to expand the use of
DADS in the simultaneous determination of several
species by FIA by exploiting all the information
obtainable from the absarption and derivative
spectra of each analyte.
Mul~~omp5nent analysis with an FIA-DAD
system involves two essential aspects, namely (a) the
recording of the spectrum under conditions of maxi-
mum sensitivity and repraducibility, and (b) the
mathematical treatment of the information gathered,
to resolve the analyte mixture.
This paper reports tw5 different procedures fur
spectrum recording, and three methods for treating
the results, namely: (a) the resolution of a system
consisting of one equation per analyte; (b) multi-
component analysis (MA) by a computation program
supplied by Newlett-Packard; (c) multi-wavelength
linear regression analysis (ML&X), a linear re-
gression graphical method. All three procedures have
been applied to the simultaneous determination
of Fe(H) and Fe(II1) by use of a mixture of
1, 10-phenanthroline and sulphosalicylic acid as
chromogenic reagent. The results obtained by the
three methods are compared and discussed.
FXPRRIMENEAL
Reagents
Stock iron solutions containing 1000 ppm Fe(H) and
Fe(W) were prepared by dissolving Fe(NH4),(S04), . &I20
988 M. BLANCO et al.
and FeNH,(SO,),. lZH,O, respectively, in 0.05M sulphuric
acid and standardized both titrimetrically with a standard
dichromate solution and p-diphenylaminesulphonate as in-
dicator and gravimetrically by weighing as Fe,O, . Working
solutions were prepared by appropriate dilution with de-
gassed distilled water. The concentration of the Fe(H)
solution was checked daily by titration with standard di-
chromate solution.
I, lo-Phenanthroline (Phen) solution 0.5%. Prepared from
1,lo-phenanthroline hydrochloride.
Suiphosalicylic acid solution (SSA), 2.5%.
Acetate buffer, IM, pH 5.7. Adjusted with sodium hy-
droxide.
The stock reagent solutions were used to prepare those
injected into the manifolds, as follows.
Manifold A: 1 mM Phen, 15 mM SSA, 0.2M acetate buffer.
Manifold B: 10 mM Phen, 25 mM SSA, 0.2M acetate buffer.
All solid reagents used were of analytical grade.
Apparatus and manifolds
A Hewlett-Packard 8451A diode-array spectrophoto-
meter furnished with a 128-K RAM HP-82909A module, an
HP-9 121 floppy disk drive, an HP-98155A keyboard and an
HP-7440A plotter was used as the detection and data-
processing system. An HP-82940 GPI/O interface was used
to control the automatic injection valve. Other equipment
included a Gilson Minipuls-2 peristaltic pump and a Hellma
178.712-QS flow-cell (inner volume 18 ~1, optical path
length 10 mm).
An automatic Tecator V-200 and a manual Rheodyne
5020 injection valve were used in manifolds A and B,
respectively. All reaction coils and sample loops were made
from PFTE tubing of 0.5 mm bore.
Normalized standard spectra
Three pure standard solutions of each component of the
mixture, with a concentration lying in the linear concen-
tration range and giving absorbance readings lower than
1.0, were prepared. The batch solutions were injected in
triplicate and a spectrum for each standard was obtained
by averaging the three spectra obtained. A normalized
spectrum for unit concentration was obtained for each
component by regression of the absorbance us. concen-
tration. These normalized spectra were used as references
for resolution of mixtures.
RESULTS AND DISCUSSION
Recording of spectra
The HP-8415A spectrophotometer is capable of
recording a whole spectrum (190-820 nm) in 0.1 set
every 0.7 sec. Such a high recording speed allows a
vast amount of information to be gathered from an
FIA peak, yet a mixture can be resolved from a single
spectrum. In order to ensure the greatest possible
sensitivity and reproducibility, the spectrum was
recorded at the maximum of the FIA peak. The first
few attempts at recording spectra at high speed failed
because of the poor reproducibility of the results
obtained in successive injections. The reproducibility
was improved by increasing the integration time, T,,
the period over which the spectrophotometer is con-
tinuously measuring the spectrum, the end of which
is marked by delivery of the spectrum. Obviously,
since FIA signals are transient peaks, T, should be as
short as possible. Relative standard deviations not
greater than 1% were achieved for integration times
of 0.3 set or longer. Thus, a T, of 0.4 set, sym-
metrically distributed about the residence time of the
FIA peak, was chosen for further experiments.
Flow-injection analsysis is characterized by the fact
that the dispersion of the sample in the manifold is
a reproducible function of time. Thus, the repro-
ducibility of the sample image obtained at a given
time will depend chiefly on the reproducibility of the
time elapsed between image recording and the refer-
ence time. Every FIA recording features two points
that are eligible as a reference time, namely sample
injection and sample arrival at the detector. A suit-
able computer program facilitates recording of spec-
tra under preselected conditions (wavelength range,
l time interval, integration time, etc.).
The simplest approach uses sample injection as
the reference time. Obviously, the injection should
be synchronized as accurately as possible with the
triggering of the program handling the spectrum
recording, in order to ensure reproducibility; this is
made possible by the use of an automatic valve, the
control scheme of which is depicted in Fig. l(A).
A suitable alternative is to record the spectra once
a preselected time has elapsed after arrival of the
sample at the detector. This needs no synchron-
ization, so the analyst can inject the sample manually
and then execute the recording program.
A computer-controlled injection system (manifold
A) allows the recording of spectra in both ways,
whereas a manual system (manifold B) can only cope
with the second of the two.
Figure 2 shows the flow diagrams of the programs
used to control the recording of the spectra.
FIARECl uses the first few injections to calculate the
residence time, TR; the spectrum is recorded from
further successive injections at TR - (T/2) set after
each.
FIAREC2 employs the first few injections to calcu-
late the time elapsed between the start and the
maximum of the FIA peak, T, the spectrum being
subsequent19 recorded at T - ( T,/2) set after arrival
of the sample at the detector.
FIARECl and FIAREC2 were used in conjunc-
tion with manifolds A and B, respectively.
Data treatment
The analytes were determined by processing the
information abstracted from the spectra of the mix-
tures, by three different procedures, as follows.
(a) By solving a system of two linear equations
based on the absorbances of the two complexes,
measured at their respective absorption maxima (510
and 486 nm). The equations concerned were:
Manifold A
A,,, = O.O682[Fe*+]+ 0.0255[Fe3+]
Ad86= O.O634[Fe*+]+ 0.0280[Fe3+]
 Diode-array detectors

COMPUTER

Fig. 1. Schemes of the flow-injection manifolds. S = sample; I.V. = injection valve; L = reaction coil;
D = diode-array detector; W = waste. (A) V,, = 120 ~1; flow-rate 0.95 ml/min; L = 270 cm; TR = 39 set,
(B) V,, = 8 ~1; flow-rate 1.07 ml/min; L = 50 cm TR = 7 sec. In manifold A the sample is injected into
the reagent stream (normal FIA mode) while in manifold B the reagent is injected into the sample stream
(reverse FIA mode).

Manifold B two species, d is the optical path-length and C, and

A,,, = 0.1658[Fe2+] + 0.0538[Fe3+]
A,,, = 0.1 548[Fe2+] + 0.0562[Fe3+]
(b) By use of the multi-component analysis (MA)
program supplied as part of the Hewlett-Packard
spectrophotometer bundled software. This is written
in machine code and is therefore modifiable only by
experienced programmers. It uses spectra from pure
standards of accurately known concentration as
reference, and yields the analyte concentrations that
give the best least-squares fit between the calculated
spectrum and that obtained for the mixture. This
program was applied to the absorption spectrum and
the first, second and third-derivative spectra over the
selected wavelength ranges. The derivative spectra
were run with the aid of the bundled software.
(c) By application of a graphical linear regression
method known as multi-wavelength linear regression
analysis (MLRA). For simultaneous spectropho-
tometric determination of several components in a
mixture to be achieved, each species must absorb
according to Beers law, and the absorbances must be
additive. For a binary mixture, both conditions can
be summed up in the equation
A,,, = a, dC, + a,dC, (1)
where a, and a2 are the molar absorptivities of the
C, are the respective concentrations.
The molar absorptivities at each wavelength can be
calculated from the absorbance of a pure standard
solution of each component:
A,, = a,dC,, (i = 1 or 2) (2)
where A,, is the standard absorbance at each wave-
length and C,, the corresponding ,concentration.
If the same cell is used to measure both species,
and C,, = C,, = 1, appropriate substitution and re-
arrangement gives
A,,/& = (l/C,) - (C,IC,)(&l&) (3)
Plotting the values obtained at different wave-
lengths as A,]/A, us. A,/A, gives a straight line
with slope and intercept which allow calculation of C,
and Cr.
Optimization of manifold parameters
Because of the lower molar absorptivity of the
Fe(III) complex, the optimization was aimed at in-
creasing the sensitivity of the determination of the
ferric ion. This was accomplished by applying the
univariate method.
The sensitivity of the reaction between Fe(II1) and
SSA increases with pH, but the injection of Fe(I1)
yields a signal when alkaline media are used. A pH
of 5.7 was therefore chosen as a compromise. It
(A) (B)
START
t
s& INPUT DATA
L,Ll,LL,Tl,TP,T(J) L,Ll,LZ,Tl,TL,T(J)
M.N M,N

SELECTION L

.. I= l,...N 4 v J=I,...N
SELECTION L =+,...LI
J
SELECTION ~1, ~2
1
REFERENCE
t
REFERENCE INJECTION
1
=l....Y
1REFERENCE 1 1
INJECTION MEASURE ABSORBANCE
1 t
1 INJECTION 1 TIME=+
I
RECORDING FIA PEAK

DETERMINATION DF T(J 1 STOP MEASURE

YEXT J 4
TIME = +
1
MEASURE FOR Tl SELECTION Ll , L2
I
L WAIT TS-T1/2
1 I
AVERAGE SPECTRA FOR ALL INJECTIONS MEASURE FOR Tl
E*T K

1 AVERAGE SPECTRA FOR ALL INJECTIONS 1

Fig. 2. Flow chart of programs for recording of spectra. A: FIARECI; B: FIAREC2. See Table I for definition of variables
 Diode-array detectors

0.6 - (A)

2
0.1 - -0.010
)&
I I I I I
0.0 -0.014
400 450 500 550 600 650 400 450 500 5x) 600 650

0 0006 (Cl o.mOO4 (D)

O.Ocm4
(u

P
E
1 A2

B o-
Z
l!s
-0.OOQ4 -

-0.0006 - I I I I ' -0.00004 I I I I 1

400 450 500 550 600 650 400 450 500 550 600 650

WOVelenQth ( nm )

Fig. 3. Spectra obtained for 20 ppm of Fe(III) (1) and 8 ppmFe(I1) (2) with a normal nFIA manifold.
A, absorption spectra; B, first-derivative spectra; C, second-derivative spectra; D, third-derivative spectra.

should he pointed out that the volume injected into were processed over the whole wavelength range
the manifold was kept rather low in order to avoid covered (40&650 nm).
the appearance of double peaks. The results obtained in the resolution of the
Fe(II)-Fe(II1) mixtures (Tables 2 and 3) are shown as
Mixture resolution the correlation between the concentrations taken and
A series of 12 mixed solutions containing various found, for the different calculation procedures used.
amounts of Fe(I1) and Fe(II1) in the ranges 0.25155 The real achievement of this work is not the
and 3-20 ppm, respectively, and [Fe(II)]/[Fe(III)] accurate calculation of the concentration of a given
ratios from 0.6 to 40, were prepared for manifold A. species in the presence of a higher concentration of
Similarly, another set of 14 solutions with Fe(I1) and another, but rather the possibility of simultaneously
Fe(II1) contents in the ranges 0.5-5.2 and l-12.4 determining both under all types of conditions.
ppm, respectively, in ratios from 0.25 to 16, were As would he expected from the degree of spectral
prepared for manifold B. The conventional and the overlap, the classical method of mixture resolution
first, second and third-derivative spectra (Fig. 3A-D) based on a linear equation system resulted in substan-

Table 1. Definition of variables used by FIARECl and FIARECZ

Variable Meaning
AS Y if residence time must be determined
BS Y if another sample or standard is injected
L Wavelength of maximum sample absorbance
Ll, L2 Lower and upper limit of the useful wavelength range
M Number of averaged spectra
Number of injections used for measuring residence time
TYJJ Residence time for each injection
T2 Average residence time
Integration time
T& Time interval between start and maximum of FIA peak
T5 Average T4
992 M. BLArwo et aL

Table 2. Least-squares statistical data for fits of calculated vs. added concentrations (manifold A)
Fe+ Fez+
Correlation Correlation
Intercept is Slope zks coeff. (IP) Intercept * s Slope & s coeff. (RZ)
Linear equations
-0.33 f 0.97 1.06 f 0.09 0.937 O.f0 If:0.08 0.95 & 0.01 0.998
Multicomponent analysis
Absorption spectrum 0.32 f 0.37 0.97 f 0.03 0.988 -0.12 & 0.08 1.02 * 0.01 0.998
First derivative 0.45 f 0.50 0.97 k 0.05 0.978 -0.05 f 0.08 0.99 * 0.01 0.999
Second derivative 0.10 &OS6 1.OOf 0.05 0.976 -0.03 If:0.07 0.99 + 0.01 0.999
Third derivative 0.02 + 0.65 I .OI & 0.06 0.967 0.01 f 0.07 0.98 + 0.01 0.999
MLRA
400-600 nm 0.42 + 0.40 0.96 & 0.04 0.986 -0.15 kO.09 1.02 f 0.01 0.998
420-580 nm 0.24 + 0.40 0.97 f 0.04 0.987 -0.11 *Oo.08 1.01 + 0.01 0.999

tial errors and hence in poor precision, as cart be seen
from the deviation from linearity and the low cor-
relation coeEcient obtained.
The use of a calculation program (MA) operating
throughout the visible spectrum provides better re-
sults for the slope, intercept and precision: However,
deviations from linearity are still significant and the
slope is rather low for Fe(III) and high for Fe@),
The application of the MA program to the deriva-
tive spectra results in considerably improved results,
particularly for Fe(II). The precision achieved is
virtually independent of the spectral mode and the
best slope and intercept are obtained for the first and
second-derivative spectra.
Although the use of derivative spectra also im-
proves the slope and intercept for FelfIIf, increasing
the derivative order results in poorer precision as a
result of the concomitantly decreasing spectral
signal.
It should be noted that manifold l3 yiefds more
precise results than manifold A, probably because of
the greater sensiti~ty it produces.
The MLRA treatment, as applied to the whole
wavelength range, yields similar results to those
obtained by multicomponent analysis, but has
two significant advantages, namely the resolution
procedure is straightforward and transparent to
the analyst-whereas that of MA is obscure
and inaccessible-and (most important), it makes
selection of the most suitable wavelength range for
resolution of the mixture quite a simple task. The
reliability of mixture resoln~on has been reported
to depend strongly on the wavelength range
selected, but up till now the selection has been made
empirically.
The MLRA program provides the user with the
wavelength range that gives the best linearity of
response. This is illustrated in Fig. 4 for a mixture
containing 0.52 ppm Fe(II) and 8.32 ppm Fe(III). In
this case, deviations from linarity are observed only
near the ends of the wavelength range, where the
absorbance values are significantly smaller. This is
consistent with the findingi that one of the requisites
for accurate mixture resolution with the aid of the
MA program is that al1 species to be determined must
absorb significantly in the selected wavelength range.
Narrowing of the spectral interval to 42G580 nm
results in a better fit, as can be seen from Tables 2 and
3. This range narrowing also resulted in slight im-
provements in the results obtained by MA from the
absorption spectra, though not from the derivative
spectra. No other wavelength range assayed yielded
better results.
CONCLUSIONS
This paper demonstrates the ~ssibili~y of accom-
plishing simultaneous determinations by FIA with

Table 3. Least-squares statistical data for fits of calculated US.added concentrations (manifold B)
Fe3+ Fez+
-..-
Correkition Correlation

Linear equations
-0.09 zt 0.26 1.l2-tO.04 0.983 0.17 rf:0.09 0.88 + 0.03 0.986
Multicomponent analysis
Normal spectrum 0.03 _t 0.08 0.9s * O.OI 0.998 0.06 + 0.02 1.036 f 0.006 0.9996
First derivative 0.09 & 0.06 1.02 i_O.Of 0.999 0.06 * 0.03 0.994 + 0.009 0.999 t
Second derivative 0.08 f 0.08 I.os+o.or 0.999 0.05 + 0.03 0.99 * 0.01 0.999
Third derivative 0.02 kO.15 1.07& 0.02 0.994 0.06 & 0.03 0.99 f OxJl 0.999
MLRA
400-600 nm 0.02 f 0.08 0.9s * 0.01 0.998 0.05 * 0.02 1.04 f 0.006 0.9996
420-580 nm 0.04 & 0.07 0.96 * O.OI 0.999 0.05 + 0.02 1.029 & 0.006 0.9996
 Diode-array detectors 993

0.30

1 E 0.20
P
\
i
f
^9; 0.10

0.00

Fig. 4. MLRA plot for a mixture of 8.32 ppm Fe(H) and 0.52 ppm Fe(B). Data obtained with manifold
B. (A) Wavelength range, 401X650 nm; Fe(II1) found, 8.22 ppm; Fe(I1) found, 0.60 ppm; regression
coefficient, r* = 0.997. (B) Wavelength range, 420-580 nm; Fe(II1) found, 8.26 ppm; Fe(I1) found, 0.58
ppm; regression coefficient, r2 = 0.9994. Absorbance measurements made at 4-nm intervals.

the aid of a diode-array detector. Most of the repro- REFERENCES

ducibility problems encountered in working at high
spectrum-recording speeds are overcome by both
procedures proposed, which afford a reproducibility
of the order of 1% at absorbance values not greater
than 1.5.
As can be seen from Tables 2 and 3, the resolution
of mixtures with highly overlapped spectra yields
poorly correlated results by the linear-equation
method, but the MA and MLRA programs result in
better fittings. The results provided by both MA and
MLRA when applied to the absorbance spectrum are
comparable in accuracy and precision, but, the best
fittings are obtained by MA and the first and second-
derivative spectra obtained by use of manifold B.
The most outstanding features of the MLRA pro-
gram are probably its transparency to the user and
the possibility of selecting the wavelength range best
suited to the particular mixture to be resolved.
Acknowledgement-The authors are grateful to the CAICyT
(Project No. 821/84) and the Fundacion Roviralta for
financial support.
1. M. D. Luque de Castro and M. Valcarcel Cases,
Analyst, 1984, 109, 413.
2. A. F. Fell, Anal. Proc., 1980, 17, 266.
3. R. S. Vithanage and P. K. Dasgupta, Anal. Chem., 1986,
58, 326.
4. F. Lazaro, A. Rios, M. D. Luque de Castro and
M. Valclrcel, Anal. Chim. Acta, 1986, 179, 279.
5. S. D. Frans and J. M. Harris, Anal. Chem., 1985, 57,
2680.
6. M. R. Ditusa and J. Schilt, J. Chem. Ed., 1985, 62, 541.
7. A. T. Pilipenko, L. I. Savranskii and A. N. Masko,
Zh. Analit. Khim., 1985. 40. 232.
8. E. J. Meehan, in Treatise oh Analytical Chemistry, 2nd
Ed., P. J. Elving, E. J. Meehan and I. M. Kolthoff (eds.),
Part 1, Vol. 7, pp. 96-104. Wiley, New York, 1981.
9. P. Levillain and D. Fompeydie, Analuris, 1986, 14, 1.
10. D. T. Rossi and H. L. Pardue, Anal. Chim. Acfa, 1985,
175, 153.
1I. T. R. Matton, P. McSwiggen and S. A. George, Metal
Finishing, 1985, 83, 83.
12. Y. R. Tahboub and H. L. Pardue, Anal. Chem., 1985,
57, 38.
13. M. Blanco, J. Gene, H. Iturriaga and M. Maspoch,
Analyst, 1987, 112, 619.