Electrochimica Acta 188 (2016) 98–102

Contents lists available at ScienceDirect

Electrochimica Acta
journal homepage: www.elsevier.com/locate/electacta

On a possible physical origin of the constant phase element

Seyed Mohammad Rezaei Niyaa , Mina Hoorfarb,*
a
Stanford University, Mechanical Engineering Department, Stanford, CA 94305, USA
b
University of British Columbia, School of Engineering, Kelowna, V1V 1V7 BC, Canada

A R T I C L E I N F O A B S T R A C T

Article history: Despite the numerous use of the constant phase element (CPE) in the modeling of the impedance
Received 8 October 2015 characteristics of the electrochemical systems, the physical reasoning of this non-intuitive element is not
Received in revised form 28 November 2015 clear. In this paper, the CPE impedance is analytically calculated using the anomalous diffusion theory.
Accepted 28 November 2015
The fractional calculus and the anomalous diffusion are first reviewed. It is shown that the chance
Available online 2 December 2015
inequality in the random walk in a porous media can result in an anomalous diffusion. Then, the
Boltzmann distribution of the particles used in the Gouy-Chapman theory of the double layer is modified
Keywords:
to determine the double layer capacitance. Finally, the impedance of the double layer is calculated which
Constant phase element (CPE)
Anomalous diffusion
is equivalent to the CPE impedance reported in literature. It is shown that this novel theory covers the
Subdiffusion interpretations previously presented for the CPE and its relation to the fractal dimension and the pore
Gouy-Chapman theory size distribution of the porous media.
Double layer capacitance ã 2015 Elsevier Ltd. All rights reserved.

1. Introduction resultant semi-circle in the Nyquist plot. The Q parameter in Eq. (1)
is referred to as the magnitude of the CPE. Unlike the plain
The Nyquist plots resulted from the impedance measurements capacitor (which has the unit of Farad (F)), the parameter Q in
of different electrochemical systems such as biological membranes Eq. (1) has the unit of Fsa1 , where s stands for seconds.
[1], muscles [2] and a solid and porous electrode/electrolyte There is no consensus about the physics behind the CPE in
interface [3–5] reveal an impedance plot which cannot be obtained literature [3]. The CPE power (a) has been reported to be a function
by a combination of ordinary electrical elements (e.g., resistances of the surface roughness [6], fractal dimension [7,8] or the pore size
and capacitances). While the parallel connection of a capacitor and distribution [9] of the porous electrode. However, it has mostly
a resistor results in a semi-circle in the Nyquist plot which is assumed to be related to different time constants of the processes
symmetric around abscissa (i.e., the center of the semi-circle is on occurring in the porous electrode [e.g.,10–18]. The Voigt model
the x-axis), the measured Nyquist plots consist of depressed semi- (representing an equivalent circuit consisting of a series of blocks
circle whose center is below the x-axis. These curves can be of resistors and capacitors that are connected in parallel) is
produced by a parallel combination of an ordinary resistance and a normally used as a measurement model to study the CPEs
non-intuitive element called ‘constant phase element’ (CPE) which [e.g.,12,13]. While the Voigt model is an effective tool, it cannot
has the impedance in the form of represent the fundamentals of the origin of the CPEs. This can be
1 explained as follows: the Voigt model can be well fitted (with any
Z CPE ¼ ð1Þ desired accuracy) to the CPE curve (in the Nyquist plot) by adding
QðjvÞa
enough elements (e.g., resistors and capacitors) to the model. Each
pffiffiffiffiffiffiffi
where v is angular frequency and j ¼ 1. Here, a is a non- combined resistor and capacitor block in the Voigt model can be
dimensional number called the CPE power and 0 < a < 1. Clearly, interpreted as a phenomenon with a specific time constant.
the CPE becomes an ordinary capacitance if a ¼ 1. Also, the CPE However, this does not mean that the CPE is originally the result of
impedance becomes independent of the frequency and CPE these phenomena and time constants. In other words, the Voigt
responds as a plain resistance if a ¼ 0. Also, a ¼ 1=2 and a ¼ model can only present the CPE mathematically, but not physically.
1 result in Warburg and inductive elements, respectively. The It is necessary to mention that any impedance measurement which
farther the a parameter from unity, the more depressed the results in a curve in the Nyquist plot can theoretically be fitted
using the Voigt model [19].
In this paper, the physics behind the origin of the CPE is
* Corresponding author. Fax: +1 250 807 9850. presented using the anomalous diffusion concept [20]. As a result,
E-mail address: [email protected] (M. Hoorfar).

http://dx.doi.org/10.1016/j.electacta.2015.11.142
0013-4686/ ã 2015 Elsevier Ltd. All rights reserved.
 S.M. Rezaei Niya, M. Hoorfar / Electrochimica Acta 188 (2016) 98–102 99

the well-known impedance relation of the CPE (Eq. (1)) is derived. which is in agreement with the normal derivative definition when
It is shown that this new approach is a comprehensive framework n is a natural number. Fig. 1 shows different natural and fractional
covering all the previously mentioned interpretations [6–9] of the derivatives of a sample function (y ¼ x3 ). While the first (and
physics of the CPE. To obtain the CPE impedance relation, the second) derivative operators are normally considered as basic tools
double-layer theory is needed to be reconsidered using the to calculate the slope (and curvature) of a function, those can also
anomalous diffusion theory which itself needs the fractional be interpreted as discrete operators which convert a function to
calculus introduction. As this framework is not normally discussed another function, and the latter function is called the derivative of
in the CPE literature, the fractional derivative concept is first briefly the former one. Roughly speaking, this concept of fractional
introduced. The anomalous diffusion and the random walk theory derivative is used to convert this discrete operator to a continuous
are then briefly reviewed. The Gouy-Chapman theory of the one. As an example, Fig. 1 depicts that all the functions resulted
double-layer concept is then revisited using the anomalous from n th-order fractional derivation of the function y ¼ x3 will lie
diffusion concept. Finally, the CPE impedance is calculated. More between the y ¼ x3 and y ¼ dxd ðx3 Þ ¼ 3x2 curves for 0 < n < 1 for
specifically, the CPE power (a) is shown to be equal to the positive x values.
anomalous diffusion power (n). Finally, the following two properties of the Riemann-Liouville
fractional derivative are specifically needed to be mentioned here
2. Fractional derivative [21]

The Fourier transform of the nth-order derivative of a function 1

Dnt 1 ¼ tn ð6Þ
f ðtÞ is written as Gð1  nÞ
 n 
d f ðtÞ
= n ¼ ðjvÞn FðvÞ ð2Þ
dt Dnt tn1  0 ð7Þ
where =ff ðtÞg ¼ FðvÞ is the Fourier transform. While n is normally
considered as a natural number, relations similar to Eq. (2) can be 3. Anomalous diffusion and random walk theory
introduced to define nth-order derivative for real numbers (also
referred to as fractional derivative). In fact, this concept (first Random walk theory is in fact a framework which relates the
presented by Leibniz in 1695 [20]) is older than Fourier transform Brownian motion to the diffusion equation. The random walk
itself, as old as the normal derivation concept. The Riemann- theory describes the transport rate of a particle which moves
Liouville version of the n th-order fractional derivative of a function completely randomly in each time step (see Ref. [20] for the history
f ðtÞ is defined as [e.g.,20] of the theory). Using this theory, the mean square displacement of
p Z t a particle at a time t can be estimated to be [22]
1 d f ðxÞ
Dnt f ðtÞ ¼ dx ð3Þ D E
Gðp  nÞdtp 0 ðt  xÞ1pþn x2 ðtÞ ¼ 2Lt ð8Þ
where p is an integer number such that p  1 < ReðnÞ  p and G
where L is the diffusion coefficient. This linear dependence of the
denotes the gamma function. For 0 < n < 1, this operator can be
mean square displacement to the time is a characteristic of the
simplified as
Brownian motion [20]. However, the physical systems do not
Z
1 d t f ðxÞ always obey the above-mentioned relation; thus, the mean square
Dnt f ðtÞ ¼ dx ð4Þ
Gð1  nÞdt 0 ðt  xÞn displacement has to be generalized in the following form:
D E
As an example, the fractional derivative of a power function can x2 ðtÞ  tn ð9Þ
be determined as [21]
This model of transport is called anomalous diffusion. Specifically,
Gðk þ 1Þ
Dnt tk ¼ tkn ð5Þ it is called subdiffusion if 0 < n < 1 (slower than normal diffusion)
Gðk þ 1  nÞ and superdiffusion if n > 1 (faster than normal diffusion). There is
an extensive range of physical systems which show subdiffusive
and superdiffusive characteristics [20]. For instance, the charge
transfer in semiconductors and transport in fractal structures and
porous media show subdiffusive characteristics. On the other hand,
turbulent diffusion, bacterial motion and quantum optics are just
examples of superdiffusion transport. For the reasons which will
be clear hereafter, the subdiffusion process is in special interest
here.
In addition to Eq. (9), other interpretations have also been
presented for the subdiffusion power n . In a fractal dimension, the
power n can be shown to be [23,24]
ds
n¼
df

which ds and df are spectral (fracton) and fractal dimensions of the

structure, respectively [24]. Using the continuous time random
walk theory, another interpretation of n can be obtained. The
continuous time random walk theory is in fact the extension of
the random walk theory in which the length of the jumps of the
particle and also the waiting time between each jump is obtained
Fig. 1. The natural and fractional derivatives of the example function y ¼ x3 .
100 S.M. Rezaei Niya, M. Hoorfar / Electrochimica Acta 188 (2016) 98–102

using a probability distribution function [23]. This theory also approach [20], the 1-D version of the Fick’s second law is modified
results in the subdiffusion transport if the characteristic waiting as follows to cover the anomalous diffusion:
time between the jumps [20] decays as a power law when t ! 1.
The subdiffusion transport in a porous media can also be @2 C
Dnt C ¼ K ð11Þ
explained by a new approach presented in this paper. The 1-D @x2
random walk transport of particles in a porous medium having the where n is the subdiffusion power introduced in Eq. (9) and K is the
delta function distribution at the origin (x ¼ 0) at t ¼ 0 is generalized diffusion coefficient [23] and has the unit of m2sn . A
considered. Each particle has equal chance of moving towards diffusion time constant (t D ) can be defined to relate the
the origin (x ¼ 0) or away from it based on the Brownian motion. generalized diffusion coefficient K to the normal diffusion
This chance equality is conserved as long as the propagation path coefficient L (with the unit of m2s1).
of the particle is not blocked by the porous medium. In this case,
the chance of the particle moving away from the origin is nullified, K ¼ Lt 1n
D ð12Þ
while it still has the 50% chance of moving towards the origin (or
stops). In practice, the 3-D random motion of the particle will 4. Revision of the Gouy-Chapman theory
eventually move the particle to another path and makes further
propagation away from the origin possible. While it is a naïve The Gouy-Chapman theory of the electrical double layer is
interpretation which needs mathematical study of the probability based on the assumption of the Boltzmann distribution of the
distribution function in the process, it can be expected that the charged particles under the potential field which is in the form of
overall chance of the particle moving towards the origin is slightly [22]
higher than the chance of the particle moving away from the
ze’
origin. This asymmetry is agglomerated over the time and causes C ¼ C 0 exp ð13Þ
the transport process becomes slightly slower than the normal kT
Brownian motion, which is referred to as subdiffusion. Similar where C,C 0, z, e, ’, k and T are the double layer charge
concept has also been presented as mismatch and relaxation, concentration, bulk charge concentration, charge magnitude of
discussing similar trends in dielectric response of solid ionic each particle, electron’s charge, potential, Boltzmann constant and
conductors [25,26]. absolute temperature, respectively. The Boltzmann distribution is
Fig. 2 shows the modeling results of the 10,000 runs of the in fact in agreement with the normal diffusion equation [22]. The
random walk propagation of the particles over 100 to 1000 time 1-D flux of the charged particles in a potential field (J) can be
steps. The mean square displacement is calculated and shown modeled through the following equation [22]
versus time (time steps). The figure shows that the normal
@C zF L @’
diffusion assumption is not further valid even when the chance J ¼ L  C ð14Þ
deviation becomes 0.1% (i.e., the chance of moving towards the @x RT @x
origin becomes 50.1% rather than 50%; clearly in this case the where the first term in the right hand side models the diffusion
chance of moving away from the origin reduces from 50% to 49.9%). process, while the second term models the migration process. The
The subdiffusion power n is plotted in Fig. 3 against the chance rate of change of the concentration can then be written as
deviation for the cases studied. The figure shows that the power n
can be considered as a linear function of the chance deviation for @C @J @2 C zF L@C@’ zF L @2 ’
¼ ¼L 2þ þ C ð15Þ
the range shown in this figure. This linearity, however, is violated in @t @x @x RT @x @x RT @x2
higher chance deviations (which have not been shown here). If the migration is neglected, the last two terms in the right hand
Various approaches have been used to study and model the side vanish and the above equation becomes the Fick’s second law
anomalous diffusion phenomena [20]. Here, the fractional deriva- [22]. The Boltzmann distribution is in fact the steady-state solution
tive approach is employed which has been extensively used of the above equation. If the diffusion coefficient (L) is large
especially in the recent years [e.g.,20,23,24,27–31]. In this enough, the time constant of the diffusion and migration processes
is small enough to neglect the time dependency of the
concentration in the double layer theory and find the steady-state
solution which is the Boltzmann distribution.
In the case of the porous media, Eq. (15) has to be generalized
using the anomalous diffusion theory.

@2 C zFK @C @’ zFK @2 ’
Dnt C ¼ K þ þ C ð16Þ
@x2 RT @x @x RT @x2
It can still be assumed that the diffusion and migration time
constants are small enough to solve the above equation in the
steady-state condition. However, the steady-state condition has to
be redefined in this equation in a way to eliminate the time
derivative term on the left hand side. As a result, the definition of
the steady-state has to be revisited. As shown in Eq. (6), the
fractional derivative of a constant is not equal to zero. Therefore, if
the concentration (C) becomes independent of time, the left hand
side derivative does not vanish, and hence, the condition cannot be
considered as steady-state . However, if the concentration can be
written as (Eq. (7))
Fig. 2. The logarithmic plot of the mean square displacement of the 10,000 runs of t
the random walk of the particles over 100, 200, . . . , 1000 time steps. The Cðx; tÞ ¼ ½CðxÞ  C const: ð Þn1 ð17Þ
subdiffusion power n is determined from the slope of the lines. In each case, the R2
tD
of the linear regression is also reported.
 S.M. Rezaei Niya, M. Hoorfar / Electrochimica Acta 188 (2016) 98–102 101

Fig. 3. The subdiffusion power n versus the chance deviation (%) for the cases studied.

then the fractional derivative on the left hand side of Eq. (16) However, it can be shown that
vanishes and the solution can be considered as steady-state. In this  
@v

equation, C const: has to be constant (time-independent) and is = t1n dl ¼ ðjvÞ= t1n vdl  ð1  nÞ=ftn vdl g ð22Þ
considered to be the volumetric average of the concentration in the
@t
domain (Cavg). It can be assumed that the concentration distribu- As a result, the current-voltage relation in the frequency domain
tion becomes equal to that presented in Eq. (17) for large ratios of can be obtained as
t=t D (which is the case considered in the Gouy-Chapman theory).
Inserting Eq. (17) into Eq. (16) results in Q0dl
IðvÞ ¼ ðjvÞj1n D1n
v VðvÞ ð23Þ
2    t 1n
D
@ CðxÞ  Cavg n1 zF @ CðxÞ  C avg n1 @’
þ
@x2 RT @x @x ð18Þ
The impedance can then be determined as
zF  n1 @2 ’
þ CðxÞ  C avg ¼ 0 VðvÞ VðvÞ
RT @x2 ZðvÞ ¼ ¼ 0 ð24Þ
IðvÞ Qdl
ðjvÞj1n D1n
t1n v VðvÞ
The above equation suggests that @’ @ ’ 2
D
@x and @x2 are not functions of
time. As a result, the total charge of the double layer is not a The Fourier transform of the voltage (VðvÞ) is a delta function
function of time [22]. However, the resulted potential from the which can be written as [32]
charged particles for any points outside of this domain follows the
VðvÞ ¼ lim12ejvje1 ð25Þ
trend similar to that described in Eq. (17). So, the double layer e!0
capacitance (Qdl ) can be expressed as The fractional derivative can then be determined as [33]
t
Qdl ¼ Q 0
dl ð
1n
Þ ð19Þ Gð2  nÞ n1
tD v VðvÞ ¼ ð1Þ v VðvÞ
1n
D1n ð26Þ
Gð1Þ
Inserting Eq. (26) into Eq. (24) results in
5. The impedance calculation
1 1
ZðvÞ ¼ ¼ ð27Þ
As the resultant double layer capacitance (Qdl ) presented in Q0dl
ðjvÞj1n ð1Þ1n Gð2  nÞvn1
Q0dl
Gð2  nÞðjvÞn
Eq. (19) is time-dependent, the current-voltage relation of the t1n
D
t1n
D

double layer has to be modified as follows which has a similar form as the CPE impedance shown in Eq. (1). As
@qdl @v @Qdl it can be seen, the subdiffusion power n is equivalent to the CPE
¼ idl ¼ Qdl dl þ v power a in Eq. (1). When n ¼ 1, the subdiffusion transport
@t @t @t dl
0 t @v 0 t
n becomes the normal diffusion transport which results in the
¼ Qdl ð Þ1n dl þ ð1  nÞQdl 1n vdl ð20Þ
tD @t tD normal capacitive behaviour rather than CPE behaviour. Also, the
Q0
where qdl , idl , and vdl are the charge, the current and the potential CPE magnitude (Q) is equal to the t 1n
dl
Gð2  nÞ term which has the
D

difference of the double layer capacitance, respectively. The same unit of Fsn1 .
impedance of the double layer element can be determined using As mentioned above, the subdiffusion power n is shown to be
the Fourier transform related to the fractal dimension (see Eq. (10)) of the porous
  medium containing the solution. It can also be expected that the
Q0dl @v Q0dl
=fidl ðtÞg ¼ IðvÞ ¼ = t1n dl þ ð1  nÞ 1n =ftn vdl g ð21Þ pore size distribution and structure affect the subdiffusion power
tD
1n @t tD as those can easily change the average chance deviation. This
102 S.M. Rezaei Niya, M. Hoorfar / Electrochimica Acta 188 (2016) 98–102

demonstrates why the CPE power has been considered to be a CPE power is equal to the subdiffusion power introduced in the
function of fractal dimension [7,8], pore size distribution [9] or anomalous diffusion.
even the surface roughness [6].
It has to be reemphasized that there is no physical understand- References
ing of CPE in literature. Even worse, it is not clear what parameters
are controlling the CPE power and magnitude in different [1] W. Schelder, J. Phys. Chem 79 (1975) 127.
[2] G. Falk, P. Fatt, P. Roy, Soc. London 160 (1964) 69.
electrochemical systems. The most-widely assumed theory is that [3] J. Jorcin, M.E. Orazem, N. Pebere, B. Tribollet, Electrochim. Acta 51 (2006) 1473.
CPE is the result of different time constants in the process. This [4] S.M. Rezaei Niya, M. Hoorfar, J. Electroanal. Chem. 747 (2015) 112.
assumption which was presented by Cole and Cole [10] in 1941 is [5] S.M. Rezaei Niya, R.K. Phillips, M. Hoorfar, ASME Power & Energy, California
USA (2015) .
considered as the best conjecture which is still an outcome of [6] R. De Levie, Electrochim. Acta 10 (1965) 113.
rearranging the CPE impedance to determine hypothetical [7] L. Nyikos, T. Pajkossy, Electrochim. Acta 30 (1985) 1533.
electrochemical systems which can lead to CPE. In other words, [8] B. Sapoval, Solid State Ionics 75 (1995) 269.
[9] H.K. Song, H.Y. Hwang, K.H. Lee, L.H. Dao, Electrochim. Acta 45 (2000) 2241.
this conjecture was not resulted from physical laws, and hence,
[10] K.S. Cole, R.H. Cole, J. Chem. Phys. 9 (1941) 341.
does not present any physical interpretation even as a conjecture. [11] G.J. Brug, A.L.G. Van Den Eeden, M. Sluyters-Rehbach, J.H. Sluyters, J.
While the discussion presented in the previous sections in this Electroanal. Chem. 176 (1984) 275.
[12] B. Hirschorn, M.E. Orazem, B. Tribollet, V. Vivier, I. Frateur, M. Musiani, J.
paper can be criticised as not being pervasive, it at least presents a
Electrochem. Soc. 157 (2010) C452.
‘possible’ explanation of the physical origin of CPE when diffusion [13] B. Hirschorn, M.E. Orazem, B. Tribollet, V. Vivier, I. Frateur, M. Musiani, J.
in porous media can cause the chance deviation which finally Electrochem. Soc. 157 (2010) C458.
results in the CPE behaviour. As the CPE behaviour is in general [14] B. Hirschorn, M.E. Orazem, B. Tribollet, V. Vivier, I. Frateur, M. Musiani,
Electrochim. Acta 55 (2010) 6218.
more extensive than what it is covered here [e.g.,3], further [15] M. Musiani, M.E. Orazem, N. Pebere, B. Tribollet, V. Vivier, J. Electrochem. Soc.
analysis and studies are required. For instance, the ion distribution 158 (2011) C424.
in a liquid solution might also cause the chance deviation which in [16] M.E. Orazem, I. Frateur, B. Tribollet, V. Vivier, S. Marcelin, N. Pebere, A.L. Bunge,
E.A. White, D.P. Riemer, M. Musiani, J. Electrochem. Soc. 160 (2013) C215.
this case finally results in the CPE behaviour. While this conjecture [17] M. Musiani, M.E. Orazem, N. Pebere, B. Tribollet, V. Vivier, Prog. Org. Coat. 77
needs special analysis to be accepted, it just depicts one of the (2014) 2076.
capabilities of the presented analysis towards more detailed [18] C.L. Alexander, B. Tribollet, M.E. Orazem, Eletrochim. Acta 173 (2015) 416.
[19] B.A. Boukamp, J. Electrochem. Soc. 142 (1995) 1885.
studies of CPE. [20] R. Metzler, J. Klafter, Phys. Rep. 339 (2000) 1.
[21] A. Carpinteri, F. Mainardi, Fractals and fractional calculus in continuum
6. Conclusions mechanics, Springer Verlag, 1997.
[22] A.J. Bard, L.R. Faulkner, Electrochemical methods, fundamentals and
applications, Second ed., John Wiley & Sons, 2001, 2015.
The physics behind the constant phase element (CPE) which [23] R. Klages, G. Radons, I.M. Sokolov, Anomalous transport: foundations and
appears in the modeling of the impedance characteristics of applications, Wiley-VCH, 2007, 2015.
[24] S. Havlin, D. Ben-Avraham, Adv. Phys. 51 (2002) 187.
various electrochemical systems was studied in this paper. The
[25] K. Funke, D. Wilmer, Solid State Ionics 136–137 (2000) 1329.
fractional calculus was first introduced. Then, the anomalous [26] K. Funke, R.D. Banhatti, S. Bruckner, C. Cramer, D. Wilmer, Solid State Ionics
diffusion was reviewed and formulated using the fractional 154–155 (2002) 65.
calculus. It was also shown that the chance inequality of the [27] A. Compte, R. Metzler, J. Phys. A: Math. Gen. 30 (1997) 7277.
[28] D.A. Benson, R. Schumer, M.M. Meerschaert, S.W. Wheatcraft, Transport
random walk in a porous media can result in an anomalous Porous Med. 42 (2001) 211.
diffusion. The Gouy-Chapman theory of the double layer capaci- [29] F. Mainardi, Y. Luchko, G. Paganini, Fractional Calc. Appl. Anal. 4 (2001) 153.
tance was then modified using the anomalous diffusion theory and [30] R. Metzler, J. Klafter, J. Phys. A: Math. Gen. 37 (2004) R161.
[31] M.M. Meerschaert, J. Mortensen, S.W. Wheatcraft, Physica A 367 (2006) 181.
the double layer capacitance was determined. Finally, the [32] G.B. Arfken, H.J. Weber, Mathematical methods for physicists, Sixth Ed.,
impedance of the double layer was calculated which is shown Elsevier Inc., 2015.
to be equal to the classical CPE impedance. It was shown that the [33] K.S. Miller, B. Ross, An introduction to the fractional calculus and fractional
differential equations, John Wiley & Sons, 1993.