25

J. Eleciroanal. Chem, 223 (1987) 25-50

Elsevier Sequoia S.A., Lausanne - Printed in The Netherlands
Review
IMPEDANCE SPECIROSCOPY AND ITS USE IN ANALYZING THE
STEADY-STATE AC RESPONSE OF SOLID AND LIQUID ELECTROLYTES *
J. ROSS MACDONALD
Department of Physics and Astronomy, University of North Carolina, Chapel Hill, NC 27514 (U.S.A.)
(Received 24th November 1986)
ABSTRACT
Impedance spectroscopy is defined and some of its applications illustrated for both liquid and solid
electrolyte situations. Particular emphasis is placed on complex least squares fitting of small-signal
frequency response data at various immittance levels. Most such response data must be fit to an
equivalent electrical circuit since a detailed microscopic model of the response is usually lacking. It is
found for most ionic response data, as well as that from purely dielectric systems, that one usually must
include in the equivalent circuit one or more distributed circuit elements in addition to the usual ideal
elements, such as resistors and capacitors. Important distributed circuit elements useful in equivalent
circuits are described. A number of actual equivalent circuits used in the past for both liquid and solid
electrolytes are presented and compared. Equivalent circuits following from detailed continuum models
of the electrical response of blocking and partly conducting systems are compared, and it is shown that
some of the same circuits may be used even in the presence of dc bias which usually makes the charge
distribution in the system very inhomogeneous. Because frequency response data have often been
inadequately analyzed, the authors general and powerful complex nonlinear least squares fitting program
is now available for use by others.
(I) GENERAL INTRODUCTION
Here I shall discuss some aspects of the non-equilibrium steady-state response of
systems which contain an electrical double layer (EDL), as nearly all not-purely-
ohmic electrically conducting systems do. Although quasi-equilibrium measure-
ments on EDLs with completely blocking electrodes yield primarily the total double
layer differential capacitance, Cr, by itself, small-signal (s-s hereafter) ac measure-
ments on EDL systems, which may or may not have blocking electrodes, yield much
more information whether we like it or not, and we do like it when we can
l Part of the present work was presented at The Robert A. Welch Foundation Conference on Chemical
Research. XXX. Electrochemistry, which was held in Houston, TX, 3-5 November, 1986.0022-
0022-0728/87/$09.10 0 1987 Else&r Sequoia S.A.
26
disentangle all the information available. The basic experiment is to apply a small
sinusoidal potential difference (p.d.) to the system and measure the resulting current
(or vice versa)_ The amplitude of the applied p.d. should preferably be less than the
thermal voltage Vr = kT/e. Sometimes a larger static p.d. is applied as well, but its
presence complicates the analysis of results greatly, especially in unsupported
situations. Measurements are carried out over as wide a frequency range as possible,
often from lop4 Hz to lo6 or lo7 Hz, usually using automatic measuring equipment,
and the impedance (p.d. divided by current) or admittance (current divided by p.d.)
is calculated at each frequency. Since there is generally a phase shift present
between current and potential, these ratios are, by definition, complex quantities,
and it is thus tautological to speak of complex impedance or complex admit-
tance as is frequently done by many electrochemists.
Although there is nothing intrinsically new in the above approach, one that has
been used in electrical engineering for seventy years or more, several new measure-
ment and analysis elements have been added in recent years which make it far easier
to carry out an experiment and to interpret its results. One such element is the
development of automatic measuring equipment [l]. The combination of the basic
frequency response experiment and some or all of the new elements has come to be
called Impedance Spectroscopy, abbreviated IS. In this section, I shall first give a
brief introduction to IS, with illustrations, then discuss some of its applications for
solid and liquid systems. If one assumes, as I shall, that the experimental data are
available, then it remains to present such data in a meaningful way and to analyze
them so that maximum understanding of the material-electrode-interface response
is gained.
(II) INTRODUCTION TO IMPEDANCE SPECTROSCOPY
Only a relatively brief background on IS will be presented because a much more
detailed introduction will soon be available [l]. First, the word impedance in IS is
a bit of a misnomer because in IS, one deals not just with impedance but with four
closely related functions which can be subsumed under the umbrella designation of
immittances. Thus IS can also stand for Immittance Spectroscopy. The four
functions are impedance: Z = Z + i Z, admittance: Y = Z-l, complex dielectric
constant; E = e - i c = Y/(i wC,), and complex modulus; it4 = e-l. Here w is the
angular frequency (w = 2?rf ), i = J -i ; and C, is the capacitance of the measuring
cell in the absence of the material of interest. Although exactly the same fitting
functions and mathematical models may be used for both intrinsically conducting
systems and for intrinsically non-conducting dielectric systems [2], we shall consider
only conducting systems here since they are more directly relevant to the response
of even completely blocking EDL situations than are pure dielectric systems.
Figure 1 shows the main elements in an IS experiment. The ideal situation, which
allows estimation of the values of microscopic quantities which characterize the
material-electrode system in detail, is to analyze the data by fitting them to a
27
SYSTEM
CHARAIXERIZATION
Fig. 1. Flow diagram for the measurement and characterization of a material-electrode system (reprinted
by permission from John Wiley and Sons, Inc., copyright 6 1987 [l]).
detailed microscopic model of the system, one which yields an explicit expression
for impedance as a function of frequency and predicts the temperature dependence
of all parameters present. Unfortunately, few such models are currently available,
even when they are derived using continuum approximations (linear differential
equations). In the absence of appropriate models one tries to make do with an
equivalent electrical circuit which lumps the main physical processes occurring into
macroscopic circuit elements such as capacitances, resistances and distributed
circuit elements (does). Even then there may still remain ambiguity about the most
appropriate way the elements should be connected together [1,3]. Such ambiguity
may often be reduced or eliminated by carrying out IS epxeriments on the same
system at several temperatures, electrode separations, pressures, oxygen tensions,
etc. This matter will be discussed further later on.
For all systems of physical interest there are two basic circuit elements which
always appear. These are the high frequency limiting geometric capacitance, Cs, and
the high frequency limiting resistance of the system, R,, the bulk or solution
resistance. They are extensive quantities and are part of the bulk response of the
system as opposed to its interfacial properties. As usual, I shall ignore the distinc-
I.---_--_---~
ZI
(a)
, ,c,
I I
(b)
Fig. 2. Two general equivalent circuits for the small-signal ac response of a material between two
conducting electrodes [4]. (a) Circuit used in early unsupported-system analysis. Here R, =R, +RE.
(b) Circuit which separates out main bulk elements explicitly.
tion between such quantities and their unit area values. The product Z&C, = mu, an
intensive quantity, is just the dielectric relaxation time of the system. When neither
Cs nor R, is distributed, they are frequency independent and the basic equivalent
circuit involving them is that shown in Fig. 2b. The impedance 2, represents the
response of all the rest of the system and is usually the quantity of primary interest.
The circuit of Fig. 2a will be discussed later.
It is worth emphasizing that Cp always spans the electrodes, as shown in the
figure [5]. It is improper when Z, Z 0 just to connect Cs in parallel with R,,
although this has often been done and frequently makes little actual difference to
the analysis. In fact, Cs is usually entirely neglected in liquid electrolyte studies
since it is generally very much smaller than the double layer capacitance, Co,. The
C, which is derived from quasi-equilibrium studies is actually Cs + Co, when no
adsorption effects occur, but it is virtually always identified as Co,. In IS studies it
proves important, however, to maintain the distinction between C, and Co,.
(I I .a) Presentation of data
The proper presentation of data can be very helpful in indicating possible
experimental errors and in suggesting the presence of various physical processes
leading to the overall response. Because IS experiments often extend over a wide
frequency range, it is usual to consider response as a function of the logarithm of
frequency (f or Y) or angular frequency. One often sees plots of -Z and/or Z
vs. log( f ), or sometimes ] Z ] and/or I#J vs. log(f) instead. Here, + is the phase
angle of the impedance; cp = tan-(Z/Z). In the dielectric literature, plots of
tan( #) vs. log(f) used to be common but are no longer.
One approach which is coming to be widely used is the plotting of Im(Z) = Z
(or - Im( Z) = - Z in capacitative systems) vs. Z, with frequency as a parametric
variable. Such complex plane plots can be very instructive. Here, however, I wish to
illustrate the usefulness of a further development in the presentation of IS data, the
three-dimensional plot with perspective [6], an approach which shows the full data
response in a single graph. The three dimensions of the plot are usually the real and
29
Fig. 3. Simple equivalent circuit and two 3-D perspective plots of its response viewed from different
directions [6].
imaginary parts of an immittance function and log(f) or log(w). The alternative of
using modulus and phase instead of real and imaginary parts is possible but usually
turns out to be less useful. Finally, when the magnitudes of the immittance
components vary by several orders of magnitude over the measured frequency
range, as they often do in solid electrolyte studies, it proves useful to use logarithms
of the real and imaginary parts in the 3-D plot.
Incidentally, many workers in the electrochemistry area use a non-standard
definition of impedance, one which amounts to writing Z = Z - iZ rather than
Z = Z + iZ. Although this usage is convenient for systems which show capacita-
tive rather than inductive response, it is an unwarranted redefinition of a long:
established quantity. In order to avoid the hubris of Humpty-Dumpty in Through
the Looking Glass, a proper alternative is to write Z* = Z - iZ, where Z* is the
complex conjugate impedance, and refer to the conjugate impedance in place of the
ordinary impedance. When the proper definition of impedance is maintained, one
may use any of the equivalent designations - Im( Z) = - Z = Im( Z * ) in plotting.
Figure 3 shows ordinary 3-D plots for the impedance of a simple equivalent
circuit, typical of those often appearing in solid or liquid electrolyte IS studies (with
Cs omitted). The two plots involve different viewing positions. The heavy line
represents the full response and the other three curves its projections in the three
planes. In addition to showing the full response, these plots thus include all three of
the different 2-D plots commonly used in the past. Note that the two parallel
30
Fig. 4. Three-dimensional perspective plot of 83 K Na /3-alumina impedance data [7-91.
elements in the equivalent circuit lead to a single time constant and thus to a
semicircle in the complex plane, a common response shape. For even more realistic
viewing, one could use a stereoscopic pair of 3-D plots.
Figures 4 and 5 illustrate the usefulness of 3-D plots in highlighting experimental
errors in an IS study of the solid electrolyte Na /3-alumina [7,8]. The main response
curve and the complex plane projection curve in Fig. 4 show that the lowest
frequency data point is inconsistent with the rest of the response. But note that the
Fig. 5. Three-dimensional perspective plot of 83 K Na @hnina complex modulus data 17-91.
31
projections in the other two planes, the most common sorts of plots in the past,
show no trace of this error! Their use alone, as in the original data analysis [9], thus
gives no clue to the presence of the error. Now the complex modulus function,
M = iwC,Z, weights the higher frequency data points more heavily than the lower
frequency ones. Since many higher frequency points in Fig. 4 are not well resolved
with the scale used, a modulus 3-D plot seems desirable.
Figure 5 shows that such a plot allows one to identify the same low-frequency
error discussed above and to discover two new problems in higher frequency
regions. Here one sees a curving back at the highest frequencies which is not
theoretically likely and thus suggests the presence of measurement errors in this
region. In addition, the bad glitch at somewhat lower frequencies arose because of
the shift during the measurements from one type of measurement apparatus to
another, evidently without adequate cross calibration in an overlap region. Here
only the low frequency error shows up in the conventional projection plot of M
vs. log( f ).
These results underline the desirability of constructing 3-D plots, preferably of all
four immittance functions, before any data analysis is carried out. With modem
computer-controlled plotting, such plots are readily produced. Of course if bad
points show up, measurements should be repeated. If that is impractical, outliers
should either be eliminated or weighted weakly, and possibly mild smoothing should
even be used [l].
(I I .b) Complex non-linear least squares data analysis
Even the most complete data presentations are only suggestive of the processes
occurring in the system investigated; a complete characterization requires that some
kind of a comparison be made between the data and a theoretical model and/or
reasonable equivalent circuit, as depicted in Fig. 1. In the past various graphical
and/or simple mathematical fitting procedures have been carried out, often involv-
ing subtractive calculations, which can be very inaccurate. Further, these approaches
generally do not analyze all the data simultaneously, and they usually yield
parameter estimates without any uncertainty measures. An approach which avoids
these difficulties and has great resolving power as well is that of complex nonlinear
least squares (CNLS) data fitting [lo]. Here all the real and imaginary data values
(or the modulus and phase values) are fitted simultaneously by weighted, nonlinear
least squares to a model or equivalent circuit, determining the best-fit estimates of
all the free parameters, as well as first-order estimates of their standard deviations.
The latter results are an essential part of the analysis since they indicate which
parameters are well determined and which, if any, should be eliminated.
An illustration of the accuracy and resolution of CNLS fitting is presented in
Fig. 6. Here the lumped-constant circuit shown was constructed using actual
elements whose values, listed at the top, were separately measured at a few
frequencies. The admittance of the full circuit was then measured with IS for many
frequencies between 0.4 and lo4 Hz. Three-D impedance and admittance plots are
32
12 nF
(12.07 f: 0.02)
I
2oG il
(2067 2.0)
-Im(ZI
-
Fig. 6. Three-dimensional perspective impedance and admittance plots of the response of the lumped
constant ladder network shown. CNLS fit values shown in parentheses [6].
presented in the figure and show little resolution of the two main time constants of
the system. CNLS led to a very good fit of the data, however, as well as the
parameter estimates and standard deviations shown in parentheses. These values
agree excellently with the individually measured ones and are, in fact, probably
more accurate [6,10].
Finally, Fig. 7 shows a 3-D plot and the results of a CNLS fit of data obtained
from IS measurements on the solid electrolyte / 3-PbF, with platinum electrodes.
Although the circuit is rather similar to that of Fig. 6, the response is of very
different character, primarily because of the needed presence in the circuit of the
impedance Z,, a constant phase element (CPE), a dce. Such elements will be
discussed in the next section. The small values of the relative standard deviations of
the parameters obtained from the fit show that it was a good one and that all the
parameters present were needed. The heavy dots in the 3-D plot are the original
data points while the fit predictions are shown by solid triangles. The projection
lines from these points down to the bottom plane begin to show slight discrepancies
(because of imperfect fitting) at the lowest frequencies, indicating that the circuit
cannot represent the data perfectly in this region. Note that seven parameters have
been well estimated here. Good estimates of even more parameters can be obtained
from CNLS fitting when data extend over a sufficiently wide frequency range.
33
p-WFP at 474 K
( 13.29 LO.241 nF
I ,
I ,
(41.6f33)nF -
I
I I
Z. = [A(iw)l- A= I2 196+_ 0.006).10-
n = (0.4025 t 0.00 16)
Fig. 7. Three-dimensional perspective impedance plot of p-PbF, data and CNLS fit response for the
circuit shown [6].
There are two further sources of possible ambiguity in CNLS fitting. First, a fit
may involve only a local minimum rather than the absolute least squares minimum.
This problem becomes worse the larger the number of parameters to be determined
in a non-linear least squares fit, but it can nearly always be circumvented by
carrying out the fitting several times with very different initial values for the
parameters. If the final parameter estimates are the same in all such fits, it is likely
that the true least squares solution has been obtained.
Second, it is not always obvious what kind of weighting to use. In the absence of
weights derived from replicating the experiment a number of times, unity weighting
(UWT) or proportional weighting (PWT - the uncertainties of the components are
taken proportional to their magnitudes at each point) is customarily used. Luckily,
the weighting employed has only a small effect on the parameter estimates when the
errors present in the fit are reasonably small [lo]. It has been suggested that when
the errors in the real and imaginary components are strongly correlated, modulus
weighting (MWT) should be used [ll]. In this case the weights for both components
are just ( II-*, where I is the immittance function being considered. Now if one fits
the data in their original form without transformation to another immittance form,
as one should, and if the errors are random, as one would hope, they should be
uncorrelated. But transformation from a given function to its complex inverse, as
from impedance to admittance, will induce some correlation in the errors of the
individual components at each frequency; then for comparison of fits with these two
forms, MWT may indeed be appropriate.
Computerized fitting has recently been criticized [12] because the deviations
between the data and the fitting results (residuals) may show systematic behavior.
Indeed they may, but this is a strength, not a weakness, of such methods as CNLS
fitting! Such results provide very important information, namely that the model or
equivalent circuit used in the fitting is not a perfect representation of the data.
When the deviations are large enough, they are a stimulus to try to discover and
34
eliminate systematic errors in the data themselves and/or to try more appropriate
models or equivalent circuits. Although residuals are not usually calculated and
examined in the older approximate (non-computerized or non-CNLS) analysis
methods, as they are as a matter of course in the CNLS approach, they are always
likely to be larger than with CNLS and thus to indicate systematic behavior more
frequently, sometimes when it is only an artifact of the inadequate analysis
procedures used.
(III) DISTRIBUTED CIRCUIT ELEMENTS, MODELS. AND CIRCUITS
No matter how accurate and extensive ones data are, it is fair to say that the
unexamined data set is not worth crowing about! Ony when one has derived
maximum enlightenment from it about the system involved has it served its purpose.
To do so nearly always requires some comparison between the data and theoretical
expectations, as discussed in the last section. A detailed model should always yield
an expression for impedance vs. frequency, but it may or may not allow a useful
equivalent circuit to be constructed. Whether or not such a circuit is available, the
model impedance can be fit to the data using CNLS. On the other hand, when no
model is available an often heuristic equivalent circuit must be used. Since real
systems are distributed in space and their properties are frequently distributed as
well [13], one usually needs to include one or more dces in the fitting circuit. These
are elements which cannot be represented by a finite number of ordinary ideal
circuit components but subsume the response of a distributed process, say diffusion,
into a single expression. I shall discuss some of them briefly in the next section, and
then move on to consider some models and equivalent circuits which have been used
to represent and analyze the s-s ac response of supported and unsupported solid and
liquid electrolyte systems. Some pertinent reviews appear in refs. 12 and 14-19.
(I I I .a) Some distributed circuit elements
The diffuse layer differential capacitance is itself a dce since it represents space
charge response over a finite region of space. Although it shows some frequency
dependence at very high frequencies (see the next section), such dependence may be
neglected in the usual experimental frequency range. Many different expressions for
its capacitance have been derived, depending on the exact physical conditions
considered. Here I shall give only an approximate but usually adequate expression
applying for the common situation of two identical, plane, parallel blocking
electrodes separated by a distance 1 which contains many Debye lengths, LD. Thus
we shall actually be concerned, in the present case of the low-frequency-limiting
differential capacitance, with two diffuse layer capacitances in series, one associated
with each interface, and with Cs in parallel with the combination. As the general
circuit of Fig. 2b shows, however, only in the low frequency limit can we take Cs in
parallel with this combination alone. Since I shall be usually dealing with two
35
identical electrodes in the following, when I am, I shall define C,, as the series
combination of the two diffuse layer capacitances without the effect of Cs included,
and take Cn as the series combination of the two inner layer capacitances. Then the
(combined) double layer capacitance, CDL, satisfies CD: = C,; + C, , and, for
sufficiently low frequencies, CT = Co, + Cs. For solid electrolytes, Cn for a single
interface will generally be large since it is then approximately the capacitance of two
parallel plates separated by an ionic radius. It may often be set infinite to good
approximation in such cases.
Although the use of two identical electrodes, a common practice for solid
materials, may seem a limitation, especially for liquid electrolyte situations where
half-cells are often employed, this is not the case when no static potential difference
appears across the cell. Because of symmetry, two-electrode s-s ac results may be
considered to be equivalent to the results which would be obtained for two identical
half-cells in series, with each half-cell involving the boundary conditions of the full
cell at one end and that equivalent to an ohmic electrode, undisturbed bulk
concentrations, at the other. Thus full-cell solutions include both cases.
Let us ignore for the moment the effect of any Cu by taking it infinite; then C,,
and Co, will be the same. Consider a situation where the continuum (igm) model is
appropriate, and a static potential difference of #, is applied across the electrodes;
thus #a/2 occurs across each diffuse layer. Then one finds for the quasi-static
differential capacitance [20,21]
CT = C,[(W ctnWf)l ~sh(W4W (1)
Here M, the number of Debye lengths in a half cell, is defined as 1/2Lo. This
expression for CT is intensive, as it should be for an interface-related quantity, when
A4 >> 1 and Cs plays a negligible role. When 1 $JV, 1 -K 1, one obtains the usual
small-signal ac expression for C, (not distinguished from CDL by many authors)
when M X- 1, and one finds the extensive result C, = Cs when M * 1.
I shall now turn to the consideration of frequency response which may be
associated with a distribution of relaxation times or activation energies and which
applies for a single, possibly wide, dispersion region. A very important dce, whose
response, or response very close to it, appears over a limited frequency range in
nearly every distributed situation and in most other dces, is the constant phase
element (CPE). Its admittance is of the form [l-3,13,22-25]
Y=A,(iw)
(2)
where 0 d n < 1. This element is not physically realizable at the extremes of
frequency and so should be used in conjunction with other limiting elements or in
truncated form. A degenerate form of the CPE, when n = 0.5, is infinite-length
Warburg response associated with uniform diffusion in a right half space [26]. It has
been widely used in electrochemical IS studies but suffers from the same lack of
physical realizability as the CPE. Its impedance is usually designated as Zw, and it
is an intensive quantity.
36
For a single dispersion region whose low frequency limiting resistance is R, and
whose high frequency limiting value is R,, it is convenient to introduce the
normalized impedance function [2]
rz=(Z-R,)/(R,-L) (3)
whose limiting real values are 1 and 0. Another dce associated with uniform
diffusion, but more physically plausible than the CPE or Z,, is the finite-length
Warburg, impedance Z,, where
Z zo = tanh(fi)/fi
(4)
and s = wrM is a normalized frequency. The time constant rM involves mobilities or
diffusion constants and the electrode separation, 1. Since all real systems involve a
finite separation of electrodes, Z,, defined by eqns. (3) and (4), should always be
used in place of Zw, although formally it reduces to Z, when R, =0 and s > 2.
In this limit, the response is intensive, as appropriate for a region near an electrode.
Note that in the zero frequency limit Z, reduces to the impedance of a capacitor
and resistor in parallel, even when R, =0. Here the response is extensive in
character because diffusion effects then extend over the entire region between the
electrodes.
The response form of eqn. (4), which is also the normalized input impedance of a
finite-length shorted transmission line [27], first appeared in the present context in
1953 [28] for the unsupported situation and in 1958 [29] for the supported one.
Comparisons of the two approaches [30] show that although the frequency response
is of the same form, the coefficients and time constants involved are generally
different, although the coefficients may be the same in one simplifying case.
Franceschetti and Macdonald [27] have considered many more complicated diffu-
sion effects in supported and unsupported small-signal response. Besides the above
particular finite-length response, associated with a fast reaction at the electrode,
another such limiting response appears when the diffusing entity is blocked (and
possibly adsorbed) at the electrode (the analog of a finite-length transmission line
with a infinite terminating resistance). Let its impedance be designated Zoo. Then
[27,311
Z zoo = ctnh(fi)/fi
(5)
which becomes purely capacitative in the low frequency limit but shows Z,
behavior again for s > 2. Responses of the above types can appear when the
diffusing entity is either charged or neutral. Both types of response are seen fairly
frequently in experimental results. It has even been suggested [31,32] that Z,,
response may arise in electrochromic thin films where the diffusing metallic ion is
supported by electronic conduction.
Another important dce which has been widely employed in equivalent circuits for
both dielectric and conductive systems leads to a complex plane curve which is a
semicircle with its center depressed below the real axis, a common shape when a
distribution is present [1,2,13,23,25,33]. Its I, representation is
I, = [1+ (is)] -I
(6)
31
where s is again a normalized frequency variable. This result, which has been
termed the ZARC or ZC dce, may either be considered as a response function in its
own right or as the parallel combination of a resistance and a CPE. There are
several other interesting combinations of a CPE and an ideal circuit element [l], but
the ZC is the most common. Unfortunately, it is not physically realistic at both
frequency extremes since it does not meet the criterion that its response reduce to
that of a system with a single relaxation time at very low frequencies and to another
single relaxation time in the limit of high frequencies [2,3,25]. In many practical
cases measurements may not extend to the regions where these deficiencies become
apparent, however.
Many other dces have been proposed over the years. Some of them are discussed
in refs. 1, 2, 13 and 34. Here, however, it will suffice to mention three of them of
current interest: the Williams-Watts (WW), Exponential Distribution of Activation
Energies (EDAE), and Gaussian Distribution of Activation Energies (GDAE) dces.
WW response has been found in many theoretical and experimental studies (many
references are listed in the present ref. 3), although most comparisons with experi-
ment have been inadequate, in part because the integral definitions of the WW Z
and Z functions are very difficult to evaluate accurately. Recently, however, an
accurate approximation for WW response has been developed [35] and incorporated
as an elective part of a powerful CNLS fitting program (available from the present
author).
Although the GDAE and EDAE responses are also defined as integrals, they are
readily evaluated numerically and are also included in the CNLS computer pro-
Fig. 8. Three-dimensional perspective impedance plot showing a comparison of accurate WW response
(solid lines) with response obtained [3] by fitting the EDAE model to the data by CNLS (dashed lines).
38
gram. The general EDAE dce has been found to be able to fit very well the
responses of nearly all the other dces which have been proposed [2,34,35]. For
example, Fig. 8 shows the result of a CNLS fit of the EDAE model to accurate
calculated WW response. When the system being investigated is thermally activated,
as it often is, and shows evidence of distributed character (wider dispersions than
arise from single-time-constant Debye response), the EDAE is probably the dce of
choice, both because it is fully physically realizable and because it leads, unlike
other models, to temperature dependence predictions in good agreement with
experiment for the fractional frequency and time power-law exponents often found.
(I I I . b) Circuits and models
(I I I . b. I ) General discussion
The presence of a supporting, or indifferent, electrolyte in supported situations
decouples the charged ionic species of interest from the rest of the charges in the
system, thus making its electrical effects very much easier to calculate in an
approximate but usually adequately accurate way. The situation is quite different
for an unsupported solid or liquid where Poissons equation couples charges of both
signs together strongly. It is thus generally much more difficult to solve electrical
response problems, particularly under large signal (non-linear) conditions, in unsup-
ported than in supported situations. Here I will primarily consider s-s solutions for
unsupported conditions, finishing with some numerical results for the highly nonlin-
ear situation where a static bias, p.d., $J,, is present as well as a small sinusoidal p.d.
When an equivalent circuit involves only ideal elements, it is found that some
circuits with the same number of elements but with different interconnections may
yield exactly the same impedance for all frequencies [3,14,18]. Three such circuits
are shown in Fig. 9. The first is a series type, the second essentially parallel, and the
third a hierarchical connection. The circuit elements are named differently in each
circuit because they must have different values in order that the impedances of all
three circuits be the same. This ambiguity may sometimes make it difficult to find
the most appropriate circuit for a given situation, but it can usually be eliminated by
considering the dependences of the circuit elements on one or more other experi-
mental variables besides frequency, as mentioned in Section II. Further, it turns out
that when the circuit being investigated requires the presence of one or more dces
which involve CPE-like behavior the ambiguity discussed here disappears [33]. For
all the circuits presented in this section, the resistances and capacitances shown are
taken to be frequency independent unless otherwise noted.
In the following circuits, I shall follow prior usage in defining R,, as the solution
resistance for liquid materials, but I shall use R, as a more general definition of the
high frequency limiting bulk resistance of either a liquid or a solid material. In
many electrolyte equivalent circuits the quantity Co, (or C,,) appears defined as
just the double-layer capacitance without distinction being made between the three
concepts here denoted by Cdl, CDL, and Cr. When an expression is given for it, it is
39
VOIGT
MAXWELL
;3* LADDER
Fig. 9. Three circuits which can have exactly the same impedance-frequency relation [18]. 0 1981 IEEE.
usually that of the CT of eqn. (1) in the s-s A4 X=1 situation: c,/47rL, for a single
electrode or er,/ 8rrL, for two identical electrodes. Here l r, is the dielectric constant
of the bulk material. Since Co, turns out to be closely associated with charge
transfer reactions at an electrode involving a reaction resistance R,,, or R,, I shall
often denote Cm as C,.
Figure lOa-e shows some representative equivalent circuits which have been
proposed over the years as appropriate representations of the response of supported
(liquid) electrolytes. The circuit of Fig. 10f is for a membrane with only charge of a
single sign present in the membrane, somewhat similar to a supported situation.
Other circuits are discussed in refs. 12, 14,42, and 43. Note that only the last of the
Fig. 10 circuits includes Cs and a few do not include R,. Although some of these
circuits have been used for data analysis, it is unfortunate that rarely have several
different circuits been used to analyze the same data in order to try to discover
which one is the more appropriate, and hardly any supported-situation data have
been analyzed with CNLS. Such fitting and comparisons are still much needed,
especially since the presence of dces in circuits of this kind eliminates most of the
possibility of ambiguity discussed above. The important circuit of Fig. 10a is known
as the Randles circuit (although it had already been discussed by Ershler). Randles
[36] calculated expressions for R,, and Z, appropriate for the simple charge
transfer reaction
Ox+n e-@Red
(7)
where n is here the total number of electrons transferred in the reaction, and Ox
and Red are oxidized and reduced species.
40
, ,COL
I I
-
cDL
zw
-lH
zw
(a) (d)
( ,COL
I I
RA CA
-
II 0
zw
(b)
I lCDL
Fig. 10. Some circuits proposed for the impedance of a supported electrode-material situation; (a)
Randles [36], (b) Laitinen and Randles [37], (c) Llopis et al. [38], (d) Barker [39], (e) Timmer et al. [40],
(f) de Levie and Vukadin [41]; a membrane situation; see text. Here R, and C, are adsorption resistance
and capacitance; AC is the capacitance difference between low and high frequency; Rww and Cww are
Warburg-l&e elements for diffusion-controlled adsorption; R, is a membrane (bulk) resistance; and R p,
is a phase transfer resistance.
An important difference between supported and unsupported conditions is
associated with the mobilities of the charges. Most supported situations are present
in liquid electrolytes, where both positive and negative species are mobile, usually
without a tremendous difference in mobilities. Unsupported situations occur, how-
ever, in fused salts and in solids with ionic or electronic conduction. Although
charges of both signs may be mobile in solids and have comparable mobilities, it is
common to encounter situations where the difference in mobility is so large that the
slower charges may be taken completely immobile over the time scale of the
experiment.
(I I I .b.2) Unsupported conditions: models and theoretical results
The earliest correct treatment of the s-s frequency response of an unsupported
situation appeared in 1953 [28]. It involved uni-univalent charges of arbitrary
mobilities, complete blocking conditions at the two identical electrodes, and no
applied static p.d. This theory and most of those discussed below apply to
semiconductors as well as to solid or liquid electrolytes, but I shall emphasize the
latter materials here.
41
No full theory of s-s frequency response for unsupported conditions with partial
discharge at the electrodes appeared for some time after the above work. Such
discharge occurs when a charge transfer reaction, such as that of eqn. (7), is present.
Although I obtained, over a period of some years, many new theoretical results
[4,15,30,44-521 incorporating the simple discharge boundary conditions of Chang
and Jaffe [53], these results were made more complicated and harder to use by their
analysis in terms of the equivalent circuit of Fig. 2a, and later by the use of the first
circuit shown in Fig. 9. The Fig. 2a circuit isolates the zero-frequency limiting
resistance of the system, R,, and R, is given by the parallel combination of R,
and R,. By contrast, the Fig. 2b circuit separates out the high frequency elements,
R, and Ca, of the total response and turns out to lead to much simpler analysis and
fitting [4,54].
The Chang-Jaffe (C-J) boundary conditions, which do not take an inner layer
into account but do involve a pure concentration overpotential, were later gener-
alized by LArryi [55] and the present author [15,49,50,54] by taking the discharge
parameters for positive and negative charges complex and frequency dependent in
such a way that the possible presence of sequential specific adsorption, as well as an
electrode reaction, could be simply included. Some resulting complex plane curve
shapes are shown in Fig. 11. Note that each semicircle involves a single time
constant, such as ~a = RRCR for the reaction arc. The negative loops in Fig. lla
imply the presence of inductive or negative differential resistance and capacitance
response. It has been shown [54] that the representation of the adsorption response
in terms of negative resistance and capacitance is preferable to the introduction of a
non-physical inductance. The rather exotic sorts of behavior shows here have
actually been observed when adsorption is present.
The most complete theory using the generalized C-J boundary conditions
appeared in 1978 [4]. It leads to a complicated expression for the Z, of Fig. 2;
Fig. 11. Complex plane response curves for adsorption/reaction (A/R) situations [15,54]. Arrows denote
direction of increasing frequency. Here R identifies a reaction arc, B a bulk arc, and D a diffusion arc.
The r,,s are rate constant parameters. The curves in (a) apply for n,,, S- 1, and those in (b) for q,, = 1 and
-
w-=r,,<-2.
42
involves arbitrary mobilities, valence numbers, and discharge parameters; and treats
both intrinsic and extrinsic conduction possibilities, with inclusion of dynamic
dissociation and recombination effects. It thus includes the five important processes:
charge separation near an interface, adsorption-desorption, charge transfer at an
electrode, mass transport (diffusion effects), and intrinsic-extrinsic generation-re-
combination. Its results are too complicated to yield a useful equivalent circuit in
the most general case, but such circuits may be found for some less general
situations. Of course when an equivalent circuit is derived from the exact solution,
expressions for the circuit elements in terms of microscopic parameters of the model
follow immediately. In the present overview I shall not define in this way all the
elements appearing in the equivalent circuits discussed since the detailed relations
are available elsewhere.
It is possible to obtain a simple equivalent circuit for the completely blocking,
intrinsic, equal valence numbers, equal mobilities case, with or without recombina-
tion. One finds [4] the exact result that Z, = Z,, is made up of a capacitance Csfi in
series with a resistance R&, where t, = [(M\I;L) ctnh(M&) - l] and J, = 1 +
iw~o. These elements are thus frequency dependent in the o = rD1 region where
bulk effects dominate. In the usual lower frequency region where w < ~6~ and
interface effects dominate, however, they are essentially frequency independent;
t, a r - 1; and r =(M) ctnh( M), a quantity usually much greater than unity. Then
the series resistance can be neglected compared to R, and the capacitance becomes
just the usual C,, = Cg(r - 1). In the limit of low frequencies, C, = Ca + C,, = rCg,
in full agreement with the result of eqn. (1) when 4, = 0.
Exact results are considerably more complicated when the mobilities are unequal;
then diffusion effects usually appear even in the completely blocking case [4,28,56].
The limiting low frequency finite-length-diffusion capacitance following from Z,,
sometimes called a pseudo-capacitance, is proportional to I and may be very much
larger than C,, [4,30,45-471. When I/J, is non-zero, exact analytic solution of the
coupled set of non-linear differential equations which determine transient response
or s-s frequency response is impossible, but Franceschetti and Macdonald [31,57,58]
have solved the equations numerically for many different cases of interest. The
static potential difference, +,, may include an applied component and/or intrinsic
Frenkel layer contributions.
The frequency response of material in finite-length half cells with one completely
blocking electrode or in full cells with two such electrodes, has been calculated [31]
for several values of the mobility ratio, or, = ~,/~P, and many values of 4: = $JVr.
Some representative complex plane impedance and admittance results for a half cell
with n;, = 5 are presented in Fig. 12. The equivalent circuit of Fig. 13a was found,
by CNLS fitting, to represent the data well with 4, zero, positive, or negative.
Compare the supported-case circuit of Fig. 10d. For J/, = 0 and or, = 1, it turns out
that Z, = 0, R, =R,, and C, = Cdl, as given above. We have used the designations
R, and C, here in place of R, and C,, to emphasize the dependences on 4, found
for these quantities and for the parameters of Z,. These dependences are in
agreement with expectations for charge accumulation and depletion layers; that for
43
2.0
z
; 1.0
I+
05
I
O.Oo.0
I I I
0.5 1.0 1.5
Re (2,)
(a)
0.5 IO
Re CY,,)
(b)
Fig. 12. Impedance (a) and admittance (b) plane plots for a half cell with a blocking electrode, A4 = 10,
and q, = 5, for several values of the normalized applied static potential difference [21].
R, is very small; and that for C, agrees very closely with
$,-dependent analytic expressions [31], such as that of
considered.
appropriate quasi-static
eqn. (l), for the cases
For the 4, = 0 full-cell situation where positive charges are taken completely
blocked at the electrodes and negative ones may react at the electrode with an
arbitrary rate constant, k,, Macdonald and Hull [59] have used CNLS fitting of an
Fig. 13. (a) Equivalent circuit appropriate for one or two blocking electrodes, no specific adsorption,
arbitrary rr,,,, and either without or with an applied static bias p.d. (b) Equivalent circuit appropriate in
the small-signal, arbitrary mobilities situation (591. (c) Equivalent circuit presented by Archer and
Armstrong [17] for one-mobile blocking conditions.
44
appropriate circuit to the exact theoretical response [4] in order to investigate how
the circuit elements depend on rr,,, and on recombination effects. The best fitting
circuit found is shown in Fig. 13b. Notice that it would be identified as a Randles
circuit (with a finite-length rather than infinite-length diffusion impedance) if
supported behavior were being considered. Attention was concentrated on the
wr,, CK 1 frequency region, and for rrm = 1 exactly the above result for C,, was
found from the fitting. But C,, increased rapidly for rr,,, > 1 and quickly dropped for
V~ < 1 to about 0.7 times its rrm = 1 value, or, more precisely, to C,(r, - 1) for
rrm -=XK 1. Here r, = (Ml) ctnh(M,) and Mi = Z/L,,. The latter result may be readily
understood. When V~ < 0.1, the effective Debye length is no longer that appropriate
when charges of both sign are mobile, L,, but is well approximated by the
one-mobile value, L ,,, = &&.
The resistance R, also showed interesting and important behavior. Although it is
usually taken to be the reaction resistance, proportional to k;, it was found to be
non-zero and dependent on M- * even when k, was taken infinite. Such pseudo-re-
action rate response, associated with the drag of charges of one sign on those of the
other sign and not with a finite reaction rate at all, can lead to entirely incorrect
estimates of reaction rate values when it is unrecognized; and even when its
presence is accounted for, it sets an upper limit on the maximum reaction rate value
which can be reliably estimated when using CNLS to fit data as accurately as
possible to the present circuit.
For the rest of the discussion I shall be concerned only with the one-mobile
situation (a single species of mobile charge present) appropriate for many solids.
Charge of one sign is taken immobile, and is uniformly distributed in the absence of
recombination, while that of the other sign is assumed to be mobile and may react
or be blocked at the electrodes. Archer and Armstrong [17] have discussed the
equivalent circuit of Fig. 13c for a blocked, one-mobile situation with specific
adsorption. Since it has no dc path, it allows no Faradaic current and thus no
charge transfer reaction occurs. Although the exact s-s solution [4] yields a relatively
simple expression for Z, for the general one-mobile case, it still does not lead to a
relatively simple equivalent circuit when recombination is possible and dissociation
is incomplete. In the full dissociation limit, however, it does yield a simple circuit,
that of Fig. 14 when all the Z,s are taken zero and C, = Cur = C,,. The resulting
hierarchical structure is then equivalent to the ladder network of Fig. 9 and is also
equivalent in form to the supported-case circuit of Fig. 1Oc when one Z, in Fig. 14
is taken non-zero and is approximated by Zw, and Cs is ignored.
Actually, the exact solution shows that the C. in Fig. 14 should be replaced by
the impedance Zso, identified above for the completely blocking situation, but with
M and r replaced by M1 and rl since L,, rather than L, is the appropriate bulk
Debye length in the present one-mobile case. Thus, the Fig. 14 circuit, with the
restrictions above, is accurate Only in the wro -=K 1 frequency region. In this region
it can, however, lead to all the kinds of complex-plane curve shapes shown in Fig.
lla. Note that when R, = co, only adsorption is present, but the structure is
different from that of the Fig. 13c circuit. When R, and Zo, are both zero, one has
45
Fig. 14. General equivalent circuit. For the one-mobile situation without diffusion of reaction products,
all Z,s are zero [4,18]. Q 1981 IEEE.
the situation of heterogeneous reaction without adsorption. As shown above, the
basic one-mobile situation involves no diffusion elements, but if a neutral reaction
product diffuses in both the electrode and in the solid, at least two non-zero Zos
may need to be included in the Fig. 14 circuit [18,27,60,61], but their effects will
usually show up only at very low frequencies.
One might ask how the quasi-static circuit of Fig. 9 of the preceding paper [62]
for diffuse and inner layer capacitances could be extended to be consistent with the
reaction/ adsorption parts of Fig. 14. A tentative suggestion is as follows. When a
reaction is present, Cs will be paralleled by a low resistance and their combined
effects could probably be ignored in the measurable frequency range. Then if one
identifies C, as Cn and Coo as Cdl, their series combination is just C,. One needs
only to add R, in series with C,, and R, in parallel with it, to obtain the pertinent
part of the Fig. 14 circuit.
We have given little attention to the effect of C, so far since its presence is
ignored when C-J boundary conditions are employed. Luckily, a transformation of
variables method has been developed [54] which allows known exact s-s solutions
using even generalized C-J reaction/ adsorption parameters to be transformed to
solutions taking proper account of C, and involving even more general overpoten-
tial-dependent, first-order electrode reaction kinetics than conventional Butler-
Volmer (B-V) kinetics. When this method is applied for the present one-mobile case
without adsorption, it leads [63,64] to the exact equivalent circuit of Fig. 15a. This
circuit shows that it is possible to separate out all Cn effects into a separate series
circuit which reduces to just C, when the mobile carrier is completely blocked.
Here Cs is the geometric capacitance of the system excluding the two inner layer
regions and R,, is the bulk resistance also excluding these regions. The exact
solution for the total impedance Z, is relatively complicated and although it could
be used directly for CNLS fitting, it is useful to derive simpler approximate results
[63,64].
46
C
, , 90
CH
I I I I
P
-I-
-
ZS
I
Z
2,
SC
, ,coa , ,cH
I I
RL
I I
Lc
P -
(b)
Fig. 15. Equivalent circuits which take the inner layer into account. Applicable for the one-mobile
situation with general reaction kinetics [64]. (a) Exact small-signal circuit; (b) first approximate circuit;
(c) second approximate circuit, appropriate with or without a Faradaic current flowing.
Such results are embodied in the equivalent circuits of Fig. 15b and 15~. That of
Fig. 15b is quite accurate even up to or,, = 1, while that of Fig. 15c is a good
approximation up to wrn = 0.1 or so. It is surprising that the complicated circuit of
Fig. 15b can be well approximated by a circuit of the same form as its left half, the
ordinary C, = 0 solution, but this is indeed the case. The exact s-s solution shows
that to a good approximation C, = Cs, and RI = R,. Further, when the kinetics
used are simplified to B-V form, it turns out [54,64] that R, is exactly R,, entirely
unchanged by the presence of Cn. The identity of the present unsupported-case R,
with the conventional R,, reaction resistance used in supported situations and
derived for B-V kinetics was pointed out long ago [47]. In addition, it has been
shown [54] that the unsupported and supported expressions for the adsorption
capacitance C, are also identical.
But C, is not generally equal to Co,, even for B-V kinetics. The results do show,
however, under what specific conditions the conventional approximation is ap-
propriate. For the general kinetics, a complicated expression for C, is obtained [64]
which involves most of the parameters of the Fig. 15b circuit. The result is much
simplified for B-V kinetics but still involves R,. It turns out, nevertheless, that for
ordinary conditions in cells with large M, it is a good approximation to set
C,=C,,, where C$=C,;+C, a s usual. Then the Fig. 15c circuit is just that
long used for both supported and unsupported conditions. For thin membranes
with small M most of these results definitely do not apply, however, and the exact
results should be used for fitting [64].
Finally, Franceschetti and I have obtained numerical solutions for the present
situation with static bias applied for both full cells and half cells [58]. Typical
complex plane results are shown in Fig. 16 for C-J and for B-V kinetics. Although
the curves look very similar for the two cases, notice the quite different biasing
currents listed. The Fig. 15c circuit was found to be quite adequate to represent the
47
1.5
I I I I I
(0.01. 001. 100.0.1:,
8.0
fioi
05 IO I.5 20 25 30
Re (2.)
(0)
I5
I I I I I
~0.01.0.01.100.0. 1:r
8.1. a.o.5
c++$q
IO I5 20 25 30
Re &I
(b)
Fig. 16. Impedance plane plots for M = 100 symmetrical cells with several biasing currents [58]. (a)
Chang-Jaffb electrode kinetics; (b) Butler-Vohner electrode kinetics. These two figures were originally
presented at the Spring 1979 Meeting of the Electrochemical Society, Inc. held in Boston, MA.
results, but except for C, the parameter values agreed only with those discussed
above under zero-bias conditions. Although R, was found to have only small bias
dependence, R, and C, varied appreciably and systematically with bias but showed
considerably less variation for B-V than for C-J kinetics.
For both supported and unsupported situations, IS s-s measurements often do
not agree with ideal theoretical results. For example, the RR--CR reaction arc
appearing in the complex impedance plane is not always found to be a perfect
semicircle with its center on the real axis but often is depressed, with its center
below the real axis. Although the exact s-s theory yields some such depression when
q,, is appreciably different from unity [47], the possible amount of depression is
insufficient to explain most results. Further, diffusion arcs often have high-
frequency power-law exponents different from the theoretical n = 0.5 value. Al-
though it appears that the general hierarchical circuit of Fig. 14 (with one or more
2,s set to zero) is an appropriate starting point for fitting either supported or
unsupported data, it must clearly be modified for use with data showing non-ideal
behavior. One approach which often helps is to replace one or more of the ideal
circuit elements or 2,s by more general dces such as ZCs, WWS, or EDAEs.
48
Although considerable s-s frequency response data, both supported and unsup-
ported, have been analyzed, the analysis employed has often fallen short of the state
of the art. A representative list of unsupported solid materials whose data have been
analyzed might include Na /&hunina (single crystal) [7,8]; (KBr),,,(KCN),, (single
crystal) [34]; polyphenylene-oxide (polymer film) [56]; fi-PbF, (single crystal) [61];
lithium nitride (single crystal) [65]; and zirconia-yttria (polycrystalline) [66]. The
data have not always been plotted in ways that show up dubious points; CNLS
fitting has not always been used; and most important, too few different models or
equivalent circuits have been fitted for a given set of data to allow a best choice to
be established with some confidence. Much yet remains to be done, both in
developing new theoretical models (e.g., see the recent work on response of
three-phase electrodes [67]), and in analyzing data sets in ways worthy of them.
ACRONYM DEFINITIONS
dce
igm
p.d.
;IsD
B-V
C-J
CNLS
CPE
EDAE
EDL
GDAE
IS
MWT
PWT
UWT
ww
distributed circuit element
ideal-gas model
potential difference
small-signal
three dimensional
Butler-Volmer
Chang- JaffC
complex non-linear least squares
constant phase element
exponential distribution of activation energies
electrical double layer
Gaussian distribution of activation energies
impedance spectroscopy
modulus weighted least squares
proportional weighted least squares
unweighted least squares
Williams-Watts
ACKNOWLEDGEMENTS
First, I wish to thank the Army Research Office for their continuing support of
my work. Second, it is a pleasure to acknowledge the important contributions of the
many associates who have worked with me in the present area over the years, most
especially D.R. Franceschetti.
REFERENCES
1 J.R. Macdonald (Ed.), Impedance Spectroscopy, Wiley-Interscience, New York, to be published.
2 J.R. Macdonald, J. Appl. Phys., 58 (1985) 1955, 1971. In the second of these papers, the exp( - N,,E)
term in eqn. (17) should be replaced by exp( - q,,E) and the f sign in eqn. (24) replaced by an equals
sign.
49
3 J.R. Macdonald, Revs. Mod. Phys., submitted.
4 J.R. MacdonaId and DR. Franceschetti, J. Chem. Phys., 68 (1978) 1614.
5 J.R. MacdonaId, J. EIectroanaI. Chem., 40 (1972) 440.
6 J.R MacdonaId, J. Schoonman and A.P. Lehnen, Solid State Ionics, 5 (1981) 137.
7 J.R. MacdonaId and G.B. Cook, J. EIectroanal. Chem., 168 (1984) 335.
8 J.R. MacdonaId and G.B. Cook, J. Electroanal. Chem., 193 (1985) 57.
9 D.P. Almond and A.R. West, Solid State Ionics, 3/4 (1981) 73; P.G. Bruce, A.R. West and D.P.
Almond, Solid State Ion&, 7 (1982) 57.
10 J.R. MacdonaId, J. Schoonman and A.P. L&ten, J. EIectroanal. Chem., 131 (1982) 77.
11 P. Zoltowski, J. Electroanal Chem., 178 (1984) 11.
12 M. Sluyters-Rehbach and J.H. Sluyters in E. Yeager, JGM. Bockris, B.E. Conway and S. Saran-
gapani @is.), Comprehensive Treatise of Electrochemistry, Vol. 9, Plenum Press, New York, 1984,
pp. 177-292.
13 R.L. Hurt and J.R. MacdonaId, Solid State Ionics, 20 (1986) 111.
14 R. Parsons, Adv. Electrochem. Electrochem. Eng., 7 (1970) 177.
15 J.R. MacdonaId in G.D. Mahan and W.L. Roth (Eds.), Superionic Conductors, Plenum Press, New
York, 1976, pp. 81-97.
16 R.D. Armstrong, M.F. Bell and A.A. MetcaIfe, Electrochemistry, 6 (1978) 98.
17 W.I. Archer and R.D. Armstrong, Electrochemistry, 7 (1980) 157.
18 J.R. MacdonaId, IEEE Trans. Electr. Insul, 16 (1981) 65.
19 R.P. Buck, Ion-Sel. Electrode Rev., 4 (1982) 3.
20 J.R. MacdonaId, J. Chem. Phys., 22 (1954) 1317.
21 D.R. Franceschetti and J.R. MacdonaId, J. EIectroanal. Chem., 100 (1979) 583.
22 H. Fricke, Philos. Mag., 14 (1932) 310.
23 K.S. Cole and R.H. Cole, J. Chem. Phys., 9 (1941) 341.
24 J.R. Macdonald, Solid State Ionics, 13 (1984) 147.
25 J.R. Macdonald, Solid State Ionics, 15 (1985) 159.
26 E. Warburg, Weid. Ann. Phys. Chem., 67 (1899) 493.
27 D.R. Franceschetti and J.R. Macdonald, J. Electroanal Chem., 101 (1979) 307.
28 J.R. Macdonald, Phys. Rev., 92 (1953) 4.
29 J. Llopis and F. Colom, Proc. 8th Meeting C.I.T.C.E., 1956, Butterworths, London, 1958, pp.
414-427; see also P. Drossbach and J. Schulz, Electrochim Acta, 9 (1964) 1391.
30 J.R. MacdonaId, J. EIectroanaI. Chem., 32 (1971) 317; 47 (1973) 182; 53 (174) 1.
31 D.R. Franceschetti and J.R. Macdonald, J. EIectrochem. Sot., 129 (1982) 1754.
32 S.H. Glarum and J.H. Marshal& J. EIectrochem. Sot., 127 (1980) 1467.
33 J.R. Macdonald and R.L. Hurt, J. ElectroanaI. Chem., 200 (1986) 69.
34 J.R. MacdonaId, J. Appl. Phys., 61 (1987) 700.
35 J.R. Macdonald and R.L. Hurt, J. Chem. Phys., 84 (1986) 496.
36 J.E.B. Randles, Discuss. Faraday Sot., 1 (1947) 11.
37 H.A. Laitinen and J.E.B. Randles, Trans. Faraday Sot., 51 (1955) 54.
38 J. Llopis, J. Fern&id=-Biarge and M. Perez-Femandez, Electrochim. Acta, 1 (1959) 130.
39 G.C. Barker, J. ElectroanaI. Chem., 12 (1966) 495.
40 B. Timmer, M. Sluyters-Rehbach and J.H. Sluyters, J. Electroanal Chem., 18 (1968) 93.
41 R. de Levie and D. Vukadin, J. Electroanal. Chem., 62 (1975) 95.
42 D.C. Grahame, J. Electrochem. See., 99 (1952) 370C.
43 P. Delahay, Double Layer and Electrode Kinetics, Interscience, New York, 1965.
44 J.R. Macdonald, Trans. Faraday Sot., 66 (1970) 943.
45 J.R. MacdonaId, J. Chem. Phys., 54 (1971) 2026; errata 56 (1972) 681. In addition, the words
intrinsic and extrinsic are improperly used in this work in place of intensive and extensive.
46 J.R. MacdonaId, J. Chem. Phys., 58 (1973) 4982; errata 60 (1974) 343.
47 J.R. Macdonald, J. Chem. Phys., 61 (1974) 3977.
48 J.R. MacdonaId in M. Kleitr and J. Dupuy (Eds.), Electrode Processes in Solid State Ionics, Reidel,
Dordrecht, 1976, pp. 149-180.
50
49 J.R. Macdonald, J. Electroanal. Chem., 70 (1976) 17.
50 J.R. Macdonald and P.W.M. Jacobs, J. Phys. Chem. Solids, 37 (1976) 1117.
51 J.R. Macdonald, D.R. Franceschetti and R. Meaudre, J. Phys., Cl0 (1977) 1459.
52 J.R. Macdonald and J.A. Garber, J. Electrochem. Sot., 124 (1977) 1022.
53 H. Chang and G. Jaff6, J. Chem. Phys., 20 (1952) 354.
54 D.R. Franceschetti and J.R. Macdonald, J. Electroanal. Chem., 82 (1977) 271.
55 S. Lanyi, J. Phys. Chem. Solids, 36 (1975) 775.
56 S.H. Glarum and J.H. Marshall, J. Electrochem. Sot., 132 (1985) 2939.
57 D.R. Franceschetti and J.R. Macdonald, J. Appl. Phys., 50 (1979) 291.
58 D.R. Franceschetti and J.R. Macdonald in S. Bruckenstein, J.D.E. McIntyre, B. Miller and E. Yeager
(Eds.), Proc. Third Symp. on Electrode Processes, 1979, The Electrochemical Society, Proceedings,
Vol. 80-3, 1980, pp. 94-114.
59 J.R. Macdonald and C.A. Hull, J. Electroanal. Chem., 165 (1984) 9.
60 D.R. Franceschetti, Solid State Ionics, 5 (1981) 613.
61 J. Schoonman, L.J. Still, J.R. Macdonald and D.R. Franceschetti, Solid State Ionics, 3/4 (1981) 365;
D.R. Franceschetti, J. Schoonman and J.R. Macdonald, ibid., 5 (1981) 617.
62 J.R. Macdonald, J. Electroanal. Chem., 223 (1987) 1.
63 D.R. Franceschetti and J.R. Macdonald, J. Electroanal. Chem., 87 (1978) 419. The term (1 - iQ - vn)
in eqn. (8) of this paper should be replaced by (1 + is2 - v,,).
64 J.R. Macdonald and D.R. Franceschetti, J. Electroanal. Chem., 99 (1979) 283.
65 J.R. Macdonald, A. Hooper and A.P. Lehnen, Solid State Ionics, 6 (1982) 65.
66 J.E. Bauerle, J. Phys. Chem. Solids, 30 (1969) 2657.
67 D.R. Franceschetti, Solid State Ionics, 18/19 (1986) 101.