Journal of The Electrochemical Society, 156 共10兲 C345-C351 共2009兲 C345

0013-4651/2009/156共10兲/C345/7/$25.00 © The Electrochemical Society

On the Sensitivity of the Kramers–Kronig Relations

to Nonlinear Effects in Impedance Measurements
Bryan Hirschorn* and Mark E. Orazem**,z
Department of Chemical Engineering, University of Florida, Gainesville, Florida 32611, USA

The Kramers–Kronig relations, which strictly apply for systems that are linear, stable, and causal, provide an essential tool for
assessing the internal consistency of impedance data. The Kramers–Kronig relations are understood to be sensitive to failures of
causality, but insensitive to failures of linearity. Numerical simulations were performed to explore the conditions under which the
Kramers–Kronig relations are sensitive to nonlinear behavior of electrochemical systems. A characteristic frequency was identified
below which the Kramers–Kronig relations are satisfied, even for very nonlinear systems. This result is observed for systems with
a small ohmic resistance. The Kramers–Kronig relations are not satisfied for measurements which include the characteristic
frequency. For systems with a large ohmic resistance, the Kramers–Kronig relations may provide a better tool for assessing the
presence of nonlinear behavior as compared to analysis of low frequency Lissajous plots.
© 2009 The Electrochemical Society. 关DOI: 10.1149/1.3190160兴 All rights reserved.

Manuscript submitted April 27, 2009; revised manuscript received July 6, 2009. Published August 7, 2009.

The Kramers–Kronig relations, derived for systems that can be
assumed to be linear, stable, and causal, have proven useful for
confirming the self-consistency of electrochemical impedance data.
Failure of impedance data to satisfy the Kramers–Kronig relations at
high frequencies can generally be attributed to instrumental artifacts,
and low frequency deviations can be attributed to nonstationary be-
havior. Instrumental artifacts and nonstationary behavior represent
violations of causality.
While assumption of linearity is essential for the derivation of
the Kramers–Kronig relations, the Kramers–Kronig relations are
generally considered to be insensitive to nonlinear behavior in elec-
trochemical systems.1 Urquidi-Macdonald et al.2 used experimental
data to show that the Kramers–Kronig transforms are highly sensi-
tive to the condition of causality and are insensitive to the condition
of linearity. Their evaluation of the effect of the linearity condition
on the Kramers–Kronig transforms was accomplished by varying
the amplitude of the input potential perturbation signal during sub-
sequent impedance scans for the corrosion of iron in a 1 M H2SO4
solution. For the largest amplitudes, the magnitude of the impedance
decreased significantly from the small amplitude case, indicating
violation of the linearity condition for their system. The data were
nevertheless shown to remain consistent with Kramers–Kronig
transforms for all input amplitudes tested. The result showed that the
Kramers–Kronig relations were insensitive to the condition of lin-
earity which was clearly violated for large perturbation inputs.
Urquidi-Macdonald et al. attributed the cause of this insensitivity to
an equal decrease in the real and imaginary components of the im-
pedance when the perturbation amplitude was increased and to the
ability of the frequency response analyzer to reject harmonics.
The issue of nonlinearity in impedance measurements is impor-
tant. Use of an input perturbation that is too large yields an incorrect
value for the charge-transfer resistance. While the Kramers–Kronig
relations have not been found useful for assessing the appearance of
nonlinearity, experimental methods, such as examination of low fre-
quency Lissajous plots, can be used to identify a nonlinear
response.3,4 Application of a random phase multisine input can be
used to resolve nonlinear contributions to the error structure of im-
pedance measurements.5-7
The objective of this work is to use numerical simulations to
Theoretical Approach
The nonlinear response in electrochemical systems typically re-
sults from the potential dependence of faradaic reactions. For ex-
ample, both Tafel and Butler–Volmer reaction kinetics display an
exponential dependence on the interfacial potential. The total current
passed through the electrode contributes to charging the interface
and to the faradaic reaction. These contributions are presented in
parallel in the circuit in Fig. 1a, where the use of a box for the
faradaic reaction is intended to emphasize the complicated and non-
linear potential dependence. The addition of an ohmic electrolyte
resistance caused the interfacial potential V to be smaller than the
applied potential U. This effect is illustrated in Fig. 1b.
The applied potential U can be expressed as a sinusoidal pertur-
bation about a steady value Ū as
U = Ū + ⌬U cos共␻t兲 关1兴
where ⌬U is the input amplitude, ␻ is the input frequency, and t is
the time. In the absence of an ohmic resistance, as shown in Fig. 1a,
the circuit potential U and the interfacial potential V are equal. In the
presence of an ohmic resistance Re, the applied cell potential is
related to the interfacial potential by
U = V + 共if + iC兲Re 关2兴
where if is the faradaic current density and iC represents the charging
current.
The faradaic current density can be expressed as
if = i0兵exp关ba共V − V0兲兴 − exp关− bc共V − V0兲兴其 关3兴
or equivalently
if = Ka exp共baV兲 − Kc exp共− bcV兲 关4兴
where ba and bc are the anodic and cathodic coefficients with units
of inverse potential and K includes the exchange current i0 and the
equilibrium potential difference V0 as Ka = i0 exp共−baV0兲 and Kc
= i0 exp共bcV0兲.
(a) d
(b) d

identify the conditions under which the Kramers–Kronig relations

are sensitive to nonlinear behavior. A related objective is to provide
an explanation for the lack of sensitivity of the Kramers–Kronig
relations to nonlinear behavior reported by Urquidi-Macdonald et e

al.2 t t

* Electrochemical Society Student Member. Figure 1. Circuit representations of kinetic models used in the present work:
** Electrochemical Society Fellow. 共a兲 faradaic system with no ohmic resistance and 共b兲 faradaic system with an
z
E-mail: meo@che.ufl.edu ohmic resistance.

Downloaded on 2019-04-02 to IP 212.118.132.162 address. Redistribution subject to ECS terms of use (see ecsdl.org/site/terms_use) unless CC License in place (see abstract).
 C346 Journal of The Electrochemical Society, 156 共10兲 C345-C351 共2009兲

Table I. Simulation results used to explore the role of the Kramers–Kronig relations for nonlinear systems with parameters ⌬U = 100 mV,
Cdl = 20 ␮FÕcm2, Ka = Kc = 1 mAÕcm2. ba = bc = 19 V−1, and V̄ = 0 V.

Re 共⍀ cm2兲 Rt,0 共⍀ cm2兲 ⌬U* Rt,obs 共⍀ cm2兲 Rt,obs /Rt,0 Re /Rt,obs

0 26.3 9.50 17.3 0.658 0

0.01 26.3 9.50 17.3 0.658 5.8 ⫻ 10−4
1 26.3 9.15 18.0 0.684 5.6 ⫻ 10−2
100 26.3 1.98 25.8 0.981 3.9

The value of the charge-transfer resistance at a given potential
V共t兲 can be calculated from the slope of the interfacial polarization
curve
Rt共t兲 = 兵Kaba exp关baV共t兲兴 + Kcbc exp关− bcV共t兲兴其−1
The linear value of the charge-transfer resistance is given as
Rt,0 = 关Kaba exp共baV̄兲 + Kcbc exp共− bcV̄兲兴−1
where V̄ represents the steady-state potential at which the impedance
measurement was made.
An analytic expression for current density as a function of ap-
pliance was achieved by comparing the predicted value from Eq. 7
to the real component of the simulated data.
Results
关5兴
The simulation results used to explore the role of the Kramers–
Kronig relations for nonlinear systems are summarized in Table I.
For each simulation, the system parameters gave rise to a linear
关6兴 charge-transfer resistance Rt,0 = 26.3 ⍀ cm2. The ohmic resistance
Re was varied from 0 to 100 ⍀ cm2. A scaled potential perturbation
can be defined following Hirschorn et al.4 as
⌬U

plied potential U = V can be obtained for the system without ohmic
resistance shown in Fig. 1a. The current and potential terms cannot
be separated in the more general case given in Fig. 1b, and a nu-
merical method must be employed. The numerical method used to
estimate the solution of the circuit shown in Fig. 1b is described in
a related paper.4 The impedance response of the circuits shown in
Fig. 1a and b was calculated directly for each frequency using Fou-
rier integral analysis.3
The simulated impedance data were tested for compliance with
the Kramers–Kronig relations using the measurement model analy-
sis. The procedure for the determination of the Kramers–Kronig
consistency recommended by Agarwal et al. was to fit the imaginary
component of impedance data to a measurement model of sequential
Voigt elements and then predict the real component of impedance
from the extracted parameters.8-12 In the present work, the
frequency-dependent charge-transfer resistance that results from the
influence of nonlinearity prevented accurate regression of the data to
the imaginary-only component. As a result of this limitation, the
analysis in the next section was based on best-fit complex regression
of simulated impedance data to a measurement model. Sequential
Voigt elements were added to the measurement model until the ad-
dition of an element did not result in an improvement of the fit
within a confidence of 95%. Because the measurement model is
inherently consistent with the Kramers–Kronig transforms, data that
fall within the confidence interval of a regressed model transformed
successfully. Nonconformity with the measurement model indicated
noncompliance and, therefore, violation of linearity.
In addition to the measurement model analysis, the simulated
impedance data were directly tested for compliance with the
Kramers–Kronig transformations. The form of the Kramers–Kronig
integrals used is given by
冑
⌬U* = 关8兴
K ab a + K cb c
0.2 3 共1 + Re /Rt,obs兲
Kab3a + Kcbc
where a value ⌬U* = 1 yields an almost linear response, resulting in
an error of less than 0.5% in the measured charge-transfer resis-
tance, and Rt,obs is the observed charge-transfer resistance measured
at low frequency. The values of the scaled potential perturbation
given in Table I reflect the influence of ohmic resistance on the
interfacial potential ⌬Vmax resulting from an applied potential ⌬U.
As discussed extensively in Ref. 4 and 13-20 the large potential
perturbation causes an error in the observed charge-transfer resis-
tance. The magnitude of the effect can be assessed by using the
dimensionless ratio Rt,obs /Rt,0. The magnitude of the induced errors
depends on the ohmic resistance,4 i.e., the ratio Rt,obs /Rt,0 ap-
proaches unity as Re /Rt,obs increases. The error in the impedance
response caused by a large input signal is shown in Fig. 2 for the
simulations with Re = 1 ⍀ cm2.
Simulated data generated from the systems presented in Table I
were analyzed for consistency with the Kramers–Kronig relations.
Under these conditions, the introduced input amplitude of ⌬U
= 100 mV caused significant errors in the impedance response. The
simulated data were analyzed using both the measurement model
approach and direct evaluation of the Kramers–Kronig integrals, as
described in the previous section.
The residual errors resulting from a measurement model fit to
simulated impedance data generated from the system with no ohmic
resistance Re = 0 are shown in Fig. 3a and b for the real and imagi-
15
306 Hz

冕
⬁
2 xZj共x兲 − ␻Zj共␻兲
Zr共␻兲 = Zr,⬁ − dx 关7兴 442 Hz
2

␲ x2 − ␻2 10
-Zj / Ωcm

where Zj共␻兲 is an analytic function of the imaginary component of

impedance and Zr,⬁ is an adjustable parameter representing the value 5 ∆U = 1 mV
∆U = 10 mV
of the ohmic resistance. The utility of Eq. 7 is that the real compo-
∆U = 100 mV
nent of impedance Zr共␻兲 can be predicted from an analytical func-
tion of the imaginary component if the conditions of linearity, sta- 0
0 5 10 15 20 25 30
bility, and casuality are not violated. The integral expressed in Eq. 7 2
was evaluated by inserting the imaginary component Zj共x兲 of the Zr / Ωcm
simulated nonlinear impedance data into the integrand and then per- Figure 2. Calculated impedance response with applied perturbation ampli-
forming a numerical integration at each frequency. This allowed for tude as a parameter for Re = 1 ⍀ cm2 and system parameters presented in
the prediction of the real component Zr共␻兲. The test for data com- Table I.

Downloaded on 2019-04-02 to IP 212.118.132.162 address. Redistribution subject to ECS terms of use (see ecsdl.org/site/terms_use) unless CC License in place (see abstract).
 Journal of The Electrochemical Society, 156 共10兲 C345-C351 共2009兲 C347

8 3

2
-13

-13
4 Figure 3. Residual error resulting from a
(Zr,s-Zr,m) / Ωcm x 10

(Zj,s-Zj,m) / Ωcm x 10
1 measurement model fit Zm to simulated
impedance data Zs for the system with
2

2
0
0 Re = 0 ⍀ cm2: 共a兲 real and 共b兲 imaginary
parts. The lines correspond to the 95.4%
-4 -1 共2␴兲 confidence interval for the regres-
sion. The system parameters presented in
-2 Table I give rise to Rt,obs /Rt,0 = 0.658 and
-8 Re /Rt,obs = 0.
-3
-5 -3 -1 1 3 5 7 -5 -3 -1 1 3 5 7
10 10 10 10 10 10 10 10 10 10 10 10 10 10
(a) f / Hz (b) f / Hz

nary parts, respectively. The dashed line represents the confidence side the confidence interval for the regressed model, is evident for
interval for the regressed model. All of the residual errors fell within frequencies greater than 103 Hz. The magnitude of the scaled re-
the confidence interval, suggesting that the Kramers–Kronig rela- sidual errors is larger than seen in Fig. 4. In addition, the residual
tions were satisfied. The magnitude of the residuals, on the order of errors fall outside the confidence interval at a lower frequency as
10−13, show that the measurement model could fit the data to within compared to Fig. 4.
12 significant digits. The data were shown to satisfy the Kramers– The residual errors resulting from a measurement model fit to
Kronig relations, even though the errors due to a nonlinear response simulated impedance data generated from the system with Re
were very large, i.e., Rt,obs /Rt,0 = 0.658. = 100 ⍀ cm2 are shown in Fig. 6a and b. Due to the large ohmic
The real and imaginary parts of the normalized residual errors
resistance, the error due to nonlinearity was small, i.e., Rt,obs /Rt,0
resulting from a measurement model fit to simulated impedance data
= 0.981. Nevertheless, the normalized residual errors fell outside the
generated from the system with Re = 0.01 ⍀ cm2 are shown in Fig.
confidence interval for all frequencies.
4a and b, respectively. The residual errors fell outside the confidence
interval at frequencies greater than 105 Hz. The scaled values of The measurement model analysis of consistency with the
10−4 at high frequencies show that the deviations from Kramers– Kramers–Kronig relations presented in Fig. 3-6 was complemented
Kronig relations are in the fourth significant digit, which may not be by an independent analysis using direct evaluation of the Kramers–
visible for experimental data. Nevertheless, the simulation results do Kronig integral Eq. 7. As shown in Fig. 7, the real part of the
not conform to the Kramers–Kronig transforms. impedance predicted from Eq. 7 agreed perfectly with the simulation
The real and imaginary parts of the normalized residual errors value for the system with no ohmic resistance Re = 0 ⍀ cm2. This
resulting from a measurement model fit to simulated impedance data result agrees with the result presented in Fig. 3, showing that even a
generated from the system with Re = 1 ⍀ cm2 are shown in Fig. 5a very nonlinear impedance response yielding Rt,obs /Rt,0 = 0.658 is
and b, respectively. The deviation from consistency with the consistent with the Kramers–Kronig relations for an ohmic resis-
Kramers–Kronig relations, marked by residual errors that fall out- tance equal to zero. For the system with Re = 1 ⍀ cm2, the real

2.5 1.4

2.0 1.2
-4

Figure 4. Normalized residual errors re-

-4
(Zr,s-Zr,m)/Zr,s / 10

1.5 1.0
(Zj,s-Zj,m)/Zj,s / 10

sulting from a measurement model fit Zm

1.0 0.8 to simulated impedance data Zs for the
0.5 0.6 system with Re = 0.01 ⍀ cm2: 共a兲 real and
共b兲 imaginary parts. The lines correspond
0.0 0.4
to the 95.4% confidence interval for the
-0.5 0.2 regression. The system parameters pre-
sented in Table I give rise to Rt,obs /Rt,0
-1.0 0.0
= 0.658 and Re /Rt,obs = 5.8 ⫻ 10−4.
-1.5 -0.2
-4 -2 0 2 4 6 8 -4 -2 0 2 4 6 8
10 10 10 10 10 10 10 10 10 10 10 10 10 10
(a) f / Hz (b) f / Hz

1.0

1.0
0.5
-2

Figure 5. Normalized residual errors re-

(Zr,s-Zr,m)/Zr,s / 10

-2
(Zj,s-Zj,m)/Zj,s / 10

sulting from a measurement model fit Zm

0.0 to simulated impedance data Zs for the
0.5 system with Re = 1 ⍀ cm2: 共a兲 real and
-0.5 共b兲 imaginary parts. The lines correspond
to the 95.4% confidence interval for the
regression. The system parameters pre-
-1.0
0.0 sented in Table I give rise to Rt,obs /Rt,0
= 0.684 and Re /Rt,obs = 5.6 ⫻ 10−2.
-1.5
-4 -2 0 2 4 6 8 -4 -2 0 2 4 6 8
10 10 10 10 10 10 10 10 10 10 10 10 10 10
(a) f / Hz (b) f / Hz

Downloaded on 2019-04-02 to IP 212.118.132.162 address. Redistribution subject to ECS terms of use (see ecsdl.org/site/terms_use) unless CC License in place (see abstract).
 C348 Journal of The Electrochemical Society, 156 共10兲 C345-C351 共2009兲

1.5
2

1.0 Figure 6. Normalized residual errors re-

-4

-2
1
(Zr,s-Zr,m)/Zr,s / 10

(Zj,s-Zj,m)/Zj,s / 10
sulting from a measurement model fit Zm
to simulated impedance data Zs for the
0 0.5 system with Re = 100 ⍀ cm2: 共a兲 real and
共b兲 imaginary parts. The lines correspond
to the 95.4% confidence interval for the
-1
0.0 regression. The system parameters pre-
sented in Table I give rise to Rt,obs /Rt,0
-2 = 0.981 and Re /Rt,obs = 3.9.
-0.5
-4 -2 0 2 4 6 8 -4 -2 0 2 4 6 8
10 10 10 10 10 10 10 10 10 10 10 10 10 10
(a) f / Hz (b) f / Hz

component of impedance predicted from Eq. 7 deviated from the The influence of the transition frequency can be seen in Fig. 10
real component of the simulated data, indicating noncompliance where the normalized real part of the impedance is presented as a
with the Kramers–Kronig transforms due to violation of linearity. function of normalized frequency for the system with Re
This result agrees with the results presented in Fig. 5. = 1 ⍀ cm2. The frequency is scaled by the frequency characteristic
The direct integration of Eq. 7 was also able to reveal an incon- of the Rt,0C time constant, and the real part of the impedance is
sistency with the Kramers–Kronig relations for the system with
Re = 100 ⍀ cm2, as shown in Fig. 8. The percent error in the low
frequency region, corrected for ohmic resistance, is 4 times greater
for the Re = 100 ⍀ cm2 system as compared to the Re = 1 ⍀ cm2
system shown in Fig. 7. For the system with Re = 0.01 ⍀ cm2, how- 20 simulated data
predicted
ever, direct integration of Eq. 7 did not reveal the inconsistencies
with the Kramers–Kronig relations shown by the measurement 16 19.2
model analysis in Fig. 4. This discrepancy may be regarded to be a 19.1
2

testimony to the sensitivity of the measurement model analysis for

Zr / Ωcm

failures of consistency with the Kramers–Kronig relations. 12 19.0

17.3
Discussion 8 17.2
The sensitivity of the Kramers–Kronig transforms to nonlinearity 17.1 Re = 1 Ωcm
2
-4 -3 -2
clearly depends on both the magnitude of the errors Rt,obs /Rt,0 and 4 10 10 10
on the ohmic resistance. The objective of the following section is to
Re = 0
identify the conditions under which the Kramers–Kronig relations
may detect errors caused by a nonlinear impedance response. 0 -6 -4 -2 0 2 4 6 8
10 10 10 10 10 10 10 10
Influence of transition frequency.— The maximum variation in f / Hz
interfacial potential ⌬Vmax corresponding to an input perturbation Figure 7. A comparison of simulation results to the real component of im-
⌬U = 0.1 V is given in Fig. 9a for the simulations presented in pedance predicted using Eq. 7 for the systems with Re = 0 ⍀ cm2 and Re
Table I. One influence of the ohmic resistance is seen at low fre- = 1 ⍀ cm2. The system parameters presented in Table I give rise to
quencies, where Rt,obs /Rt,0 = 0.658 and Rt,obs /Rt,0 = 0.684, respectively. In the absence of
ohmic resistance, the simulated data and the predicted values are equal.
⌬U
lim ⌬Vmax = 关9兴
␻→0 共1 + Re /Rt,obs兲
130
The presence of the ohmic resistance further decreases the perturba- simulated data
tion amplitude at higher frequencies where the interfacial impedance 120 126.5 predicted
becomes small and the role of the faradaic current is diminished. 126.0
As previously reported,4 the frequency dependence of the inter- 110 125.5 2
Re = 100 Ωcm
facial potential causes a corresponding change in the apparent
19.5
2
Zr / Ωcm

charge-transfer resistance, as shown in Fig. 9b. The charge-transfer

resistance was calculated using Eq. 5 for each time-dependent value 100 19.0
of V generated during the development of synthetic data. At each 18.5 -4 -3 -2
frequency, the charge-transfer resistance was averaged over a com- 10 10 10
20
plete sinusoidal cycle yielding the effective charge-transfer resis-
2
tance, which at low frequency is approximately the observed charge- Re = 1 Ωcm
transfer resistance Rt,obs. The frequency characteristic of the
10
transformation shown in Fig. 9 from low frequency behavior to high
frequency behavior was given by Hirschorn et al. to be4 0 -6

冋 册
-4 -2 0 2 4 6 8
10 10 10 10 10 10 10 10
1 1 1 Rt,obs
ft = + = 1+ 关10兴 f / Hz
2␲Rt,obsCdl 2␲ReCdl 2␲Rt,obsCdl Re
Figure 8. A comparison of simulation results to the real component of im-
where Rt,obs is the observed charge-transfer resistance at the given pedance predicted using Eq. 7 for the systems with Re = 1 ⍀ cm2 and Re
perturbation amplitude. The transition frequency depends on the di- = 100 ⍀ cm2. The system parameters presented in Table I give rise to
mensionless ratio Rt,obs /Re and is given in units of hertz. Rt,obs /Rt,0 = 0.684 and Rt,obs /Rt,0 = 0.981, respectively.

Downloaded on 2019-04-02 to IP 212.118.132.162 address. Redistribution subject to ECS terms of use (see ecsdl.org/site/terms_use) unless CC License in place (see abstract).
 Journal of The Electrochemical Society, 156 共10兲 C345-C351 共2009兲 C349

0.10
ft = 10
5.9 26
3.9
Re/Rt,obs = 0 ft = 10
0.08 Re/Rt,obs = 0
-4
Re/Rt,obs = 5.8×10 -4
Re/Rt,obs = 5.8×10
-2 24 Figure 9. Interfacial parameters as func-

2
Re/Rt,obs = 5.6×10
∆Vmax / V

Rt / Ωcm
-2
0.06 Re/Rt,obs = 5.6×10
Re/Rt,obs = 3.9 tions of frequency for the simulations pre-
Re/Rt,obs = 3.9
ft = 10
3.9
sented in Table I: 共a兲 maximum variation
0.04
22
ft = 10
5.9
in the interfacial potential and 共b兲 the ef-
2.6 2.6
ft = 10 ft = 10 fective charge-transfer resistance. Vertical
0.02 lines correspond to the transition fre-
quency given by Eq. 10.
20
0.00

-3 -1 1 3 5 7 -3 -1 1 3 5 7
10 10 10 10 10 10 10 10 10 10 10 10
(a) f / Hz (b) f / Hz

corrected for the ohmic resistance and scaled by the Rt,obs seen at
low frequency. At the transition frequency given by Eq. 10, the
effective charge-transfer resistance changes from the low frequency
value Rt,obs, which is affected by the nonlinear response, to the linear
value Rt,0.
The change in apparent charge-transfer resistance has an influ-
ence as well in the imaginary part of the impedance. The normalized
impedance responses are presented as functions of normalized fre-
quency for the systems with Re = 0.01, 1, and 100 ⍀ cm2 in Fig.
11a and b for the real and imaginary parts, respectively. The solid
curve is the ideal linear response and the dashed curves are the
nonlinear impedance responses arising from a large input amplitude
of ⌬U = 100 mV for system parameters presented in Table I. As
shown in Fig. 11a, the real component of impedance is distorted
from the ideal linear response for the Re = 1 ⍀ cm2 and Re
= 0.01 ⍀ cm2 systems. Distortion is also present for the Re
= 100 ⍀ cm2 system; however, it is not visually evident in Fig. 11a
due to the small deviation of the observed charge-transfer resistance
from the linear value, i.e., Rt,obs /Rt,0 = 0.981. The distortion from
the linear response occurs at the transition frequency described by
Eq. 10. As shown in Fig. 11b, the imaginary component of imped-
ance is distorted from the ideal linear response for the Re
= 1 ⍀ cm2 and Re = 100 ⍀ cm2 systems. Distortion is not evident
for the Re = 0.01 ⍀ cm2 system. For the case with Re
= 0.01 ⍀ cm2 the transition frequency was f t = 8 ⫻ 105 Hz which
was in the calculated range of frequencies, but the transition to Rt,0
takes place in a frequency range where the current is predominately
charging and the value of the charge-transfer resistance is inconse-
quential.
In contrast, no distortion of the impedance response is seen for
the case in which Re = 0. The normalized impedance response for
the system with Re = 0 is presented as a function of normalized
frequency in Fig. 12a and b for the real and imaginary parts, respec-
tively. Both the ideal linear response and the nonlinear impedance
response are superposed in spite of the large potential amplitude
applied, yielding a value Rt,obs /Rt,0 = 0.658.
lations are also satisfied if the departure from linear behavior is
sufficiently small that Rt,obs /Rt,0 ⬇ 1. The Kramers–Kronig relations
are violated for conditions where Rt,obs /Rt,0 ⫽ 1 and the transition
frequency given by Eq. 10 falls within the experimental frequency
range. In such a case, the varying effective charge-transfer resistance
is the mechanism that causes the Kramers–Kronig relations to fail,
just as a time-dependent Rt would cause the Kramers–Kronig rela-
tions to fail due to violation of the condition of causality.
There exists an interesting balance of effects for the systems with
a large ohmic resistance. The presence of a large ohmic resistance
reduces the portion of the applied potential perturbation that contrib-
utes to the interfacial potential and therefore reduces the departure
of Rt,obs /Rt,0 from unity. At the same time, the transition frequency
approaches 1/Rt,obsCdl, thus making any departure from linear be-
havior detectable by use of the Kramers–Kronig relations. In these
cases, the use of Lissajous figures at low frequencies may be less
sensitive to nonlinear behavior as compared to the use of the
Kramers–Kronig relations. For example, the Lissajous analysis of
the Re = 100 ⍀ cm2 system, shown in Fig. 7 of the work by Hir-
schorn et al.,4 did not detect the presence of nonlinearity, while the
Kramers–Kronig analysis shown in Fig. 8 did detect nonlinear be-
havior. A further increase in the ohmic resistance eventually leads to
an approximately linear response that is in compliance with the
Kramers–Kronig relations.
For small values of ohmic resistance, the transition frequencies
are significantly greater than 1/Rt,obsCdl and the transition to Rt,0
takes place in a frequency range where the value of the charge-
transfer resistance has a negligible influence on the imaginary im-
0
10
-2 2
10 Rt,obs/(1+(ωRt,obsC) )
(Zr-Re)/Rt,obs

The work presented here demonstrates that the sensitivity of the 10

-4

Kramers–Kronig relations on the nonlinearity of an electrochemical

2
system depends on both the magnitude of the potential perturbation Rt,0/(1+(ωRt,0C) )
and the value of the transition frequency given by Eq. 10. The criti- 10
-6

cal parameters are ⌬U*, given by Eq. 8, Rt,obsCdl, and Re /Rt,obs. ft/fRt,0C
-2
When Re = 0, the transition frequency given by Eq. 10 is equal to Re/Rt,obs= 5.6×10
-8
infinity, and the impedance response is given by 10
∆U = 100 mV
Rt,obs
Z= 关11兴 -10
1 + j␻Rt,obsC 10 -2 0 2 4
10 10 10 10
where Rt,obs differs from Rt,0 but is independent of frequency. In this
case, the Kramers–Kronig relations are satisfied. When the ohmic f/fRt,0C
resistance is small and the transition frequency is outside the experi- Figure 10. The normalized real part of the impedance as a function of
mentally assessable range, the effective charge-transfer resistance is normalized frequency for the system with Re = 1 ⍀ cm2 共solid line兲. The
approximately independent of frequency, as shown in Fig. 9b, and dashed lines represent the ideal linear responses for systems with Rt,0
the Kramers–Kronig relations are satisfied. The Kramers–Kronig re- = 26.3 ⍀ cm2 and with Rt,obs = 18.0 ⍀ cm2.

Downloaded on 2019-04-02 to IP 212.118.132.162 address. Redistribution subject to ECS terms of use (see ecsdl.org/site/terms_use) unless CC License in place (see abstract).
 C350 Journal of The Electrochemical Society, 156 共10兲 C345-C351 共2009兲

2
Re/Rt,obs = 5.6×10
-2 10
0 Re/Rt,obs = 3.9
10 -4
Re/Rt,obs = 5.8×10 -2
Re/Rt,obs = 5.6×10
0
10
10
-2 Figure 11. The normalized impedance re-
sponse as functions of normalized fre-
(Zr-Re)/Rt,obs

Ideal ideal
-2
quency for the systems with Re = 0.01, 1,

-Zj/Rt,obs
-4 ft/fRC 10
10
and 100 ⍀ cm2: 共a兲 real and 共b兲 imaginary
-4 parts. The solid curve is the ideal linear
-6 10
10 response and the dashed curves are the
nonlinear impedance responses arising
10
-8 Ideal
-6
10 from a large input amplitude of ⌬U
ft/fRC
= 100 mV for system parameters pre-
-10 -8 sented in Table I.
10 -4 -2 0 2 4
10 -8 -6 -4 -2 0 2 4 6 8 10
10 10 10 10 10 10 10 10 10 10 10 10 10 10 10

(a) f/fRt,obsC (b) f/fRt,obsC

2
10
0
10
0
-2
10
10

-2
Zr/Rt,obs

Figure 12. The normalized impedance re-

-Zj/Rt,obs

-4 10
10
sponse as functions of normalized fre-
-4 quency for the system with Re = 0: 共a兲
-6 10
10 real and 共b兲 imaginary parts. Both the
ideal linear response and the nonlinear im-
-6
10
-8
10 pedance response are superposed.

-10 -8
10 -4 -2 0 2 4
10 -8 -6 -4 -2 0 2 4 6 8 10
10 10 10 10 10 10 10 10 10 10 10 10 10 10 10

(a) f/fRt,obsC (b) f/fRt,obsC

pedance. Therefore, it is difficult to detect discrepancies at low fre- earity associated with large perturbation amplitudes, was approxi-
quencies between the data and the predicted values using the mately frequency independent. As a result, the data complied with
Kramers–Kronig transform expressed by Eq. 7 . This agrees with the Kramers–Kronig relations.
results presented in Fig. 8, where the Re = 100 ⍀ cm2 system with It is worth asking, for systems for which the measured frequency
f t = 4 ⫻ 102 Hz is more sensitive to nonlinear behavior than is the range includes the transition frequency, whether the distortions as-
Re = 1 ⍀ cm2 system with f t = 8 ⫻ 103 Hz. sociated with nonlinear behavior are sufficiently large to be discern-
ible in experimental measurements. Normally distributed additive
stochastic errors with a mean value of zero and a standard deviation
Application to experimental systems.— The transition frequency of 0.1% of the modulus of the calculated impedance response were
given by Eq. 10 is presented in Fig. 13 as a function of RC time applied to the system with Re = 1 ⍀ cm2. This level of noise has
constant with Re /Rt as a parameter. The time constant for fast reac- been reported to be typical of impedance measurements.21,22 The
tions, such as the reduction in ferricyanide on a platinum electrode measurement model analysis for this system with an input potential
at an appreciable fraction of the mass-transfer-limited current den-
sity, can be on the order of 10−5 s. For these systems, the transition
frequency may fall outside the experimentally accessible frequency 5
range. The RC time constant for reactions near the equilibrium po- 10
tential may, however, be significantly larger. For these systems, the
transition frequency may fall within the experimental range, even 4
10
for small values of Re /Rt. Re/Rt = 0.001
Urquidi-Macdonald et al. reported, based on experimental obser-
vations, that the Kramers–Kronig relations are not sensitive to a 10
3

nonlinear system response.2 Their conclusions were based on ex- 0.01

ft / Hz

periments performed with different perturbation amplitudes on an

2
0.1
iron electrode in a 1 M H2SO4 electrolyte. They found that the 10
Kramers–Kronig relations were satisfied even for potential perturba- 1
tion amplitudes sufficiently large to cause measurable distortions in ∞
1
the impedance response. Their results can be placed into the context 10
of Fig. 13. System parameters Re = 2 ⍀ cm2, Cdl = 10 ␮F/cm2, and
Rt,obs = 14 ⍀ cm2 were estimated from the small-amplitude imped- 0
ance data from the published experimental results shown in Fig. 4 of 10
-4 -3 -2 -1 0 1 2
their work.2 The corresponding transition frequency was approxi- 10 10 10 10 10 10 10
mately 9000 Hz. Their experimental frequency range extended only τR C / s
to 5000 Hz, as shown in Fig. 5 of their work. Therefore, the transi- t

tion frequency was not in the measured frequency range, and the Figure 13. The transition frequency given by Eq. 10 as a function of RC
measured charge-transfer resistance, although in error due to nonlin- time constant with Re /Rt as a parameter.

Downloaded on 2019-04-02 to IP 212.118.132.162 address. Redistribution subject to ECS terms of use (see ecsdl.org/site/terms_use) unless CC License in place (see abstract).
 Journal of The Electrochemical Society, 156 共10兲 C345-C351 共2009兲 C351

3 3

2 2
Figure 14. Normalized residual errors re-
sulting from a measurement model fit Zm
-3

1 1

-3
to simulated impedance data Zs with nor-
(Zr,s-Zr,m)/|Z| / 10

(Zj,s-Zj,m)/|Z| / 10
mally distributed additive stochastic errors
0 0 with standard deviation of 0.1% of the
modulus for the system with Re
-1 -1 = 1 ⍀ cm2: 共a兲 real and 共b兲 imaginary
parts. The lines correspond to the 95.4%
-2 -2 confidence interval for the regression. The
input potential perturbation amplitude was
⌬U = 1 mV.
-3 -3
-4 -2 0 2 4 6 -4 -2 0 2 4 6
10 10 10 10 10 10 10 10 10 10 10 10
(a) f / Hz (b) f / Hz

8 6

4 4 Figure 15. Normalized residual errors re-

sulting from a measurement model fit Zm
-3

2
-3

to simulated impedance data Zs with nor-

(Zr,s-Zr,m)/|Z| / 10

(Zj,s-Zj,m)/|Z| / 10

0 2 mally distributed additive stochastic errors

with standard deviation of 0.1% of the
-2
modulus for the system with Re
-4 0 = 1 ⍀ cm2: 共a兲 real and 共b兲 imaginary
parts. The lines correspond to the 95.4%
-6
confidence interval for the regression. The
-8 -2 input potential perturbation amplitude was
-10
⌬U = 100 mV.
-4 -2 0 2 4 6 -4 -2 0 2 4 6
10 10 10 10 10 10 10 10 10 10 10 10
(a) f / Hz (b) f / Hz

perturbation amplitude of ⌬U = 100 mV is presented in Fig. 5 in the tal tests involving either repeated measurements with different per-
absence of added noise. The measurement model analysis is pre- turbation amplitudes or observation of nonlinear responses in low
sented in Fig. 14 for simulated data using an input potential pertur- frequency Lissajous plots.3,4 Inspection of Lissajous plots may be
bation amplitude of ⌬U = 1 mV. The normalized residual errors for less useful for systems with large values of Re /Rt,obs. In this case, the
both real and imaginary parts of the impedance, shown in Fig. 14a Kramers–Kronig relations may provide a more useful tool for detec-
and b, respectively, are distributed around zero, indicating that the tion of nonlinear responses to large potential perturbations.
Kramers–Kronig relations are satisfied.
The corresponding measurement model analysis using an input References
potential perturbation amplitude of ⌬U = 100 mV is presented in 1. D. D. Macdonald, Electrochim. Acta, 51, 1376 共2006兲.
Fig. 15. The normalized residual errors for both real and imaginary 2. M. Urquidi-Macdonald, S. Real, and D. Macdonald, Electrochim. Acta, 35, 1559
共1990兲.
parts of the impedance, shown in Fig. 15a and b, respectively, are 3. M. E. Orazem and B. Tribollet, Electrochemical Impedance Spectroscopy, John
not distributed around zero, indicating that the Kramers–Kronig re- Wiley & Sons, Hoboken, NJ 共2008兲.
lations are not satisfied. Thus, a Kramers–Kronig analysis based on 4. B. Hirschorn, B. Tribollet, and M. E. Orazem, Isr. J. Chem., 48, 133 共2008兲.
the measurement model detects nonlinearity of systems for which 5. E. V. Gheem, R. Pintelon, J. Vereecken, J. Schoukens, A. Hubin, P. Verboven, and
O. Blajiev, Electrochim. Acta, 49, 4753 共2006兲.
the measured frequency range includes the transition frequency,
6. O. Blajiev, T. Breugelmans, R. Pintelon, and A. Hubin, Electrochim. Acta, 51,
even when reasonable experimental error is present. 1403 共2006兲.
7. E. V. Gheem, R. Pintelon, A. Hubin, J. Schoukens, P. Verboven, O. Blajiev, and J.
Conclusions Vereecken, Electrochim. Acta, 51, 1443 共2006兲.
While the results presented here are consistent with the observa- 8. P. Agarwal, M. E. Orazem, and L. H. García-Rubio, J. Electrochem. Soc., 139,
1917 共1992兲.
tions reported by Urquidi-Macdonald et al.2 that the Kramers– 9. P. Agarwal, O. D. Crisalle, M. E. Orazem, and L. H. García-Rubio, J. Electrochem.
Kronig relations were insensitive to failures of linearity, this work Soc., 142, 4149 共1995兲.
also shows that, under appropriate conditions, the Kramers–Kronig 10. P. Agarwal, M. E. Orazem, and L. H. García-Rubio, J. Electrochem. Soc., 142,
relations provide a useful tool for detection of nonlinear system 4159 共1995兲.
11. M. E. Orazem, J. Electroanal. Chem., 572, 317 共2004兲.
responses. The sensitivity of the Kramers–Kronig relations on the 12. P. Shukla, M. Orazem, and O. Crisalle, Electrochim. Acta, 49, 2881 共2004兲.
nonlinearity of an electrochemical system depends on both the mag- 13. G. Popkirov and R. Schindler, Electrochim. Acta, 40, 2511 共1995兲.
nitude of the potential perturbation and on whether the transition 14. K. Darowicki, Electrochim. Acta, 40, 439 共1995兲.
frequency given by Eq. 10 falls within the experimental frequency 15. K. Darowicki, Electrochim. Acta, 42, 1781 共1997兲.
16. J. Diard, B. LeGorrec, and C. Montella, J. Electroanal. Chem., 432, 27 共1997兲.
range. The value of the transition frequency depends on Rt,obsCdl and 17. J. Diard, B. LeGorrec, and C. Montella, J. Electroanal. Chem., 432, 41 共1997兲.
Re /Rt,obs. The Kramers–Kronig relations will be violated for condi- 18. J. Diard, B. LeGorrec, and C. Montella, J. Electroanal. Chem., 432, 53 共1997兲.
tions where Rt,obs /Rt,0 ⫽ 1 and the transition frequency given by Eq. 19. J. Diard, B. LeGorrec, and C. Montella, Electrochim. Acta, 42, 1053 共1997兲.
10 falls within the experimental frequency range. 20. R. Milocco, Electrochim. Acta, 44, 4147 共1999兲.
21. M. E. Orazem, T. E. Moustafid, C. Deslouis, and B. Tribollet, J. Electrochem. Soc.,
For small values of Re /Rt,obs, the Kramers–Kronig relations may 143, 3880 共1996兲.
be of limited utility for detecting errors associated with a nonlinear 22. M. E. Orazem, M. Durbha, C. Deslouis, H. Takenouti, and B. Tribollet, Electro-
response. In this case, it will be more appropriate to use experimen- chim. Acta, 44, 4403 共1999兲.

Downloaded on 2019-04-02 to IP 212.118.132.162 address. Redistribution subject to ECS terms of use (see ecsdl.org/site/terms_use) unless CC License in place (see abstract).