ARTICLE IN PRESS

Physica B 358 (2005) 201–215

www.elsevier.com/locate/physb

Theory of dielectric relaxation in non-crystalline solids:

from a set of micromotions to the averaged collective
motion in the mesoscale region
R.R. Nigmatullin
Department of Theoretical Physics, Kazan State University, Kremlevskaya str., 18, Kazan, 420008, Tatarstan, Russian Federation
Received 23 October 2004; received in revised form 8 January 2005; accepted 8 January 2005

Abstract

Based on the rather general decoupling procedure that reduces a set of micromotions to the averaged collective
motion in the mesoscale region, a consistent and general theory of dielectric relaxation describing a wide set of dielectric
spectroscopy (DS) data measured in a certain frequency/temperature range can be developed. The new theory explains
and generalizes the Vogel–Fulcher–Tamman (VFT) equation and leads to the new type of kinetic equation, containing
non-integer operators of differentiation and integration combined in the specific triads. Each triad combines three
operators: the first non-integer operator has a real exponent, the other two form a pair of non-integer operators having
the complex conjugate power-law exponents. This approach explains naturally the ‘universal response’ (UR)
phenomenon discovered by Jonscher in a wide class of heterogeneous materials and confirms the justified data-curve
fitting approach developed previously as a phenomenological tool for the analysis of raw complex permittivity data. It
explains and generalizes some well-established experimental facts and contains also new predictions that can be verified
in the experiment. This general approach helps to find the proper place for the DS as a science studying the different
types of the ‘reduced’ collective motions and their various interactions with each other in the mesoscale region. This new
theory explains also the fact why kinetic equations containing non-integer operators are realized in reality. They are
necessary for description of slow collective motions which are probably realized in the intermediate range of scales not
only in DS. Reology and mechanical relaxation represent also interesting and not well-understood phenomena, where
the ‘fractional’ kinetics will find its proper place.
r 2005 Elsevier B.V. All rights reserved.

PACS: 61.25.Em; 77.22.Gm; 02.60.Ed; 06.20.Dk; 07.05.Kf

Keywords: Theory of dielectric relaxation of non-crystalline materials; The generalized reind/recap element; Kinetic equations with
non-integer real and complex-conjugate exponents

Tel.: +7 8432360612.
E-mail address: nigmat@knet.ru.

0921-4526/$ - see front matter r 2005 Elsevier B.V. All rights reserved.
doi:10.1016/j.physb.2005.01.173
 ARTICLE IN PRESS
202 R.R. Nigmatullin / Physica B 358 (2005) 201–215
1. Introduction
Despite substantial progress in experimental
measurements in dielectric spectroscopy (DS) [1]
the modern state of the theory of dielectric relaxa-
tion remains unsatisfactory. At present (as before) a
number of empirical relationships exist in order to fit
the broadband dielectric spectra of a wide class of
heterogeneous materials. The most popular empiri-
cal expression which is used to describe the general-
ized broadened asymmetric relaxation loss peak is
the Havriliak–Negami (HN) expression [2]


HN ðjoÞ ¼
0 ðoÞ  j
00 ðoÞ

s 
1
¼
1 þ . ð1Þ
ð1 þ ðjotÞn Þb
Here,

HN ðjoÞ is the HN complex permittivity with
real and imaginary components
0 ðoÞ and
00 ðoÞ;
respectively. Traditionally, the measured permittiv-
ity-frequency data are interpreted and analyzed
quantitatively using expression (1) or its linear
combinations. However, in this description the
fitting parameters n and b are empirical and the
desired relationship with structural or microscopic
motion parameters of the material considered is not
known.
Other approaches such as the mode-coupling
theory [3] and the concentration fluctuation model
[4] allow one to understand qualitatively only a
part of the dielectric spectra of glass-forming
systems and at present they cannot be used for
the description of the measured complex permit-
tivity in a wide frequency/temperature range for
other heterogeneous materials.
In our opinion, any new theory providing a
description of the given dielectric spectrum in a
broad frequency range should satisfy the following
conditions:
1. It should explain how the numerous micro-
scopic motions observed in disordered materials
may be reduced to a few power-law exponents
occurring in analytical expressions for the
complex permittivity chosen for the fitting of
the measured data.
2. It should explain the empirical Vogel–Fulcher–
Tamman (VFT) equation, which is observed in
a wide class of glass-forming systems and other
non-crystalline materials.
3. It should explain why empirical expressions
suggested previously by Cole and Cole or Cole
and Davidson [2] chosen for the analytical
description of complex permittivity fit closely
certain DS data.
4. It should explain the ‘universal’ response phe-
nomenon discovered by Jonscher [5] and ob-
served in a wide class of heterogeneous materials.
5. This theory should describe the measured
complex permittivity of numerous heteroge-
neous materials in a wide frequency/tempera-
ture range.
6. The suggested theory should not contradict the
basic principles of the existing theory of non-
equilibrium processes.
Only if these points are satisfied, the new theory
can be valid for the treatment of experimental data
and practical applications in the modern broad-
band dielectric spectroscopy.
Thus, the construction of such a theory (which
should be derived from the first principles) is the
basic problem of the DS.
The main object of this paper is to develop a new
theory of dielectric relaxation satisfying the require-
ments mentioned above. Our consideration is based
on the Mori–Zwanzig formalism [6]; however, the
decoupling procedure for the memory function is
new and original. It uses a hypothesis connecting the
self-similar cluster structure of a medium and is valid
for some intermediate (mesoscale) region located
between omin n and oMxx frequencies, respectively.
One can prove mathematically that summation over
partly correlated self-similar clusters constructing a
possible structure for a heterogeneous medium,
where microprocesses of relaxation/exchange (they
will be specified below) with thermostat occur, leads
in the frequency domain to the generalized element
coinciding with the memory function and having the
following analytical expression:
K 1 ðjoÞ ¼ K 0 þ C 0 ðjoÞ n
þ C 1 ðjoÞ nþjhOi þ C n1 ðjoÞ njhOi . ð2Þ
It is natural to define this function as the general-
ized (when hOia0) impedance function describing
 ARTICLE IN PRESS
R.R. Nigmatullin / Physica B 358 (2005) 201–215
the behavior of the so-called reind (resistance+in-
ductance, n40) and recap (resistance+capacitance,
no0) two-pole element, respectively. Below we shall
use the abbreviation, the generalized recap/reind
element (GRE), for the memory function satisfying
relationship (2). Here, the frequency o is located in
the interval omin ooooMxx : The temperature-
dependent constants K0, C0, C1 are determined by
a microscopic function, describing the relaxation/
exchange process of a correlated cluster of dipoles
with a thermostat. The power-law exponent n with
possible complex part hOi determines the complex
fractal dimension of the self-similar region con-
sidered. This basic mathematical result allows one
to justify linear kinetic equation for the macro-
scopic polarization P(t) containing non-integer
operators of differentiation and integration:
X
n able decoupling ansatz for the memory function in
tnkk Dnt0k ½PðtÞ  Pðt0 Þ
þ PðtÞ ¼ 0: (3) order to truncate the infinite chain of integro-
k¼1
Here, P(t) is the total polarization, Dnt0 the
Riemann–Liouville fractional integral (no0) or
derivative (n40) [7]. Initially, this type of linear
kinetic equation was developed as a phenomen-
ological approach. It follows from a data-curve
fitting approach (DCFA), developed initially as a
phenomenological tool for analysis of raw di-
electric data [8–10]. The procedure of reduction of
a set of partly correlated micromotions to the
averaged collective motion described by the mem-
ory function (2) helps one to understand many
hidden features of dielectric relaxation for a wide
range of heterogeneous materials and to prove that
dielectric relaxation in disordered materials takes
place on an intermediate (mesoscale) interval
(omin ooooMxx ). It is described in many cases
by kinetic equations of form (3) or its general-
izations containing a set of non-integer operators
with real and complex-conjugate exponents.
2. Decoupling of the memory function and
reduction of a set of micromotions to the averaged
collective motion in the region of mesoscale
It is well-known that the Mori–Zwanzig theory
[6] has many important applications in description
203
of kinetic properties of complex systems [11–14].
In DS this approach was applied recently to the
description of the properties of water molecules
[15]. The chain of equations for an arbitrary time
correlation function (in our case proportional to
the total polarization) can be written as [15,16]
Z t
dPðtÞ
¼ k1 ðt  uÞPðuÞ du,
dt t0
Z t
dk1 ðtÞ
¼ k2 ðt  uÞk1 ðuÞ du,
dt t0
::::::::::::::::::::::::::::::::::::::::::::::::::, ð4Þ
where k1(t), k2(t) are the memory functions of the
corresponding orders including the equilibrium
frequency moments. The basic problem in using
the Mori–Zwanzig formalism is to make a reason-
differential equations.
Having in mind the confirmation of the DCFA
developed in Refs. [8–10] from the basic statistical
principles, we present the first equation in the form
Z t
PðtÞ  Pðt0 Þ þ Mðt  uÞPðuÞ du ¼ 0,
t0
Z t
MðtÞ ¼ ðt  uÞk1 ðuÞ du. ð5Þ
t0
The reason for such presentation is that the
Mori–Zwanzig equations cannot describe broa-
dened dielectric spectra. Any attempt to recover
the broadened spectra from the Mori–Zwanzig
equations presented in form (4) for different
memory functions was unsuccessful [17]. Probably,
it is related to the fact that in nature there is a wide
class of processes which are changed more slowly
in comparison with the velocity of a relaxation
process, which is usually associated with the first
derivative. Such relaxation process should be
described by kinetic equation with a derivative
less than the first order
Dn0 tn PðtÞ þ PðtÞ ¼ 0. (6)
Eq. (6) leads in the frequency domain to the well-
known Cole–Cole expression for the complex
susceptibility [18]. Such type of kinetic equations
cannot be derived directly from the conventional
 ARTICLE IN PRESS
204 R.R. Nigmatullin / Physica B 358 (2005) 201–215
Mori–Zwanzig formalism [19] and new investiga-
tions in this direction are necessary.
In this paper, we suggest a rather general
decoupling procedure related to the calculation
of the memory function M(t). Thus, it is con-
venient to consider the memory function M(t) as a
function of the Laplace parameter s. Putting t0 ¼ 0
and applying the Laplace transform to Eq. (5) we
obtain
Pð0Þ
PðsÞ þ MðsÞPðsÞ ¼ . (7)
s z PK
In order to find an analytical expression for the
Laplace image of the memory function M(s)
describing a relaxation/exchange process with
thermostat we suppose that:
A1: We consider a heterogeneous material
having a self-similar structure. This structure can
be presented by a set of electrically active clusters.
Each cluster includes a group of strongly corre-
lated dipoles. However, different clusters are
weakly correlated with each other.
A2: The relaxation/exchange process with a
thermostat in some volume Vn is described by a
microscopic function f ðstn Þ: Here, tn is a char-
acteristic relaxation time describing the relaxation/
exchange process with the thermostat for a group
of strongly correlated dipoles located in a cluster
having the volume Vn. If the nth cluster contains
Nn number of dipoles, then the relaxation/
exchange process with the thermostat of a set of
weakly correlated clusters is described by the
function
X
MðsÞ ¼ N n f ðstn Þ. (8)
n for the cluster having a minimum number of
This expression increases the limits (when N n a1)
of applicability of the additivity hypothesis widely
used in statistical mechanics for a set of weakly
correlated physical values.
A3: Without loss of generality we may suppose
that the Laplace image of the function f(z)
depending on the complex variable z and describ-
ing the microscopic act of interaction of an electric
dipole with the thermostat has the following form:
a0 þ a1 z þ    þ aK z K
f ðzÞ ¼ (9)
b0 þ b1 z þ    þ bP z P
with KpP+1 and the polynomial in the denomi-
nator has only a negative and complex-conjugate
root.
Case (a): ReðzÞ51ðc0 ¼ a0 =b0 ; c1 ¼ a1 =b0 
b1 =b20 Þ
f ðzÞ ¼ c0  c1 z þ c2 z2 þ    . (10a)
Case (b): ReðzÞb1ðA1 ¼ aK =bP ; A2 ¼
aK bP1 =b2P Þ
A1 A2
f ðzÞ ¼ þ þ . (10b)
z PKþ1
For K ¼ P+1 we define f(z) as a relaxation
function describing the process of interaction of a
dipole with a thermostat. If KoP+1 and the
denominator of polynomial (9) has divisible roots,
then we define f(z) as an exchange function
describing the interaction process of a dipole with
thermostat. The reason for such division is that the
LT
minimal value of the function f ðtÞ ¼ f ðsÞ in the
first case is f ð0Þa0 and moreover f(t) goes to zero
as t ! 1 monotonically. In the exchange case,
however, the value f ð0Þ ¼ 0 and the microscopic
function f(t) has at least one maximum and so may
tend to zero monotonically or non-monotonically
as t ! 1:
A4: We assume that the distributions of the
values Nn and the set of relaxation times tn obey
the following scaling conditions:
N n ¼ N 0 bn ,
tn ¼ t0 xn ðNonoN; Nb1; b; x40Þ. ð11Þ
At n ¼ 0 we define the number of dipoles N0 and
the characteristic relaxation time t0 ; respectively,
correlated dipoles. Without loss of generality one
can assume that the volume of the minimum
cluster coincides with one dipole. So, N 0 ¼ 1 and
t0 defines the characteristic time of relaxation/
exchange process with thermostat of a single
dipole.
The above suppositions mean that the expres-
sion for the Laplace image of the memory function
M(t) assumes the form
X
N 1
MðzÞ  SðzÞ ¼ bn f ðzxn Þ; ðNb1Þ: (12)
n¼ðN1Þ
 ARTICLE IN PRESS

R.R. Nigmatullin / Physica B 358 (2005) 201–215 205

Here z  st0 : Moreover, the function S(z) at any By analogy with these calculations one can
fixed N satisfies the functional equation consider other conditions imposed on parameters
1 b and x: All these cases have been considered and
SðzxÞ ¼ SðzÞ þ bN1 f ðzxN Þ presented in Table 1.
b
Analyzing these cases one can say that relaxa-
 bN f ðzxNþ1 Þ. ð13Þ
tion/exchange processes with thermostat taking
We consider the asymptotic solution of this scaling place in a system of self-similar clusters are
equation at Nb1 for b and xo1: reduced to the GRE (coinciding with the corre-
Taking into account the asymptotic behavior of sponding memory function (2)) if the Laplace
the function f(z) at small (10a) and large (10b) image parameter is identified with the complex
values of Re(z), the contribution of the last two frequency s  jo:
terms in Eq. (13) becomes negligible for the
interval:
zmin 5jzj5zMx . (14) 3. The meaning and evaluation of the log-periodic
function pm (ln z)
Here
 1=ðPKÞ Solution (17) of scaling equation (16) of the
A1 b
zmin ¼ xN1 ; zMx ¼ . (15) previous section and all other possible solutions
bN ðbxÞN
(presented in Table 1 for a wide class of functions
For interval (14) at Nb1 the scaling equation (13) f(z) with variable z taking on real or complex
is simplified and takes the form values) allow one to interpret the meaning of the
1 fractional integrals with the complex exponents.
SðzxÞ ¼ SðzÞ. (16) Following the ideas developed in Ref. [20], the
b
periodic function with unit period can be ex-
The general solution of this scaling equation can
panded as the Fourier series
be written as [19]
  X1  
lnð1=bÞ lnðzÞ lnðzÞ
SðzÞ ¼ pn ðln zÞzn ; n¼ . (17) pn ¼ C n exp 2pnj . (20)
lnð1=xÞ lnðxÞ n¼1
lnðxÞ

Here pn ðln z ln xÞ ¼ pn ðln zÞ is a complex log- Taking into account definition (17) for n and the
periodic function with real period lnðxÞ: The last expression one can present expression for S(z)
geometrical/physical meaning of this function and in the form
methods for its calculation will be considered in the X1     
1 lnðzÞ
next section. If we approximately replace S(z) in SðzÞ ¼ C n exp ln þ 2npj
b lnðxÞ
Eq. (12) by integration, then one can evaluate the n¼1
zeroth Fourier component of the function pn ðln zÞ: X1
 C n exp½ðn þ jOn Þ lnðzÞ . ð21Þ
We define this component using condition
n¼1
hpn ðln zÞizn Here, the real exponent n is defined by expression
Z N1
(17), On ¼ 2pn=lnðxÞ defines a set of frequencies
 bu f ðzxu Þ du. ð18Þ leading to a period lnðxÞ for sum (17). Let us
u¼Nþ1
suppose that this infinite series can be replaced
Taking into account the asymptotic decomposi- approximately by three terms
tions (10) and condition b, xo1; one can obtain the
following expression from (18): SðzÞ ffi zn ðC 0 þ C 1 expðjhOi ln zÞ
hpn ðln zÞi  C 0 ðnÞ þ C
1 expðjhOi ln zÞÞ
Z 1 ¼ zn ðC 0 þ jC 1 j cosðhOi ln z  cÞÞ
1
¼ xn1 f ðxÞ dx. ð19Þ
lnð1=xÞ 0 ¼ C 0 zn þ C 1 znþjhOi þ C
1 znjhOi . ð22Þ
 ARTICLE IN PRESS

206 R.R. Nigmatullin / Physica B 358 (2005) 201–215

Table 1
Various solutions of the asymptotic scaling equation (16) for particular values of the scaling parameters b and x

Case Values of b Values of x omin n oMxx Functional equation Solution

1. bo1 xo1  1=ðPKÞ b 1 SðzÞ ¼ pn ðln zÞzn

A1 SðzxÞ ¼ SðzÞ
xN1 ðbxÞN b
bN
lnð1=bÞ
n¼
lnð1=xÞ
2. b41 xo1  1=ðPKÞ b 1 br0
A1 SðzxÞ ¼ SðzÞ+r0 SðzÞ ¼ pn ðln zÞzn þ
xN1 ðbxÞN b b1
bN
r0 ¼ c0 bN1 is finite value n¼
lnðbÞ
lnð1=xÞ
3. bo1 x41 ðA1 bN1 Þ1=ðPKÞ ðbxÞN 1
SðzxÞ ¼ SðzÞ+r0 SðzÞ ¼ pn ðln zÞzn 
br0
xN c1 x b 1b
r0 ¼ c0 bN1 is finite value lnð1=bÞ
n¼
lnðxÞ
4. b41 x41 ðA1 bN1 Þ1=ðPKÞ ðbxÞN 1 SðzÞ ¼ pn ðln zÞzn
SðzxÞ ¼ SðzÞ
xN c1 x b
lnð1=bÞ
n¼
lnð1=xÞ
5. b¼1 xo1 ðA1 Þ1=ðPKÞ xN1 c1 xN SðzxÞ ¼ SðzÞ þ c0 lnðzÞ
SðzÞ ¼ p0 ðln zÞ þ c0
lnðxÞ

Here, hOi is the averaged frequency (mode) the ECs for the function
referring to the leading term with the averaged yðzÞ ¼ C 0 þ jC 1 j cosðhOi lnðzÞ  cÞ,
value of n and the complex amplitude C 1 ¼
SðzÞ  zn yðzÞ. ð24Þ
jC 1 j expðjcÞ: These parameters determine the con-
tribution of the leading term of the corresponding In accordance with the ECs method expression
series (20). The average value of the mode hOi is (24) initially including some nonlinear fitting
defined as parameters, (hOi; c) can be transformed identically
into the basic linear relationship (BLR)
2phni
hOi ¼ . (23) Y ðxÞ ¼ D1 X 1 ðxÞ þ D2 X 2 ðxÞ þ D3 X 3 ðxÞ. (25)
ln x
Here
It is natural to denote expression (22) as a one-
mode approximation (OMA) used for the evalua- Y ðxÞ ¼ y  h  i,
Z x
tion of the exact expression (20), which contains in
X 1 ðxÞ ¼ ðx  uÞyðuÞ du  h  i,
principle an infinite set of modes. x0
For verification of expression (22) one can use
D1 ¼ hOi2 ,
the eigen-coordinates (ECs) method and consider
the values C0, |C1|, hOi and c as a set of unknown X 2 ðxÞ ¼ x2  h  i,
fitting parameters. hOi2 C 0
The basic principles of the ECs method with D2 ¼ ,
2
application to analysis of DS data have been X 3 ðxÞ ¼ x  h  i,
considered in papers [21–24]. So, it is not necessary
to repeat here the basic ideas. Here, we give only D3 ¼ ðhOi2 C 0 x0 þ y0 ðx0 ÞÞ. ð26Þ
 ARTICLE IN PRESS

R.R. Nigmatullin / Physica B 358 (2005) 201–215 207

The latter expression defines a linear combination b=0.1

Ft.function
depending on the variable x ¼ lnðzÞ and enters into
the desired BLR. Relationship (25) helps one to 0.535
identify two important fitting parameters C0 and
hOi: Other two parameters |C1| and c are found 0.530
from another BLR

y(x)
0.525
UðxÞ ¼ C 1 cosðhOixÞ þ C 2 sinðhOixÞ, (27)
where 0.520
UðxÞ ¼ yðxÞ  C 0 ,
0.515
C 1 ¼ jC 1 j cos c; C 2 ¼ jC 1 j sin c. ð28Þ
-4 -2 0 2 4 6 8 10
With the help of these expressions one can verify (a) x=ln(z)
numerically the OMA and calculate the values of
the fitting parameters C0, hOi; |C1| and c: b=1
For numerical verification of the applicability of 19.250 Ft.function
the OMA we chose the Laplace image correspond-
ing to the microscopic relaxation function f(t) in 19.245
the form
19.240
1
y(x)

n
f ðzx Þ ¼ ,
1 þ zx þ ðzxn Þ2
n
19.235
  pffiffiffi 
LT 2 t 3 t
f ðzÞ ¼ f ðtÞ ¼ pffiffiffi exp  sin . ð29aÞ 19.230
3 2t0 2 t0
19.225
In order to extract the pure oscillations caused by
-4 -2 0 2 4 6 8 10
the function pn ðlnðzÞÞ we fixed the parameter b in
(b) x=ln(z)
the interval (0.05pbp1) and put x ¼ 0:1: The
variable z was located in the interval Fig. 1. (a) Verification of log-periodic oscillations calculated
(0.1p|z|p10,000). for b ¼ 0:1; x ¼ 0:1 (n ¼ 1). The solid line is the fitting function
calculated in a OMA and (b) verification of oscillations for
The numerical calculations realized have the
b ¼ 1:0; x ¼ 0:1 (n ¼ 0). Unfortunately these two plots cannot
following objects: be displayed in one figure. The remaining plots also demon-
strate the effectiveness of a OMA. All fitting parameters for
1. Direct calculation of sum (12) and its verifica- different b are collected in Table 2.
tion with approximate solution (17) including
conditions (15).
2. Calculation of the functions S(z) and y(x) in
accordance with definitions (22) and (24). lines) for different values of b is presented in
3. Calculation of the fitting constants Dk ðk ¼ Fig. 2. Fig. 3 show the influence of oscillations
1; 2; 3Þ in accordance with the BLR (25) by the on behavior of S(z), caused by another probe
linear least-square method (LLSM). They function for b ¼ 0:1:
should present a set of sloping lines if the
OMA is correct. sin2 ðzxn Þ
f ðzxn Þ ¼ . (29b)
4. Calculation of the necessary set of fitting 1 þ zx þ ðzxÞ2
parameters (C0, hOi; |C1|, c). The verification
of log-periodic oscillations defined by Eq. (24) It is also reasonable to introduce other solutions of
at different b is presented in Fig. 1. The fitting scaling equation (13) which were found in numer-
of the function (12) (shown by points) by the ical verifications. If the limiting values of the
approximate expression (22) (expressed by solid function f(z) in (12) are not negligible, we have a
 ARTICLE IN PRESS

208 R.R. Nigmatullin / Physica B 358 (2005) 201–215

6 b=0.1
Ft.function
5
0.40
S (b=0.1)
4 Ft.function 0.35
b=0.1 S (b=0.5) 0.30
3
S(z)

Ft.function
0.25

y(x)
2 b=0.5 0.20
1 0.15
0.10
0
0.05
-4 -2 0 2 4 6 8 10 0.00
(a) ln(z)
-4 -2 0 2 4 6 8 10
21 (a) x=ln(z)
20
19 3.5
18
17 3.0
16
15 b=1.0 2.5
14 S (b=0.1)
S(z)

13 S (b=1.0) 2.0 Ft.function

12 Ft.function
S(z)

11 1.5
b=0.9 S (b=0.9)
10 Ft. function
9 1.0
8
7 0.5
6
0.0
-4 -2 0 2 4 6 8 10
(b) ln(z) -0.5
-4 -2 0 2 4 6 8 10
Fig. 2. (a) The calculated values of the sum S(z) and its fit (b) x=ln(z)
(shown by solid lines) in the OMA for two values of b ¼ 0:1 and
0.5 and (b) the calculated values of the sum S(z) and its fit in the Fig. 3. (a) The verification of the OMA for the function defined
OMA for two values of b ¼ 0:9 and 1.0. For b ¼ 1 the behavior by expression (29b) and containing oscillations (b ¼ 0:1). The
of S(z) collapses in a straight line. quality of the fitting is not so good in comparison with previous
plots calculated for function (29a). So for this case the OMA is
not sufficient and (b) the calculation of the sum S(z) and its fit
functional equation with solution by the function y(x) shows that log-periodic oscillations reflect
the influence of some random components. Probably, in real
1 experiments the discrete fractal structure is masked by such
SðzxÞ ¼ SðzÞ þ a,
b ‘random’ behavior.
ab
SðzÞ ¼ pn ðln zÞzn þ ,
b1
lnðbÞ
n¼ . ð30aÞ
lnðxÞ numerical calculations prove that the OMA is
correct and physically reflects the discrete structure
For b ¼ 1; the solution of the scaling equation
of the fractal considered. It is interesting to note
(30a) is given by
from analysis of the parameters given in Table 2
SðzxÞ ¼ SðzÞ þ a, that the basic contribution to the approximate
lnðzÞ expression (23) comes from the first Fourier
SðzÞ ¼ p0 ðln zÞ þ a ; n ¼ 0. ð30bÞ components (hni ffi 1).
lnðxÞ
Based on Laplace image (22) it is easy to
The values of the fitting parameters for various b obtain in time domain the generalized value of
and x ¼ 0:1 are collected in Table 2. These the fractional Riemann–Liouville integral for a
 ARTICLE IN PRESS

R.R. Nigmatullin / Physica B 358 (2005) 201–215 209

Table 2
The values of the fitting parameters obtained in the verification of sum (19) with function (29a) by the ECs method (x ¼ 0:1)

Case b n hOi C0 C1 C2 c hni Std. dev.

0 0.05 1.30103 2.71757 0.60239 0.00862 0.00574 0.58798 0.9959 0.05935

1 0.1 1 2.7176 0.52519 0.01036 3.98995E4 0.03851 0.99591 0.05175
2 0.2 0.69897 2.71757 0.60239 0.00815 0.00639 0.66521 0.9959 0.05935
3 0.3 0.52288 2.71754 0.75679 0.0053 0.00889 1.03337 0.99589 0.07457
4 0.4 0.39794 2.71746 0.97866 0.00281 0.00996 1.29561 0.99586 0.09643
5 0.5 0.30103 2.71447 1.29843 6.3437E4 0.01032 1.50941 0.99476 0.12793
6 0.6 0.22185 2.71751 1.78592 9.79814E4 0.01031 1.47607 0.99588 0.17596
7 0.7 0.1549 2.71751 2.60713 0.00242 0.01008 1.33525 0.99588 0.25688
8 0.8 0.09691 2.71752 4.26176 0.00363 0.00972 1.21327 0.99588 0.4199
9 0.9 0.04576 2.71753 9.24966 0.00465 0.00928 1.10571 0.99589 0.91135
10 1 0 2.71753 19.2374 0.00553 0.00879 1.00953 0.99589 0.99543

The last column expresses the value of the standard deviation calculated for absolute value between S(z) and its fitting value calculated
in the OMA.

complex fractional exponent. Using relation- pears (C 1 ¼ 0; hOi ¼ 0) and so the last equation
ship [25] reduces to Eq. (6) yielding in the frequency domain
the generalized Cole–Cole expression for the
ta1 LT a
¼p , (31) complex susceptibility.
GðaÞ
one can present the memory function in time
domain as 4. The confirmation of the justified data-curve
fitting approach and new expressions for complex
tn1 tnþjhOi1
MðtÞ ¼ C 0 þ C1 susceptibility
GðnÞ Gðn þ jOÞ
njhOi1
t In papers [8–10] based on analysis of raw
þ C
1 . ð32Þ
Gðn  jOÞ available DS data obtained from different inter-
national laboratories, we developed the special
The first term in the last expression represents by
recognition procedure and suggested the phenom-
itself the evaluation of the kernel M(t) in the
enological approach, which leads to the new
continuous approximation and is correct for
kinetic equation (3) for the total polarization
collective motions distributed on random fractals;
containing fractional integral and derivatives.
other two terms in the OMA reflect the existence
The recognized complex permittivity expressions
of log-periodic structure existing for true discrete
fit DS data well and allow a generalization of the
fractals and modifying the collective motion
VFT equation for the description of the tempera-
taking place in such structures. Using the defini-
ture dependence of a loss peak. In the previous
tion of the Riemann–Liouville integral, kinetic
section, we reduced a set of micromotions dis-
equation (7) for the total polarization with one
tributed over different scales and described in the
collective motion distributed over a discrete fractal
frequency domain by the function f ðjoÞ to a single
structure takes the form
collective motion, corresponding to a single GRE
nþjhOi
ðPðtÞ  Pð0ÞÞ þ ½C 0 Dn
0 þ C 1 D0 (Eq. (2)). It is reasonable to assume that in the
structure considered one can have several types of
þ C
1 DnjhOi
0 PðtÞ ¼ 0. ð33Þ
collective motions described by the different GRE.
For a random fractal structure, the log-periodic If, for example, we have two types of collective
effect evoked by discrete fractal structure disap- motions, then we have two possibilities: they can
 ARTICLE IN PRESS

210 R.R. Nigmatullin / Physica B 358 (2005) 201–215

be connected in series or in parallel. For the first 100

case we have the following expression for complex
susceptibility [8,9] 80
Re(ε) in series

ð0Þ 
1 60 Re(ε) in parallel

Reε

ðjoÞ ¼
1 þ ,
1 þ RðjoÞ 40
RðjoÞ ¼ ½ðjot1 Þn1 þ ðjot2 Þn2 1 , ð34Þ
20
corresponding to a linear combination of two
0
memory functions. In order to derive expression
for
ðjoÞ connecting two GRE in parallel we -2 0 2 4 6 8 10 12 14 16
suppose that the Laplace function describing an (a) lg(ω)
exchange process with thermostat has the form
'alpha' peak
f ðsÞ ¼ sf 1 ðsÞ. (35) 10

For this case the sum S(z) is defined by the 8 Excess wing
expression
Im(ε) 6 Im(ε) in series
X
N 1 Im(ε) in parallel
n n n 4
SðzÞ ¼ b zx f 1 ðzx Þ
n¼ðN1Þ
2
X
N1
¼z ðbxÞn f 1 ðzxn Þ ¼ zS1 ðzÞ: ð36Þ 0
n¼ðN1Þ
-2 0 2 4 6 8 10 12 14 16
If the sum S1(z) satisfies again conditions (15), then (b) lg(ω)
for an additive combination of two memory
Fig. 4. (a) The calculated values of the real part of complex
functions satisfying condition (36) we obtain an permittivity presented correspondingly by expressions (34) and
expression for the complex susceptibility describing (37) and (b) the imaginary parts of these functions mimic the so-
two collective motions connected in parallel, viz. called ‘excess wing’ observed in glass-forming systems. So, these
functions can be used for the fitting of the measured DS data in

ð0Þ 
1 frequency range for glass-forming materials. The chosen

ðjoÞ ¼
1 þ ,
1 þ RðjoÞ parameters are as follows: n1 ¼ 0:21; n2 ¼ 0:95; t1 ¼ 0:5 
102 ; t2 ¼ 0:4  105 ;
1 ¼ 3;
s ¼ 100 (two GRE in series);
RðjoÞ ¼ ½ðjot1 Þn1 þ ðjot2 Þn2
. ð37Þ
n1 ¼ 0:21; n2 ¼ 0:9; t1 ¼ 0:5  105 ; t2 ¼ 0:4  103 ;
1 ¼ 3;
From analysis of available DS data given in papers
s ¼ 100 (two GRE in parallel).
[8–10], it follows that the first combination of two
collective motions (34) is more preferable than the In this case the Laplace variable is shifted on the
combination (37), which has been identified only in value z þ lt0 : If we suppose that only one
one case [26]. It is interesting to note that the collective mode exists, then the reduction proce-
complex parts of expressions (34) and (37) reproduce dure described above leads to the following
the so-called ‘excess wing’ observed in many glass- expression for the complex permittivity:
forming materials [27]. The corresponding plots of

ð0Þ 
1
functions (34) and (37) imitating ‘excess wing’ are
ðjoÞ ¼
1 þ
presented by Figs. 4a and b, correspondingly. ðlt þ iotÞn þ 1
Suppose now that the initial memory function
ð0Þ 
1

1 þ ,
M(t) in Eq. (5) is modified by an exponential ½1 þ að1 þ io=lÞn

damping function and is written as a ¼ ltn ; tn ¼ hpn ðlnðz þ lt0 Þi. ð39Þ
Z t
MðtÞ ¼ ðt  uÞk1 ðuÞelu du. (38) The last expression at ab1 becomes the
0 empirical Cole–Davidson expression for complex
 ARTICLE IN PRESS
R.R. Nigmatullin / Physica B 358 (2005) 201–215
permittivity. If a medium, has a memory about a set
of scales forming discrete fractals that can be
generated in this medium, then one can expect that
the OMA will be acceptable for description of such a
quasi-regular structure. In this case, following the
procedure described in the previous section, the
GRE with real exponent should be replaced by the
corresponding expression (2) containing a linear
combination of a pair of the complex-conjugate
exponents. For example, the usual Cole–Cole
expression for the complex permittivity is replaced by

ð0Þ 
1

ðjoÞ ¼
1 þ ,  d f
1 þ RðjoÞ
RðjoÞ ¼ ½C 0 ðjoÞn
þ C 1 ðjoÞnþjhOi þ C
1 ðjoÞnjhOi
. ð40Þ where df is the fractal dimension, G is the
So the reduction procedure realized for one memory
function or for their linear combinations helps to
understand the DCFA and predicts new kinetic
equations for the total polarization containing non-
integer derivatives or integrals combined in the
specific triads. Each triad contains a linear combina-
tion of non-integer operators: one with real exponent
and a pair with the complex-conjugate exponents.
These equations are applicable to the description of
relaxation phenomenon in heterogeneous media
having a specific memory about the countable set
of scales generating by a true fractal structure. The
complex-conjugate exponents disappear if the corre-
sponding medium becomes completely random and
the set of scales generating the corresponding fractal
structure is continuous.
5. The temperature dependence of the power-law
exponents and the generalized VFT equation
In Ref. [28] the expression for the power-law
exponent n (in notation (11) of this paper) can be
rewritten as
lnðbÞ
n¼ . (41)
lnðxÞ
Following the basic ideas of this paper one can
write the scaling parameter x as
t for a set of correlated clusters and are confirmed
x¼ , (42)
t0
211
where t is the mean relaxation time, characterizing
some collective motion, t0 is the cutoff scaling time
related to the collective motion. Parameter b is
connected to a scaling factor associated with the
relative number of dipoles located in some
correlated cluster. One can write the following
expression for b keeping in mind that the mean
relaxation time can be related to the possible
quasi-diffusion motion inside a group of corre-
lated dipoles forming an isotropic cluster of the
size R. One can write the following relationship for
the value b:
R
b¼G , (43)
R0
geometrical coefficient equaled unity and, more-
over, depending on the geometrical form of the
cluster considered. R0 is the cutoff size of a cluster
in the 3D-space. For bo1, the cluster of the size R0
should be chosen as the maximum size cluster, in
the opposite case b41 this cluster should be
chosen as the minimum. One can generalize the
relationship between R and t; taking into account
the anomalous diffusion process
R2 ¼ Dy ty , (44)
where the exponent y ¼ 1=2 corresponds to
normal diffusion process, for y41=2 we obtain
superdiffusion and yo1=2 corresponds to sub-
diffusion [29]. We should mark also the recent
paper [30], where anomalous diffusion process has
been considered and some anomalous relaxation
functions have been derived from kinetic equa-
tions. The parameter Dy; in Eq. (44) can be
associated with the self-diffusion coefficient. Com-
bining the latter expressions, one can write the
relationship between the power-law exponent n
and mean relaxation time t in the form
 2=y
lnðt=ts Þ yd f R0 D2
y
n ¼ n0 ; n0 ¼ ; ts ¼ .
lnðt=t0 Þ 2 G 1=d f
(45)
In spite of simplicity of these equations, they admit
new interpretation and generalization (with ya12)
by experimental measurements (at y ¼ 12) [28].
 ARTICLE IN PRESS
212 R.R. Nigmatullin / Physica B 358 (2005) 201–215
Following these simple ideas one can consider the
dependence of the average mode hOi on t: It
follows from expression (23) that
2phni 2phni
hOi ¼ ¼ . (46)
ln x lnðt=t0 Þ
We suppose [10] that the possible connection
between mean relaxation time t and temperature
T is given by the Arrhenius relationship
tðTÞ ¼ tA expðE=TÞ, (47)
where E defines the mean reorientation energy of
the fractal subsystem considered and characterizes
a collective motion. Based on expressions (45) and
(47) in paper [10] we derived mathematically the
VFT equation and its possible generalizations
from expressions (41) and (47). The parameters
of the VFT equation in a new possible interpreta-
tion reflect a specific transition from one type of
collective motion to another one. So, from this
point of view, glass–gel transition can be consid-
ered as an intermediate case between two types of
collective motions expressed in terms of the VFT
equation [27]. In the case of a simple collective
motion, the two power-law exponents describing
the left and right branches surrounding an
extremum point will coincide with each other
(for example, if the DS spectrum is described by
Cole–Cole or Cole–Davidson expressions). The
temperature-dependent peak for one type of
collective motion will be described by Eq. (47)
written for op ðTÞ ¼ 1=tðTÞ and for this case the
Vogel–Fulcher temperature T VF ¼ 0 and D ¼ E:
This approach allows one to explain naturally
the so-called ‘universal response’ (UR) observed
by Jonscher with co-authors in many heteroge-
neous materials. In accordance with the Jonscher’s
observations [5], the behavior of the complex
susceptibility at frequency range omin ooooMxx is
wðjoÞ ¼ C n ðjoÞ1þn ,
wðjoÞ ¼ Am  Bm ðjoÞm ; 0om; no1 ð48Þ
and may be defined as the UR. If we compare
expressions for the memory function MðjoÞ given
in Table 1 with expressions (48), we see that these
expressions formally coincide with each other.
Our approach allows one to understand the
values of power-law exponents when they combine
a triad with a pair of complex-conjugate values
(similar to expression (2) at K 0 ¼ 0) and to
interpret the exponent-dependent constants Am,
Bm and Cn. The VFT equation and the UR
phenomenon help in the interpretation of DS
spectra over a wide range of frequencies. The UR
behavior at certain frequency range characterizes
in the reduced (averaged) form a collective motion
on mesoscale and is observed presumably in the
wings (branches) of a spectrum surrounding a
transition region. In turn, the transition region is
located in the vicinity of an extremum character-
ized by a local peak or a minimum. As it was
mentioned above, the VFT equation in a new
interpretation reflects a transition from one
collective motion to another for a certain range
of temperatures. Thus, the VFT equation should
be correct not only for the peaks of dielectric losses
(usually only the so-called ‘alpha peak’ is verified)
but for the minima points also, for example, lying
between the branch belonging to the low-fre-
quency dispersion located on the left at low
frequencies and the beginning of the alpha peak
located on the right at higher frequencies. We
received the measured complex permittivity data
from Dr. G. Smith (De Monfort University,
Leicester, UK). These data concern the liquid melt
of polymethylmethacrylate without placticizer
(Eudragit RS Melt) measured in the frequency
range (0.1–106 Hz) and temperature range
(353.15–403.15 K with step 2.5 K). The complete
treatment of these data has been given in the Ph.D.
Thesis of Dr. S.I. Osokin. However, here it is
reasonable to give only the temperature depen-
dence of the minimum frequency located between
the low-frequency dispersion (LFD) branch at low
frequencies and the beginning of the so-called
‘alpha’ peak at higher frequencies. The tempera-
ture dependence is well described by the VFT
equation with parameters T VF ¼ 262:6 10:6 ðKÞ;
o0 =2p ¼ 4:4768  104 ðs1 Þ D ¼ 4:823: The valid-
ity of the corresponding VFT plot is confirmed by
Fig. 5. We expect that the validity of the VFT
equation for local minima will be confirmed also
for other DS data. This consideration allows one
to identify number of new processes in the analysis
of DS spectra. Usually each process is identified
with a possible peak, but this identification will be
 ARTICLE IN PRESS
R.R. Nigmatullin / Physica B 358 (2005) 201–215
correct if the branches surrounding the peak are
symmetrical and described by the same power-law
exponent. It is natural to associate a collective
motion with the branch that can form a possible
peak or minimum. The detection of the number of
processes associated with a number of branches
will be correct if any two branches surrounding an
extremum are asymmetrical and the temperature
dependence of transition region follows the VFT
describing the relevant micro-
equation or its generalizations.
6. Results and discussion
In conclusion, we want to stress some basic
points that help us to construct a consistent and
more or less general theory of dielectric relaxation
in non-crystalline materials.
1. Based on the supposition that the structure of
non-crystalline material is composed of a set of
weakly correlated clusters and each cluster, in
turn, is composed of a group of strongly
correlated dipoles, it is possible to prove that
memory function in frequency domain is
reduced to the GRE (Eq. (2)).
fmin(Hz)
1 and only the real power-law exponent from
data
the VFT_fit
0.1
2.45 2.50 2.55 2.60 2.65 2.70 2.75
1000/T(K)
Fig. 5. Confirmation of the VFT equation obtained for a
minimum frequency located between LFD branch (on the left)
and the beginning of the alpha peak (on the right). These
data belong to the liquid melt of polymethylmethacrylate
without placticizer (Eudragit RS melt) measured in frequency
range (0.1–106 Hz) and temperature range (353.15–403.15 K
with step 2.5 K). The fitting parameters of the VFT equation
are as follows: T VF ¼ 262:6 10:6 ðKÞ; o0 =2p ¼ 4:4768 
104 ðs1 Þ D ¼ 4:823:
213
2. Each GRE reflects in a mathematical form a
collective relaxation/exchange process of a set
of self-similar microscopic motions with ther-
mostat (representing other degrees of freedom
existing in the material considered). Each
microscopic motion is characterized in fre-
quency domain by the function f ðjoÞ; repre-
senting the Fourier component of the function
FT
f(t), ½f ðtÞ ¼ f ðjoÞ
motion.
3. The results of such a reduction of the micro-
scopic motions is presented in Table 1 and can
be considered as theoretical evidence for the
UR phenomenon discovered exper imentally in
many heterogeneous materials.
4. Each local extremum (max/min) located be-
tween the frequency branches of a dielectric
spectrum can be considered as a transition
region separating two types of collec tive
motions. This transition region is described by
the generalized VFT equation that receives
extended interpretation in this approach.
5. This theory predicts the existence of a specific
triad resulting from one real and a pair of the
complex conjugate power-law exponents. This
triad should appear in cases where a discrete
fractal structure is predominant and only a
countable set of scales is needed. For a
completely random fractal structure a contin-
uous set of scales is possible. The sum (12) can
be replaced approximately by an integral (18)
and here the log-periodic function is averaged
each triad survives.
6. This theory predicts generalizations of the
Cole–Davidson (39) and Cole–Cole expressions
(40) for complex susceptibility that are neces-
sary in experimental verifications.
2.80 7. The theory predicts also the temperature
dependence of the leading mode hOi (46) needed
for experimental verification in the case of
detection of a discrete fractal structure.
8. The most important prediction is related to the
explanation of the previously developed DCFA
leading to new kinetics equation containing
non-integer derivatives or in tegrals forming
triad from real and pairs of the complex-
conjugate exponents. Probably, the rather
 ARTICLE IN PRESS
214 R.R. Nigmatullin / Physica B 358 (2005) 201–215
unusual form of kinetic equation (3) explains
why previous attempts to derive a consistent
and a general theory of dielectric relaxation
using conventional ‘integer’ calculus have been
unsuccessful.
9. From this theory it follows that a wide class of
micromotions are not directly detectable. The
dielectric spectrum in frequency domain con-
tains only information about a set of the
‘reduced’ (averaged) collective motions. So, this
theory helps one to identify the DS as an ideal
tool for studying collective motions and their
various interactions with each other in the
mesoscale region.
We suppose that this theory, which at the
present stage cannot be considered as pure
microscopic and derived directly from the first
principles, nevertheless can receive wide applica-
tions in description and interpretation of different
DS spectra, representing the behavior of different
collective motions in a wide frequency/tempera-
ture range.
In conclusion, it is necessary to recall an
important problem that should be considered in
further investigations.
Following Ref. [31], one can write the relation-
ship connecting the complex conductivity with
permittivity in the form
sðjoÞ ¼ sð0Þ þ jo½
ðjoÞ 
1
. (49)
If we combine Eq. (49) with the expression for
complex permittivity used in [9,10]

ð0Þ 
1

ðjoÞ ¼
1 þ , (50)
1 þ RðjoÞ
then we obtain
sðjoÞ ð
ð0Þ 
1 Þ½jo=sð0Þ
Dielectric Press, London, 1983;
¼1þ
sð0Þ 1 þ lRðCjo=sð0ÞÞ
 
jo
F C . ð51aÞ
sð0Þ
The parameter l figuring in Eq. (51a) is found
from the identity
lRðCjo=sð0ÞÞ  RðjoÞ. (51b)
Expression (51a) is correct if the function RðjoÞ
can be represented only by linear combinations of
power-law functions as follows from the DCFA.
So this approach helps to find independently the
function RðjoÞ from conductivity measurements
also and confirms the Taylor–Isard scaling phe-
nomenon [32]. However, relationship (49) cannot
explain the low-frequency dispersion (LFD) phe-
nomenon, when in the low-frequency domain it
modifies the expression for the complex permittiv-
ity by the additive term sy =ðjoÞy
sy
ð0Þ 
1

ðjoÞ ¼
1 þ y
þ . (52)
ðjoÞ 1 þ RðjoÞ
In our opinion, this LFD phenomenon is gener-
ated by strongly correlated collective motion
taking place in groups of ions/dipoles and merits
special consideration.
Acknowledgements
The author wants to thank sincerely Prof. W.T.
Coffey for attentive reading of the manuscript
during the submission process. The author also
wants to express his sincere thanks to Prof. Yu.
Feldman for sending his recent publications and
constant stimulating discussions related to the
construction of the general theory of dielectric
relaxation.
References
[1] F. Kremer, J. Non-Cryst. Solids 305 (2002) 1.
+
[2] C.J.F. Bottcher, P. Bordewijk, Theory of Electric Polar-
ization, Elsevier, Amsterdam, 1992.
+
[3] W. Gotze, +
L. Sjogren, Rep. Progr. Phys. 55 (1992) 241.
[4] G. Katana, et al., Macromolecules 28 (1995) 2714.
[5] A.K. Jonscher, Dielectric Relaxation in Solids, Chelsea
A.K. Jonscher, Universal Relaxation Law, Chelsea Di-
electric Press, Oxford, 1996.
[6] H. Mori, Progr. Theor. Phys. 30 (1963) 578;
R. Zwanzig, in: W.E. Brittin, B.W. Downs, J. Downs
(Eds.), Lectures in Theoretical Physics, vol. III, Inter-
science Publ. Inc., New York, 1961, pp. 106–141.
[7] K. Oldham, J. Spanier, The Fractional Calculus, Academic
Press, New York, 1974.
[8] R.R. Nigmatullin, S.I. Osokin, J. Signal Proc. 83 (2003)
2433.
[9] R.R. Nigmatullin, S.I. Osokin, G. Smith, J. Phys. D 36
(2003) 2281.
 ARTICLE IN PRESS
R.R. Nigmatullin / Physica B 358 (2005) 201–215
[10] R.R. Nigmatullin, S.I. Osokin, G. Smith, J. Phys. C.:
Condens. Matter. 15 (2003) 3481.
[11] R.M. Yulmetuev, N.R. Khusnutdinov, J. Phys. A 27
(1994) 5363.
[12] V. Yu. Shurygin, R.M. Yulmetuev, Zh. Eksp. Teor. Fiz. 96
(1989) 938
V. Yu. Shurygin, R.M. Yulmetuev, Sov. JETP 69 (1989)
532 and references therein.
[13] Y.P. Kalmykov, S.V. Limonova, J. Mol. Liq. 43 (1989) 71.
[14] R.M. Yulmetuev, Phys. Lett. 56A (1976) 387.
[15] V.I. Arkhipov, A. Yu. Zavidonov, J. Mol. Liq. 106 (2–3)
(2003) 155.
[16] V.I. Arkhipov, J. Non-Cryst. Solids 305 (2002) 127.
[17] V.I. Arkhipov, private communication.
[18] R.R. Nigmatullin, Ya.E. Ryabov, Phys. Solid State 39
(1997) 87;
R.R. Nigmatullin, Ya.E. Ryabov, Russ. Phys. J. 40 (1997)
314.
[19] A. Le Mehaute, R.R. Nigmatullin, L. Nivanen, Fleches du
Temps et Geometrie Fractale, Hermez, Paris, 1998.
[20] D. Sornette, Phys. Rep. 297 (1998) 239;
D. Sornette, A. Johansen, A. Arneodo, J.-F. Muzy, H.
Saleur, Phys. Rev. Lett. 76 (1996) 251.
[21] R.R. Nigmatullin, Appl. Magn. Reson. 14 (1998) 601.
215
[22] R.R. Nigmatullin, Physica A 285 (2000) 547.
[23] R.R. Nigmatullin, M.M. Abdul-Gader Jafar, N. Shinya-
shiki, S. Sudo, S. Yagihara, J. Non-Cryst. Solids 305
(2002) 96.
[24] R.R. Nigmatullin, G. Smith, Physica A 320 (2003) 291.
[25] G.A. Korn, T.M. Korn, Mathematical Handbook for
Scientists and Engineers: Definitions, Theorems and
Formulas for Reference and Review, McGraw-Hill, New
York, 1961.
[26] M.M. Abdul-Gader Jafar, R.R. Nigmatullin, Thin Solid
Films 396 (2001) 280.
[27] P. Lunkenheimer, Dielectric Spectroscopy of Glassy
Systems, Shaker Verlag, Augsburg, 1999.
[28] Ya.E. Ryabov, Yu. Feldman, J. Chem. Phys. 116 (2002)
8610.
[29] E.W. Montroll, G.H. Weiss, J. Math. Phys. 6 (1965) 1967;
B.A. Carreras, V.E. Lynch, D.E. Nowman, G.M. Zaslavs-
ky, Phys. Rev. E 60 (1999) 4770.
[30] Y.P. Kalmykov, W.T. Coffey, S.V. Titov. Phys. Rev. E 69
(2 Part 1) (2004) Art. No. 021105.
+
[31] J.C. Dyre, T. Shroder, Rev. Mod. Phys. 72 (2000)
873.
[32] H.E. Taylor, J. Soc. Glass Technol. 43 (1959) 124T;
J.O. Isard, J. Non-Cryst. Solids 4 (1970) 357.