J. Electrochem. Soc., Vol. 142, No. 12, December 1995 9 The Electrochemical Society, Inc.

4149

13. D. S. Chan and C. C. Wan, J. Power Sources, 50, 261

REFERENCES (1994).
1. J. A. Appleby and E. B. Yeager, Energy, 11, 137 (1986). 14. T. Mort, J. Imahashi, T. Kamo, K. Tamura, and Y. Hish-
2. E. A. Ticianelli, C. R. Derouin, and S. Srinivasan, inuma, This Journal, 133, 896 (1986).
J. EIectroanal. Chem., 251, 275 (1988). 15. D. S. Cameraon, J. Mol. Catal., 38, 27 (1986).
3. E. A. Ticianelli, C. R. Derouin, A. Redondo, and 16. A. Pebler, This Journal, 133, 9 (1986).
S. Srinivasan, This Journal, 135, 2209 (1988). 17. K. Mitsuda, H. Shiota, H. Kimura, and T. Murahashi,
4. S. Srinivasan, E. A. Ticianelli, C. R. Derouin, and J. Mater. Sci., 26, 6436 (1991).
A. Redondo, J. Power Sources, 22, 359 (1988). 18. T. Kenjo and K. Kawatsu, Electrochim. Acta, 30, 229
5. M. S. Wilson and S. Gottesfeld, J. Appl. Electrochem., (1985).
22, 1 (1992). 19. T. Maoka, ibid., 33, 379 (1988).
6. M. S. Wilson and S. Gottesfeld, This Journal, 139, L28 20. K. Kordesh, S. Jahangir, and M. Schautz, ibid., 29, 1589
(1992). (1984).
7. M. S. Wilson, U. S. Pat. 5,211,984 (1993), 5,234,777 21. R. Holze and J. Ahn, J. Membrane Sci., 73, 87 (1992).
(1993). 22. T. Sakai, H. Takenaka, and E. Torikai, This Journal,
8. M. Uchida, Y. Aoyama, N. Eda, and A. Ohta, This Jour- 133, 88 (1986).
nal, 142, 463 (1995). 23. Y. Aoyama, M. Uchida, N. Eda, and A. Ohta, Abstract
9. M. Watanabe, M. Tomikawa, and S. Motoo, J. EIec- IC08, p. 62 of Annual Meeting of The Electrochemical
troanal. Chem., 195, 81 (1985). Society of Japan, Sendal, 1994.
10. M. Watanabe, K. Makita, H. Usami, and S. Motoo, ibid., 24. Y. Aoyama, M. Uchida, N. Eda, and A. Ohta, Abstract
197, 195 (1986). 2K03, p. 257 of Fall Meeting of The Electrochemical
11. M. Giordano, E. Passalacqua, V. Recupero, M. Vivaldi, Society of Japan, Tokyo, 1994.
E. J. Taylor, and G. Wilemski, Electrochim. Acta, 35, 25. G. Petrow and R. J. Allen, U.S. Pat. 4,044,193 (1977).
1411 (1990). 26. M. Watanabe, M. Uchida, and S. Motoo, J. Electroanal.
12. M. Giordano, E. Passalacqua, V. Alderucci, P. Staiti, Chem., 229, 395 (1987).
L. Pino, H. Mirzaiani, E. J. Taylor, and G. Wilemski, 27. M. Uchida, Y. Aoyama, M. Tanabe, N. Yanagihara,
ibid., 36, 1049 (1991). N. Eda, and A. Ohta, This Journal, 142, 2572 (1995).

Application of Measurement Models to

Impedance Spectroscopy
II. Determination of the Stochastic Contribution to the Error Structure
Pankaj Agarwal, *'a Oscar D. Crisalle, and Mark E. Orazem**
Department of Chemical Engineering, University of Florida, Gainesville, Florida 32611, USA

Luis H. Garcia-Rubio
Department of Chemical Engineering, University of South Florida, Tampa, Florida 33620, USA

ABSTRACT
Development of appropriate models for the interpretation of impedance spectra in terms of physical properties re-
quires, in addition to insight into the chemistry and physics of the system, an understanding of the measurement error
structure. The time-varying character of electrochemical systems has prevented experimental determination of the
stochastic contribution to the error structure. A method is presented by which the stochastic contribution to the error
structure can be determined, even for systems for which successive measurements are not replicate. Although impedance
measurements are known to be heteroskedastic in frequency (i.e., have standard deviations that are functions of frequency)
and time varying over the duration of the experiment, the analysis conducted in the impedance plane suggests that the
standard deviations for the real and imaginary parts of the impedance have the same magnitude, even at frequencies at
which the imaginary part of the impedance asymptotically approaches zero. On this basis, a general model for the error
structure was developed which shows good agreement for a broad variety of experimental measurements.

This paper is part of a series intended to present the foun-
dation for the application of measurement models to
impedance spectroscopy. The basic premise behind this
work is that determination of measurement characteristics
is an essential aspect of the interpretation of impedance
spectra in terms of physical parameters. The influence of
the error structure on interpretation of impedance spectra
is discussed elsewhere for various electrochemical and
electronic systems. 1-7In the previous paper of this series, it
was shown that a measurement model based on Voigt cir-
cuit elements can provide a statistically significant fit to
typical electrochemical impedance spectra. BHere a method
is demonstrated in which the measurement model is used to
* Electrochemical Society Student Member.
** Electrochemical Society Active Member.
Present address: Department of Materials, Swiss Federal In-
stitute of Technology (Lausanne), Lausanne, Switzerland.
identify the stochastic component of the frequency-de-
pendent error structure of impedance data. A preliminary
model for the stochastic component of the error is pro-
posed. The third paper of this series addresses the use of the
measurement model for identification of the bias compo-
nent of the error structure. ~
Background
The objective of impedance measurements is identifica-
tion of physical processes or parameter values appropriate
for a given system. Interpretation of impedance spectra re-
quires, in addition to an adequate deterministic model, a
thorough understanding of the error structure for the
measurement. The concept that the error structure plays an
important role in the interpretation of experimental data is
well established in the scientific literature. I~ While its ap-
4150 J. Electrochem. Soc., Vol. 142, No. 12, December 1995 9 The Electrochemical Society, Inc.
plication to the interpretation of optical spectra is under
development, knowledge of the error structure has been
shown to allow enhanced interpretation of light scattering
measurements in terms of particle size distribution or even
particle classification, n'~2
Identification of the error structure for most radiation-
based spectroscopic measurements such as light scattering
can be accomplished by calculating the standard deviation
of replicate measurements. The error analysis approach has
been successful for light spectroscopy measurements be-
cause these systems lend themselves to replication and,
therefore, to the independent identification of the different
errors that contribute to the total variance of the measure-
ments. In contrast, the stochastic contribution to the error
structure of electrochemical impedance specfroscopy mea-
surements generally cannot be obtained from the standard
deviation of repeated measurements because even a mild
nonstationary behavior introduces a nonnegligible time-
varying bias contribution to the error.
The discussion of error structures in impedance spec-
troscopy, therefore, has been limited to a priori predictions
of measurement noise based on instrument noise ~3 and to
standard assumptions concerning the error structure such
as constant or proportional errors. ~4-~6Zoltowski reports an
experimental assessment of the stochastic noise at selected
frequencies which showed that the standard deviation of
the real and imaginary parts of the impedance (and admit-
tance) are correlated. ~7 This result was used to defend the
use of modulus weighting for regression of models to exper-
imental data. ~749 Attempts to weight regressions by stan-
dard deviations determined from repeated impedance
scans (e.g., Ref. 20) have not been successful because the
standard deviation obtained from repeated impedance
measurements includes both bias and stochastic contri-
butions to the error structure. Our objective here is to
present a procedure for assessing the stochastic contribu-
tion to the error structure of impedance measurements and
to develop a model for the stochastic noise of impedance
measurements.
Knowledge of the error structure is essential in assessing
experimental technique and can play a role in determining
the influence of instrumental strategies. For example, use
of matched filters for the input of the frequency response
analyzer has been proposed to be an appropriate technique
for reducing the noise in impedance spectra. 2~ While the
reduction of the noise in input signal can be readily seen by
the display of an oscilloscope, a method for assessing the
noise in the ultimate measurement is essential in deciding
whether the filters contribute to reduction of noise in the
impedance spectra.
Knowledge of the error structure is also needed for anal-
ysis of experimental data. For example, proper weighting
during nonlinear regression of a model to impedance data
is necessary to get unbiased parameter estimates. ~~ A
weighted least squares strategy that includes the errors in
the real and imaginary parts of the impedance is given by
minimization of
(Zr, k - Zr, k) 2 (Zj,k -- Zj,k) 2
J= E ~ +~ -' 2 '
k O'r,k O'j,k
where Zr,k and Zi,k represent the real and imaginary part of
the data, respectively, the caret signifies the corresponding
model value, and cry,k and %,~ are the real and imaginary part
of the standard deviation at each frequency. The use of the
variance to weight data ensures that data points with "low
noise" content are emphasized and the data points with
"high noise" content are de-emphasized. Use of the experi-
mentally determined variance for weighting has been
shown to increase the amount and quality of the informa-
tion that can be obtained from impedance measurements3
Classification of Measurement Errors
The residual errors (e~J that arise when a model is re-
gressed to experimental data can be described as being
composed of deterministic (systematic) and stochastic
(randomly distributed) contributions, Esystand es~ooh,respec-
tively. Thus, at any given frequency
Z - Z = ere~= esy~t+ e~toch [2]
where the caret signifies the model value for the complex
impedance Z. The presence of stochastic errors estoohin any
experimental data is inevitable. In this work, the system-
atic errors that result from model inadequacies are distin-
guished from the experimental errors that are propagated
through the model and that could arise from a changing
base line or from instrumental artifacts. Systematic errors
are therefore defined to consist of contributions from the
lack of fit of the model to the data (elo~)and a bias (%~as) in
the experiment, i.e.
%s~ = elo~+ eblas [3]
In principle, improvement of a stationary model reduces
the error associated with a lack of fit, but nonstationary
behavior (%s) and instrumental artifacts ( e J can still con-
tribute to the bias errors, i.e.
eb~ = en~ + ei~ [4]
The nonstationary contribution to the bias usually is ob-
served most easily at frequencies that require the longest
time for measurement. Instrumental artifacts may be seen
at high frequency resulting from equipment limitations.
Many, if not most, electrochemical systems show at least a
mild nonstationary behavior due to changes in electrode
properties during the course of an experiment. In contrast,
solid-state systems, as a first approximation, may be as-
sumed to be stationary. Impedance data can be corrupted
by instrumental artifacts for both electrochemical and
solid-state systems.
The emphasis in regression of models to impedance data
is on reducing the error associated with a lack of fit, but it
is evident that the residual errors must contain contribu-
tions associated with phenomena that are independent of
the adequacy of the model, i.e.
Z - 2 = e~o~ + (ens + ei~) + e~o~h [5]
The object here is to quantify the stochastic errors in
impedance measurements. The quantification of bias er-
rors associated with experimental issues such as instru-
mental artifacts or nonstationary phenomena will be ad-
dressed in a subsequent paper. 9
Identification of Error Structure
The method for assessing the stochastic part of the error
structure is based on using a measurement model as a filter
for nonreplicacy of impedance data. The measurement
model is composed of a superposition of line shapes which
can be chosen arbitrarily. A model composed of Voigt ele-
ments in series with a solution resistance has been shown to
be a useful and general measurement model (see Fig. 1 in
Ref. 8). While the line shape parameters may not be associ-
ated unequivocally with a set of deterministic or theoreti-
cal parameters for a given system, the measurement-model
approach has been shown to represent adequately the
impedance spectra obtained for a large variety of electro-
[11 chemical systems.8 The characteristic time constants for
the line shapes and their dispersion are low frequency as
compared to the noise. The line shape models therefore can
be said to represent the low-frequency stationary compo-
nents of the impedance spectra (in a Fourier sense). Re-
gard]ess of their interpretation, the measurement model
representation can be used to filter and thus identify the
nonstationary (drift) and high frequency (noise) compo-
nents contained in the same impedance spectrum.
It is not obvious that such an approach should work. It is
well known, for example, that the impedance spectrum as-
sociated with an electrochemical reaction limited by the
rate of diffusion through a stagnant layer (the Warburg
impedance) can be approximated by an infinite number of
resistance-capacitance (RC) circuits in series (the Voigt
model). In theory, then, a measurement model based on the
Voigt circuit should require an infinite number of parame-
ters to describe adequately the impedance response of any
J. Electrochem. Sac., Vol. 142, No. 12, December 1995 9 The Electrochemical Society, Inc. 4151

electrochemical system influenced by mass transfer. In 9 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18

ilIOIIOOOOOOIIIOIIIOIIIOOIO|(IIO|IfI!
practice, stochastic errors (or noise) in the measurement 8 .~888~8oo i% ....
IoO o~f~gOno 9 o ~zu~ 16
limit the number of Voigt parameters that can be obtained : o o0~ 9 9400K
from experimental data. An infinite number of Voigt .1~ o
parameters cannot be obtained even from synthetic data
0 6 ~ 9 12 8
because round-off errors limit the information content of
the calculation. ~ The residual errors associated with fitting 5 lO~
a Voigt model to experimental impedance data which are
influenced by mass transfer, with appropriate weighting,
" ;" ~
can be made to be of the order of the stochastic noise in the
measurement. 9 A Voigt circuit therefore can yield an 6 9 9 6 co
appropriate measurement model for electrochemical 2 ,a 8 A 9 I 9 4 "~
impedance spectra.
It is evident that the measurement model composed of
="
E 1 .9 |~ ~s ",, ,9 ". 2 E
=
~" o ~9 .
Voigt elements may not be the most parsimonious or effi-
cient model for a given spectrum. However, it does have the
advantage that the maximum number of elements and,
therefore, parameters to be estimated, is limited by the 1 10 100 1000 10000 100000
noise frequency. In effect, in using the measurement model, Frequency, Hz
one takes advantage of the noise present in any measure-
ment, which limits the number of parameters that can be Fig. 1. Real and imaginary parts of the impedance of single-crystal
resolved, and the large number of measured frequencies as n-GaAs/Ti Schottky diode held at 320 K (open symbols) and 400 K
compared to the number of resolvable parameters. (closed symbols). Five replicate measurements were made at each
temperature. The circles represent the real part of the impedance and
Stochastic Contributions to the Error Structure the triangles represent the imaginary part of the impedance.
The approach taken to assess the error structure of
impedance data is to identify the measurement character- N
istics experimentally rather than to assess the noise level ,0.
r^2 : ~ (Eres,~ __ ires,r)2 [Sa]

from the published specifications of the component in- k=l

strumentation. Pseudo-replicate impedance scans are ob-
tained, recognizing that even stable electrochemical ~ =~ [8b]
N 1
systems are usually nonstationary such that measure- k=l

ments cannot be repeated exactly. Two cases must be where ~r2and ~ are the calculated variances for the real and
distinguished. imaginary components of the residual errors, respectively,
1. For stationary systems, the standard deviation (at a N is the number of data points at a given frequency, and
given frequency) of the residual errors resulting from a sin-
gle fit of a measurement model to successive measurements e~es= mean (elof+ e~) [9]
can provide an estimate for the frequency-dependent stan-
dard deviation of the stochastic contribution to the error The above procedure yields the same results as obtained by
structure. calculating directly the standard deviation of the real and
2. For nonstationary systems, the standard deviation of imaginary components of the data at each frequency. The
the stochastic contribution to the error structure is esti- discussion presented here is intended to emphasize the as-
mated from the standard deviation of the residual errors sumptions made and to illustrate the difference between
resulting from a fit of a measurement model to each indi- the treatment used for stationary systems and that used for
vidual successive measurement. Because a separate fit is nonstationary systems.
used for each repeated spectrum, the measurement model The results of the calculation of the stochastic contribu-
can be said to act as a filter for the contribution of the lack tion to the error structure for the measurements presented
of replicacy of successive measurements. in Fig. 1 are shown in Fig. 2 for data taken at 320 and 400 K.
The standard deviation of the impedance measurement is a
Stationary systems.--The procedure for calculation of
the standard deviation of the stochastic error estoch in sta-
1000000 ! ........ , ........ , ........ , ........ , .......
tionary systems is illustrated by its application to im-
pedance data for n-GaAs single-crystal wafer with a Ti ~~176 ~ ^
, oO~o~O A~e
Schottky contact and an Au/Ge/Ni ohmic c o n t a c t : Experi- I00000 ~ A ~~176 o o 320K
mental data for five replicate experiments conducted at m ! a ., ~ 9 400K
320 and at 400 K are shown in Fig. 1. For stationary sys-
10000 oi
tems, e~ is equal to zero. Therefore Eq. 5 reduces to o 2~
Z - ~ = e~ = e,o~ + em~ + es~oc~ [6]
6 ~o
.o I000 ma~
..~
where Z is calculated by regression of a measurement r A

model to the combined set of replicate data. Under the as- r~ 100 @I
A oa4 9
9
sumption that the stochastic errors e~oc~ = %oc~,r + jesto~,~ 9 - =- - _~ e=I &O 9 T T Io__ ~ 9 4
follow a normal distribution~
c 10~= 9 9 9 9 =o8~
e~too~: ~ N(E,, ~r) [7a] u~ , =o t I
esto~hj ~ N(% ~j) [7b] 1 % A
where the mean stochastic-error components er and ej are
each equal to zero. The standard deviations ~r and ~ are not 0.1 ...... '" . , ,,,,.i ..... .d , , ,,,,,,t , ......
known a priori. For replicate experiments, the errors due to
instrumental artifact can be assumed to be constant from 1 10 100 1000 10000 100000
one experiment to another. As one model was regressed to Frequency, Hz
all data sets, the lack of fit is also constant. The frequency-
dependent standard deviation ~, therefore, can be esti- Fig. 2. Unfiltered standard deviation of the data presented in Fig. 1
mated by the standard deviation of the departure of the as a function of frec.luency. Circles represent the real part, and the
residual error eres from the mean value at each frequency triangles represent the imaginary part of the impedance.
4152 J. E l e c t r o c h e m . Soc., Vol. 142, No. 12, December 1995 9 The Electrochemical Society, Inc.

20 ' ' ' ' I ' ' ' ' I ' ' ' ' 1 ' ' '
0.2 '"'"'~ '"'"'~ '"'"'q ' " ' " ' , '"'""1 '"""1

15

7"-1o ! I
o . o

t
0 5 10 15 20
Z~, kfl
-0.2 , ......1 . . . . . . .I ....... l .......= i,..,J ...... J
Fig. 3. Six successive impedance measurements for a copper disk
rotating at 1000 rpm in an alkaline 1 M chloride solution. The data 10 -1 10 0 101 10 2 10 3 10 4 10 5
were collected after 3 days of exposure. Frequency, Hz

Fig. 4. Real residual errors }or the regression o} a single-measure-

strong function of frequency, and the real and imaginary
ment model to the data shown in Fig. 3.
parts are indistinguishable at 320 K and almost indistin-
guishable at 400 K. The extent of the correlation of the real
and imaginary standard deviations at 400 K is comparable stainless steel shaft) was cleaned with alcohol before and
to that reported by Zoltowski. 17The correlation of real and after each impedance scan without interrupting the rota-
imaginary noise and the observation of heteroskedasticity tion of the disk. The long (1% closure error) autointegration
with respect to frequency are important results, and their cycle of the frequency-response analyzer was used.
significance is discussed in later sections. Each spectrum shown in the figure was found by the
Nonstationary systems.--For most electrochemical sys- method presented in Ref. 9 to be consistent with the
tems, the nonstationary contribution to the bias error ens is Kramers-Kronig relations, but a careful analysis of the
not equal to zero. Since the system evolves with time, ens data using the measurement-model approach showed that
changes from one experiment to another as a set of repli- the data were not replicate, i.e., the system changed meas-
cate (or consecutive) experiments are performed. Hence, urably from one experiment to another2 To demonstrate
Eq. 8 leads to inaccurate estimates for r because the contri- that the impedance scans were not replicate, a measure-
bution of end to the standard deviation cannot be ignored. ment model with eight Voigt elements was regressed (using
One can identify three time scales over which an electro- a complex nonlinear least squares regression algorithm 7'8
chemical system can change during the course of consecu- under modulus weighting) to the combined data set (i.e., a
tive impedance experiments in which data are collected at single measurement model was regressed to all six data sets
several frequencies for each scan. The smallest time scale simultaneously). The normalized residual errors obtained
considered here is the time to collect a datum point at one from the regression are shown in Fig. 4 and 5. The sinu-
frequency. The time required to collect one complete im- soidal character of the residual errors is caused by
pedance scan constitutes the next time scale, and the third
time scale is the total time elapsed from the start of the first
experiment to the end of the last experiment.
If the system is evolving rapidly, changes can occur dur-
ing the time in which one datum point is collected.
Impedance spectroscopy may not be a feasible experimen-
tal technique for such systems. For systems showing a
slower rate of change, the impedance at each frequency
may be measurable, but significant change can occur be-
tween the start and end of a complete frequency scan.
These types of nonstationarities result in the data being
inconsistent with the Kramers-Kronig relations. The issues
arising from these inconsistencies are discussed in the next _ 0.0
paper in this series. 9 I
The third time scale considered here appears when s
system under investigation is evolving slowly. In this case
the change in the system during one complete scan is small
L
and can be ignored, b u t nonnegligible differences can be
seen between successive spectra. Such pseudo-stationary
impedance scans are typically observed for even the most
stationary electrochemical systems. The need for a method
for analysis of such pseudo-stationary systems is illus-
trated below. -0.2 5
Six replicate (consecutive) impedance spectra obtained 10
for a copper disk rotating at 1000 rpm in an alkaline (pH Frequency, Hz
11.5) 1 M chloride aqueous solution are shown in Fig. 3.3
The electrodes were held at the open-circuit condition. The Fig. 5. Imaginary residual errors for the regression of a single-
electrical contact to the disk electrode (carbon brush and measurement model to the data shown in Fig. 3.
J. Electrochem. Soc., Vol. 142, No. 12, December 1995 9 The Electrochemical Society, Inc.
| ,
1000
I 0.2
500
r--
-500
:
:. oo
oo
-1000 , , , I , ,
-1000 -500 0 500
%, Q
Fig. 6. The imaginary departures from the mean residual error
ere,i -- ere, for the regression shown in Fig. 4 and 5 as a function of
the' corresponding real values.
errors associated with the lack of fit of the model. The time-
varying character of the measurements is not readily ap-
parent in Fig. 3, but examination of the residual errors
shows that the system changed from one experiment to the
other and that the residuals for the six experiments can be
distinguished from one another.
The measurement-model analysis shown in Fig. 4 and 5
shows that the standard deviation obtained using Eq. 8
must contain, in addition to the desired stochastic contri-
bution to the error structure, a significant contribution
from a nonstationary bias component. The standard devia-
tions that are corrupted by such bias errors show several
distinctive features. The real and imaginary departures
from the mean residual error are correlated, as shown in
Fig. 6. The standard deviations calculated using Eq. 8 are
shown in Fig. 7 as a function of frequency. Examination of
real and imaginary values suggests that they are not equal.
Both observations are counter to results obtained for sta-
104 . . . . . . . . . ' ........ ' ........ ' ........ ' ....... '
10 2
- -2--;'--
b
I 0 - 4 _ 'I ....... I ........ j ........ i ........ i . . . . . .
10 1 0 0 1 0 1 1 0 2 1 0 `5 1 0 4
Frequency, Hz
Fig. 7. Unfiltered standard deviation of the data shown in Fig. 3 as
a function of frequency. Circles represent the real, and the triangles
represent the imaginary part of the standard deviation. The solid line
represents the error structure for the n-GaAs sample held at 320 K
(see Eq. 14). The dashed line represents the contribution that is pro-
portional to I Z.iI, the dashed-dot line represents the contribution that
is proportional to I Zrl, and 2the dashed-dot-dot-dot line is the cantri-
bution proportional to IZI /R,. The discontinuity apparent in the
solid line and the dash-dot-dot-dot line is caused by a change of the
value of the current measuring resistor at 100 Hz.
4153
, , , .
. . . . ''I ....... '1 " ...... I ' " ' " 1 ' ....... I ' " ' " "
~ L
N
N
I
0 . 0 ~
I L.
N
, ,
1 )00
-0.2 ........
I .... ,,,I .......J , ,,,,,,,I , ,,,,,.I ......
10 -1 100 101 102 103 104 105
Frequency, Hz
Fig. 8. Real residual errors for the separate regression of measure-
ment models to the data shown in Fig. 3.
tionary data, where the real part of the standard deviation
was observed to be equal to the imaginary part.
For nonstationary systems, the stochastic component of
the error cannot be calculated by taking the standard devi-
ation of the residual errors directly. A procedure to filter
out the nonstationary component of the error is outlined in
the next section. The significance of the lines in Fig. 7 is
discussed in later sections.
Identification of the noise m o d e L - - F o r nonstationary
data, a measurement model is regressed to each data set
using the m ax i m u m number of parameters that can be re-
solved from the data. For this work, a complex nonlinear
least squares program was used that was developed in-
........ house? '8 As a model for the stochastic contribution to the
error structure is not known a priori, modulus weighting is
typically used for the regression. The regressed parameters
for the measurement model for each data set are slightly
different because the system changes from one experiment
to the other. Hence, by regressing the measurement model
to individual data sets separately, the effects of the change
of the experimental conditions from one experiment to an-
other are incorporated into the measurement model
parameters. The standard deviations of the real and imagi-
nary residual errors therefore can be obtained as a function
of frequency and provide a good estimate for the standard
deviation of the stochastic noise in the measurement.
The data set from Fig. 3 is used to illustrate the tech-
nique. The normalized real and imaginary residual errors
for six regressions are shown in Fig. 8 and 9. The real and
imaginary residual errors are randomly distributed about
the mean value at a given frequency. The plot of the imagi-
nary vs. real departures from the mean residual error,
10 5 shown in Fig. 10, further suggests that the residual errors at
a given frequency are not correlated. The cyclic behavior of
the residual errors with frequency is caused by the lack of
fit of the model. The technique of estimating the standard
deviation of the stochastic contribution by calculating the
standard deviation of the residual errors is in effect a filter
for lack of fit. Using the measured error structure to weight
the regression of data that are free of bias errors, the
residual errors for the measurement model can be made to
be of the same order as the standard deviation of the
measurement. 9
The real and imaginary standard deviations of the resid-
ual errors are shown as a function of frequency in Fig. 11.
4154 J. Electrochem. Sac., Vol. 142, No. 12, December 1995 9 The Electrochemical Society,Inc.

0 . 2 [ ' ..... "I ...... "I ....... '1 . . . . '"'1 ........ I ....... 10 4 ' ''"'"1 ........ I ........ I ....... I ' '"""1 ' " .....

10 2 .
O O

0o
b
IN--
0.0
I
t 9 10 -2
N--
v

-4
1 0 , ,,,,,,,I ........ i ........ I ........ I ........ I ......

10 - 1 10 0 101 10 2 10 5 10 4 05
-0.2 ., ...... I ,, ..... ,I ........ I , ,,,,,.I , ,,,,,,,I , ,,,,,
Frequency, Hz
10 -1 10 0 101 10 2 1 0 `5 10 4 10 5 Fig. 11. Standard deviation of the residual errors presented in
Frequency, Hz Fig. 8 and 9 as a function of frequency. Circles represent the real and
the triangles represent the imaginary part of the standard deviation.
Fig. ?. Imaginary residual errors for the separate regression of The solid line represents the error structure for the n-GaAs sample
measurement models to the dote shown in Fig. 3. held at 320 K. The dashed-dot line represents the contribution of the
imaginary term to the error. The dashed-dot-dot line represents the
contribution of the real term, and the dashed line is the measuring
The s t a n d a r d deviations are m u c h smaller t h a n seen in resistor term.
Fig. 7, and the real and i m a g i n a r y values are n o w equal.
The m e t h o d of regressing a m e a s u r e m e n t m o d e l to indi- and ~j are the real and i m a g i n a r y p a r t of the s t a n d a r d devi-
v i d u a l spectra serves as a filter for l a c k of replieaey, and, as ation. S o m e symbols used in Eq. 10 have been changed to
m e n t i o n e d above, the calculation of the s t a n d a r d d e v i a t i o n c o n f o r m w i t h the n o t a t i o n of this paper. Other authors h a v e
of the residual errors for the i n d i v i d u a l fits serves as a filter s u p p o r t e d use of modulus w e i g h t i n g on the basis of the
for l a c k of fit. observed correlation b e t w e e n the s t a n d a r d deviations of
real and i m a g i n a r y c o m p o n e n t s of impedance. :722
M o d e l for the Error Structure For nonzero values of 6,, Eq. 10 does not c o n f o r m to the
In this section a p r e l i m i n a r y m o d e l for s t a n d a r d devia- e x p e r i m e n t a l evidence (e.g., Fig. 2 and 11) t h a t the real part
tion of the error, ~, is proposed. Relatively little w o r k has of the s t a n d a r d d e v i a t i o n is e q u a l to the i m a g i n a r y part. If,
Been done in d e v e l o p i n g p r o p e r models for the error struc- however, 6r is e q u a l to zero, Eq. 10 yields a s t a n d a r d devia-
ture of i m p e d a n c e data. ~a ~6 M a c d o n a l d has p r o p o s e d a tion that is i n d e p e n d e n t of frequency, a result that is also in
p o w e r - l a w m o d e l for the f r e q u e n c y - d e p e n d e n t v a r i a n c e v is conflict w i t h Fig. 2 and 11.
l) r = 0"~. ---- OL~ 4- ~I2r 12~0 Theoretical development.--While it is evident t h a t the
[10] stochastic c o n t r i b u t i o n to the error structure is a f u n c t i o n
~ = ~ = ~ + ~[2~1 ~-0
of frequency, the most general f o r m u l a t i o n for the error
where ar, ~r, and ~0 are p a r a m e t e r s of the m o d e l for the error structure can be w r i t t e n in terms of the m e a s u r e m e n t itself
structure, Z~ and Z i are the real and i m a g i n a r y p a r t of the (as was done, for example, in Eq. 10). 15 U n d e r the a s s u m p -
impedance, respectively, v,, vi are the v a r i a n c e of the real tion t h a t the f u n d a m e n t a l i m p e d a n c e m e a s u r e m e n t in the
and i m a g i n a r y p a r t of the impedance, respectively, and ~r i n s t r u m e n t a t i o n is the m a g n i t u d e IZl and the phase angle
(b, the s t a n d a r d d e v i a t i o n ~ for the real and i m a g i n a r y com-
ponents can be expressed as
200 ' ' '
8Zr 8Zr
(rr = ~ Ezi (Y6 + 0 Z 6 (rlzl
[11]
oz~
% (~J = Odp ~zl if* + OZ , (~z'
A + o o
where a,z, and ~, are the standard deviation of the magni-
- 0 tude and phase angle, respectively.
The development of a preliminary model for the standard
deviation of measurements in the impedance plane was
based on published instrument specifications. The error in
A the phase angle was assumed to be a constant, and the error
in the magnitude was assumed to be proportional to the
signal with a term added to account for the poor signal-to-
noise ratio experienced when there is mismatch between
--200 i i i i i i
the system impedance and the measuring resistor. Thus, the
-200 0 200 initial postulate for the model development was
e r, f~
Fig. 10. The imaginary departures from the mean residual error [12]
ere, - ~,~, for the regressions shown in Fig. 9 and 10 as a function of O'rz j + Lzq izl
the'corresponding real values.
J. Electrochem. Soc., Vol. 142, No. 12, December 1995 9 The Electrochemical Society, Inc. 4155

where e, ~, and ~ are constants, and R= is the value of the 1000000 : , .,,.,. , ,,,,,., ........ . , ,,,,,,,j . ,,,,,,j 10
current-measuring resistor. Parameters ~, ~, and % in prin- !%o __ o 00%0 o
ciple, depend on the specific instruments being used for I00000 ~ ~oo o
@;
impedance measurements. The expressions for the errors in Oo ~a
the real and imaginary components become ,~ a~ o
~ 10000 ,'2 ,
9~ Q; a aa A o% ~l:
O'r = ~rlZjl + ~rlZrl + % ~ iZ r] o m a a 1
~. "~
~ 1000 A A A a A %0 CO
[13] 0 ~ o

IZI 0.~ ~.
~ri = %lZil + ~jlZrl + "y~~ IZil m'~a~
.~ 100 a A a A~ co =
'0
" a,,s 0% o
To reconcile Eq. 13 with the observation that the real and I0 ~ o o "N
imaginary standard deviations are equal, the following re- "~ : '.'........:'..:.... a % o 01 ,-,|
vised error structure was proposed A 0 AO

IZl ~
% = ~= = ~ = ~lZil + ~lZrl + ~y Rm [14]
o.1 9 ":"'" 4
9 eee 9/
While the form of Eq. 14 was suggested by the assump-
tions given in Eq. ii, a recasting of Eq. 14 in polar coordi- 0,01 ............................ 0.01
A ....... -~' ~1~'""1

nates; i.e. 1 10 100 1000 10000 100000

r IZ~I ~31Zrl i)
iZ12+ ~+'y~ (IZ~l+lZ~l)
( _ ,zq
O-lz== et ~IZI+ 131Z~-+ 'Y Rm/ (IZrl + IZ~l)
shows that the initial assumptions about the errors in
phase angle and the magnitude are incorrect and that the
errors in phase angle are not independent of frequency.
These conclusions can be confirmed experimentally. The
"stationary" data set of Fig. 1 is presented in Bode format
(phase angle and modulus) in Fig. 12. The standard devia-
tion of the phase angle and the modulus are presented in
Fig. 13. The standard deviation of the phase angle reaches
a maximum of 2 ~ at a frequency between i0 and 60 Hz
(corresponding to a phase angle between -20 and -70 ~
respectively) and reaches a value as low as 0.02 ~ at a phase
angle of -90 ~. The standard deviation of the modulus
tracks the value for the modulus only approximately. The
standard deviation of the modulus is as high as 400,000 ~at
low frequencies (4% of the modulus) and reaches a value of
0.01% of the modulus at high frequencies.
Equation 14 is preliminary, but it has the desirable fea-
tures of frequency dependence embedded in the measured
values for the impedance and of implicit agreement with
the experimental observation that the noise in the imagi-
nary and real parts of the impedance is equal. The validity
of Eq. 14 as a model for the stochastic noise was established
by comparison to experimental data, as discussed in the
subsequent section.
1.0E+07
~ i
1.0E+06
= ! % -40
O
1.0E+04
-10
1.0E+03 ......................... 0 weighting or proportional weighting in the regression gave
1 10 100 1000 10000 100000
Frequency, Hz
Fig. ] 2. Impedance response in the Bode Formot of a single-crystal
n-GaAs/~ Schoffky diode held at 320 K. Five replicate measure-
men~ were made.
Frequency, Hz
Fig. ] 3. Unfiltered standord deviation of the data presented in
[15] Fig. 12 as a function of frequency. Open circles represent the stan-
dard deviation of the modulus in ohms, closed circles represent the
standard deviation of the modulus as a percentage of the measured
modulus, and triangles represent the standard deviation of the phase
angle in degrees.
Comparison to experiment.--The error structure is de-
veloped here for impedance data collected for an n-GaAs/
Ti Schottky diode. This solid-state system was chosen for
this analysis because the standard deviation of the meas-
ured impedance had a broad range of values and because,
to a first approximation, the sequential measurements
could be assumed to be replicate. Previous work 1'7 showed
that, while use of modulus weighting, proportional weight-
ing, or no weighting gave only one electronic state, regres-
sion of these data using the error structure to weight the
regression allowed determination of four electronic states.
The number of states and their energy level were confirmed
by independent measurementsY
Experiment.--Values for ~, ~, and ~/were calculated for
impedance data collected on a Solartron 1286 potentiostat
and a Solartron FRA1250 frequency-response analyzer
(FRA). The data used for the estimation of the error struc-
ture parameters were collected for the n-GaAs Schottky
diode discussed in Fig. 1 at temperatures ranging from
320 K to 420 K. i The frequency range used for the experi-
ment was 1 Hz to 65 kHz. Data were collected frequency by
frequency using the long channel integration feature of the
FRA, which completed a measurement at each frequency
-100
on reaching a 1% closure error. The wide range of tempera-
-90 ture and frequency ensured a wide range of impedance val-
-80 ues. Five replicate experiments were conducted at each
temperature. The results of the experiments at 320 and
-70 400 K are shown in Fig. i. As a first approximation it was
-60 assumed that the system was stationary, and the stochastic
component of the error e=toc~ and its standard deviation, cr
were calculated for data sets for the GaAs sample at 320,
340, 360, 380, and 400 K, as described in the section on
Stationary systems,
-3o
Regression procedure.-- Equation 14 was regressed to the
-20 standard deviation values to obtain the values of cr ~, and
~. The data were detrended to ensure that the mean residual
error for the regression was equal to zero. 27 Use of no-
poor results. Following Eq. i, the regression was weighted
by the estimated variance. The standard deviation of the
standard deviation for impedance measurements was ob-
tained by considering the standard deviation of the real
and imaginary parts of the impedance to be independent
observations of the standard deviation at a given frequency.
4156 J. Electrochem. Soc., Vol. 142, No. 12, December 1995 9 The Electrochemical Society, Inc.
Table I. Parameter estimates for the error-structure model (Eq. 14) obtained by regression to replicate data for a
single-crystal n-GaAs/Ti Scholtky diode.

Unfiltered Filtered
No moving Three-point moving Five-point moving No moving Three-point moving Five-point moving
Parameter Average Average Average Average Average Average

cr 3.29-+ 0.13 10 -4 2.46_+ 0.48 10 -4 8.2-+ 5.6 10 3 8.12_+ 0.020 10 -4 9.45-+ 0.50 10 4 9.43 -+ 0.64 I0 -4
(_+4.0%) (-+20%) (-+69%) (_+0.23%) (-+5.3%) (+_6.8%)
1.20-+ 0.01 10 3 1.24_+ 0.03 x 10 -3 1.59-+ 0.07 10 -3 9.33-+ 0.011 10 4 1.49_+ 0.045 10 -3 1.12 _+ 0.052 10 -3
(-+0.88%) (+-2.8%) (-+4.1%) (-+1.1%) (+-3.0%) (-+4.7%)
2.833 2.832 2.785 2.306 2.209 2.268 -+ 0.010 10 .4
-+0.0021 10 -4 _+0.0062_+ 10 ~ +-0.011 10 4 -+0.0017 _+ 10 -4 _+0.0090 10 4 (0.46%)
(0.074%) (0.22%) (0.39%) (0.072%) (0.41%)
X3/~ 4.74 1.77 1.64 15.2 2.18 1.83

The square of the standard deviation of the standard devia-
tions for the real and imaginary components of the im-
pedance was used, therefore, to weight the regression for
identification of the error structure. In addition, three- and
five-point moving averages were used to increase the sam-
ple size for the standard deviation while retaining the gen-
eral trends. The weighted X2 statistic (normalized by the
degrees of freedom for the regression)
1 (Yexpt~i-- Ymodel,i)2
i weighting strategies for regression. This information also
was improved for regressions using a moving average value
f o r t h e v a r i a n c e . T h e u s e of a m o v i n g a v e r a g e d i d n o t a p -
p e a r to i n f l u e n c e t h e fit of t h e m o d e l to c a s e s w h e r e t h e
s t a n d a r d d e v i a t i o n f o r t h e i m p e d a n c e w a s l a r g e , b u t t h e fit
was qualitatively improved for cases where the standard
d e v i a t i o n of t h e i m p e d a n c e w a s s m a l l . T h e u t i l i t y of t h e
moving average should depend on the sampling rate. The
five point moving average worked well for the ten points/
decade sampling rate used here. The increased standard
deviation for the parameter estimates obtained using the
moving average reflects the corresponding increased value
for the average standard deviation used to weight the re-
gression. B y giving a better estimate for the standard devi-
ation of the fitted quantity, the moving-ayerage approach
yielded more reliable estimates for the confidence intervals
of the error-structure model parameters.
Results of regression.--The parameter estimates obtained
are given in Table I. The results of the regression for G a A s
at 320 K are shown in Fig. 14, and the results for 400 K are
presented in Fig. 15. The solid line represents the model for
the error structure given by Eq. 14. The dashed lines repre-
sent the contribution of the different terms in Eq. 14. The
jog in the line for the model corresponds to the frequency at
which a change was m a d e in the value of the current meas-
uring resistor.
Discussionand Conclusion
The measurement model provides m u c h more than a pre-
liminary analysis of impedance data in terms of the number
of resolvable time constants and asymptotic values, as sug-
gested, for example, by Zoltowski. 23'24As shown here, the
measurement model can be used as a filterfor lack of repli-
cacy that allows accurate assessment of the standard devi-
ation of impedance measurements. As is discussed by m a n y
authors, 1'15,2~this information is critical for selection of
provides a quantitative basis for assessment of the quality
of fits and can guide experimental design. In the subse-
quent paper of this series,9 the measurement model is used
to assess the bias component of the error structure.
The results presented here show that impedance mea-
surements are heteroskedastic (in the sense that the stan-
dard deviations are funetior~s of frequency). In spite of the
apparent complexity, a definite structure for the errors is
resolved. In the impedance plane, the standard deviations
for the real and imaginary parts of the impedance are
equal, even at frequencies sufficientlyhigh or low that the
imaginary part of the impedance asymptotically ap-
proaches zero. This result is consistent with the results pre-
sented by Zoltowski. 17
Aside from the obvious impact on the parsimony of the
model for the error structure (only three parameters are
10 6 ' '""9 ........ I . . . . . . . '1 ' '"'"'1 ' ' .....
10 4
-

The filteringalgorithm described in the section Identifi- 10 2

cation of the noise model was applied to the above data set
to give a better estimate for the stochastic errors. Equa-
tion 14 was regressed to the filtered errors. The resulting b"
10 0
parameters for the error structure model are shown in
Table I. The filtered errors for G a A s at 320 and 400 K, with
"%
the n e w model, are shown in Fig. 16 and 17, respectively.
-2
The standard deviations for the filtered data (Fig. 14) were 10
smaller than for the unfiltered case (Fig. 12), and the real
and imaginary parts of the standard deviations are closer.
The regression results in Fig. 16 and 17 show that the agree- -4 i % .
ment of the model for the error structure (Eq. 14) to the
10 , * ,iliil[ i I. l i l l J d , *liil,,l , ililild , ,,,i,

experimental standard deviations is good. Similar agree- I0 ~ I01 I02 I0 ~ 104 105
ment was found at all temperatures. The parameters from Frequency, Hz
Table I, for filtered errors, were used to predict the errors
for corrosion of copper data shown in Fig. 3. The results are Fig. 14. Unfiltered standard deviation of the 320 K data presented
presented in Fig. ii. The model shows a good agreement in Fig. I as a function of frequency. Circles represent the real and the
with the experimentally obtained standard deviations. The triangles represent the imaginarypart of the standard deviation. The
validity of the equation is supported since a three-parame- solid line represents the error structure with unfiltered parameters
ter model provides a good agreement for solid-state sys- shown in Table I. The dashed-dot line represents the contribution of
tems as well as for electrochemical systems, for data col- the imaginary term to the error. The dashed-dot-dot line represents
lected under various experimental conditions, and for the contribution of the real term, and the dashed line is the measur-
errors ranging in magnitude from 10 -3 to i0 ~ ~. ing-resistor term.
J. Electrochem. Soc., Vol. 142, No. 12, December 1995 9 The Electrochemical Society, Inc. 4157

106 ' ''"'"I ' '''""I ' ';'""I ' ''"'"I ' ''"I 106 ' ''"'"1 ' ''"'"t ' ''"'"l ' ''""~

104

102,
104

102
1
100 100 .4
,i i %.--

s
i

10 - 2 f\ 14
10 - 2
t\ -"
!

-4 f
10 - 4 I i llillll j I ' illliii t I ill'ltJ t I lllilJ i I ilail
10 = ,,,,.,I , ,,,,,,,i , ,,,,,ul , ,,,,,,,I , i , . ...

100 101 102 105 104 05 1 )0 101 10 Z 105 104 05

Frequency, Hz Frequency, Hz
Fig. 17. Filtered standard deviation of the 400 K data presented in
Fig. 15. Unfiltered standard deviation of lhe 400 K data presented Fig. 1 as a function of frequency. Symbols and lines as defined in
in Fig. 1 as a function of frequency. Symbols and lines as defined in Fig. 16.
Fig. 14.

ance spectroscopy (EHD)] is, to a first approximation, in-

needed in Eq. 14 as compared to six in Eq. 13), the equality dependent of frequency. 6'7
of the real and imaginary standard deviations has implica- Selection of inappropriate weighting strategies may have
tions for the regression of models to impedance data. That a severe impact on the amount of information that can be
the information content of the imaginary part of the obtained by regression of models to impedance data since
impedance can be obscured by noise at the asymptotic tails the weighting strategies discussed here can differ by many
influences the manner in which the Kramers-Kronig rela- orders of magnitude. To illustrate the influence of weight-
tions can be applied to assess the bias contribution to the ing, normalized weightings, defined by
measurement. ~ In addition, the equality of the real and
imaginary standard deviations becomes a criterion for se- [ o'/IZl ~2
lection of appropriate weighting strategies. Among the to,z, = \(r l}~vJ [16a]
commonly applied weighting strategies, for example, pro-
portional weighting does not conform to this observation, for modulus weighting or by
but "no-weighting" and modulus weighting do conform
and may be useful weightings for preliminary regressions. ~o = [16b]
While the heteroskedastic nature of the measurements
shown here suggests that the no-weighting strategy is inap-
propriate for experiments under potentiostatie modula-
tion, this is not a general result since the standard deviation 1.0E+01 100
of some measurements [e.g., electrohydrodynamic imped-
0 +00
m
1,0E-0

104~ ~ I0 o
106 ........ , ........ I ........ , ........ I ....... I .OE-02 " ~"

.oE-o3 ..
i ~
~1.0E-04 1 "
J %',%
102 "~.
N 1.0E-06 ~.
%,%,
E1.0E_07 0"1 i
10 ~ ~
Z
1.0E-08 .'_'_iiii'iiiin ~
No Weighting
t -
Z

1.0E-09 ........................................... 0.01

10 - 2 "-
t 10 100 1000 10000 100000
Frequency, Hz
10 4 ....... J ........ , ........ i ........ , .... Fig. 18. Comparison of modulus weighting and no-weighting
10 U 101 102 103 10 4 strategies to the weighting by the error structure given in Eq. 14. The
terms plotted are defined by Eq. 16. The solid lines represent the
Frequency, Hz comparison of the modulus weighting strategy to weighting by the
standard deviation of the experiment determined by the methods of
Fig. 16. Filtered standard deviation of the 320 K data presented in this paper. The dashed lines represent the corresponding comparison
Fig. 1 as a function of frequency. Symbols and lines as defined in for the no-weighting weighting strategy. The upper set of lines were
Fig. 14 with exception that model parameters are those given as obtained for the data collected at 320 K, and the lower set of lines
Filtered in Table I. were obtained for the data collected at 400 K.
4158 J. Electrochem. Soc., Vol. 142, No. 12, December 1995 9 The Electrochemical Society, Inc.
for no weighting, are shown in Fig. 18 for the semiconduc-
tor d a t a of Fig. 1. Modulus and no-weighting options are
thereby compared to the weighting by ~2 where a is given
by Eq. 14. Weighting according to Eq. 14 has been shown to
yield more detailed information about the Schottky diode
than could be obtained by modulus- or no-weighting op-
tions, and the structural information so obtained was con-
firmed by independent measurements} Thus, Fig. 18 can be
said to provide a comparison of the modulus- and no-
weighting options to the optimal weighting strategy for
this system. The jog in the line for the mode] corresponds to
the frequency at which a change was made in the value of
the current measuring resistor. The data taken at 320 K
show a fully resolved spectrum over the measured fre-
quency range (see Fig. 1 and 12). In this case, the no-
weighting option underweights the high-frequency re-
sponse by nine orders of magnitude as compared to the low
frequency data. The modulus weighting is better, but still
underweights the high-frequency end by three orders of
magnitude. The consequence is that information contained
in the high-frequency data is not extracted by the regres-
sion procedure. The spectrum is less fully resolved at 400 K,
and the no- and modulus-weighting options are closer to
the optimal weighting strategy for this system. In this case,
the modulus weighting overweights the high frequency end
by a factor of only ten; whereas the no-weighting option
under weights the high frequency end by the same factor.
The model presented as Eq. 14 provides good agreement
to the experimentally measured stochastic noise for a
broad range of experimental systems. The systems have
included, in addition to those discussed here, studies of
electrochemistry at metal hydride electrodesy corrosion of
copper, 3'7 transport across biological membranes, 2~ and
electronic transitions in large bandgap materials such as
Z n O and ZnS. 26 The stochastic noise seems to he primarily
a function of the instrumentation used, since use of
parameter values in Eq. 14 that were obtained with the
semiconducting system provided a good prediction of the
error structure for the electrochemistry at a rotating copper
disk. The carbon electrical contact to the rotating electrode
and the reference electrode did not m a k e a significant addi-
tional contribution to the noise. With the exception of ex-
periments in which gas evolution was seen on the elec-
trodes, similar agreement was seen for other experimental
systems mentioned above.
The model for the error structure is stillunder develop-
ment. The role, for example, of the amplitude of the poten-
tial perturbation has not been incorporated into the model.
The lack of importance of the perturbation amplitude ap-
parent in the work to date m a y be a testament to the effi-
ciency of the autointegration feature of the FRA. The error
structure has been developed here for potentiostatic modu-
lation and m a y need to be refined for experiments con-
ducted under galvanostatic control. Finally, all systems
studied here have had a small ohmic contribution to the
impedance. A large ohmic resistance m a y have a striking
impact on the error structure of impedance measurements
given the observed equality of the standard deviation for
the real and imaginary parts of the impedance.
The model parameters presented in Table I are not neces-
sarily those that would be obtained on all systems or in
other laboratories. The parameters can depend on the ex-
perimental design, the type of instruments used, the man-
ner in which the instruments are used, and, in some cases,
the system under study. The contribution of this work is a
systematic model development strategy for impedance
measurements that incorporates a quantitative assessment
of the stochastic contribution to the error structure.
Acknowledgment
The work performed at the University of Florida (RA.
and M.E.O) was supported by the Office of Naval Research
and by Gates Energy Products. The work performed at the
University of South Florida (L.H.G.R.) was supported by
the National Science Foundation under Grants No. RII-
8507956 and INT-8602578. This paper was written while
one author (M.E.O) was on sabbatical leave at the U P R 15
du C N R S "Physique des Liquides et Electrochimie," in
Paris, France. The use of their facilitiesto prepare portions
of this work and the helpful suggestions of Dr. Claude
Deslouis and Dr. Bernard Tribollet are greatly appreciated.
Manuscript submitted April i0, 1995; revised manuscript
received Aug. 4, 1995.
University of Florida assisted in meeting the publication
costs of this article.
REFERENCES
i. M. E.Orazem, R Agarwal, A. N. Jansen, R T. Wojcik,
and L. H.Garcia-Rubio, Electrochim. Acta, 38, 1903
(1993).
2. P. Agarwal, M. E. Orazem, and L. H. Garcia-Rubio, in
Electrochemical Impedance: Analysis and Interpre-
tation, ASTM STP 1188, p. 115, J. Scully, D. Silver-
man, and M. Kendig, Editors, American Society for
Testing and Materials, Philadelphia (1993).
3. P. Agarwal, O. C. Moghissi, M. E. Orazem, and L. H.
Garcia-Rubio, Corrosion, 49, 278 (1993).
4. M. E. Orazem, P. Agarwal, L. H. Garcia-Rubio, J. of
Electroanal. Chem. Interfac. Electrochem., 378, 51
(1994).
5. P. Agarwal, M. E. Orazem, and A. Hiser, in Hydrogen
Storage Materials, Batteries, and Chemistry, D. A.
Corrigan and S. Srinivasan, Editors, PV 92-5, p. 120,
The Electrochemical Society Proceedings Series,
Pennington, NJ (1991).
6. R Agarwa], M. E. Orazem, C. Deslouis, and B. Tribollet,
This Journal, Submitted.
7. P. Agarwa], Ph.D. Thesis, University of Florida,
Gainesville, FL (1994).
8. P. Agarwal, M. E. Orazem, and L. H. Garcia-Rubio,
This Journal, 139, 1917 (1992).
9. P. Agarwal, M. E. Orazem, and L. H. Garcia-Rubio,
This Journal, 142, 4159 (1995).
10. G. E. P. Box and N. R. Draper, Empirical Model-Build-
ing and Response Surfaces, John Wiley & Sons, Inc.,
New York (1987).
11. R. W. Christy, Am. J. Phys., 4O, 1403 (1972).
12. A. Jutan and L. H. Garcia-Rubio, Process Control and
Quality, 4, 235 (1993).
13. G. Spinolo, G. Chiodelli, A. Moghistris, and U. A. Tam-
burini, This Journal, 135, 1419 (1988).
14. J.R. Macdonald and L. D. Potter, Jr., Solid State Ionics,
23, 61 (1987).
15. J. R. Macdonald, Electrochim. Acta, 35, 1483 (1990).
16. J.R. Macdonald andW. J. Thompson, Commun. Statist.
Simula., 20, 843 (1991).
17. P. Zoltowski, J. Electroanal. Chem., 178, 11 (1984).
18. P. Zoltowski, ibid., 260, 269 (1989).
19. P. Zoltowski, ibid., 260, 287 (1989).
20. B. Robertson, B. Tribollet, and C. Deslouis, This Jour-
nal, 135, 2279 (1988).
21. Harold W. Sorenson, Parameter Estimation: Principles
and Problems, Marcel Dekker, Inc., New York (1980).
22. B. A. Boukamp, Solid State Ionics, 20, 31 (1986).
23. P. Zoltowski, Pol. J. Chem., 68, 1171 (1994).
24. P. Zoltowski, J. Electroanal. Chem., 375, 45 (1994).
25. M. Membrino, Ph.D. Thesis, University of Florida, In
preparation.
26. A. N. Jansen, Ph.D. Thesis, University of Florida,
Gainesvilte, F L (1992).
27. G.A. E Seber, Linear Regression Analysis, pp. 330-334,
John Wiley & Sons, New York (1977).