LOVELL
LOVELL
LOVELL
This article has been downloaded from IOPscience. Please scroll down to see the full text article.
(http://iopscience.iop.org/0022-3719/7/23/024)
View the table of contents for this issue, or go to the journal homepage for more
Download details:
IP Address: 130.120.230.222
The article was downloaded on 21/10/2010 at 10:37
R Love11
Department of Physics, Chelsea College, University of London, Pulton Place,
London SW6 5PR
Abstract. The Kramers-Kronig transforms relating the real and imaginary parts of the
dielectric susceptibility have been used both on empirical functions and on experimental
data. Good agreement is found between measured values of the imaginary part and those
calculated from measured values of the real part of the susceptibility.
1. Introduction
Under certain conditions that usually apply at low fields, the real and imaginary parts
of the dielectric susceptibility (x = 1' - if') are related by the Kramers-Kronig trans-
forms. Hence, if only one component is known, the other may be found by applying a
transform.
Dielectric loss peaks in solids are rarely of the classic Debye form but are usually
much broader and often asymmetric when plotted with a logarithmic frequency axis.
An empirical relation that fits a variety of materials (particularly organic polymers) is of
the form (A K Jonscher 1974 private communication) :
1
=
(O/OJm + (0/w2)l-" (1)
where m and n lie between 0 and 1, and CO is positive. This increases as omat low frequen-
cies and decreases as con-' at high frequencies (ie it reduces to a Debye peak if m = 1
and n = 0).
This paper describes the application of analytical and numerical transforms to
empirical functions of this form and to experimental data.
2. Mathematical properties
The polarization of an isotropic dielectric in an electric field that varies sinusoidally with
time may be described by a complex susceptibility
P = E,,xE (2)
provided that: (i) the amplitude of the polarization varies linearly with the amplitude of
the field; and (ii) the polarization varies sinusoidally with time.
4378
Application of Kramers-Kronig relations 4379
These requirements are satisfied if the polarization and the field obey a linear
differential equation with constant coefficients. This is true for most materials at low
fields.
The Kramers-Kronig relations may be derived by assuming that the principle of
causality holds (Landau and Lifshitz 1960)
where 9 means that the Cauchy principal value of the integral is used. x' and x" are
therefore Hilbert transforms of one another and belong to the class of conjugate func-
tions (Titchmarsh 1959).
Now from the definition (2), together with the requirement that the values of P and
E are real, it may be shown that ~ ' ( o
is an
) even function whereas ~ " (is
oan
) odd function.
Hence the relations may also be expressed as
x"(0) = - -9
2w
n sp
which are more useful in practice.
____
U' -0
' du
3. Analytic solutions
A number of materials give loss-peaks similar to this with n N 0.7 (A K Jonscher 1974
private communication). The real part of the susceptibility is
1
X' (02
1
4. Numerical solutions
The empirical expression for broad, asymmetric loss-peaks (1)may be given in the reduced
4380 R Love11
form
I-
I I
c
X
0.1 - 0.1
Log w Log w
Figure 1. Exact transformation from x" to x' for Figure 2. Numerical transformation from f to
x" = o"/(l+ (11) with n = 0.7. x' for x''
= om/(l+ m m - " + l ) with m = 0.8 and
n = 02.
in xi’. Smoothing the data would reduce the effect but could hide true features. Various
criteria may be used to find true errors:
(i) for frequencies less than 10” Hz, all loss processes are likely to be relaxations, that
is, dx’/do should be negative;
(ii) a gradient of log x‘ against log o steeper than -2 is unlikely, as is a gradient of
log f ’ against log o lying outside the range - 1 to + 1 ; and
(iii) data are also suspect if the gradient of lines joining consecutive points changes
sharply from one interval to the next. Random errors are less important for data that
are closely spaced in frequency.
Since the value of X” (say) at a given frequency depends on values of 1’over a range of
frequencies near the given one, it is necessary to extrapolate X’ outside the range of
measurements to calculate x” near the extremes of the range. The method used in the
1 I I I I 1
-2 0 2 4
Log f (Hz)
present work is to fit a least-squares straight line to the first (or last) few points on a log x
against log o plot and this line is used over the decade of frequency below the first point
(or above the last point). x‘ is taken to be constant (or x’’is taken to be zero) outside this
range and X” (or x’)is only calculated for frequencies within the range of the measure-
ments. For accurate results, the points should extend over at least three decades of fre-
quency. An outline of the method of computation is given in the Appendix.
Transformations of results for a STAG glass are shown in figure 3. It can be seen that
there is good agreement with the experimental x” at high frequencies and that the AC
component of 2’’ is recovered at low frequencies.
4382 R Love11
In organic polymers, the loss usually varies much less with frequency. A transfor-
mation of typical polymer data (nylon 610: Curtis 1961) is shown in figure 4. It can be
seen that at low frequencies the difference between the measured and calculated x'' gives a
good straight line with a slope of - 1 (log x'' against log f ) , as expected for the DC com-
ponent of x".
IO I I I I I 1
x" I
0.1
5. Conclusions
The work has shown how the Kramers-Kronig relations can be used in practical studies
of dielectrics at low frequencies. Numerical transforms may be used for empirical
functions that cannot be transformed exactly. Experimental data may also be trans-
formed numerically provided that they extend over at least three decades of frequency.
Acknowledgments
The author wishes to acknowledge the tenure of an SRC Research Assistantship and to
thank M Frost for the use of the data on STAG glass and Professor A K Jonscher and
other members of Chelsea College for helpful discussions.
Appendix
0 o-u20+u,
(u2 - u l ) +-ln
+
~~
u2 - w 2 2 0 - u1 0 u2
Application of Kramers-Kronig relations 4383
A subroutine is used to integrate over the rest of the range. Since X” is significant over
a wide frequency range, the integral is used in the form
X”(Uid(ln
1-olu
U).
This is evaluated using Simpson’s rule with equal increments of In U. The subroutine
integrates over one decade (or part of a decade) at a time in the following order:
(i) the rest of the decade below 100 using 100 increments;
(ii) the rest of the decade above 0.10 using 100 increments;
(iii) the decades above 100 using 10 increments, continuing until a decade gives a
negligible contribution ; and
(iv) the decades below 0.1 o using 10 increments, continuing until a decade gives a
negligible contributjons.
All the contributions are then added and multiplied by 2/71 to give ~’(0).
The subroutine for transforming to z’(f)is similar to that used before, whereas for
transforming to x”u) the integral is used in the form
The integrals are evaluated using Simpson’s rule with equal increments of In U,the number
of increments being chosen to give about 50 per decade. We define an interpolating
interval to be the region in which a single interpolating equation is used (eg the interval
between the first and third points or the interval between the highest frequency and ten
times the highest frequency). The subroutine integrates over one interval (or part of an
interval) at a time in the following order:
(i) the remainder of the interval below u1 (=0.8 f);
(ii) the remainder of the interval above u2 ( = 1.25f) ;
4384 R Love11
(iii) the rest of the intervals above f,stopping if an interval gives a negligible con-
tribution; and
(iv) the rest of the intervals below f, stopping if an interval gives a negligible con-
tribution.
For transforming to ~ ' ( f ) ,we assume that ~"(f) is zero outside O.lf, to lOf, (where
f, is the lowest frequency and f, is the highest frequency). Hence the sum of all the
contributions so far multiplied by 2/71 gives x'cf).
For transforming to ~"0, we assume that x'cf) takes the end of range values outside
O e l f , to lOf, and hence there is an extra contribution:
where ~'(O.lf,) and x'( lOf,) are obtained by extrapolation using the least-squares
straight lines. Adding this to 2/71 times the sum of all the other contributions gives ~"(f).
References