Journal of Electroanalytical Chemistry, 378 (1994) 165-175 165

A hollow fibre supported liquid membrane system for metal speciation

and preconcentration: calculation of the response time *

O.C. Keller a, S. Poitry b, J. Buffle a

n Department of Inorganic, Analytical and Applied Chemistry, University of Geneva, Geneva (Switzerland)
b Experimental Ophthalmology Laboratory, University of Geneva, Geneva (Switzerland)
Received 21 January 1994; in revised form 17 March 1994

Abstract

We report a method for calculating the response time of a hollow fibre supported liquid membrane (SLM) for metal
preconcentration. The SLM system is composed of three distinct compartments: an internal stripping solution (filling the hollow
fibre lumen), a hydrophobic liquid membrane (adsorbed in the hollow fibre pores), and an external source solution (containing the
analyte).
The calculation consists of resolving the diffusion equation which describes the evolution of the concentration of the metal ion
in the system. The Laplace transform method is used for this purpose, and the inverse Laplace transformation is performed using
discrete Fourier transforms.
Variation in the concentration of the strip solution with time, after a given change in concentration of the source solution, can
be predicted from these calculations. The numerical results obtained by considering various geometrical (membrane thickness,
lumen diameter), hydrodynamic (Nernst layer thickness) and chemical (diffusion and partitioning constants) conditions, are
presented in order to allow optimization of the system, i.e. reproducibility, fast response time, and high metal preconcentration.

1. Introduction tion (if the charge of the diffusing and co-diffusing

species have the same sign, e.g. M’+ in Fig. 1). The
Separation and preconcentration of cations, anions co-diffusing species is generally H+, a cation M’“+ or
or organic molecules from aqueous solutions can be an anion A- [l-4]. This co-transport is the driving
accomplished by using supported liquid membranes force of the enrichment process.
(SLMs) as extractants [l-6]. An SLM (Fig. 1) consists SLMs have found applications in water and waste
of an organic solvent containing a specific carrier R water treatment [5,6], Pt4+ [7] and Au3+ [8] recovery,
immobilized on an inert microporous support. The amine 191 and herbicide [lo] separations, gas sensors
liquid membrane is interposed between two aqueous [ll], alkali metal sensors [12-141, Cd’+ and Cu2+ ISE
solutions. The aqueous solution containing the target [l&16] and optrodes for Pb*+ detection [17]. Up to
metal ion is called the source, and the solution into now, very few SLM applications to trace metal analysis
which we want to concentrate the metal is called strip have been reported [4,18]. This is probably due to
solution (containing a stripping agent L). The transport difficulty in obtaining a stable membrane system.
mechanism is diffusion controlled (passive transport); if We have recently reported an SLM system for Cu*+,
the diffusing species is charged, there must be a co- Cd*+ Pb*+ and Zn*+ preconcentration [4,18]. Under
transport in the same direction (if the co-diffusing the cbnditions used for the source-strip volume ratio,
species has opposite charge), or in the opposite direc- the attained preconcentration factor, i.e. the strip metal
concentration over the source metal concentration, is
about 20. When coupled to square wave stripping
l Dedicated to Professor R.P. Buck on the occasion of his 65th voltammetric techniques using a mercury-plated irid-
birthday. ium microelectrode, this SLM system allows the mea-

0022-0728/94/$7.00 0 1994 - Elsevier Science S.A. All rights reserved

SSDI 0022-0728(94)03481-H
166 0. C. Keller et al. / Hollow fibre supported liquid membrane system

surement of free metal concentrations down to lo-l2

I

cz@
r

M in complex media such as natural waters. This

technique has several advantages:
6
(1) it allows trace metal analysis at very low concentra- 2
b
tion (down to lo-l2 M); a
(2) it allows trace metal speciation in natural waters 1 0
without sample perturbation;
(3) it prevents electrode surface from fouling.
The limitation of the system described in Refs. 141
(a) (b)
and [181 is its relatively long response time and its I = detector components

limited preconcentration factor F which can be esti- P:.4 = wzlating parts

mated from: Fig. 2. (a) Section of a hollow fibre system and of compartments 1, 2
and 3: 1, strip solution, 0 _< r < a; 2, membrane, a < r < b; 3, Nernst
F = Kource/%trip when a/P > Kource/Grip layer in source solution, b < r < b + 6; D,, D,, D,, diffusion coeffi-
F=(Y/@ when a/P < KourcJKstrip cients in compartments 1, 2, 3 respectively; (Y= [ML],,,, /[MR],,,,,,
where L, R are complexants in strip and membrane respectively;
where P = [M” + Isource /IMRl,,,,,: (b) Longitudinal section of the hollow
fibre system. The source solution (compartment 3) irrigates the
a = [MLI,t,ip/[MRImembrane hollow fibre which supports the liquid membrane (compartment 2).
The strip solution (compartment 1) is in the lumen of the hollow
P = [MlSO”Tce/[MRlmembTane fibre, which is closed on each side by the detector components.
With this system, the time t,, required to reach 98% of
the equilibrium value is about 3 h. The value of t,,
(4) increasing the ratio of the membrane surface area
may be decreased by
to the strip solution volume.
(1) decreasing l&, compared with QUrCe,
Based on item (41, the preconcentration factor can be
(2) improving the stirring efficiency in the source solu-
drastically increased by using a hollow fibre as a mem-
tion,
brane, with the strip solution in its lumen and the
(3) decreasing the membrane thickness, or
source solution percolating at its external surface (sys-
tem composed of three concentric cylindrical compart-
SLM ments; Fig. 2).
Source Membrane Strip In this paper we report the calculation of the re-
sponse time of this system, i.e. variation in metal
concentration in the strip solution as a function of time
after a step concentration change in the source solu-
tion. The response time is computed by numerical
methods as a function of various geometrical, hydrody-
namical and chemical parameters in order to allow
optimization of the system.
In a second stage, the rigorous results are compared
with those obtained with a much simpler model, giving
approximate solutions but for which an analytical equa-
tion can be used. The conditions of application of this
model are discussed.

2. Hollow fibre SLM system description

ML
Figure 2 shows transverse and longitudinal sections
Fig. 1. Summary of diffusion process in metal-driven facilitated
transport in SLM: path a, diffusion of metal ions M towards the
of the hollow fibre SLM system. Hollow fibres with
membrane; path b, formation of neutral complex at source-mem- different lumen diameters and wall thickness are avail-
brane interface and partitioning of the metal between the source able. For instance Celanese (Charlotte, NC) produces
solution and the membrane; path c, diffusion of the metal complex in hollow fibres with 25 Frn wall thickness and 100, 200,
the membrane; path d, partitioning of metal ion at the membrane- 240 and 400 ,um lumen diameters; fibres with larger
strip interface by reaction with the ligand (L) of the strip solution;
path e, diffusion of metal complex in the strip solution. The same
lumen diameters are available and find applications in
processes in the opposite direction take place for the metal M’ analytical methods requiring larger analyte volumes
(countertransport). such as FIA systems. Fibres with smaller lumen are
 O.C. Keller et al. / Hollow fibre supported liquid membrane system 167

strip membrane SOWX

useful for the development of microanalytical systems, (3)
A (1) (2)
in combination with microdetectors. At any rate, the
length of the fibre is assumed to be much larger than
the lumen diameter.
For microsystems such as that schematically de-
picted in Fig. 2(b), any type of detector sensitive to the
target substance (e.g. M”+) in the strip solution can be
used provided that it does not perturb the diffusion
process in the hollow fibre system. If the detector is
positioned as shown in Fig. 2(b), the response time t,,
will be the time at which the metal concentration in the
radial centre of the fibre will be 98% of equilibrium
concentration. The same definition is valid for hollow
fibres coupled for instance to classical FIA systems.
The values of the diffusion coefficients of the differ-
ent chemical species in the three compartments (com- Fig. 3. (a) Schematic representation of concentration profiles in the
partments 1, 2, 3 see Fig. 2(a)) are generally in the three compartments as a function of time according to the simplified
range 10-6-10-5 cm2 s-‘. In particular, for many free model. (b) Schematic representation of concentration profiles in the
three compartments as a function of time according to the general
and complexed metal ions D, and D, are close to
model.
(5-7) x lo-’ cm2 s-r. In the membrane, D, = (2-3) X
10e6 cm’ s-l [4]. The values of the partition constants
LYand p may vary by several orders of magnitude
depending on the stability constants of complexes ML, culations based on a simplified model considering that
RMA and RM’A and the concentration of the ligands the only rate-limiting step is the diffusion in the mem-
L and R. brane. The concentration profiles in compartments 1, 2
and 3 for the simplified and rigorous calculation are
3. Theory schematically depicted in Figs. 3(a) and 3(b) respec-
tively. The calculations below are valid in principle for
Steady state diffusion treatment for carrier-media- any chemical substance that can form complexes with L
ted facilitated transport has been used by many re- and R. Metal transport (Ml is taken as an example for
searchers (for instance Refs. [19-241). Non-steady-state demonstration. In compartment 3, M may be either
diffusion through membranes has been applied for free or complexed (inert complexes).
planar diffusion conditions; in particular Buffle and In the subsequent text, for the sake of simplicity,
Spoerri [25] have derived an equation to calculate the [Ml in compartments 1, 2 and 3 will be referred to as
response time of a potentiometric CO, electrode con- cr, c2 and cg respectively.
sisting of a planar membrane, a thin internal solution A simple analytical expression can be derived for
and a potentiometric sensor. Mancy et al. [26] have the case of Fig. 3(a) (Appendix A). This expression can
derived an equation to describe the 0, reduction cur- be useful provided that its conditions of validity are
rent for a detector consisting of a planar membrane, a clearly defined by comparison with rigorous computa-
thin internal solution and a voltammetric sensor (Clark tions. This will be discussed in Section 5.
electrode). The rigorous calculation performed here corre-
Non-steady-state heat diffusion through a system of sponds to the case of Fig. 3(b), by considering radial
two concentric cylinders is treated in Ref. [27]. Mitlin diffusion through three concentric cylindrical compart-
and Bromberg [28] have solved analytically the problem ments (Fig. 2).
of diffusion through a system of an infinite number of (1) Compartment 1 refers to the strip solution in the
concentric cylinders where the value of the diffusion lumen of the hollow fiber.
coefficient of the diffusing substance is different in (2) Compartment 2 refers to the liquid membrane.
each cylinder. They considered variable concentrations (3) Compartment 3 refers to the stagnant Nernst layer
along the composite membrane longitudinal axis, but in the source solution.
all partition coefficients between cylindrical compart- In this model, the values of the diffusion coefficients in
ments were equal to unity, and the external concentra- the compartments and those of the partition coeffi-
tion (source solution) was assumed to be constant. cients between adjacent compartments may be arbitrar-
In this paper rigorous calculations are performed as ily chosen. In addition, the concentration change of the
explained below. These results are compared with cal- external solution may be freely chosen and varied
168 O.C. Keller et al. / Hollow fhre supported liquid membrane system

The diffusion processes in compartments 1 and 2

1 (1)
1 are described by Eqs. (1) and (2):

I
ac,(r, t) a*c,(r,t> %(r, t>
=D, +r-’
at ar* ar

1
4
2
5
ac2(r,t) a2c2(r,f> ac,(r, t>
0, =D, +r-' (2)
at at-* ar
The boundary conditions are the following:
O-
6) no net flux at r = 0, i.e.
I I I I I
0 N/4 (3N/4)-1 N-l [Dlacl~f)]~~O=O (3)
”
Fig. 4. Concentration profile as a function of time in the bulk source
(ii) partitioning of metal at interfaces a and b respec-
solution.
tively, i.e.
(step, exponential, linear changes are possible) in order a =c,(a, t)/c,(a, t) (4)

to mimic any hydrodynamic experimental conditions. P = G(b, t)/c*(b, t) (5)

The numerical results will be presented in later
(iii) flux conservation at r = a, i.e.
sections using a square pulse change in the concentra-
tion of the external solution (Fig. 4). In the following
section, equations describing the diffusion of the sub- [Draci:f)]r=~= [DisC?(:r~~~]~~~ (6)
stance in the system as well as resolution of these
equations by Laplace transforms method will be given. (iv) flux conservation at r = b, i.e.
The following assumptions have been made.
(1) A linear gradient in the Nernst layer of solution 3 [ D2ac*:P 91r&

(7)
is assumed. This simplification is justified when the

I
diffusion in the source solution is fast and the cs(m, t) -Pc,(r, t)
Nernst layer thin. The first condition is often valid = D,
] S r=b
for aqueous phase as a source solution. The second
condition is valid under conditions of strong agita- where ~~(03,I) = g(t).
tion. The function g(t) describes the evolution of the
(2) Anion or metal co-transport (see Section 1 and Fig. concentration in the well mixed bulk source solution.
1) is not a limiting step. This condition is experi- The expression on the right-hand side term in Eqn. (7)
mentally fulfilled when [M’lstrip> [Ml,,,,, where reflects the assumption that a linear concentration
M’ is the co-transported metal, and M is the ana- profile is instantaneously reached in the Nemst layer
lyte (metal), which is a typical condition used in of constant thickness 6. After performing a dimension-
practice 14,181. less variables transformation, Laplace transformation,
(3) The free metal concentrations in membrane and and solving the resulting system of equations (see Ap-
strip solution are negligible compared with that of pendix B), one can express the Laplace transform of
the complexed forms. This condition is usually ful- concentration in the strip solution as follows:
filled in SLM systems because strong complexing cl( r’,s>
agents are used in the membrane and strip solu-
tion. = - (d4MW/4/~4)
(4) The metal in the source solution is free or in an x(D;“*Z,(h3)[IO(h4)P-K0(A4)Q]
inert complex form.
Rigorous calculations are made numerically. How- -a-%OJ [W,V’ + WdQ])-’
ever, we did not apply directly a numerical method of
with
resolution (e.g. finite difference methods) to Eqs. Cl)-
(7); indeed, mathematical transformations allow a much A, = u’h,
faster computation to be performed. They are reported A, = urA2
in detail in Appendix B. The basic equations describing
P = PK,(A,) - 6’A,K,(h2)
the diffusion through the hollow fibre are the follow-
ing. Q = PMA,) + S’A,Z,(A,)
 O.C. Keller et al. / Hollow fibre supported liquid membrane system 169

The results of the plots in Fig. 6 show the following.

(1) t$s and tj, increase with increasing (Y and/or
decreasing j3. This is consistent with the fact that
s high a//3 values correspond to high preconcentra-
z 70. tion factors, i.e. more substance has to be trans-
6
ported across the membrane so more time is
8 needed.
f 50-
E (2) t& and t&, increase with increasing (Y,even if the
g 40- ratio a/P is maintained constant (for instance see
$ U/P = 1000, for (Y= 1000, 100, 10, on Fig. 6). This
B- 30-

ap is consistent with the fact that, when p-’ is small,

20-
the concentration gradient across the membrane is
g 60-J,&\
10 small, so the diffusion transport is slow. From this
-too -200 0 200 4w 800 1000 1200
example we see that it is preferable to have (Y as
t/s small as possible at constant a/P value.
Fig. 5. Evolution of strip concentration as a function of time after a (3) For (Y2 10, log t& and log t&, are linear functions
square pulse change in the bulk concentration of the sourcesolution. of log cy.
a = 100 km, b = 125 pm, S = 20 pm, a = 20, p = 0.5, D, = D, = 6~
10m6 cm* s-l, D, = 2x 10m6 cm2 s-l. Equilibrium concentration at
4.2. Dependence on a /b and 6 /b
r = 0: c,(O, teq,) = (Y/pc,(m, t).
The values of the following parameters are kept
constant: D, = D, = 6 X lop6 cm* s-l, D, = 2 X lop6
where I, and K, are modified first-order Bessel func-
tions.
Since the analytical inversion of this function (Eqn.
(Bll)) is very complex we have used a numerical
2.0-
method of inversion: fast Fourier transform (see Ap-
pendix 0. 1.5-

An example of the calculation of c1 by applying the 10

procedure described in Appendix C to Eqn. (A13) is
given in Fig. 5, with typical conditions used in the rest
-z
of the paper. -g
$ 0.5
0.0;::--;::

-0 5-
4. Results of the general model
-l.O-

The dependence of the response time on the most

-l.5! ,I,.,., . I. I. I
important experimental conditions is examined in de- 0.0 0.5 1.0 1.5 2.0 2.5 3.0

tail as it is the important parameter to optimize for (a) log a

analytical applications. The following results report the
30-
dimensionless times t’ at 30% (t&J and 98% (t&> of
equilibrium concentration in the strip solution: c&O,
03)= (ay/p)c&co, t). tie is still in the pseudo-linear part
of the response time curve (Fig. 5) whereas t& gives an
idea of the time necessary to reach a value close to
equilibrium concentration C(0, eq). The real time t can
be computed from t’ using Eq. (Bl) (Appendix B):
t’ = tD,/b2

4.1. Dependence on (Y and p -0.5! , ( I ( . , I , I I - I 1

0.0 0.5 1.0 1.5 20 2.5 3.0
All parameters other than a and p are kept con- log a
(b)
stant at a = 100 pm, b = 125 pm, S = 20 pm, D, = D,
Fig. 6. Logarithmic plots of (a) tie and (b) t& as a function of (Yfor
= 6 X 10P6, D, = 2 X 10T6 (Fig. 6). If V~urce/~/strip is various /3 values (D, = D, = 6~ low6 cm2 cm2 s-l, D, = 2~ 10m6
smaller than (Y/P, the preconcentration factor is given cm2s~‘a=100~m,b=125~m,6=20~m):~,/3=1;~,~=0.1;
by (y/P. A, p = 0.01.
170 O.C. Keller et al. / Hollow fhre supported liquid membrane system

negligible when a/b is small. This result has an

interesting experimental implication: if repro-
ducible source solution stirring is difficult to reach
(reproducible S), because of uncontrollable param-
eters such as edge effects or fibre rigidity, it is then
preferable to choose a fibre with walls rather thick
compared with the lumen size; in this way, the
influence of S (i.e. stirring) on the response time is
minimized. At large a/b values (very small thick-
ness of the membrane), the effect of S becomes
01
very significant; Fig. 8 shows that this effect in-
02 04 06 08 '1 0
I
creases dramatically when the partition coefficient
(a) a/b of M between the source solution and the mem-
brane is more and more in favour of the latter.

4.3. Dependence on D, /D, and D, /D,

In Fig. 9 all parameters are kept constant at a = 100
pm, b = 125 pm, S = 20 pm, LY= 20, p = 0.5.
The results of Fig. 9 shows that the influence of
D,/D, on t& and t&, is medium, compared with that
of D,/D,. This is partly due to the simplification we
made in our initial assumption i.e. the formation of a

1
o! I
’ 1
II 1
1
12-
02 0.4 0.6 08 10

(b) a/b
IO-
Fig. 7. (a) tie and (b) t& as a function of a/b for various 6/b
values (D, = D, = 6~ 10e6 cm2 s-l, D, = 2x 10e6 cm2 s-l, a = 20, B-
p = 0.5). (a) n , 6 = 0.0033; 0, 6 = 0.01; A, 0.0333; V, 6 = 0.1. (b) n ,
6’= 0.0033; 0, Is= 0.001; A, 6’= 0.0333; V, 6’= 0.1.

4-
cm2 s-l, LY= 20, p = 0.5 (Fig. 7). The results reported
in Fig. 7 show the following.
(1) t’ decreases with decreasing a/b for a/b < 0.65.
This is due to the smaller amount of metal trans-
0.2 0.4 0.6 0.8 1.0
ferred into the strip solution (strip solution smaller)
which decreases with a/b. In fact, two factors must (a) a/b
140
be considered when a/b decreases.
- The thickness of the diffusion-limiting compart- 120

ment (compartment 2) increases. This should

100
cause an increase in f;, and tbs.
- The volume of strip solution decreases and
60
therefore a smaller amount of metal is trans-
?a
ferred from source to strip solution. This should -- 60
result in a decrease in r;, and (,a.
40
The global decrease shown in Fig. 7 indicates that
the latter effect is more important than the former. 20

For a/b > 0.6-0.7, tj, and tbs decrease. This is

01
because the diffusion step in the membrane be- 0'2 0'4 Oi 0'8 1'0
comes less and less rate limiting owing to the (b) a/b
decreasing value of the membrane thickness. Fig. 8. (a) t&, and (b) tbs as a function of a /b for various /3 values
(2) The striking feature revealed in Fig. 7 is that the (D,=D3=6X10M6 cm2 s-l, D2=2x10W6 cm2 s-l, a=20, i?=
influence of the Nernst layer thickness S becomes 12.5pm): n,p=2;0, p=O.l; A,p=0.05; T, p=O.Ol.
 0. C. Keller et al. / Hollow fibre supported liquid membrane system 171

linear concentration profile in the Nernst layer after a

3000 -
concentration change in source solution. With this as- .

sumption, the predicted response time is somewhat 25OO-

overestimated, but the corresponding error is likely to

be small. Two regions can be discriminated in the zooo- .
In
changes in tbs and t& with D,/D,. B
$ 1500-
(1) For DJD, > 1, the influence of D,/D, on t$a and .
.
t&, is very small, because the rate-limiting compart- IOOO-
.

v---------v
ment is the membrane (compartment 2). This situa- v-v
.
0

tion corresponds to the experimental conditions soo-

*
A--h t *- f
encountered in many SLM systems. In such cases tx_8e
o-------o
I Y
the influences of other factors such as 6/b, a/b, a, 0
0 2 4 6 8 10
and p on tbs and t& are more important than IO6 D/cm’ SC’
diffusion coefficient ratios between compartments. Fig. 10. T,,* predicted by the simple model (0, 0, A, v) and
(2) D,/D, < 1, a large increase in tbs and tie is ob- general model Cm, 0, A, r> as a function of various a values
served when DJD, decreases, e.g. if compartment (b-a=25 pm, D,=2~10-~ cm2/s-‘, a=20, /3=0.5): n , 0,
3 is diffusion limiting, transport across the other a=lOO~m;O,o,a=2OO~m; A, A,a=4OOpm; V, V,a=lOOO
compartments will be significantly slowed down. pm.
So, for obtaining fast response times, low viscosity
source and strip solutions (i.e. dilute solutions)
must be used. This point is unimportant when working with aqueous samples such as natural wa-
ters (low viscosity), but it should not be neglected
when biomedical samples are used.

5. Comparison between the simple and general model

predictions

In this section we compare the results given by the

simple and general models and determine under which
conditions the former is applicable. In the following we
discuss the differences in the values of 71/Z predicted
by the two models for various values of D, and a, D,
and 6, LYand p (r1,2 is the time at which the metal
concentration in the radial centre of the fibre will be
50% of the equilibrium concentration).
o!
-1.0 -0.5 0.0 0.5 1.0 1.5 2.0
5.1. Znfluence of the lumen parameters: D, and a
(a) lag(D3/D2)
Since the simplified model considers that diffusion
through the membrane is rate limiting, in this model

I
B D, does not play any role in the response time (Fig.
60 10); only the lumen diameter a plays a role (Eqn. (A2)).
i The differences in the values of 71/Z predicted by
60- the two models are rather small when a < 400 pm and
6 = 0 (AT,,~ < 60 s). The discrepancies increase dra-
%
L matically when a > 400 pm and D, ID,. Obviously
40-

B
the differences between the predictions of the two
models are even greater when 6 # 0 which is a realistic
zo-
condition.
8 8 0 B
o! ,.,.,.I. I .,,I 1 5.2. Influence of source solution parameters: D, and S
-1.0 -0.5 0.0 0.5 1.0 1.5 2.0
Since D, and 6 are not considered by the simple
(b) lag(D3/D2)
model (Eq. (A2)), the predicted values of r1,2 are
Fig. 9. (a) t&, and (b) t& as a function of D, /D2 for various D, /D,
values (a = 100 pm, b = 125 pm, S = 20 pm, Q = 20, /3 = 0.5): q, independent of these parameters (Fig. 11). Comparison
D, /D, = 0.1; o, D, /D2 = 0.3; A, D, /D2 = 1; v, D, /D, = 100. with the general model shows that, when D, x-D,, 6
172 O.C. Keller et al. / Hollow j&e supported liquid membrane system

is small (less than 10 pm> and a is small (100 pm), the

differences in the values of pi,* predicted by the two
models are small (less than 20 s>. When D, = D, and 6
has a realistic value (above 10 pm), the differences in
the predicted ~i,~ values are significant (more than 50
s). It should be noted that the differences in 71/z
reported in Fig. 11 are based on small cy and l/p
*
values (a = 20, p = 0.5); as we shall see, for greater
values of (Y and l/p, the differences in the predicted
71/z may be much larger.

5.3. Influence of partitioning constants: (Y and p

In Sections 5.1 and 5.2, the differences in the T,,~
2ooj
;; . , ; , . , ; 40 60 80 100
values are quite important; these differences increase cr.
drastically when the preconcentration factor o//3 in- Fig. 12. 7 1,2 predicted by the simple (- ) and general models
creases ((w= 20 and p = 0.5 in Sections 5.1 and 5.2). (symbols) as a function of o for different p values (a = 100 km,
(Y/P may reach values of 300 or more in actual systems b=125 pm, S=O, D,=D,=6X10e6 cm* s-t, D,=2~10-~
[4,18]. It should be noted, in particular, that the simple cm2 s-l).
model fails to predict any role of /3 in ~r,~ (Eqn. (A2)).
The differences between the r1,2 values predicted by tence of a Nernst layer in the source solution, variable
the two models are then particularly important for high source concentrations, and variable partition coeffi-
values of l/p. This is confirmed by Fig. 12: for l//3 > cients. Comparison with a simplified model shows the
100 and cy> 30, Ar1,2 > 50 min. necessity for a general treatment of the problem. The
On the basis of the above calculations, it can be general model allows prediction to be made of the
concluded that the simplified model makes realistic concentration change in the strip compartment after
predictions provided that the following conditions are any kind of concentration change (step, exponential,
satisfied: a I 400 pm, 6 I 10 pm, D,, D, 2 2D,, linear function, positive or negative change) in the
p > 0.5. source solution (see Fig. 4).
For practical application, the exact treatment of the
6. Conclusion diffusion in the hollow fibre SLM system allows one to
decide whether to wait for equilibrium (to obtain a
In this paper we have calculated the response time larger preconcentration, and therefore sensitivity), or
of a cylindrical SLM system. We have used a general to make a measurement far away from equilibrium (to
model considering cylindrical symmetry, non-steady- increase the rapidity of the measurement). Prediction
state conditions in the three compartments, the exis- of the time required to purge the system (the time
necessary to decrease the strip concentration to zero
before making a new analysis) is also possible. This has
practical implications: if the purge time is long, a strip
purging system has to be used. Finally, the general
model also allows the best geometrical and hydrody-
namical conditions for designing the system to be de-
fined, as discussed in the text.

Acknowledgements

1
200 . .
.
. We are grateful to Dr. N. Parthasarathy and Mrs.

1
.
100 . i x M. Keller for patient reading of the manuscript. This
0 = work was supported by Swiss National Science Founda-
1 I I I, I II
0 2 4 6 8 10 tion Grant FNRS Project 20-31192.91.
IO6t&/cm2 s-’
Fig. 11. ~t,z predicted by the simple (0) and general Cm, 0, A, V) References
models as a function of D, for different 6 values (a = 100 pm,
b = 125 pm, D, =6x 10e6 cm’ s-l, D, = 2x 10m6 cm2 s-l, (Y= 20, [ll R.M. Izatt, J.D. Lamb and R.L. Bruening, Sep. Sci. Technol., 23
p=O.5): n,6=O~m;~,6=10~m; r,6=20wm; r,s=40I*m. (1989) 1645.
 O.C. Keller et al. / Hollow fibre supported liquid membrane system 173

[2] R.M. Izatt, G.A. Clark, J.S. Bradshaw, J.D. Lamb and M. With these conditions, the problem is reduced to the
Christensen, Sep. Purif. Methods, 15 (1) (1986) 21. diffusion through the membrane. We can then express
[3] R.M. Izatt, G.A. Clark, J.S. Bradshaw, J.D. Lamb and M.
Christensen, Pure Appl. Chem., 58 (11) (1986) 1453.
the internal concentration C, in a very simple way:

1
[4] N. Parthasarathy and J. Buffle, Anal. Chim. Acta, 254 (1991)
1-19.
acl(t)
-=-- 4 Al”,” c3
---
Cl(f)

--cl(t)
at b - a v,“nm

t-41)
[5] R.D. Noble and J.D. Way, ACS Symp. Ser. 347 (1987) 110. P ff

a1
[6] H. Hirathani and Y. Yamaguchi, in T. Araki and H. Tsukube
(Eds.1, Liquid Membrane: Chemical Applications, CRC Press, 202 c3
Boca Raton, FL, 1990. =a(b-a)
[7] T. Nishiki and R.C. Bautista, AIChE J., 31 (1985) 2093.
[-p
[8] H. Tromp, M. Burgard and M.J.F. Leroy, 38 (1988) 295.
where 4”Im” and ~“Im are the surface area of the
[9] J.A. Johnsson and L. Mathiasson, Trends Anal. Chem., 11 membrane in the lumen and its volume respectively.
(1991) 107.
Integration of Eqn. (Al) yields
1101L. Knutsson, G. Nilve and B. Ussen, Int. J. Environ. Anal.
Chem. 45 (1991) 117.
1111D. Doetz, Proc. 4th Int. Conf. on Indoor Air Quality and
c,(t) = 7 [I - ,-2Waa(b-a)]
(Ml
Climate, FRG, 1987.
[121 J. Koryta, Anal. Chim. Acta, 206 (1980) 1. with
I131 R.L. Solsky, Anal. Chem., 62 (1990) 21.
iI41 P. Ammann, Ion Selective Microelectrodes, Principles and Ap- aa(b -u)
plications, Springer, Berlin, 1986. 7t/2 = 142)
24
I151 S. Kamata, A. Bhale, Y. Fukunaga and H. Murata, Anal.
Chem., 60 (1988) 2465.
I161Y. Umezawa, H. Katoaka, W. Taum and E. Kimura, Anal.
Appendix B: Calculation of the inverse Laplace trans-
Chem., 60 (1988) 2392.
I171M. Lerchi, E. Bakkar, B. Rusterholz and W. Simon, Anal. form of metal concentration in the strip solution
Chem., 64 (1992) 1534.
[181N. Parthasarathy and J. Buffle, Anal. Chim. Acta, 284 (3) (1994) The following dimensionless variable transforma-
649. tions are introduced:
I191P.R. Danesi, J. Membr. Sci., 20 (1984) 231-248.
Dl J.P. Behr, M. Kirch and J.M. Lehn, J. Am. Chem. Sot., 107 r’=r/b,a’=a/b
(1985) 241.
D'l = D,/D,, oj = D,/D,
ml L.F. Reush and E. Cussler, AIChE J., 19 (1973) 736.
D21 R. Mohapatra, S.B. Kanungo and P. Sarma, Sep. Sci. Technol., t’ = tD2/b2 (Bl)
27 (1982) 765.
1231T.B. Stolwijk, E. Sudholter and D.N. Reinoud, J. Am. Chem. After performing Laplace transformation [Bl], and with
Sot., 109 (1987) 7042. these dimensionless variables, Eqs. (l)--(7) become
[241P.R. Danesi, Sep. Sci. Technol., 19 (1984) 857.
[251J. Buffle and M. Spoerri, J. Electroanal. Chem., 129 (1981) 67.
WI K.H. Mancy, D.A. Okun and C.N. Reilley, J. Electroanal.
4 (1962) 65.
Chem.,
a+-‘,
art2
S)
+ r,-l acm
- ar, (s/D;)c,(r’, s) = 0

D71 H.S. Carslaw and J.C. Jaeger, Conduction of Heat in Solids, W)

Oxford University Press, Oxford, 1959, p. 346.
LW V. Mitlin and L. Bromberg, Chem. Eng. Sci., 47 (3) (1992) a2c2(rf, S) k,(r’, s>
695-703. I2
+ rr-’ ar, -x2( r’, s) = 0 (w
Appendix A: Simplified model k,(r’, s)
[ D;ir
ar’ ] 0 w
The following approximations are made in addition r’=O=
to assumptions (2)-(4) of Section 3 (see also Fig. 3(a)). Cl(U’,s>
(1) Concentration gradient is zero in the strip solution. (Y= (w
c2(a’, s>
If the lumen diameter is small enough (see Section
5) and the diffusion in the membrane is rate limit- db’, s)
ing, this condition is valid. P= WI
(2) The Nernst layer in the source solution (compart- cz(b’, s>
ment 3) is non-existent, i.e. concentration in com-
partment 3 is constant. This condition becomes
increasingly valid with better agitation of the source
yr,==,
[Df?,:: =I”2::“1.,==, W)
solution. s>
(3) The concentration profile in the membrane is con-
sidered to be linear at any time. I k2(r’,
ar I r’= 1
=a’-‘[go-kqqa, P8)
174 O.C. Keller et al. / Hollow fibre supported liquid membrane system

with 6’ = S/bDj. Appendix C: Inversion of Eqn. (Bll)

The general solutions for Eqs. (B2) and (B3) are
cr(r’, s) =A,Z,(h,r’) +B,K,(A,r’) Inverting Eqn. (Bll) is a rather involved matter.
(B9)
Instead a simpler approach has been used below.
c*( r’, s) =&I,( A&) + B,K,( h,r’) (BIO) If g(t’ I 0) = 0, the Laplace transforms of Eqns.
with (B2)-(B8) can be replaced by the corresponding Fourier
transforms. In practice, the Laplace parameter s of
A, = (s/D;)“* Eqns. (B2)-(b12) can be replaced by iw with i = \/-1
A, = sl/* [Cl]. Working with Fourier transforms is simpler, be-
cause these and their inverse transforms can be com-
Z, and K, are modified Bessel functions or order 0 puted numerically (discrete Fourier transform).
[B21. Consider a general function f(io) sampled with N
Since K,(O) = 03 and c1 has to be finite at r’ = 0 points in the frequency domain -TN/T I w I TN/T,
(Eqn. (B4)), B, must be equal to zero. Expressions for where T is the corresponding time domain for which
A,, A, and B, can be found by using the conditions the inverse function is calculated (in our case T = t’).
(B5), (B7) and (B8) in Eqns. (B9) and (BlO) and solving Its discrete inverse Fourier transform is defined as
the resulting system of three equations.
[Cl,C21
As our main interest is to study the evolution of
N-l
metal concentration in the strip solution, we shall give
f( tn) = (Aw’J2~r) c f( w’J2~r) e(‘“;‘n) (Cl)
only the result obtained in this way for cl( r’, s) (for
n=O
0 I r < a):
with
cr(r’, s)
2rk/T ifO<k<N/2
= - (g(s)Z,(A,V)/A,) (C2)
@‘= 2r(k-N)/T ifN/2+1<k<N-1
i
x(~;‘*z,(A~)[I,(~,)P-K,(~,)QI and
-a-‘l,(h,)[I,(A,)P+K,(h,)Ql)-’ Wl) Aw; = 2r/T (C3)
with t,=nT/N withO<n_<N-1 (C4)
A3 = a’Al
In our case, the only difficulty that arises by using
A, = afA2 Fourier transforms is that substitution of iw for s in
f’ = PUA,) - a’A,K,(A,) Eqn. (Bll) leads to modified Bessel functions with fi
in the argument, i.e. Kelvin functions [C3] ber, bei, ker
Q = Pb(A2) + 6’A,Z,(h2) and kei. However, this is a minor problem: good poly-
and where I, and K, are modified first-order Bessel nomial approximations of these functions [C31, and
functions. also scientific program packages (e.g. ASYST), are avail-
Eqn. (Bll) can be rewritten in the following abridged able for computing modified Bessel functions of com-
form: plex argument. Hence, we can now calculate F( r’, io;)
(Eq. (B12)) for w; values as defined in Eqn. (C2).
ci( r’, s) = g( s)F( r’, s) (BI2)
Three distinct domains are considered.
where F(r’, is called the transfer function of the (a) For 0; > 0, F( r’, iwi) can be calculated from the
problem. Knowing this function, it is possible to obtain corresponding Kelvin functions calculated by means
an expression for ci( r’, s), irrespective of the value of of the polynomial approximations given in Ref.
g(s). However, to obtain ci(r, t), the inverse Laplace [C3]. These polynomial approximations are defined
transform of the right-hand side of Eqns. (Bll) or for arguments below 8; this implies that N < 20T.
(B12) must be calculated. This is the main difficulty of (b) For w; < 0, some of the properties of Fourier
this method; however, as shown below, this problem transforms were used, i.e. the real part of the
can be circumvented using discrete Fourier transforms. Fourier transform of a real function is symmetric,
whereas its imaginary part is antisymmetric. In our
References for Appendix B case this implies that
[Bl] M.N. Ozisik, Heat Conduction, Wiley, New York, 1980, p. 246. F( r’, -iw;) =F*(r’, iw;) (C5)
[B2] M. Abramowitz and I. Stegun, Handbook of Mathematical
Functions, Dover Publications, New York, 1965, p. 374. where * indicates complex conjugate.
 0. C. Keller et al. / Hollow fibre supported liquid membrane system 175

(c) For 0; = 0 D;, D; ratios D,/D, and D,/D,

lii[F( r’, io;)] = cu/p F( r’, s> transfer function of the problem
(C6)
g(t) substance bulk concentration of compart-
ment 3 (M)
For the function g(t,), a square pulse function was
value of g at point n (M)
chosen (Fig. 4):
Laplace transform of g(t)
g(t,)=O forOIn<N/4-1 \/-1
g(t,)=l forN/4<nI3N/4-1 modified Bessel functions of the first
(CT)
kind of order 0 and 1
g(t,) =0 for3N/4<nIN--1
modified Bessel functions of the second
Using these conditions, the concentration of metal in kind of order 0 and 1
the strip solution as function of time after a change number of sampling points in discrete
g(t,> in the source solution is readily obtained from the Fourier transform
formula P term in Eqn. (Bll)
e term in Eqn. (Bll)
cr( r’, t’) = IFT(FT[ g( t,)]F( r’, iw;))
r distance from the axis of the cylinder
where FT is the discrete Fourier transform and IFT km)
the discrete inverse Fourier transform. Fourier trans- r’ ratio r/b
form and inverse transform algorithms are available on s Laplace variable
most mathematical programs as fast algorithms [C2]. t time (s)
t’ dimensionless time, D2t/b2
References for Appendix C 4l time corresponding to point n of the
[Cl] S. Poitry, Th&se, Section de Physique, Universitt de Genke, discrete inverse Fourier function (9
1993.
T time domain of the discrete inverse
[C2] E.O. Brigham, The Fast Fourier Transform, Prentice-Hall, En-
glewood Cliffs, NJ, 1974.
Fourier transform function (t’ in our
[C3] M. Abramowitz and I. Stegun, Handbook of Mathematical case)
Functions, Dover Publications, New York, 1965, pp. 384-385. partition coefficient between strip and
membrane
Appendix D: Nomenclature partition coefficient between source and
membrane
a radius of internal cylinder (strip solution) Nernst layer thickness (cm)
(cm> ratio 6/b
I
ratio a/b ( s/D;)“~
:,, A,, 4 “constants” of the general solutions sw
b radius of internal cylinder + membrane a’h 1
(cm) a’A2
Cl, c2, c3 substance concentration in compart- time at which the concentration in the
ments 1, 2, 3 (M) centre of the fibre lumen is half of equi-
c3(m, t) substance bulk concentration of compart- librium concentration
ment 3 (M) Fourier variable (frequency) (s - ’ >
D,, 4, D, diffusion coefficients in compartments 1, sampling frequency for discrete Fourier
2, 3 (cm2 s-l) transform