Professional Documents
Culture Documents
4149
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- identify the stochastic component of the frequency-de-
dation for the application of measurement models to pendent error structure of impedance data. A preliminary
impedance spectroscopy. The basic premise behind this model for the stochastic component of the error is pro-
work is that determination of measurement characteristics posed. The third paper of this series addresses the use of the
is an essential aspect of the interpretation of impedance measurement model for identification of the bias compo-
spectra in terms of physical parameters. The influence of nent of the error structure. ~
the error structure on interpretation of impedance spectra
is discussed elsewhere for various electrochemical and Background
electronic systems. 1-7In the previous paper of this series, it
was shown that a measurement model based on Voigt cir- The objective of impedance measurements is identifica-
cuit elements can provide a statistically significant fit to tion of physical processes or parameter values appropriate
typical electrochemical impedance spectra. BHere a method for a given system. Interpretation of impedance spectra re-
is demonstrated in which the measurement model is used to quires, in addition to an adequate deterministic model, a
thorough understanding of the error structure for the
* Electrochemical Society Student Member. measurement. The concept that the error structure plays an
** Electrochemical Society Active Member.
Present address: Department of Materials, Swiss Federal In- important role in the interpretation of experimental data is
stitute of Technology (Lausanne), Lausanne, Switzerland. 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 (randomly distributed) contributions, Esystand es~ooh,respec-
development, knowledge of the error structure has been tively. Thus, at any given frequency
shown to allow enhanced interpretation of light scattering
measurements in terms of particle size distribution or even Z - Z = ere~= esy~t+ e~toch [2]
particle classification, n'~2 where the caret signifies the model value for the complex
Identification of the error structure for most radiation- impedance Z. The presence of stochastic errors estoohin any
based spectroscopic measurements such as light scattering experimental data is inevitable. In this work, the system-
can be accomplished by calculating the standard deviation atic errors that result from model inadequacies are distin-
of replicate measurements. The error analysis approach has guished from the experimental errors that are propagated
been successful for light spectroscopy measurements be- through the model and that could arise from a changing
cause these systems lend themselves to replication and, base line or from instrumental artifacts. Systematic errors
therefore, to the independent identification of the different are therefore defined to consist of contributions from the
errors that contribute to the total variance of the measure- lack of fit of the model to the data (elo~)and a bias (%~as) in
ments. In contrast, the stochastic contribution to the error the experiment, i.e.
structure of electrochemical impedance specfroscopy mea- %s~ = elo~+ eblas [3]
surements generally cannot be obtained from the standard
deviation of repeated measurements because even a mild In principle, improvement of a stationary model reduces
nonstationary behavior introduces a nonnegligible time- the error associated with a lack of fit, but nonstationary
varying bias contribution to the error. behavior (%s) and instrumental artifacts ( e J can still con-
The discussion of error structures in impedance spec- tribute to the bias errors, i.e.
troscopy, therefore, has been limited to a priori predictions eb~ = en~ + ei~ [4]
of measurement noise based on instrument noise ~3 and to
standard assumptions concerning the error structure such The nonstationary contribution to the bias usually is ob-
as constant or proportional errors. ~4-~6Zoltowski reports an served most easily at frequencies that require the longest
experimental assessment of the stochastic noise at selected time for measurement. Instrumental artifacts may be seen
frequencies which showed that the standard deviation of at high frequency resulting from equipment limitations.
the real and imaginary parts of the impedance (and admit- Many, if not most, electrochemical systems show at least a
tance) are correlated. ~7 This result was used to defend the mild nonstationary behavior due to changes in electrode
use of modulus weighting for regression of models to exper- properties during the course of an experiment. In contrast,
imental data. ~749 Attempts to weight regressions by stan- solid-state systems, as a first approximation, may be as-
dard deviations determined from repeated impedance sumed to be stationary. Impedance data can be corrupted
scans (e.g., Ref. 20) have not been successful because the by instrumental artifacts for both electrochemical and
standard deviation obtained from repeated impedance solid-state systems.
measurements includes both bias and stochastic contri- The emphasis in regression of models to impedance data
butions to the error structure. Our objective here is to is on reducing the error associated with a lack of fit, but it
present a procedure for assessing the stochastic contribu- is evident that the residual errors must contain contribu-
tion to the error structure of impedance measurements and tions associated with phenomena that are independent of
to develop a model for the stochastic noise of impedance the adequacy of the model, i.e.
measurements. Z - 2 = e~o~ + (ens + ei~) + e~o~h [5]
Knowledge of the error structure is essential in assessing
experimental technique and can play a role in determining The object here is to quantify the stochastic errors in
the influence of instrumental strategies. For example, use impedance measurements. The quantification of bias er-
of matched filters for the input of the frequency response rors associated with experimental issues such as instru-
analyzer has been proposed to be an appropriate technique mental artifacts or nonstationary phenomena will be ad-
for reducing the noise in impedance spectra. 2~ While the dressed in a subsequent paper. 9
reduction of the noise in input signal can be readily seen by Identification of Error Structure
the display of an oscilloscope, a method for assessing the The method for assessing the stochastic part of the error
noise in the ultimate measurement is essential in deciding structure is based on using a measurement model as a filter
whether the filters contribute to reduction of noise in the for nonreplicacy of impedance data. The measurement
impedance spectra. model is composed of a superposition of line shapes which
Knowledge of the error structure is also needed for anal- can be chosen arbitrarily. A model composed of Voigt ele-
ysis of experimental data. For example, proper weighting ments in series with a solution resistance has been shown to
during nonlinear regression of a model to impedance data be a useful and general measurement model (see Fig. 1 in
is necessary to get unbiased parameter estimates. ~~ A Ref. 8). While the line shape parameters may not be associ-
weighted least squares strategy that includes the errors in ated unequivocally with a set of deterministic or theoreti-
the real and imaginary parts of the impedance is given by cal parameters for a given system, the measurement-model
minimization of approach has been shown to represent adequately the
impedance spectra obtained for a large variety of electro-
(Zr, k - Zr, k) 2 (Zj,k -- Zj,k) 2
J= E ~ +~ -' 2 ' [11 chemical systems.8 The characteristic time constants for
k O'r,k O'j,k the line shapes and their dispersion are low frequency as
where Zr,k and Zi,k represent the real and imaginary part of compared to the noise. The line shape models therefore can
the data, respectively, the caret signifies the corresponding be said to represent the low-frequency stationary compo-
model value, and cry,k and %,~ are the real and imaginary part nents of the impedance spectra (in a Fourier sense). Re-
of the standard deviation at each frequency. The use of the gard]ess of their interpretation, the measurement model
variance to weight data ensures that data points with "low representation can be used to filter and thus identify the
noise" content are emphasized and the data points with nonstationary (drift) and high frequency (noise) compo-
"high noise" content are de-emphasized. Use of the experi- nents contained in the same impedance spectrum.
mentally determined variance for weighting has been It is not obvious that such an approach should work. It is
shown to increase the amount and quality of the informa- well known, for example, that the impedance spectrum as-
tion that can be obtained from impedance measurements3 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
Classification of Measurement Errors resistance-capacitance (RC) circuits in series (the Voigt
The residual errors (e~J that arise when a model is re- model). In theory, then, a measurement model based on the
gressed to experimental data can be described as being Voigt circuit should require an infinite number of parame-
composed of deterministic (systematic) and stochastic ters to describe adequately the impedance response of any
J. Electrochem. Sac., Vol. 142, No. 12, December 1995 9 The Electrochemical Society, Inc. 4151
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
1000
I 0.2 . . . . ''I ....... '1 " ...... I ' " ' " 1 ' ....... I ' " ' " "
500
~ L
r--
N
-500
:
:. oo
oo
N
I
0 . 0 ~
I L.
N
-1000 , , , I , , , ,
104 . . . . . . . . . ' ........ ' ........ ' ........ ' ....... ' ........ 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
10 2 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-
- -2--;'-- 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
b
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-
I 0 - 4 _ 'I ....... I ........ j ........ i ........ i . . . . . . nary vs. real departures from the mean residual error,
10 1 0 0 1 0 1 1 0 2 1 0 `5 1 0 4 10 5 shown in Fig. 10, further suggests that the residual errors at
a given frequency are not correlated. The cyclic behavior of
Frequency, Hz the residual errors with frequency is caused by the lack of
fit of the model. The technique of estimating the standard
Fig. 7. Unfiltered standard deviation of the data shown in Fig. 3 as
deviation of the stochastic contribution by calculating the
a function of frequency. Circles represent the real, and the triangles
represent the imaginary part of the standard deviation. The solid line standard deviation of the residual errors is in effect a filter
represents the error structure for the n-GaAs sample held at 320 K for lack of fit. Using the measured error structure to weight
(see Eq. 14). The dashed line represents the contribution that is pro- the regression of data that are free of bias errors, the
portional to I Z.iI, the dashed-dot line represents the contribution that residual errors for the measurement model can be made to
is proportional to I Zrl, and 2the dashed-dot-dot-dot line is the cantri- be of the same order as the standard deviation of the
bution proportional to IZI /R,. The discontinuity apparent in the measurement. 9
solid line and the dash-dot-dot-dot line is caused by a change of the The real and imaginary standard deviations of the resid-
value of the current measuring resistor at 100 Hz. 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
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- Discussionand Conclusion
tions for the real and imaginary components of the im-
pedance was used, therefore, to weight the regression for The measurement model provides m u c h more than a pre-
identification of the error structure. In addition, three- and liminary analysis of impedance data in terms of the number
five-point moving averages were used to increase the sam- of resolvable time constants and asymptotic values, as sug-
ple size for the standard deviation while retaining the gen- gested, for example, by Zoltowski. 23'24As shown here, the
eral trends. The weighted X2 statistic (normalized by the measurement model can be used as a filterfor lack of repli-
degrees of freedom for the regression) cacy that allows accurate assessment of the standard devi-
ation of impedance measurements. As is discussed by m a n y
1 (Yexpt~i-- Ymodel,i)2 authors, 1'15,2~this information is critical for selection of
i weighting strategies for regression. This information also
provides a quantitative basis for assessment of the quality
was improved for regressions using a moving average value
of fits and can guide experimental design. In the subse-
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 -
quent paper of this series,9 the measurement model is used
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
to assess the bias component of the error structure.
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 The results presented here show that impedance mea-
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 surements are heteroskedastic (in the sense that the stan-
moving average should depend on the sampling rate. The dard deviations are funetior~s of frequency). In spite of the
five point moving average worked well for the ten points/ apparent complexity, a definite structure for the errors is
decade sampling rate used here. The increased standard resolved. In the impedance plane, the standard deviations
deviation for the parameter estimates obtained using the for the real and imaginary parts of the impedance are
moving average reflects the corresponding increased value equal, even at frequencies sufficientlyhigh or low that the
for the average standard deviation used to weight the re- imaginary part of the impedance asymptotically ap-
gression. B y giving a better estimate for the standard devi- proaches zero. This result is consistent with the results pre-
ation of the fitted quantity, the moving-ayerage approach sented by Zoltowski. 17
yielded more reliable estimates for the confidence intervals Aside from the obvious impact on the parsimony of the
of the error-structure model parameters. model for the error structure (only three parameters are
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 10 6 ' '""9 ........ I . . . . . . . '1 ' '"'"'1 ' ' .....
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 10 4
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. -
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 , . ...
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