The determination of apparent memory functions from dielectric relaxation data for amorphous polymers and glassforming liquids:

Theory and experiment

Graham Williams and Jrme Fournier Citation: J. Chem. Phys. 104, 5690 (1996); doi: 10.1063/1.471770 View online: http://dx.doi.org/10.1063/1.471770 View Table of Contents: http://jcp.aip.org/resource/1/JCPSA6/v104/i14 Published by the American Institute of Physics.

Additional information on J. Chem. Phys.

Journal Homepage: http://jcp.aip.org/ Journal Information: http://jcp.aip.org/about/about_the_journal Top downloads: http://jcp.aip.org/features/most_downloaded Information for Authors: http://jcp.aip.org/authors

Downloaded 23 Feb 2012 to 193.136.139.253. Redistribution subject to AIP license or copyright; see http://jcp.aip.org/about/rights_and_permissions

The determination of apparent memory functions from dielectric relaxation data for amorphous polymers and glass-forming liquids: Theory and experiment
Graham Williams and Jerome Fournier
Department of Chemistry, University of Wales, Swansea, Singleton Park, Swansea, SA2 8PP, United Kingdom

Received 15 August 1995; accepted 20 December 1995 Interrelationships between time-correlation functions for molecular reorientation, their apparent memory functions and dielectric permittivity for the relaxation in glass-forming liquids and amorphous polymers are considered briey and a method is given for the determination of an apparent rst-order memory function K 1 ( ) in the frequency domain from dielectric relaxation data. For the KohlrauschWilliamsWatts KWW function and the ColeCole function and for experimental data for three amorphous polymers it is shown that the real and imaginary components of K 1 ( ) increase without limit as frequency f ( /2 ) is increased. The physical signicance of such apparent memory functions is considered in view of this behavior and the recent experimental studies of the relaxation in polymers by SchmidtRohr and Spiess and in glass-forming liquids by Cicerone and Ediger, who suggest that a dynamic heterogeneity is present that leads to a broad distribution of relaxation times where each elementary process has no proper memory. For such cases it appears that the apparent memory functions are simply related to the parameters process in characterizing the distribution and may have no physical signicance for the glass-forming materials. 1996 American Institute of Physics. S0021-9606 96 51012-6

I. INTRODUCTION

Molecular motions in amorphous polymers and glassforming liquids have been studied for many years over a frequency range 10 5 to 1011 Hz by different experimental methods including dielectric relaxation spectroscopy DRS ,13 dynamic mechanical relaxation,1,4 Kerr electrooptical relaxation,5 quasi-elastic light scattering6 and neutron scattering,7 multidimensional multinuclear NMR,8,9 timeresolved uorescence depolarization,10 and time-resolved optical relaxation.11,12 Multiple relaxations , ,... are commonly observed which may be assigned13 to limited local process of chain units or whole molecules for motions glass-forming liquids and large-scale microbrownian moand tions -processes of such units. Studies of the processes have been made below, through and above the glass-transition temperature T g and provide an understanding of the molecular dynamics of the amorphous state and also of the time dependence of macroscopic thermodynamic quantities in the T g range for systems away from thermodynamic equilibrium e.g., see Refs. 1315 for studies of the nonequilibrium behavior of volume, enthalpy, and specic heat of glass-forming materials . The relaxation behavior of such materials has been critically reviewed17,9,14 18 and was a major theme at two recent major international meetings.19 It is shown that the dielectric and processes coalesce at high temperatures to form the process.20,17 The pattern of , , processes may be rationalized using a general model21,17 in which partial relaxation of dipoles gives the process at short times/high frequencies and total relaxation gives the process at long times/low frequencies. A simple phenomenological model embracing these concepts is able to rationalize all the qualitative features of the dielectric ,
5690 J. Chem. Phys. 104 (14), 8 April 1996

and processes in glass-forming materials17,21 and the same model may be applied to the complementary processes observed by the different relaxation, scattering and spectroscopic techniques indicated above. An important feature of process in such systems, observed by any of the principal the techniques, is that its relaxation function in time is of a stretched-exponential form that may be represented, approximately, by the KohlrauschWilliamsWatts empirical function2224 t exp t/
0

where 0 is a correlation time and is a spread parameter 0 1. Experimental values of commonly lie in the range 0.4 0.6.2,17,21 Although Eq. 1 represents data fairly well over several decades of time, or frequency in its transformed form,23,24 it fails to t data at short times/high frequencies, as was made clear in the original publications.23,24 The quest for realistic models that lead to KWW behavior has been a major topic in condensed-matter physics in recent years. Many models have been proposed including defect diffusion by controlled random walks on lattices, free volume relaxation, dynamics of Ising chains, multisite hopping in barrier models and, recently modemode coupling theories see, e.g., Refs. 17, 2528 and references therein . The present paper is concerned with the nature and use of certain memory functions in this connexion. It is prompted by the works of Gotze and co-workers27,28 and Mazenko,29 who describe a modemode coupling approach involving higher-order memory functions, and by Douglas and Hubbard30 who proposed a new semiempirical rst-order memory function for general application to polymer-melt dynamics. We outline a simple method for the determination of
1996 American Institute of Physics

0021-9606/96/104(14)/5690/9/$10.00

Downloaded 23 Feb 2012 to 193.136.139.253. Redistribution subject to AIP license or copyright; see http://jcp.aip.org/about/rights_and_permissions

G. Williams and J. Fournier: Memory functions from dielectric relaxation

5691

a rst-order memory function K 1 ( ) in the frequency do. main from data for the complex dielectric permittivity We illustrate the method for the special cases of the KWW and ColeCole empirical relaxation functions and for experimental data for the dielectric relaxation in poly vinyl acetate , poly ethyl acrylate and poly vinyl octanoate .
II. DIELECTRIC RELAXATION FUNCTIONS AND t -DEPENDENT ORIENTATION DISTRIBUTION FUNCTIONS

obtained, normally, from each experiment in order to describe the mechanism for motion. Complementary experiments e.g., Kerr-effect and dielectric relaxation as in Ref. 5 may further specify f ( ,t/ 0 ,0) or it may be measured directly, as described in the NMR studies of Schmidt-Rohr and Spiess.8 10 We now consider the relationships between the time-correlation functions J (t) and certain memory funcmn tions.
III. MOMENTS OF f ( FUNCTIONS ,t/ ,0) AND MEMORY

The complex dielectric permittivity at a frequency f /2 for an ensemble of equivalent dipolar species contained in a macroscopic spherical volume V may be related to the macroscopic relaxation function (t) see Ref. 1 and (t) see Refs. 3133 the molecular relaxation function by the following equations 1 i F
0

The TCF for a homogeneous ergodic system obey a set of coupled integro-differential Volterra equations, that contain successively higher-order memory functions. Writing J 31 mn (t) C(t) for convenience it follows that dC t dt dK n t dt
t 0

t t

2 3

K 1 t,
t 0

d ,

7a

p
0

1 i F

Kn

t,

Kn

d ,

7b

are the limiting low- and high-frequency where 0 and permittivities respectively, p( ) is a local-eld factor and F indicates a one-sided Fourier transform. We may write3234 t
i

where we have written K 1 (t) as K 1 (t, ) following Douglas and Hubbard.30 By Fourier transformation and use of the convolution theorem the Mori-Zwanzig continued fraction is obtained from Eqs. 7 as31 C K0 0 i K1 0 i K2 0

0
2 i

t
j i

0
j

where the autocorrelation functions refer to the motions of a reference dipole i having dipole moment i and the crosscorrelation functions include orientational correlations between i and further dipoles j. The strength of the crosscorrelation terms, i (0) j (0) , depend, for polymers, on the conformational properties of a chain.17 The motions of a reference unit are conveniently expressed using the conditional orientational distribution function31 f ( ,t/ 0 ,0) where and 0 refer to the nal and initial orientations, in Euler space, of the dipole moment vector. Expanding this distribution in terms of Wigner rotation matrix elements D J ( ), D J ( 0 ) and the time-correlation functions mn mn J 31,5 mn (t), we obtain f where
J mn

i 8 ( ) F [C(t)], K ( ) F [K (t)], F indicates a where C n n one-sided Fourier transform and the fraction continues innitely. Two special cases relevant to the present work are where the fraction is truncated at rst and second-order terms giving C C i K0 0 K
1

9a

,t/ aJ mn

0 ,0

J,m,n

aJ DJ mn mn

D J* mn

J mn

t ,

K0 0 i K1 0 i K2 . 9b

are known coefcients. Hence f

0

t DJ mn

,t/
0

0 ,0

D J* mn .

DJ mn

D J* mn

6
1 00 (t)

diDielectric experiments give information on for pole groups undergoing isotropic reorientations while Kerreffect relaxation, NMR relaxation, uorescence and optical relaxations give information on 2 (t). Such experiments mn therefore give information on individual moments or weighted sums of individual moments of the orientational distribution function in Eq. 5 , so insufcient information is

Equations 9 are exact and as valid as Eq. 8 : i.e., the continued fractions are being terminated at the level chosen and each may be used as a starting point in our considerations of relaxation processes. Douglas and Hubbard30 take Eq. 9a as their starting point. They proposed a new empirical equation for K 1 (t, ) that involves two spread parameters. They show that the KWW, DavidsonCole,35 ColeCole,36 and HavriliakNegami37 empirical functions are obtained from their memory-function equation by choice of these param eters. Gotze and co-workers27,28 start with the second-order memory function K 2 ( ). They assume that K 2 (t) is given by particular functions involving C(t) itself, thus closing the mathematical problem in Eq. 9b and allowing C ( ) to be

J. Chem. Phys., Vol. 104, No. 14, 8 April 1996

Downloaded 23 Feb 2012 to 193.136.139.253. Redistribution subject to AIP license or copyright; see http://jcp.aip.org/about/rights_and_permissions

5692

G. Williams and J. Fournier: Memory functions from dielectric relaxation

determined for each assumed relation between K 2 (t) and C(t) giving a variety of forms for C(t). Although memory functions may be proposed in this way and hence C ( ) and C(t) determined, it is important to emphasize that for the process in glass-forming materials the memory function may not be applicable. The broad processes of KWW-type may arise from independent parallel processes which each have no memory. In the next section we show that apparent rst order memory functions K 1 ( ) may be determined from model dielectric functions or from experimental data and we shall examine these functions to see if they are acceptable physically.
IV. APPARENT MEMORY FUNCTIONS DETERMINED FROM EXPERIMENTAL DATA

In Appendix A we show that a rst-order memory func tion K 1 ( ) in the frequency domain may be determined for dielectric relaxation from a knowledge of the normalized complex dielectric permittivity R( ) at the frequency of /2 . Writing R( ) R ( ) iR ( ) we have the def nitions R
0

; .
0

10a 10b

For a single-relaxation time process K 1 ( ) 1/ 0 ( ) constant, K 1 0 at all since the memory function K 1 (t) is a function in the t domain in this case. We illustrate the use of the method for the KWW function2224 and ColeCole function36 which are widely used for dielectric relaxation in glass-forming liquids and polymers. We have also made similar calculations for the DavidsonCole function35 but they will not be presented here.
A. KWW function

This is of interest since plots of R ( ) vs log f and R ( ) vs log f for the dielectric relaxation in many glassforming materials have been tted using this function.23,24,17 From Eq. 1 and using values of R ( ), R ( ) tabulated by Koizumi and Kita38 for a range of values we have calculated the real and imaginary parts of the memory function K1( ) i K 1 ( ). Figures 1 a 1 c show repreK 1( ) equal ( ) and K ( ) vs log sentative plots of K 1 0 for 1 to 0.9, 0.7, and 0.5. Also included in Figs. 1 are the plots of R ( ) and R ( ) vs log 0 so the range of the dispersion and absorption are seen in relation to the variations in K 1 ( ) and K 1 ( ). Several features are evident in Figs. 1 a 1 c . i In all cases K 1 ( )constant as 0 and the constant level is reached for 0 values where the absorption (R ( )) is small compared with its maximum value. Since the KWW function can be expressed in terms of a distribution of relaxation times39 it is possible to deduce the limiting low-frequency behavior of K 1 ( ) and

FIG. 1. Plots of the real part K 1 ( ) and the imaginary part K 1 ( ) against log for the KWW function for 0.9, 0.7, and 0.5. The ordinate is 0 scaled so that K 1 (0) 1. Also shown are R ( ) - - - and R ( ) for comparison.

K 1 ( ). In Appendix B we deduce this limiting lowfrequency behavior for the special case of a discrete distribution of relaxation times giving the result for the KWW function that K1 0 1 1 1
0

11a

J. Chem. Phys., Vol. 104, No. 14, 8 April 1996

Downloaded 23 Feb 2012 to 193.136.139.253. Redistribution subject to AIP license or copyright; see http://jcp.aip.org/about/rights_and_permissions

G. Williams and J. Fournier: Memory functions from dielectric relaxation

5693

same exponent. This is expected since, as Douglas and Hubbard show30 the KWW function, Eq. 1 corresponds to a memory function K 1 t, t

12

1 and is a linear time variable. Hence where ( ) 1 K 1 (t) varies as 1/t 1 and we have K ( ) for large values of , as is observed in Figs. 13. Although the formal application of Eq. 7a and Eqs. A2 A4 in Appendix A to C ( ), and hence R( ), gives ( ) and its equivalent in the timethe memory function K 1 domain K 1 (t, ) Eq. 12 it should be emphasized that the KWW function is empirical so no physical signicance should be given to these memory functions at this stage.

FIG. 2. Plots of K 1 ( ) and K 1 ( ), on a logarithmic scale, against log 0 for the KWW function for 0.56. Also shown are R ( ) - - - and R ( ) for comparison.

B. The ColeCole function

The empirical frequency-dependent relaxation function introduced by Cole and Cole36 takes the form R 1 1 i , 13

K1

0 0.

11b

The K 1 ( ) values in Figs. 1 are scaled to give K 1 ( 0) 1). The 1 for the single-relaxation time process ( steady decrease in K 1 ( 0)-values with decrease in is as calculated from Eq. 11a . ii K 1 ( ) shows a positive increase from the low-frequency is increased. As is decreased the plateau level as onset of the deviation in K 1 ( ) moves to lower frequen cies and, at the same time K 1 ( ) increases markedly at a given frequency. If the calculations are extended to high values of 0 , as we show in Figs. 2 and 3 for 0.56 and 0.38, respectively, then we see that K 1 ( ) and K 1 ( ) increase as a power law with the

1. Using the method described previously in where 0 Sec. IV A above and in Appendix A together with values of R ( ) and R ( ) calculated from Eq. 13 for chosen values of , we have determined plots of K 1 ( ) and K 1 ( ) vs log . These are shown in Figs. 4 a 4 c for equal to 0.9, 0.7, and 0.5, respectively. No low frequency plateau is increases, K 1 ( ) and K 1 ( ) observed for K 1 ( ). As ( ) 0.5 we have K 1 increase monotonically. For K 1 ( ) since, from Eq. 13 we may write K1 R 1 R
1

cos

i sin

14

Thus both K 1 ( ) and K 1 ( ) vary as 1 ( )/K ( ) tan /2 which is unity for K1 1

C. Experimental memory functions

and the ratio 0.5.

FIG. 3. Plots of K 1 ( ) and K 1 ( ), on a logarithmic scale, against log 0 for the KWW function for 0.38. Also shown are R ( ) - - - and R ( ) for comparison.

As examples of the determination of K 1 ( ) from experimental dielectric data, we show in Figs. 57 results for the -process in poly ethyl acrylate , poly vinyl acetate and poly vinyl octanoate as calculated from the master curves of permittivity and loss factor shown in Refs. 21 poly ethyl acrylate and 23 poly vinyl acetate , poly vinyl octanoate . These data for R ( ) and R ( ) vs log / max , where f max max/2 is the frequency of maximum loss factor, had been tted fairly well by the KWW function23,24 except at the highest frequencies where the experimental loss values exceeded the calculated values. We had tted the data for PVAc and PVOct with 0.38. For comparison, with Figs. 57 we show in Figs. 8 and 9 the components of K 1 ( ) equal to 0.56 and calculated using the KWW function with 0.38. Several points should be noted: rst the components of the memory function depend on both R ( ) and R ( ) which is a severe test of the accuracy of the experimental data at low frequencies where R ( ) approaches R ( 0).

J. Chem. Phys., Vol. 104, No. 14, 8 April 1996

Downloaded 23 Feb 2012 to 193.136.139.253. Redistribution subject to AIP license or copyright; see http://jcp.aip.org/about/rights_and_permissions

5694

G. Williams and J. Fournier: Memory functions from dielectric relaxation

FIG. 5. Plots of K 1 ( ) and K 1 ( ) against log 1 for the dielectric process in polyethyl acrylate. Also shown are R ( ) and for comparison. 0 1/ m where m is the frequency R ( ) of maximum loss factor.

input quantities R ( ) and R ( ). In consequence we suggest that rather than t experimental data for R ( ) and R ( ) with an assumed form of memory function or one derived from a theoretical approach as in modemode coupling theory it would be appropriate to convert experimental R ( ), R ( ) data into memory function components K 1 ( ), K 1 ( ) and to then compare the experimentally derived memory functions with those proposed from theory, e.g., that of Douglas and Hubbard,30 with its adjustable pa rameters, or those proposed by Gotze.27,28
V. DISCUSSION

It is apparent that the memory functions K 1 ( ) determined for the KWW function and ColeCole function are

FIG. 4. Plots of K 1 ( ) and K 1 ( ) against log 0 for the ColeCole function with equal to 0.9, 0.7, and 0.5. Also shown are R ( ) - - - and R ( ) for comparison.

Direct current conduction leads in practice to dielectric absorption at low frequencies for all materials and this adds to R ( ) and slightly to R ( ). Any slight changes in experimental values of [R ( ),R ( )] at low frequencies lead to marked changes in the derived values for K 1 ( ) and K 1 ( ). In Figs. 57 K 1 ( ) and K 1 ( ) increase markedly with increasing frequency and resemble the curves shown in Figs. 8 and 9 for KWW behavior. It is evident that experi mental values for K 1 ( ) and K 1 ( ) are very sensitive to the

FIG. 6. Plots of K 1 ( ) and K 1 ( ) against log 1 for the dielectric process is polyvinyl acetate. Also shown are R ( ) and R ( ) for comparison. 0 1/ m as in Fig. 5.

J. Chem. Phys., Vol. 104, No. 14, 8 April 1996

Downloaded 23 Feb 2012 to 193.136.139.253. Redistribution subject to AIP license or copyright; see http://jcp.aip.org/about/rights_and_permissions

G. Williams and J. Fournier: Memory functions from dielectric relaxation

5695

FIG. 7. Plots of K 1 ( ) and K 1 ( ) against log 1 for dielectric process and R ( ) in polyvinyl octanoate. Also shown are R ( ) for comparison 0 1/ m as in Fig. 5.

FIG. 9. Plots of K 1 ( ) and K 1 ( ) against log 1 for the KWW function with 0.38. Also shown are R ( ) and R ( ), - - - and , respectively, for comparison.

badly behaved at high frequencies. Similarly the rst-order memory functions determined from the experimental data for the dielectric relaxation in the three amorphous polymers are badly behaved at high frequencies. Similar behaviour for relaxation in K 1 ( ) will be obtained for the dielectric other amorphous polymers and glass-forming liquids that ex hibit KWW-type behavior.2,17 The rapid increase in K 1 ( ) ( ) with increasing frequency implies a dependence and K 1 1 as for the KWW-function and hence a memory func tion K 1 (t 1 ) t 1 (1 ) which is badly behaved. Thus although apparent rst-order memory functions can be determined from the experimental dielectric data the resultant behavior is such that the meaning of the memory function should be questioned and alternative explanations should be sought for stretched-exponential behavior of the kind described here. It

is important to remember that C(t) for the KWW39 and ColeCole functions may be expressed in terms of distributions of relaxation times where each process has no proper memory so the apparent memory function has no physical meaning in these cases. The question arisesdo the apparent memory functions derived from experimental data have a physical signicance? The fact that K 1 ( ) in Figs. 57 are badly behaved at high frequencies suggest that they do not. However, experiments that give information on only certain moments of f ( ,t/ 0 ,0) are not able to fully answer this question since individual J (t) lack spatial information. mn Such information has been obtained by Schmidt-Rohr and Spiess8 10 and Cicerone and Ediger12,40 for the relaxations in amorphous polymers and glass-forming liquids, respectively. These studies indicate that a dynamic heterogeneity is present in such systems so the relaxors span a distribution of relaxation times where individual processes do not have intrinsic memory functions for their correlation functions. Thus the apparent memory functions determined above for the relaxation in three polymers may not have a physical signicance but are simply determined by the parameters of the distribution of relaxation times. In Appendix B we force the relaxation function C(t) for a distribution of relaxation times to obey the memory function representation, Eq. 7a and we show that K 1 ( ) is expressed in terms of the relaxation times i and their weightings a i in the distribution. In Appendix C we further show that for a distribution of relaxation times the apparent rst-order memory function is given by K 1 t, ai t
i

ci

a ic i

15

so at a chosen time t 1 say we have

FIG. 8. Plots of K 1 ( ) and K 1 ( ) against log 1 for the KWW function with 0.56. Also shown are R ( ) and R ( ), - - - and , respectively, for comparison.

K1 t1

a ic i t 1
i i i

a ic i t 1 .

16

J. Chem. Phys., Vol. 104, No. 14, 8 April 1996

Downloaded 23 Feb 2012 to 193.136.139.253. Redistribution subject to AIP license or copyright; see http://jcp.aip.org/about/rights_and_permissions

5696

G. Williams and J. Fournier: Memory functions from dielectric relaxation

The apparent memory function weighs contributions from c i (t i )/ i and from c i (t i ). It is not a true memory function since i each process in the distribution is independent of the other processes and, ii each process has no proper memory. This demonstrates that while it is possible to determine apparent memory functions from experimental data or to t experimental data using assumed forms of memory functions, with their adjustable parameters, such functions may not have a physical signicance. As we have seen the works of Schmidt-Rohr8,9 and Spiess and Cicerone and Ediger40 suggest that the process is made up of a distribution of parallel processes each having no proper memory. The present work, coupled with that of the above authors, brings into question the use of memory functions for the process as described by Douglas and Hubbard and by Gotze. In the absence of the new experimental evidence8,9,40 discussed above, how might a memory function be envisaged relaxations? For simple molecular liquids for the slow memory functions of correlation functions are associated with momentummomentum correlations in the picosecond range.31,33 However for slow molecular reorientational motions such correlations will be unimportant so position position correlations are sought as a source of a memory function. If such a memory existed its origins may lie in the correlated motions of molecules during cooperative relaxation. In the usual models of small-step reorientational motions a relaxor moves as a result of the random forces exerted by neighboring molecules. Thus a single exponential correlation function is obtained for the motion of dipoles in a two-site barrier model using random noise theory.41 We speculate that if the forces acting on the relaxor are correlated at short times then this would lead to correlated motions at short times thus generating a memory function in the TCF. While random noise f 0 noise has no memory, pink noise f 1 noise , brown noise f 2 noise and black noise f 3 noise all have short-time correlations in the noise trajectory in time see Ref. 42, Figs. 5, p. 111, and Fig. 1, p. 123 . Thus if stretched-exponential functions obtained for the relaxations in glass-forming materials imply a true memory function then a possible explanation would be that the motions of relaxors are correlated at short times due to the correlation of the impacting forces at short times. Note that K 1 (t) is related to the forceforce correlation function F 0 F(t) the second uctuationdissipation theorem31 . However, in view of the recent experimental studies of Schmidt-Rohr and Spiess and by Cicerone and Ediger, and the unphysical behavior of K 1 ( ) seen in Figs. 57 it seems likely that stretched-exponential behavior for the process in glass-forming materials arises from a distribution of parallel processes, each having no proper memory, and not from a process having an intrinsic memory.
VI. CONCLUSIONS

outlined for the determination of a rst-order apparent memory function from dielectric permittivity data. The apparent memory functions derived from experimental dielectric data for the process in three amorphous polymers appear to be badly behaved at high frequencies and this result brings into question the physical meaning of such functions for this class of relaxation processes. Taken together with recent ndings8,9 from multinuclear multidimensional NMR process in glass-forming relaxation experiments for the polymers it seems likely that apparent rst- and second-order memory functions have no physical meaning in these cases but are simply related to the relaxation parameters contained in a broad distribution of relaxation times resulting from a weighted sum of parallel processes.43
ACKNOWLEDGMENT

We are pleased to acknowledge an Erasmus Grant to J. Fournier.

APPENDIX A

We seek to obtain relationships between the complex for a material and the memory dielectric permittivity functions K 1 (t), K 1 ( ) in Eqs. 7 and 9 of the text. Reand (t) and (t), the macrolationships between scopic and molecular relaxation functions have been described.1,32,33 For the present work we shall set the internal eld factor p(i ) equal to unity32 so that (t) (t) C(t) say. It follows that the normalized complex permittivity R( ) R ( ) iR ( ) is given by the relation R iR
0

i
0

1 i C

A1

where C ( ) F [C(t)] and F indicates a one-sided Fourier transform. We require that C(t) obeys the integrodifferential equation, Eq. 7a , of the text dC t dt
t 0

K 1 t,

d ,

A2

where is the time variable. Fourier transformation of Eq. A2 and using the convolution theorem gives a relation be tween K 1 ( ) F [C(t)] and C ( ) as follows K1 1 C i . A3

/2 then Thus if C ( ) is known at a given frequency f ( ) a complex quantity generally is determined at that K1 frequency. Combining Eq. A1 to A3 we obtain the re quired relations between (R ,R ) and K 1 at : K1 K1 R 1 R R
2

The relationships between complex permittivity, dipole moment correlation function, orientational distribution function, and its moments, and rst- and higher-order memory functions have been described briey and a method has been

R
2

, R
2 2

A4a . A4b

1 R 1 R

J. Chem. Phys., Vol. 104, No. 14, 8 April 1996

Downloaded 23 Feb 2012 to 193.136.139.253. Redistribution subject to AIP license or copyright; see http://jcp.aip.org/about/rights_and_permissions

G. Williams and J. Fournier: Memory functions from dielectric relaxation

5697

Thus a knowledge of both R ( ) and R ( ) at gives values for K 1 ( ) and K 1 ( ) at that value of . The memory function K 1 (t) is obtained by the relations K1 t F
c

K1

K1

A5

where subscripts c and s indicate cosine and sine transforms respectively. If K 1 ( ) or K 1 ( ) are known over the entire relaxation range then K 1 (t) is determined at any value of t in the relaxation region. The quantities [R ( ),R ( )] may be determined from experimental permittivity and loss data see Eq. A1 or may be provided by model theories or simulations of dielectric behavior. Similarly such values may be calculated using empirical equations for ; e.g. the equations of Cole and Cole,36 Davidson and Cole35 see Ref. 1 for example or of Havriliak and Negami37 or the KohlrauschWilliamsWatts function2123 expressed in the frequency domain, as tabulated by Koizumi and Kita.38 Note that C(t) may be determined from either R ( ) or R ( ) through the transform relations32,33 C t R F
c

but from Eqs. A4 and A5 we see that K 1 (t) obtained from K 1 or K 1 appears to require a knowledge of both R ( ) and R ( ) over the relaxation range. However a knowledge of R ( ) over the relaxation range gives values of R ( ) and vice versa) via the familiar KronigKramers relationships.

APPENDIX B

Consider the case where C(t) is given by a weighted sum of single exponential decay functions (t) exp (t/ i ). i C t ai
i

t .

B1

We require that C(t) obeys the memory function equation A2 and hence Eq. A3 . From Eq. A3 and B1 we write K1 Writing K 1 ( )
ia i / ia i

1 R F
s

1 i i/ 1 i

i i

B2

A6

K1( )

iK 1 ( ) we obtain

K1 K1

i a i /p i

i a i x i /p i i a i x i /p i 2

2 i a i x i /p i i a i x i /p i 2 2 ia ix i / p i 2 i a i x i /p i i a i x i /p i 2 2 ia ix i / p i

B3a

i a i /p i

i a i x i /p i i a i x i /p i 2

B3b

2 2 where x i i and p i (1 i ). The limiting behavior for 0 and follows from Eq. B3 as

2 2 a i /(1 i ) etc., and the averaged quantities in Eqs. , etc. B4 would correspond to P( ) d

K1 K1 K1 K1

0 0

;
2 2 2

B4a 0,

APPENDIX C

B4b

We enquire of the nature of K 1 (t, ) when C(t) is given by a weighted sum of single exponential time-functions i (t) Eq. B1 . Since dc i t dt 1
i t

; 1
2

B4c 1
2

ci t

t
i

ci

d .

C1

B4d

Note: here is a general time variable and should not be confused with i which is the relaxation time for process i. It follows that dC t dt ai dc i dt
t 0 i

m m m and m 1,2. where m i a i i , 1/ ia i i For a single relaxation process, (t) exp t/ 0 , we 1 0 for all values of see have K 1 ( ) 0 and K 1 ( ) Eq. A8 . This corresponds to a memory function K 1 (t, ) (t )/ 0 in Eq. A2 . Thus for a discrete distribution of relaxation times, K 1 ( ) and K 1 ( ) may be calculated from Eqs. B3 . The above treatment may be generalized to the case of a continuous distribution of relaxation times P( ). Thus in Eq. B3 2 2 one would write P( )/(1 )d in place of

ai

t
i

ci

d . C2

In order for C(t) to conform with Eq. A2 we write dC t dt

t 0

K 1 t,

a ic i

d .

C3

In order to force the equivalence between Eqs. C3 and C2 we have

J. Chem. Phys., Vol. 104, No. 14, 8 April 1996

Downloaded 23 Feb 2012 to 193.136.139.253. Redistribution subject to AIP license or copyright; see http://jcp.aip.org/about/rights_and_permissions

5698
ia i

G. Williams and J. Fournier: Memory functions from dielectric relaxation

K 1 t,

t
ia ic i

ci

C4

For a chosen time t 1 , say, it follows that K 1 is a delta function at t 1 and takes the value K1 t1 ai
i i

c i t 1

a ic i t 1 .

C5

In Eq. C5 a i and (a i / i ) are static and static dynamic weighting terms. Thus the apparent memory function K 1 (t) varies with time by differentially weighing the components of the distribution of relaxation times, according to Eq. C5
N. G. McCrum, B. E. Read, and G. Williams, Anelastic and Dielectric Effects in Polymeric Solids Dover, New York, 1991 . 2 G. Williams, in Dielectric and Related Molecular Processes, Spec. Period. Reports The Chemical Society, London, 1975 , p. 151. 3 G. Williams, in Comprehensive Polymer Science, edited by G. Allen and J. C. Bevington Pergamon, Oxford, 1989 , Vol. 2, Chap. 7, p. 601. 4 J. D. Ferry, Viscoelastic Properties of Polymers Wiley, New York, 1980 , 3rd ed. 5 V. Rosato and G. Williams, in Molecular Interactions, edited by H. Ratajczak and W. J. Orville-Thomas Wiley, New York, 1980 , Vol. 3, Chap. 8, p. 373. 6 G. Meier and G. Fytas, in Optical Techniques to Characterize Polymer Systems, edited by H. Bassler Elsevier, Amsterdam, 1989 . 7 J. S. Higgins and H. Benoit, High Polymers and Neutron Scattering Oxford University Press, Oxford, 1994 . 8 K. Schmidt-Rohr and H. W. Spiess, Phys. Rev. Lett. 66, 3020 1991 . 9 K. Schmidt-Rohr and H. W. Spiess, Multi-dimensional Solid State NMR and Polymers Academic, London, 1994 . 10 L. Monnerie, in Static and Dynamic Properties of the Polymeric Solid State, NATO ASI, edited by R. A. Pethrick and R. W. Richards Reidel, Dordrecht, Holland, 1982 , p. 383. 11 M. T. Cicerone, F. R. Blackburn, and M. F. Ediger, J. Chem. Phys. 102, 471 1995 . 12 P. D. Hyde, T. E. Evert, M. T. Cicerone, and M. D. Ediger, J. Non. Cryst. Solids 131133, 42 1991 . 13 S. Matsuoka, G. Williams, G. E. Johnson, E. W. Anderson, and T. Furukawa, Macromolecules 18, 2652 1985 . 14 S. Matsuoka, Relaxation Phenomena in Polymers Hanser, Munich, 1992 . 15 I. M. Hodge, J. Non Cryst. Solids, 169, 211 1994 . 16 J. Wong and C. A. Angell, Glass: Structure by Spectroscopy Marcel Dekker, Basel, 1976 . 17 G. Williams, Adv. Polym. Sci. 33, 60 1979 . 18 G. Williams, in Materials Science and Technology, edited by E. L. Tho1

mas VCH, Weinheim, 1992 , Vol. 12, Chap. 11, p. 471. The proceedings of the 1st and 2nd International Meetings on Relaxations in Complex Systems, organized by K. L. Ngai and G. B. Wright in Crete in 1989 and by K. L. Ngai, G. B. Wright, and E. Riande in Alicante in 1993, were published as special issues of J. Non Cryst. Solids 131133 1991 and 172174 1994 respectively. These issues contain many papers pertaining to the multiple relaxations in glass-forming liquids and polymers as studied by different techniques. 20 G. Williams, Trans. Faraday Soc. 62, 2091 1966 . 21 G. Williams and D. C. Watts, in Nuclear Magnetic Resonance, Basic Principles and Progress, Vol. 4, NMR of Polymers Springer-Verlag, Berlin, 1971 , p. 271. 22 R. Kohlrausch, Ann. Phys. Leipzig 12, 393 1847 . 23 G. Williams and D. C. Watts, Trans. Faraday Soc. 66, 80 1970 . 24 G. Williams, D. C. Watts, S. B. Dev, and A. M. North, Trans. Faraday Soc. 67, 1323 1971 . 25 G. Williams, Trans. Electr. Insul., IEEE E1-20, 843 1985 . 26 G. Williams, Trans. Electr. Insul., IEEE E1-17, 469 1982 . 27 W. Gotze, in Liq. Crist. Vitreuse, edited by J. P. Hansen, D. Levesque, and J. Zimm-Justin North-Holland, Amsterdam, 1988 , p. 287. 28 W. Gotze, Rep. Progr. Phys. 55, 241 1992 . 29 G. F. Mazenko, J. Non Cryst. Solids 131133, 120 1991 . 30 J. F. Douglas and J. B. Hubbard, Macromolecules 24, 3163 1991 . 31 B. J. Berne, in Physical Chemistry, An Advanced Treatise, Vol. VIIIB, The Liquid State, edited by H. Eyring, W. Jost, and D. Henderson Academic, New York, 1971 , p. 540. 32 G. Williams, Chem. Rev. 72, 55 1972 . 33 G. Williams, Chem. Soc. Rev. 7, 89 1978 . 34 M. Cook, D. C. Watts, and G. Williams, Trans. Faraday Soc. 66, 2503 1970 . 35 D. W. Davidson and R. H. Cole, J. Chem. Phys. 18, 417 1950 . 36 K. S. Cole and R. H. Cole, J. Chem. Phys. 9, 341 1941 . 37 S. Havriliak and S. Negami, J. Polym. Sci. C 14, 99 1966 . 38 N. Koizumi and Y. Kita, Bull. Inst. Chem. Res. Kyoto Univ. 56, 300 1978 . 39 G. D. Patterson, C. P. Lindsey, and J. R. Stevens, J. Chem. Phys. 70, 643 1979 . 40 M. T. Cicerone and M. D. Ediger, J. Chem. Phys. submitted . 41 V. Daniel, Dielectric Relaxation Academic, London, 1967 , p. 57. 42 M. Schroeder, Fractals, Chaos and Power Laws Freeman, New York, 1991 . 43 Note that although the dynamic heterogeneity model implies independent parallel processes within the overall time-scale for an -relaxation process, all relaxors must be equivalent dynamically for T T g i.e., it is an ergodic system when proper statistical averaging is made for the ensemble . This means that relaxors that are slow at t 0 become fast at later times, fast relaxors at t 0 become slow at later times and the entire distribution of relaxation times is spanned, on average, by each relaxor.
19

J. Chem. Phys., Vol. 104, No. 14, 8 April 1996

Downloaded 23 Feb 2012 to 193.136.139.253. Redistribution subject to AIP license or copyright; see http://jcp.aip.org/about/rights_and_permissions