Vous êtes sur la page 1sur 7

An eigenvalue method for determination of multicomponent diffusion coefficients.

Application to NaOH NaCl HzO mixtures + +


DEREKG. LEAIST'A N D ROBERT A. NOULTY
Departrnerlr qf' Chemisrry, Urliver.si~of' Wesrerrl O n a r i o , Lor~rlorr.Otlt.. Carlacla N6A 5B7
Received June 25. 1984

DEREKG. LEAISTand ROBERTA. NOULTY.Can. J. Chem. 63, 476 (1985).


A general method for the determination of multicomponent diffusion coeff~cientsis developed using the algebraic technique
of matrix diagonalization. When linear combinations of measurements from several multicomponent diffusion experiments
performed with different initial concentration gradients (but with the same final composition) are analyzed as simple binary
data, particular combinations may be found that transform the multicomponent diffusion coefficient matrix D to diagonal form
and thus yield time-invariant. pseudo-binary diffusion coefficients: the eigenvalues of D. Since the matrix that diagonalizes
D is given by the coefficients used to form the linear combinations, D is easily recovered by the inverse transformation. The
Can. J. Chem. Downloaded from www.nrcresearchpress.com by 84.78.191.92 on 04/09/11

advantages of the eigenvalue method are briefly discussed. For testing purposes, ternary diffusion coefficients are determined
from conductance measurements for d~luteaqueous NaOH + NaCl mixtures. Diffusion of NaOH in aqueous NaCl is
significantly more rapid than in pure water, and large coupled flows of NaCl are observed. The results are in close agreement
with behavior predicted by Onsager-Fuoss theory.

DEREKG. LEAISTet ROBERTA. NOULTY.Can. J. Chem. 63, 476 (1985).


Faisant appel i une technique algkbrique de diagonalisation des matrices, on a developpe une mCthode gkntrale permettant
de determiner les coefficients de diffusion i plusieurs composantes. Lorsqu'on analyse, sous forme de donnkes binaires
simples, des combinaisons linkaires de mesures provenant de plusieurs expCriences de diffusion i plusieurs composantes,
rkaliskes avec des gradients diffkrents de concentrations initiales (mais avec la meme composition finale), on peut trouver des
combinaisons particulikres qui permettent de transformer en une forme diagonale la matrice D du coefficient de diffusion i
plusieurs composantes; on peut ainsi obtenir des coefficients de diffusion peudo-binaires qui ne varient pas avec le temps.
Puisque la matrice qui diagonalise D est obtenue i partir des combinaisons utilisees pour former Ies combinaisons lineaires.
on peut facilement rtcupkrer D par la transformation inverse. On discute brikvement des avantages de la mCthode des valeurs
For personal use only.

de eigen. Pour fins de verification, on a determink les coefficients de diffusion ternaires i partir de mesures de conductivitt
sur des solutions aqueuses diluees de mClanges de NaOH + NaCI. La diffusion du NaOH dans le NaCl aqueux est beaucoup
plus rapide que dans I'eau pure et on observe d'importantes migrations couplCes de NaCI. Les rksultats obtenus sont en bon
accord avec le comportement prCdit par la thCorie de Onsager-Fuoss.
[Traduit par le journal]

I. Introduction tial conditions as simple binary diffusion data. Particular com-


Multicomponent diffusion is of theoretical interest and con- binations diagonalize (18, 19) the multicomponent diffusion
siderable practical importance ( 1 , 2). In many cases the dif- equations and yield time-invariant, pseudo-binary diffusion co-
fusivity of a solute in a multicomponent mixture is significantly efficients - the eigenvalues of the multicomponent diffusion
different from its binary diffusivity, especially if the solutes are coefficient matrix D. The eigenvalues provide D in diagonal
electrolytes or they react chemically (3-8). Moreover, because form. We show that the coefficients used to form the linear
diffusional flows interact, multicornponent diffusion of a com- combinations define the matrix that diagonalizes D, and that D
ponent produces countercurrent or cocurrent flows of other is easily reconstructed by the inverse transformation using ma-
components in the mixture. If the interaction is strong, a gra- trix inversion and multiplication.
dient in concentration of one component can effectively control To evaluate the eigenvalue method, ternary diffusion coeffi-
transport of a second component. cients have been determined for NaOH NaCl + H?O mix- +
Multicomponent diffusion is measured by classical gradient tures. This system was chosen in order to test recent predictions
techniques (8- 16). Typically, a system is prepared with initial (21) that aqueous hydroxides diffuse much more rapidly in
concentration differences"across a sharp boundary. As the sys- supporting electrolytes than in pure water. Another purpose of
tem relaxes toward thermodynamic equilibrium, changes in a the study was to determine whether or not pH gradients pro-
concentration-dependent property (such as the refractive index) duced by hydroxide gradients, like pH gradients produced by
are measured along the diffusion path. Diffusion coefficients acid gradients (6-8), can generate large coupled flows of salts.
are then obtained by fitting mathematical solutions of the mul- Coupled transport driven by pH gradients has applications to
ticomponent diffusion equations to the measurements. Because separation technology and plays vital roles in biological sys-
these solutions are complicated functions which contain the tems (22-24).
diffusion coefficients as non-linear parameters, determination
of diffusivities for multicomponent systems by this procedure 2. Theory
is substantially more difficult than for binary systems (17). n + 1 comporzerzts
The purpose of this work is to develop a simpler method for Provided the concentration gradients are not too large, iso-
the analysis of multicomponent diffusion data. Rather than fit thermal diffusion in a system containing n +
l components
solutions of the diffusion equations to measurements from indi- may be described by rz coupled equations (25)
vidual experiments, we propose to treat linear combinations of n

measurements from experiments performed with different ini- [I] ac,/at = C D;~V?C~ i = I , 2, . . . , n
P= l

' To whom all correspondence should be addressed. where c, is the concentration of component i in moles per unit
LEAIST A N D NOULTY 477

volume, t is the time, and V' denotes the Laplacian operator. performed with initial concentration gradients in solutes 1 and
Multicomponent diffusion coefficient D, measures the flux of 2, and 1 diffuses rapidly compared to 2, values of D,,(t) will be
component i produced per unit concentration gradient in com- large initially but will decrease toward the end of the experi-
ponent k. In matrix notation ment as the initial gradient in 1 decays rapidly and diffusion of
slower solute 2 predominates.
Although apparent diffusion coefficients are generally time-
where c denotes the column vector of n component concen- dependent, consider an experiment with particular initial condi-
trations and D is the n X n matrix of diffusion coefficients. tions X,"' such that the column vector of initial concentration
Consider linear combinations (18, 19) u; of the component differences Ac"" (0) is an eigenvector of D
concentrations defined by

[3] ci= z AiPuP


n

P= l

Upon substitution of c = Au into eq. [2] and multiplying by


Can. J. Chem. Downloaded from www.nrcresearchpress.com by 84.78.191.92 on 04/09/11

A-', the diffusion equations may be rewritten as


[4] du/at = (A-'DA) V'u Under these conditions

If coefficients A, are chosen so that the columns of A are


eigenvectors of D, the transform of D is diagonal (20) and hence

Consequently, the experiment behaves as a simple binary dif-


fusion experiment with time-invariant, pseudo-binary diffusion
coefficient equal to eigenvalue DL.Once values of Xi'" and Dk
are established, the multicomponent diffusion coefficients are
easily determined using
For personal use only.

and the transformed diffusion equations


and the inverse transformation
are de-coupled. D; are eigenvalues of D and ui are solutions of
the binary diffusion equations.
Reduction to diagonal form is a powerful technique (18, 19)
that is frequently used to solve coupled differential equations.
Can the same technique be used to measure multicomponent
diffusion coefficients? Consider n diffusion experiments run at
the same final composition of the system but with different
initial concentration gradients. 'The idea is to use measurements Since D is unknown at the start of the experiments, it is of
from these runs as a "basis set" of data from which suitable course virtually impossible to choose the exact initial condi-
linear combinations may be taken that diagonalize D and thus tions that diagonalize D. Nevertheless, by taking linear combi-
simplify the analysis. nations of data from experiments with different (specifically,
Let Y represent the refractive index (4-6, 8- 13), conduc- linearly independent values of Ac(0)) initial conditions, mea-
tance ( 1 4 , 2 I), density (26), optical absorbance (27), or another surements corresponding to arbitrary initial conditions may be
concentration-dependent property measured along the path of obtained. For instance, if AY;(t)/AY;(O) symbolizes nor-
diffusion during an experiment. For initial concentration differ- malized measurements of property Y for an experiment in
ences Aci(0) across a sharp diffusion boundary, the corre- which the initial concentration gradient is entirely in coni-
sponding difference in Y is " ponent i (Xi = I ) , measurements corresponding to arbitrary
n initial conditions Xi may be constructed from
[7] AY(0) = 2 SiAci(0) n
,= 1
[Is] AY(t)/AY (0) = 2 X,AY,(t)/AY;(O)
where Si are derivatives i= l

A detailed example is given later.


In order to establish the initial conditions that reduce the
evaluated at the final composition of the system. Xi will be used diffusion equations to diagonal form, the standard deviation of
to denote the fraction of AY(0) contributed by component i. measured values of the apparent diffusion coefficient
n
[9] Xi = S,Ac,(O)/z SI;Ack(0)
I=I
Apparent binary diffusion coefficients D,(t) may be obtained may be evaluated as a function of Xi. D, is the mean apparent
from multicomponent experiments by analyzing measured val- diffusion coefficient
ues of Y as if they were from a binary experiment (8, 1 1, 2 1, N
28, 29). Apparent diffusion coefficients obtained in this man- [I71 B, = N - ' 2 &(ti)
ner usually vary with time. For example, if an experiment is ,=I
478 CAN. J. CHEM. VOL. 63. 1985

For the particular initial conditions Xi'"', the apparent diffusion the solutions contained less than 0.05% carbonate based on total
coefficient is effectively constant over the duration of a run, alkali. Water used to make up the solutions was doubly-distilled and
and hence u,-,is minimized. Therefore, values of Xi".' may be deionized (conductivity 0.8 X 10." ohm-' m- I). Corrections for
obtained from the positions of the minima in u,]. Eigenvalue DL solvent conductance were negligible.
is obtained from the mean value of the apparent diffusion Data arzcr!\~.si.s
coefficient evaluated at x , ' ~ ' . For binary diffusion in a Harncd cell, the difference in electrolyte
concentration between the top (T) and bottom (B) electrode positions
Tenjury cllffilsiorl (11 = 2) is given by (30)
When the standard deviation of the apparent diffusion coef-
ficient for a ternary system is plotted against X I , minima are [20] Ac(r) = c,(r) - c.,.(r) = C exp (-T'D~/N')
found at XI"' and XIe'. Since XI + X? = 1, we have whcre C is a constant, D is the binary diffusion coefficient, r is the
time elapsed since introduction of electrolyte, and n is the height of the
diffusion channel. For strong electrolytes. Ac(r) is directly propor-
tional to the conductance difference defined by (14, 21, 30)
The mean diffusion coefficients evaluated at XI"' and XI"' equal
Can. J. Chem. Downloaded from www.nrcresearchpress.com by 84.78.191.92 on 04/09/11

eigenvalues D l and D,, respectively. The ternary diffusion


coefficients are evaluated from eq. [ 141. where kB/kT denotes the ratio of cell constants of the top and bottom
electrode pairs. Values of D are conveniently determined from the
difference equation

3. Experimental
by pairing each conductance reading with the corresponding reading
Ternary diffusion coefficients of dilute aqueous NaOH + NaCl
taken on the following day.
mixtures were measured at 25°C using a simplified version (14) of
In order to determine ternary diffusion coefficients, conductance
Harned's conductimetric procedure (30). In this technique, initial
differences AKl(r) and AKZ(r)were measured for experiments run
concentration gradients are formed by introducing a small volume of
with the initial concentration gradient entirely in NaOH (XI = I) or
electrolyte into the bottom of a column of solution. The rate of dif-
NaCl (X, = O), respectively. Apparent diffusion coefficients for initial
For personal use only.

fusion is followed by monitoring changes in electrical conductance


condition XI were calculated using the difference equation
measured near the top and bottom of the column.
Because Harned's technique employs very small concentration gra-
dients. changes in the transport coefficients due to compositional
changes along the diffusion column are negligible, and the measured
diffusion coefficients correspond to differential values defined by eq. wherein
[ I ] . (For diaphragm ccll results (31). and to a lesser extent for optical
results (32, 33). small but troublesome corrections arc required to
allow for the concentration-dependence of the diffusion coefficients.)
Moreover, the conductirnetric technique is reliable at low ionic
strength where the transport coefficients can be estimated from lim- 4. Results and discussion
iting ionic conductances (14). Comparison of measured diffusion co-
efficients with theoretical predictions will provide a check on the Diffusion coefficients for NaOH(c,) + NaCI(c,) + H,O
reliability of the proposed eigcnvaluc method. mixtures were measured at a total electrolyte concentration of
0.020 mol L-I for ratios of c, :c2 equal to 1 :3, 1 : 1, and 3 : I .
Eq~tiprnerlrarid procedure Paired experiments with the initial concentration gradient in
The diffusion cells used in this work had cylindrical diffusion chan-
nels (height 0.04560 m, diameter 0.0126 m). The cells were con-
NaOH or NaCl were carried out in triplicate at each of the three
structed from high density polyethylene rather than glass in order to compositions studied.
resist leaching by hydroxide. Pairs of circular platinum electrodes Typical results for a pair of experiments for e l = 0.0150 and
-
(diameter lo-' m, ccll constant k 10' m-I) werc positioned at levels c, = 0.0050 mol L-' are shown in Figs. I and 2. In Fig. I,
one-sixth of the distance from the top and bottom.of the channel. At values of the apparent diffusion coefficient computed from eq.
the start of each experiment, 0.1-0.3 mL of electrolyte was carefully [23] are plotted against t for various initial conditions. It is
injected with a calibrated syringe into the bottom of each solution- evident from the figure that the apparent diffusion coefficient
filled chamber. Initial concentration differences across the diffusion varies substantially over the duration of a run when the initial
boundary were 0.05-0.2 mol L-I. Conductances were measured six concentration gradient is entirely in NaOH or NaCl (XI = I and
times per day for four consecutive days beginning one day aftcr X, = 0 , respectively). It may be noted, however, that values of
injection of electrolyte. Further details of the equipment and procedure
have been reported (14. 34). Cell constants were obtained by cali-
D,(t) are essentially constant for values of XI near 0 . 6 and 1.2.
bration with 0.01 mol L-' aqueous KC1 (conductance 0.01413 m' S The "stationary" values of D;,(t) (the eigenvalues of D) were
mol-' (35)). Values of S I and S2 were determined using an auxiliary found by plotting the standard deviation of the apparent dif-
conductance cell. fusion coefficients against X I , the fraction of the conductance
difference across the initial diffusion boundary due to NaOH.
Reagents From the two minima in u,, shown in Fig. 2, initial conditions
Reagent grade NaCl was dried in a vacuum oven at 140°C and used
without further purification. Stock solutions of NaOH werc prepared
XIo' = 1.21 and X,"' = 0.64 that diagonalize D were obtained.
by diluting saturated solutions that had been filtered to remove in- From the mean apparent diffusion coefficients evaluated at
soluble darbonate. The NaOH solutions werc stored in polyethylene these two values of X I , eigenvalues D l = 2.58 X 1 0 - 5 n d D,
bottles and protected from the atmosphere. Concentrations of NaOH = 1.71 X lo-' m' s-I were obtained.
were determined by titration against potassium hydrogen pthalate. The results of the experiments are summarized in Tables 1
Winkler analyses (36) before and aftcr the experiments indicated that and 2. As shown in Table 2, values of D , , , the ternary dif-
LEAIST AND NOULTY
Can. J. Chem. Downloaded from www.nrcresearchpress.com by 84.78.191.92 on 04/09/11

FIG. 1. Apparent binary diffusion coefficients of aqueous NaOH(1) + NaCl(2) for el = 0.015 rnol L-' and cz = 0.005 rnol L-' for selected
initial conditions. For the experiment with the initial concentration gradient entirely in NaCl (XI = 0). 0.32 mL of solution containing 0.015
rnol L-' NaOH and 0.0712 rnol L-' NaCl was injected into a cell filled with solution containing 0.015 rnol L-' NaOH and 0.001 rnol L-' NaCI.
For the experiment with the initial gradient in NaOH (XI = I ) , 0.32 mL of solution containing 0.254 mol L-.' NaOH and 0.005 mol L-' NaCl
was injected into the same cell filled with 0.001 rnol L-' NaOH and 0.005 rnol L-' NaCI. The curves for X I = 0.4, 0.64, 0.80, 1.15, 1.21,
and 1.25 were obtained by analysis of linear combinations of data from the two experiments run with X I = 0.00 and X I = 1.00. The volume
of the cell was 5.78 mL.
For personal use only.

FIG.2. Standard deviation of the apparent diffusion coefficient of aqueous NaOH(1) + NaCl(2) for el = 0.015 mol L-' and c1 = 0.005 rnol
L-'. See caption of Fig. I for details.

fusivity of NaOH in aqueous NaCI, increase steadily as the down OH- also generates counterflow of C1-. Highly mobile
system becomes richer in NaCI. At the same ionic strength, in OH- can thus diffuse down the NaOH gradient more rapidly
contrast, the binary diffusivity of aqueous NaOH is only 1.99 than Naf while still maintaining electroneutrality. This mech-
x lo-' mm? s-' (34). In binary solutions of NaOH, however, the anism accounts for the increased diffusivity of the NaOH com-
condition of electroneutralitv reauires that Nat and OH- dif-
d .
ponent in aqueous NaCl as well as the large negative values of
fuse at the same rate, even though OH- has approximately four Dm?, (the cross-diffusion coefficient measuring coupled flow of
times the mobility of Naf (35); the electric field produced by NaCl produced per unit concentration gradient in NaOH). Sim-
flow of the electrolyte slows down OH- and speeds up Naf so ilar diffusional behavior has been observed for HCI + NaCl +
that they travel at the same rate and the electric current is zero. H 2 0 and other acid-containing mixed electrolytes (6, 8, 21). In
In ternary NaOH(1) + NaCl(2) + Hm?O mixtures, on the other those systems, the electric field generated by diffusion of high-
hand, the electric field along the NaOH gradient which slows ly mobile protons produces counterflow of cations.
C A N . J. CHEM. VOL. 63. 1985

TABLEI. Parameters used to calculate experimental diffusion coefficients for NaOH( I ) +


NaCl(2) + HZ0 at 25°C
(.I(( c.2 SI/SZ x,"' XI"' D," D:

" I n lo-' mol L-'


lo-.*J ",? 1.

2. Observcd and predicted ternary diffusion coefficients for NaOH(1) + NaCl(2) + H20
TABLE
at 25°C
Can. J. Chem. Downloaded from www.nrcresearchpress.com by 84.78.191.92 on 04/09/11

5.0 15.0 Obsd. 3.72 (k0.02) -0.30 (20.04) - 1.34 (20.13) 1.76 ("0.03)
Calcd. 3.70 -0.23 - 1.29 1.7 1
10.0 10.0 Obsd. 2.95 (k0.19) -0.33 (20.07) -0.71 (20.21) 1.70 (20.10)
Calcd. 2.93 -0.37 -0.7 1 1.82
15.0 5.0 Obsd. 2.42 (k0.07) -0.52 (20.01) -0.31 (20.10) 1.90 (k0.01)
Calcd. 2.39 -0.47 -0.30 1.89
"In lo-.' mol L - '
l'ln 10-9 ",'

TABLE
3. Comparison of obscrved and predicted LiLcoefficients

(.;' RTL, ," RTL Iz" RTL?," RTL~?~


For personal use only.

5.0 15.0 0bsd. 1 .69 (20.02) -0.94 (20.04) -0.90 *O 1.94 (50.05)
Calcd. 1.67 -0.87 -0.87 1.89
10.0 10.0 Obsd. 2.18 (20.16) -0.94 (k0.09) -0.94 (k0.18) 1.47 (k0.12)
Calcd. 2.33 -0.97 -0.97 1.55
15.0 5.0 Obsd. 2.27 (k0.09) -0.65 (k0.05) -0.64 (k0.10) 0.89 (k0.02)
Calcd. 2.41 -0.625 -0.625 0.89
" I n lo-.' mol L-I.
" I n 10.'' niol m-' s-I

The difference in mobility between Na' and C I is not as TABLE


4. Molar ionic conductances and
great as the difference in mobility between Nai and OH-: C1- other parameters" used to calculate the-
is about 50% more mobile than Na ' (35). The electric field per oretical valucs for the transport coefficients
unit concentration gradient in NaCl is therefore weaker than for
NaOH, and values of D l , for aqueous NaOH(1) + NaCl(2) are Electrolyte A , AO,,,' P"
proportionally smaller than for D 2 ]Also, the increase in the NaOH 50.10 198.3 0.08"
diffusivity D,, of NaCl in NaOH-rich mixtures is not as dra- NaCl 50.10 76.35 0.16'
matic as for D l , in NaC1-rich mixtures.
Onsager's equations (25) "Ion size parameter (14) a = 4 X lo-"' ni;
n
solvent viscosity (35) q = 8.90 X lo-' kg m-'
s-I; solvent dielectric constant (35) E = 78.4;
[251 J , = - 2 L,LVFL Debye-Hiickel constant (35) a = 1.17.
I= I "Reference 34.
relate molar fluxes, J i , of components to gradients in the chem- ' Reference 39.
"In lo-' m' s mol'.
ical potentials of the components. Conveniently, values for the " I n L mol-I.
L,, phenomenological coefficients can be estimated (14) for
dilute mixed electrolytes from limiting ionic conductances and
Onsager-Fuoss theory (37, 38) of the electrophoretic and re-
laxation effects. Details of the calculations have been reported
(14). Given L, values, diffusion coefficients are calculated
from (25) for the chemical potentials of the solutes. Mean ionic activity
n coefficients (on the molar concentration scale) were evaluated
[261 Dik = 2 L,,,, p,,,/a using Guggenheim's expression (39)
111 = I

The parameters used to predict the transport coefficients of [29] In y,; = -a


vet + c2 + p l c l + pzcz + p,(el + c*)
+
aqueous NaOH(1) NaCI(2) are listed in Table 4. Values of I +-
dp,,,/dc, were calculated by differentiation of expressions where ci is the Debye-Hiickel constant and are small empir-
LEAIST A N D NOULTY

TABLE 5. Experimental diffusion coefficients for NaOH(I) + NaCl(2) + H,O at 25OC


determined by non-linear least squares

" I n lo-' mol L- '


10-0 m' s - l ,

ical constants which allow for specific ion interactions. 1. E. L. CUSSLER.Multicomponent diffusion. Elsevier, Am-
Experimental transport coefficients are compared with pre- sterdam. 1976.
dicted values in Tables 2 and 3. Agreement is very good. The 2. L. ONSACER. Ann. N. Y. Acad. Sci. 46, 241 (1945).
3. V. VITACLIANO and R. SARTOKIO. J . Phys. Chem. 74, 2949
Can. J. Chem. Downloaded from www.nrcresearchpress.com by 84.78.191.92 on 04/09/11

observed values of the L,, coefficients are consistent with the


Onsager reciprocal relation (2, 25) L,? = L?, which lends fur- ( 1 970).
4. R. P. WENDT.J . Phys. Chem. 69, 1227 (1965).
ther confidence to the reliability of the eigenvalue method. 5. P. J . DUNLOP. J . Phys. Chem. 63, 612 (1959).
A general procedure for the determination of ternary dif- 6. H. KIM,G. REINFELDS, and L. J . GOSTING. J . Phys. Chem. 77,
fusion coefficients from conductance measurements using non- 934 ( 1973).
linear least squares has been reported (40). In practice, we 7. D. G. LEAIST. J . Chem. Soc. Faraday Trans. I. 78, 3069 (1982).
found that the eigenvalue method was much easier to use. The 8. A. REVZIN. J . Phys. Chem. 76, 3419 (1972).
non-linear least-squares procedure, for instance, requires 9. Ref. I, Chapt. 4.
lengthy computations in order to simultaneously optimize val- 10. P. J. DUNLOP, B. J . STEEL,and J . E. LANE.Experimental meth-
ues of the four diffusion coefficients. Experimental diffusion ods studying diffusion in liquids, gases, and solids. In Physical
coefficients determined for NaOH + NaCl mixtures by the methods of chemistry. Vol. I. Edited by A. Weissberger and
B. W. Rossiter. Wiley, New York. 1972. p. 205.
non-linear least-squares method are given in Table 5. Com-
l I. H. FUJITA and L. J. GOSTINC. J . Phys. Chem. 64, 1256 (1960).
parison of Tables 2 and 5 indicates that the eigenvalue and
For personal use only.

12. J . G. ALBRICHT and B. C. SHERRIL. J. Solution Chem. 8, 201


non-linear least-squares procedures give essentially identical (1979).
results. With the eigenvalue method, however, the problem of 13. D. G. MILLER. J . Solution Chem. 10, 83 1 ( 1 98 1).
locating the minimum in the variance in four dimensions is 14. D. G. LEAISTand P. A. LYONS.Aust. J . Chem. 33, 1869 (1980).
replaced by the more manageable problem of locating two 15. E. L. CUSSLERand P. J . DUNLOP. J . Phys. Chem. 70, 1880
minima in one dimension (i.e. minima in cr, as a function of ( I 966).
XI). 16. G. K O ~ A N O V and I C HH. T. CULLINAN. Can. J . Chem. Eng. 49,
In effect, the eigenvalue method reduces the analysis of 753 (1971).
coupled, multicomponent transport equations to an analysis 17. H. KIM.J . Phys. Chem. 73, 1716 (1969).
18. H. L. TOOR.A. 1. Ch. E. J . 10, 448 (1964); 10, 460 (1964).
based on de-coupled, well-understood binary equations. This
19. W. E. STEWART and R. PROBER.lnd. Eng. Chem. Fundam. 3,
transformation enables the determination of ternary dif- 224 ( 1 964).
fusivities without explicitly solving multicomponent diffusion 20. H. MARCENAU and G. M. MURPHY. The mathematics of physics
equations or resorting to non-linear curve fitting procedures. and chemistry. 2nd ed. Van Nostrand, New York. 1956. p. 320.
The eigenvalue method holds especial promise for the mea- 21. D. G. LEAIST and P. A. LYONS.J . Phys. Chern. 86, 564 (1982).
surement of quaternary diffusion (17). For those systems, the 22. E. M. CHOY,D. F. EVANS, and E. L. CUSSLER. J . Am. Chem.
non-linear least-squares method requires simultaneous opti- SOC.96, 7085 (1974).
mization of nine parameters (a formidable task), while the 23. R. M. IZAIT, D. V. DEARDEN, D. W. MCBRIDE,J . L. OS-
eigenvalue method would involve searching for three minima CARSON,J . D. LAMB,and J . J . CHRISTENSEN. Sep. Sci. Technol.
in only two dimensions. 18, I I I3 (1983).
24. N. N . LI and A. L. SHRIER. It1 Recent advances in separation
The present study has employed time invariance of the ap-
science. CRC Press, Cleveland. 1972. p. 163.
parent diffusion coefficient to locate the eigenvalues of D. This 25. S. R. DE GROOTand P. MAZUR.Non-equilibrium thermo-
criterion is well suited to the Harned technique and other re- dynamics. North-Holland, Amsterdam. 1962.
stricted diffusion experiments ( 1, 4 1) in which the concen- 26. F. T. WALL,P. F. GRIEGER, and C. W. CHILDERS. J . Am. Chem.
trations are measured at two positions along the path of dif- SOC.74, 3562 (1952).
fusion. On the other hand, optical methods (4-6, 8- 13) such 27. M. TANICAKI, K. KONDO,M. HARADA,and W. ECUCHL.
as Rayleigh o r Gouy experiments employ free diffusion bound- J . Phys. Chem. 87, 586 (1983).
ary conditions. In those experiments the position of inter- 28. E. L. CUSSLER. Ph.D. Thesis. University of Wisconsin. 1965.
ference fringes is followed as a function of time. The apparent 29. E. L. CUSSLER and E. N. LIGHTFOOT. J . Phys. Chem. 69, 736
diffusion coefficient obtained by analysis of the position of a ( 1965).
30. H. S. HAKNED and R. L. NUITAL.J . Am. Chem. Soc. 69, 736
single fringe will generally vary from fringe to fringe. By
(1947).
locating the minimum in c r as~ a function of fringe number the
31. R. H. STOKES.J . Am. Chem. Soc. 72, 2243 (1950).
eigenvalue technique can also be used to analyze multi- 32. J . A. RARDand D. G. MILLER.J . Solution Chem. 8, 701 (1979).
component optical data. 33. J . G. ALBR~CHT and D. G. MILLEK. J . Phys. Chern. 79, 2061
(1975).
Acknowledgement 34. R. A. NOULTYand D. G. LEAIST.J . Solution Chem. 13, 767
This work was supported by the Natural Sciences and (1984).
Engineering Research Council of Canada. 35. R. A. ROBINSON and R. H. STOKES.Electrolyte solutions. 2nd
482 CAN. J. CHEM. VOL. 63. 1985

ed. Academic Press, New York. 1959. App. 6. I, 6.2. 39. E. A. GUCCENHEIM and J . C. TURCEON. Trans. Faraday Soc. 51,
36. W. C. P l ~ ~ c ~ E.
a nL.dHAENISCH. Quantitative analysis. Wiley, 747 (1955).
New York. 1948. p. 136. 40. D.G. L ~ ~ l s ~ P.
a nA.d LYONS.J . Phys. Chem. 85, 1756 (1981).
37. L. ONSACER and R. M. FUOSS.J . Phys. Chem. 36, 2689 (1932). 41. A. F. COLLINCS, D. C. HALL,R. MILLS,and L. A. WOOLF.
38. M. CHENand L. ONSACER. J . Phys. Chem. 81, 2017 (1977). J. Phys. E. Sci. Instr. 4, 425 (1971).
Can. J. Chem. Downloaded from www.nrcresearchpress.com by 84.78.191.92 on 04/09/11
For personal use only.

Vous aimerez peut-être aussi