PolymerVol. 37 No. 18, pp.

4107 4110, 1996

S0032-3861 (96)00274-1 Copyright 1996 Elsevier Science Ltd Printed in Great Britain. All rights reserved 0032-3861/96/$15.00 + 0.00

ELSEVIER

Comparison of the Havriliak-Negami and stretched exponential functions

S. Havriliak, Jr* Rohm and Haas Research. Bristol Research Park. PO Box 219, Bristol PA 19007. USA and S. J. Havriliak Havriliak Software Development Co., Huntingdon Valley, PA 19006, USA (Received 23 December 1994; revised 16 January 1996)
Alvarez, Algeria and Colmenero compared the stretched exponential (KWW) and Havriliak-Negami (H-N) functions by comparing their distributions of relaxation times. In the case of the KWW function, the distribution of relaxation times was calculated numerically. The numerical results were then represented in terms of the closed form distribution of relaxation times available for the H - N function. They found that a specific KWW k parameter corresponds to a specific H - N a,/3 pair. In this paper the time-dependent dielectric constant, calculated from the stretched exponential for various ks and over a time range sufficiently long to define the t = 0 or cc limits, was transformed to the complex dielectric constant using the extended Schwarzl method. The complex dielectric constant was fitted directly to the H - N function. The results of this procedure are compared with those of Alvarez et al. The two different methods give the same results and the consequences of this agreement are discussed. Copyright 1996 Elsevier Science Ltd.
(Keywords: time to frequency transforms; stretched exponential; H - N function)

INTRODUCTION An expression often used to represent the time decay (or rise) of physical process is the so-called stretched exponential or K W W 1 3 function, given in the normalized form for the dielectric case by equation (1):

from equation (2) and the corresponding integral equation relating it to a distribution of relaxation times. The distribution function, F ( y ) , is given by

F ( y ) = (1)y~(sin/30)(yZ'~ + 2y~ cosTra + l ) - /3/2

e.(t) = (e(,)_-- eo~ = 1 _ exp(_(t/ro)k)

\ ~o - ~ /

(1) In this expression, y = 7-/r0 and 0 = arctan

(3)

In this expression c,(t) is the time-dependent dielectric constant and k is the K W W parameter, To is the relaxation time, while e0 and Eo~ represent the equilibrium and instantaneous dielectric constants, respectively. Another function used to represent dielectric relaxation data in the frequency domain is the H - N function 4'5 defined by e~,(w) = (g*(w) -_~oc,.'~ = {1 +(iwro)~} =J \ So-e~ / (2)

y~ + cos 7ra J

(4)

RESULTS AND DISCUSSION Alvarez et al. 6 calculated numerically the distribution of relaxation times from equation (1) for given values of k in the range of 0.1 to 1.0 and then fitted the numerical results to equation (3) using a non-linear regression routine. Plots of their results are given in Figures 1, 2 and 3. N o statistical information was supplied by these authors to describe parameter confidence intervals or the model standard error of estimate. Koizumi and Kita 7 transformed e,(t) to e~(w), calculated over a log(time) range of - 3 to +3 from the stretched exponential K W W function for given values of the k parameter from 0.29 to 1.0 incremented by 0.01. Dishon et al. 8 also treated the same problem but did not mention the results of Koizumi and Kita. Their parameter range was from 0.02 to 1 and they used the same time range. Their procedure was similar and the results were reported to six places.

In this expression, e*(w) is the normalized complex dielectric constant (or complex relative permittivity) measured at radian frequency w = 27rf, f is in Hz and i is v / S T . The parameters a and/3 represent the width and skewness of the dielectric loss (en(w)) when viewed in a log(w) plot. These parameters also describe the distribution of relaxation times which can be obtained in a closed form from equation (2). The relaxation time is represented by 7- . o The distribution of relaxation times can be obtained
* To whom correspondence should be addressed

POLYMER Volume 37 Number 18 1996

4107

Comparison of H - N and KWW functions. S. Havriliak, Jr and S. J. Havriliak

1.2

.......
p,.

1.0
//J

. ~ . ~ ~

P- 1.0 e~ 0 0.8 n" n IuJ o.s

0.4

i
Z
./-"

O O

o8

=
<

nl iii ~ j 0.2 ./

/~"

_J

I 0,0 0
0.2 0.4 0.6 0.8 1

lj'"

0.0 ~ - - - -

40

-20

L 0

2O

KWW 'k' P A R A M E T E R

LOG(TIME) j cua.E.'r DISHO. ETAL

Figure I Plot of the c~i~ p r o d u c t as a function of thc K W W k parameter. The legend indicates the source of the data

Figure 4 Plot o f t ( t ) as a function of time for k = 0.04. The heavy line represents the time range reported in reference 8

1.2

1.0 fl: ul 0.8

IE
0.6 O. 0.4
0.2

0.0

0.2

0,4

0.6

0.8

calculated from c(t) using the modified Schwarzl method 9. Examples of the complex dielectric constant, either as a complex plane plot or as ~(~), c"(a;) vs log(~;) plots, are given in Figures 5 and 6 respectively for the case k - 0.04. The transformed complex dielectric constants can be fitted to equation (2), using non-linear regression techniques based on rigorous statistical techniques m J3 to determine the parameters ~, '3 and 7- directly for 0 different k values. Although nonlinear regression can be carried out in a number of different ways, the software chosen here is PROC NLIN, available through SAS R14. The convergence criterion for this software is given by d~+lO 6 < 10 -~ (5)

ALPHA PARAMETER
i PRESENT LIMITS AAC i

I
Figure 2

Plot of a as a function o f ,4 t~r diflkrent K W W k p a r a m e t e r s

12.0

~.. uJ

10.0
8.0

In this expression, i is the ith iteration and i - 1 is the preceding one. This software also reports the parameter confidence intervals, determined at convergence, an important feature when making comparisons. Parameters, their confidence intervals as well as the

_z
<

I'Z O

6.0 4.0

P < s Z~ Z
<

o.0 1
~a

< ,,J ~1 ('~ O -1

2.0 0.0
-2.0

0.6}

"

0,4 0.2 0.4 0.6 0.8 1

..4.0

KWW 'k' PARAMETER

PRESENT AACo I

UJ

~' t=,

t
o

0.2 }

i
0 20 40

Figure 3 Plot o f the relaxation time ratio as a function of K W W k parameters

LOG(TIME)

In Figure 4 we have plotted c,,(t) over the log(time) range of - 4 0 to +30 for k = 0.04. The log(t) range o f 3 is simply inadequate to define the KWW function. Instead of relying on published tables, 2"(~,,) was

Figure 5 Plot of the t r a n s f o r m e d ~'(~) and 40 S ( ~ ' ) as a function of Iog(~.,) for the case of the K W W p a r a m e t e r k = 0.04. The symbols represent t r a n s f o r m e d values a n d the lines represent expectation values. The two lines h o v e r i n g a b o u t zero are the real and i m a g i n a r y residuals

4108

POLYMER Volume 37 Number 18 1996

Comparison of H - N and KWW functions: S. Havriliak, Jr and S. J. Havriliak

0.30 0.25 0.20
U) Cn O 0.1S ..I
0.10

0.05
0.00

-4

-2

LOG(FREQUENCY)

TRANSFORMED FII"I'ED
o

ERROR i

Figure 6 Plot of the transformed and fitted s ' ( a 0, as well as the error defined in the text, given here as a function of log(~) for the case of the K W W parameter k = 0.5. The definitions are given in the legend

standard model errors of estimate are given in Table 1 for a few k values covering the range of 0.04 to 1.0 to illustrate the statistics describing the fit. An overlay plot of transformed and fitted results is given in Figure 5. Plots of the various parameters are also given in Figures 1-3. In Figure 1, the dimensions of the solid circles do not represent confidence intervals, which are unknown, but the width of the line represents about 0.5or as obtained in this work. In Figure 2 the diameter of the circles and the two solid lines represent 2or, while the dashed line has no significance in relation to the parameter confidence intervals. It is evident from Figures 1 and 2 that although the ~fl product was determined by AAC without much scatter, the individual parameters showed considerable scatter. In Figure 3, the width of the line represents the parameter confidence intervals. The agreement between Alvarez et al. 6 and the present results is within the scatter of the former case. The conclusion drawn from both studies is that for a given K W W k parameter there exists a specific a,/3 pair. The converse is not true since functions like the Cole-Cole (set r = 1 in equation (2)), or other symmetrical functions like the Fuoss-Kirkwood z5 and Gaussian ~6 ones, cannot be represented by the K W W function. Also, some authors discuss the merits of the Cole-Davidson function (set a = 1 in equation (2)) vs the K W W function. We see from the data shown in Figure 2 that, over a fi parameter range of 0.8 to 1.0, the range of a <_ 0.95. We have observed that for most analyses of experimental data the
Table 1

confidence interval for c~ is about 0.02, so that in this range of/3 the two functions are for all practical purposes the same. As an example of the problem that experimentalists have when comparing different functional forms, consider the case of k = 0.5. The transformed and fitted results for the loss are given in Figure 6. In this case only the loss was used for the regression, so that the fit is somewhat better than that seen in Figure 5. The parameters are listed in Table I under e"(a;). There is a slight difference in the parameters because the fit is not exact and the parameters depend on the data used for determining the parameter. The line representing error in Figure 6 was calculated from the standard deviation at the particular frequency divided by the loss mean estimated over the entire frequency range. Over most of the range the error is less than 2% and only in a limited range is it about 3.5%. The frequency range in that figure is 8 decades, and the error greater than 2% is limited to the log frequency range of -2.5 to -1.5. A 2% error is about the limit of experimental error because of the practical problems of not only making the electrical measurements but keeping the temperature constant. In other words, these small differences are not expected to be readily observed except when extreme care is exercised. In Figure 7 we have represented the same data in the form of log-log plots. It is evident that the separation between the two curves increases with frequency as the frequency changes from the central or relaxation frequency. We can calculate and plot the slope of these curves as a function of log(frequency), see Figure 7. It can be shown that the low frequency slop (m) for the H - N function is a and the high frequency slope (1 - n) is aft. For the specific case shown in Figure 7 the values of rn = ct = 0.776 (0.002) and 1 - n = -(x/= -0.3982 (0.0002), and finally fl = 0.512, are obtained by averaging the extreme I0 points in Figure 8. These results are within 3cr of those listed in Table 1, indicating that the process is an H - N function, as expected. The same calculations for the transformed data, i.e. for the K W W function, yield m = c ~ = l . 0 0 (0.01) and l - n = - a f t = 0.499 (0.003) and finally/3 = 0.50 (0.01). These results suggest that the limiting high/low frequency behaviour is best represented by the Davidson-Cole z7 function, i.e. a = 1 in equation (2). These results, i.e. the high and low frequency limits of the K W W function, are that of the Davison-Cole function which has been reviewed by B6ttcher and Bordewijk zS. In other words, at low frequencies the K W W function is that of the Debye process.

H - N parameters and their limits for representing the dielectric relaxation behaviour of the K W W function for various levels of b k = 1.0 s*(w) 1.332 0.008 0.933 0.002 0.677 0.004 0.7 3,4 k = 0.7 1.332 0.008 0.933 0.002 0.677 0.004 0.7 3.4

Parameter In fo a cr c~ ~r d ~r ~r ~ ~r"

e*(w)

k = 0.5 e*(co) 0.86 0.02 0.831 0.004 0,516 0.005 1.1 5.7

k = 0.5 0.78 0.01 0.784 0.003 0.508 0.005 2.0

e"(a0

k = 0.3 ~*(~) 0.06 0.03 0,624 0.005 0.401 0.005 1.3 6.6

k = 1.0 e*(co) 0.06 0.03 0.624 0,005 0.401 0.005 1.3 6.6

k = 0.07 -4.6 0.1 0.158 0.002 0,327 0,007 0.25 8.7

e*(w)

k = 0.04 e*(~o)
- 11.0 0.2 0.099 0.001 0.306 0.006 0.25 8.7

"In f0 = - I n r 0

POLYMER Volume 37 Number 18 1996

4109

Comparison of H - N and KWW functions. S. Havriliak, Jr and S. J. Havrifiak

0.00

-0.50

~ , (/)

-1.00

U) O 3 O ,-I

-1.50 -2.00

-2.50

-3.00

-4

-2

LOG(FREQUENCY) TRANSFORMED FIr'rED

frequency slopes for 100 materials. These slopes are plotted in an m - 1 (~x/3)vs n (c~) plane. Most of the data are scattered throughout this plane with only a few of the material parameters falling on the Debye coordinates, i.e. ~x = ,~f = 1. In other words, he found no experimental evidence to support the universality of the K W W function at low frequencies. In another study 24 the c~/~ parameters were determined for nearly 1000 compounds which included polymers, their solutions, and polar liquids. Once again the data points were scattered throughout the H - N equivalent of Jonscher's m - 1 vs n plane, i.e. the c~ - log (fl) plane. These two independent studies, using two different analytical methods, came to the same conclusion, i.e. there is no evidence to support the generality of the K W W function.

Figure7

Plot of transformed and fitted log(~"(~)) as a function oI log(~) for the case of the KWW parameter k = 0.5
1,50
. . . . . i

REFERENCES
! 2 Kohlrausch, R. Pogg. Ann. Phys. Chenz. 1854, 91, 179 Kohlrausch, F. Pogg. Ann. Phys. Chem. 1963, 119, 337 Williams, W. and Watts, D. G. Trans. Faraday Soc. 1970, 66, 80 Havriliak, S., Jr and Negami, S. J. Polym. Sci. C. 1966, 14, 99 Havriliak, S., Jr and Negami, S. PoO,mer 1967, 8, 161 Alvarez, F., Algeria, A. and Colmenero, J. Phys. Rev. B. 1991, 11 ) 7306 Koizumi, N. and Kita, Y. Bull. hzst. Chem. Res. (Kyoto Univ.) 1978, 56(6), 300 Dishon, M., Weiss, G. H. and Bendler, J. T..I. Res. Nail. Bzzr. Stand. 1985, 90(l), 27 Havriliak, S., Jr and Havriliak, S. J. Poh'mer 1995, 36, 2675 Havriliak, S., Jr and Watts, D. J. in 'Design, Data, and Analysis. by some Friends of Cuthbert Daniels' (Ed. Mallows), Wiley. New York, 1986 Havriliak, S., Jr and Watts, D. G. Polymer 1986, 27, 1509 Havriliak, S., Jr and Havriliak, S. J. J. Mol. Liquids 1993, 56, 49 Havriliak, S., Jr. and Havriliak, S. J. "Dielectric and Mechanical Relaxation in Materials: Analysis and Application to Polymers, their Solutions and other Systems', Hanser Verlag, New York, 1996 SAS", SAS" Institute, SAS" Circle, Box 8000, Cary, NC 27512-8000, USA Fuoss, R. M. and Kirkwood, J. G. J. Am. Chem. Soc. I941, 63, 385 B6ttcher, C. J. E. and Bordewijk, P. "Theory of Electric Polarization: Vol. II. Dielectrics in Time-dependent Fields'. Elsevier, New York, 1978 Davidson, D. W. and Cole, R. H. J. Chem. Phys. 1951, 19(12), 1484 B6ttcher, C. J. F. and Bordewijk, P. loc. tit., p. 80 Shlesinger, G. and Montroll, E. W. Proc. Natl. Acad. Sci. USA 1984, 81, 1280 Palmer, R. G., Stein, D. L., Abrahams, E. and Anderson. P. W. Phys. Rev. Lett. 1984, 53, 589 Yoshihara, M. and Work, R. N. J. Chem. Phys. 1980, 72(11 ). 5909 Berberian. J. G. and Cole, R. H. J. Chem. Phys. 1986, 84(12), 6927 Jonscher, A. K. in 'Dielectric Relaxation in Solids. Chelsea Dielectric Press, London, 1983, p. 199 Havriliak, S., Jr and Havriliak, S. J. J. Non-Ct3,st. Soli~Z~ 1994, 172-174, 297

4
I,M a. O
,.I

0,50

6 7

44(

0,00

-0.50

9 10
-4 -2 0 2 4

-1.00

LOG(FREQUENCY) TRANSFORMED FITTED

II [2 3

8 Plot of the transformed and fitted slopes as ~ function oi log(~,,) for the case of the KWW parameter k 0.5

Figure

CONCLUSIONS Despite a lack of unbiased evidence, K W W has been cited, in some cases at least, as a 'universal model "j9 or as the 'universality of Kohlrausch's law "2. The results of comparing K W W and H - N functions to represent dielectric relaxation data of poly(4-chlorostyrene), by Yoshihara and Work 2~, or the study on 3-bromopentane by Berberian and Cole 22. showed that the universality claim is probably not true. Although these comparisons are technically sound, they do have two major shortcomings. First, the studies did not use unbiased statistical methods to evaluate the parameters, nor were residuals compared. Second, the studies are limited in extent because poly(4-chlorostyrene) 21 or 3-bromopentane 2-~ were assumed to be representatives of polar materials. In contrast to these methods, consider the work or" Jonscher 23 who determined the high (m) and low ( 1 n)

4 5 (i

19 20 21 22 23 24

4110

POLYMER Volume 37 Number 18 1996