DIPOLEMOMENTS IN POLYVINYL CHLORIDEDIPHENYL SYSTEM
Feb., 1941
or in other words, a t the transition from brittle to elastic. Furthermore, if we put on the log ET curves the temperature a t which the loss factor for a given frequency reaches its peak, we find that the modulus a t the temperature of maximum loss varies by about a factor of two for this range of composition. This is about what would be expected from Maxwell’s equation6 v = (1/2)
(R)1
(12)
BOA
where q is viscosity, p is Poisson’s ratio, Eo is the coefficient of Young’s modulus in the form E = E0e(”” and X is the time of elastic relaxation. We previously showed (Table I) that the d. c. conductance was constant at the temperature of maximum absorption for a given frequency for different concentrations of diphenyl and argued that this means equal viscosities. Then if 7 is constant, we have EoX constant, and since X decreases with increasing temperature, the modulus should increase with increasing temperature, as is actually the case according to the data shown in (6) Maxwell, Phil. Mag J. Sci.. IV. 35. 134 (1867).
[CONTRIBUTION FROM THE
385
Fig. 13. We hope to find further correlations between electrical and mechanical properties of these systems, which may lead to a clearer understanding of the molecular mechanism involved.
Summary 1. The system polyvinyl chloridediphenyl has been studied over the ranges 70 to +lOO”, 60 cycles to 10 kilocycles and 02075 diphenyl. 2. Further evidence that the electrical properties of polar polymers are due to a viscosity controlled relaxation mechanism is given. 3. As a function of composition, the loss factor for a given frequency and temperature goes through a maximum at a concentration which is characteristic of a given plasticizer. 4. The low temperature absorption in polar polymers is reduced and eventually eliminated by the addition of a second component. 5. The electrical properties are markedly nonlinear in composition in the low concentration range of compositions. SCHENECTADY, NEWPORK RECEIVED AUGUST3, 1940
RESEARCH LABORATORY O F THE GENERAL ELECTRIC CO.]
Electrical Properties of Solids. VIII. Dipole Moments in Polyvinyl ChlorideDiphenyl Systems* BY RAYMOND M. Fuoss
I. Introduction.It is now well known that the electrical properties of polar systems containing giant molecules differ markedly from those in which the dipoles are on small molecules. The properties of the latter are well accounted for by Debye’s theory of anomalous dispersion in which the fundamental assumption is a finite time of relaxation, specified by the size and shape of the polar solute molecule and the temperature and viscosity of the solvent. Three characteristic differences between the polymeric systems and the systems describable in terms of a single relaxation time appear: (1) the inflection slopes of the dielectric constant logfrequency curves are much lower in the former; ( 2 ) the corresponding loss factor logfrequency curves have a much larger halfwidth; and (3) the maximum loss factor is always much less than half the difference between * Presented at the Fifth Annual Symposium of the Division of Physical and Inorganic Chemistry of the American Chemical Society, Columbia University, New York, December 30, I940 t o January 1, 1941.
AND JOHN
G. KIRKWOOD
the static dielectric constant and the square of the index of refraction. These differences can all be accounted for by assuming that many times of relaxation rather than a single one are involved. Wagner assumed a Gaussian distribution of relaxation times in order to account for discharge curves, and Yager2 has developed a method for applying the Gauss distribution to an analysis of a. c. data. We have, however, no actual proof of the necessity for the existence of a distribution of relaxation times for a given system, nor have we any right to assume that a possible distribution should be Gaussian. In fact, for data on polyvinyl chloride systemsS it has been shown that the Gauss distribution will not reproduce the experimental results. It is, of course, qualitatively plausible that a system containing a polar polymer should exhibit a distribution of relaxation times, because every di(1) Wagner, An* Physik, 40, 817 (1913). (2) Yager, Physics, 7, 434 (1936). (8) FUOM, TEIS JOURNAL, 63, 369, a78 (1941).
RAYMONDM. Fuoss AND Jom G.KIRKWOOD
386
pole in a given chain is coupled to neighboring dipoles of the same chain by primary valence bonds, so that the motion of any dipole affects the motion of its neighbors, and they in turn influence its response to a torque. Furthermore, in the various configurations which a chain molecule can a s s ~ m e we * ~ can ~ find now one, now another segment of the chain acting effectively as a cooperating electrical unit, and these segments will, of course, vary in length between the two improbable extremes of a single monomeric unit and the whole extended chain. I n a later paper we shall show that a linear polymer must exhibit a distribution of relaxation times as a necessary consequence of its structure. In this paper, we present a solution of the following problem: given the lossfactor logfrequency curve, to find the distribution function for the relaxation times. A unique solution is presented for the case where the data can be empirically represented by an integrableanalyticfunction. For polar polymers of the type described so far in this series of papers, we find 8 LJ A sech CY x (1) where e" is the loss factor, A is a constant characteristic of the system and derivable from dielectric constant data, a is a parameter which measures the width of the distribution and x is the natural logarithm of the ratio of the frequency at maximum absorption to the frequency at which e' is measured. (For a single relaxation time, a obviously is unity and Eq. 1 reduces to the Debye formula.) The approximate equality sign appears because a function H(x) is actually used in place of er', which is, for the cases treated here, closely proportional to er', The function H(x) was introduced in order to eliminate much of the needless awkward algebra usually encountered in separating a complex quantity (the a. c. vector) into its real and imaginary parts. It is convenient to use a logarithmic scale for the relaxation times. Let T , be the relaxation time corresponding to maximum H and
Vol. 63
tion of relaxation times and the electrical properties. These properties of a polar system are in last analysis an average over the dipoles present in the system, and we should therefore be able to derive the average moment ,liper monomer unit from measurements of volume properties of polar polymers. This calculation will be made for the case of a polymer of the alternating structure (CH&HX),, where X is a simple polar group. It is based on the modifieda Onsager7 field, which must be used to replace the Lorentz field in condensed systems. If free rotation at each bond is permitted (or else restricted rotation with three equally probable energy minima), the average moment is given by 7.j
= &Pd2
(5)
where is the moment of the equivalent group isolated in free space. Comparison with experimental data for the system polyvinyl chloridediphenyl verifies Eq. 5 and, furthermore, confirms the conclusion previously reached on empirical grounds that the dipoledipole interaction usually described as association in polar liquids is negligible for the type of polar polymers considered here. 11. The Reduced Polarization.In previous treatments of electrical properties of matter, the generalized dielectric constant E
=
6'
 is"
(6)
introduced by Debye has usually been the basis of the calculation. Here e' is the ordinary dielectric constant, and E' is the loss factor. Phenomenologically, however, we are concerned with the polarization P per unit volume. It is possible to set up a relationship between P and e, but since they are both complex quantities, most calculations become needlessly complicated on account of the algebraic manipulations involved in separating their components. A quantity, Q(w), the reduced polarization, will be introduced: as will be seen below, it is intermediate between P and e, and either is very simply derivable from it. Mathes = In T / T (2) Define matical analysis of data and subsequent theoretiF(s) = TG(T) (3) cal treatment lose much of their previous trivial where G(T) is the distribution function for the re awkwardness, when calculations are based on the laxation times. It will then be shown that components of Q(w). t F ( s ) s= H ( s h / 2 ) H(s k / 2 ) For a two component system, with mole frac(4) is a general relationship connecting the distribu tions x1 and q of components of molecular weight
+
+
(4) Kuhn, KolloidZ., 68,2 (1934). (6) Gut& nnd Mark, Mondrh., 66, 93 (1034).

(6) Kirkwood. J. Chtm. Phys., 7, 911 (1989). (7) Onspgu, Tars JOOWAL, S8, 1486 (1936).
DIPOLE MOMENTS IN P O L ~ I N Y CHLORIDEDIP~NYL L SYSTEMS
Feb., 1941
MI and Mz,we have, by an obvious extension of the modified Onsager theory8 (e
 1)(2e + 1) MIXI+ MZXZ 9C
P
plxl + p p , ( 7 )
If we assume that the first component is nonpolar and the second has permanent dipoles, we may set PI = 4nNa1/3 (8) and
+
P I = ( 4 ~ N / 3 ) [ ( ~ 2(~*F)ao/3RTl
(9)
where a1 and a2 are the electronic polarizations and ( p ~ l . / lis) ~averaged ~ over all internal configurations of the molecule. In (9), p is the moment of a molecule of species 2 while ,iiis the vector sum of p and the moment induced by the molecule in its environment. Substituting (8) and (9) into (7), we find
 1)(2€+ 1)  (ne  1)(2n2 + 1) 9na
PZ'XdV
where n is the index of refraction, Pz' = 4 ~ N p * F / 9 k T and the molar volume Vis given by
+ MZXZ For further simplification, let dx) ( x  1)(2x + l ) / x PV = Mixi =
(10)
(13)
 9(n2)1V= 9Pa'xa
P(€) =
2E'
 i(2a")
 1  I/€'
(14)
The above equations derived for a static field may be extended to the general a. c. case. We denote by P2'(w) the polarization at frequency f = w/2n and define the reduced polarization as
J ( x ) = [2d
 1  l/d  9(n')1/9Z
and
Q(w)
=Jma 1 + iw7 d7 0
where G(T) is a distribution function in relaxation time, such that G(r)dr specifies the fraction of the total dipole moment of a molecule associated with relaxation times in the interval between T and r dr. From its definition, it follows that G(r)satisfies the following normalization condition JmG(r)d7 = 1
(17)
(18)
with w, as 2 ?r times the frequency a t maximum H. Separating (p(e), using (6),we find
&(7)
For the cases to be considered in this paper 0
>> €"I
(20)
c2(sx)]
(26)
+cWx)]
(27)
= F(s)
(28)
(29)
If we define a kernel K by the equation K(1) = d / ( 1
+ en)
= (sech f ) / 2
(30)
 x ) F ( s ) ds
(31)
(27) becomes H(x)
(8) Ref. 6, Eq. la.
+
sp6,
x = In w,/w
(25)
When G(T)is a delta function S(r  T O ) , the result reduces to the familiar Debye theory for a single time of relaxation 70. Separating (24) into its components, we find

 H(x)
(23)*
H ( x ) = 2en/9Z
=I:
Q(w) = / ( x )
(22)*
From the form of (22) and (23), we see that 'E and E" are easily determined if J and H a r e known, and conversely; provided, in both cases, that 2 is known. Q ( w ) in turn is very simply related to P'z by (15), and the latter to the moment by (11). It will also turn out that Q ( w ) is a very convenient function to use when distributions of relaxations are involved. 111. Determination of the Distribution Function.By an extension of the Debye theory of anomalous dispersion to a molecule with a large number of internal rotatory degrees of freedom, it can be demonstrated that
where PZ'(0)is the value of Ps' a t zero frequency, J(x) F(s)ds/[l corresponding to the static dielectric constant EO. and For a given system, the quantity H(x) = e s  x F(s)ds/[l = [da) d n * ) 1 / 9 (16) is a normalizing constant independent of fre where by definition quency. We can separate Q into its components s = In 7/7,,, as follows and where
(21)*
Comparing real and imaginary parts of (17) and ( 19), we find
+
Then (10) becomes [~(e)
so that to a rather close approximation
(11) (12)
387
=fm
K(s
The physical problem stated in the introduction : "given the loss factorfrequency curve, t o Equation8 marked with an asterisk are subject to the approxima
tion (20).
RAYMOND M. Fuoss AND
388
find the distribution of relaxation times” can now be restated as a mathematical problem. Given a function H(x), to find a function F(s) when the two are related by the integral equation (31) containing the kernel K . The integral equation can be solved by the use of Fourier transform^.^ The Fourier transform LH(u)of H ( x ) is defined as
Era
LH(u)
H(x)eiur dx
f
=
~
 x)
(
S
F(s) eiux dx ds
By introducing a new variable t = s  x
and carrying out the integration over t holding s constant, (33) separates as follows
which gives the result = L E ( u)
LF(U)
(34)
Then by the inversion theorem
We next compute L K (u). tour integral
I
=
$
eiu5
Consider the con
sech z dz
(36)
around the rectangle bounded by the real axis, two verticals at x = =!=Aand a parallel to the real axis at y = iB. It can easily be shown that the integrals along all but the real axis vanish as A and B approach infinity. Hence (35) becomes
I=Km
cos ux sech x dx
If now we substitute this result in (35), we obtain F(s)
=

2:
pm
cosh ( u r / 2 ) LX ( u ) eius du
(41)
Rewriting the transform of H ( x ) as an integral, we find
eiu(Sxiw/Z)]
? ~ F ( s ) H(s
(33)
 m
LH(U)
Vol. 63
dx du
But this is simply the Fourier transform of the inverse function and, by inspection, we see that
m
J
G. KIRKWOOD
(32)
Substituting (31) for H(x), we have &(u)
JOHN
(37)
+ i ~ r / 2 +) H ( s  in/2)
(42)
Our result now is that if we are given H as a function of x , i. e., the experimental e’’  logf curve, and know fm, and can represent H(x) by an integrable analytic function, we can then compute directly from the experimental data the function F(s) which gives the distribution of the times of relaxation by (29). (The conditions analytic and integrable are sufficient for the application of the Fourier inversion.) The function G(r) so obtained will be as good an approximation of the real distribution as the empirical interpolation function H(x) is for the data. On account of the exponential convergence of the functions naturally1° involved in this general type of problem, the accuracy will be best if the data contain H,,,,,, because then the large part of the integrals (26) and (27) near x = 0 are accurately determined, and the error in the empirical function for the exponential wings becomes a small part of the total integral. IV. The Empirical Distribution Function.Of a number of empirical interpolation formulas which might have been chosen to represent the data, the following was found to be the most satisfactory H ( x ) = H ( 0 ) sech a x
(43)
which is twice the Fourier transform of K(t). Here H(0) is, of course, the maximum value of The integrand in (36) has simple poles on the im H ( x ) ,corresponding tof = fm and a is a parameaginary axis a t the points ter which determines the width of the Hcurve at 2% = (n + l/2)7ri = y n (38) halfheight. The form of (43) was suggested by the fact that for a single time of relaxation, and therefore by the Cauchy residue theorem where G(7) is a Dirac delta function, the Debye I = 2?ri C iem/(sin y,,) theory gives d’/e’,,, = sech x . = 2r (1 e  u ( n + l / Z ) r (39) If we substitute (43) in (42), we find Summing the geometric series in (39), we have .. . . . finally LR(u) = ( r / 2 ) sech (m/2)= LK( u )
(40)
(9) Titchmarsh, “Fourier Integrals and Transforms,” Chap. XI, Oxford, 1937.
+ io+) in do) leadi to + w W ) on separating into real and imaginary parts, and conver
(10) “Naturally,” because the term (1
(1
sion to D logscale of frequency thus automatically introduce@a hyperbolic cosine in the denominator.
DIPOLEMOMENTS IN P O L ~ I N YCHLORIDEDIPHENYL L SYSTEMS
Feb., 1941
389
From the condition (25), which is equivalent to
Lam
F ( s ) ds = 1
(45)
we obtain by use of (44) H(0) = a/2
(46)
But by (23), we then have the result 4€Hm/9Z=
In order to determine
cy,
(47)
01
we plot
against the logarithm of the frequency; according to (43), a is determined by the slope of the resulting straight line (Fig. 1) and the intersection of the line with the horizontal axis determines fm. Then with the experimental values of e l f m and of a, we can calculate 2 by (47). This in turn gives us a value of the static dielectric constant, according to (16), which finally may be used to compute the mean moment per molecule. For comparison with experiment, several other useful relationships will next be derived. First, it is of interest to determine the halfwidth of the distribution function for the relaxation times. To find this, we determine s* such that F ( S * ) / F ( O ) = 1/2
(49)
This is equivalent to solving the transcendental equation 2 = cosh
+ sh
(.
th
r tnza r / 2
(50)
where
cy
(52)
is small, to 2 = cosh
r f r2a2r2/4
(53)
Otherwise, successive approximations must be used. A number of solutions of (50) are given in Table I. TABLE I SOLUTIONS OF EQ. (50)
a
r
0.0 .1 .2 .3 .4 .5 ,519
1.317 1.296 1.232 1.130 0.992 .831 .8
Fig. 1.Test
plots for pViC1Phe; 80:20 a t 5' intervals; lowest curve 35', highest 70'.
decade of rvalues above or below the value of r corresponding to maximum H when a = 0.42. The area under F(s) between the limits fs* accounts for a large fraction of the total integral, and is not very sensitive to a: for a = 0.1, the integral of F(s) between fs* is 0.635, for a = 0.5 is 0.587, and for cy = 0.9 is 0.505. The value x* satisfying
S*
P
12.96 6.16 3.76 2.48 1.662 1.541
0.576 .634 ,692 ,751 .812 ,873 .936 1.000
H(x*)/H(O) = 1/2
is very simple to find: it corresponds to
a = (2/r)tn1 (l/t)
and when
4
logf.
(51)
{=as*
When l is small, (50) is equivalent to
3
2
1
I
S*
0.7 .6 .5 .4 .3 .2 .1
1.215 0.946 ,723 .533 ,370 ,229 ,107
.o
.ooo
For example, a value of a = 0.42 gives s* = 2.303 by interpolation in Table I; in other words, F(s) drops to onehalf i t s peak value in the first
X*
= cosh'2
= 1.317
(54)
For x* = 2.303, a = 0.572; that is, for H to drop to one half its peak value in one decade of frequency above or below f m requires a distribution parameter equal to 0.572. For a = 0.42, the example of the previous paragraph, x* = 3.14; in other words, the Hcurve is always blunter than the Fcurve for the same value of a. For comparison with experiment, it is useful to have some values of the integral J ( x ) defined by (17) and (26). With the explicit form (44) for F(s) which is derived from the empirical form (43) of H(x), it was not found possible to carry out the integration in closed form, except for certain rational values of cos 4 2 . These results are summarized in the following list of equations: details of computation are omitted in order to save space. Let J ( a,x) be the value of J ( x ) for a given value of a. Then it can be shown that
RAYMOND M. Fuoss AND JOHN G. KIRKWOOD
390 J ( a , 0) = 1/2
+
/ ( a , % ) J(a, x ) = 1 J ( a , ) * 1 ; J(a,O) = 0 J ( 0 , x ) = 1/2 1 2 J ( 1 / 3 , ~ )=
sinh(x/3) 1 + e2r di2 cosh(2x/3)  1
(58)
1 1  eZx  4 sinh(x/2)
(59)
J(1/2,x) = 1 J(1,x) = 1/(1
+ ezx)
Vol. 63
action between different segments of the same chain for the present. Later we hope to treat more complicated models, and to remove some of the approximations involved in this calculation, which is to be considered as a first approximation. We can now define an average moment ji per monomer unit as follows
(60)
ii=dGiZn (65) It will be derivable from experimental data if we determine Pz'(0) as outlined in Section IV. Let k and I be two arbitrarily selected dipoles [aJ(o,x)/azI,a = 0 (61) along the chain and Y k l be the angle between the [ ~ J ( I / ~ , x ) / ~ x ]=~ =1/2 o  2 &/9 (62) individual CX dipoles pk and pl in a specified (63) [ ~ J ( I / ~ , X ) / ~ X=I ~9/48 O configuration and let cos Y k l denote the average [a/(i,x)/axix~ = 1/2 (64) of the cosine taken over all configurations obSome discussion of Equations (55)(64) seems tained by rotations which preserve tetrahedral appropriate. The function J ( x ) , which is closely bond angles. It is obvious that only rotations at related to the dielectric constant (cf. Eq. 22) is a bonds between pk and p1 can affect cos Ykl; as k symmetrical sigmoid through the point (1/2,0), and I are given all possible values, all chain congoing asymptotically to zero for infinite frequency figurations will be included. For p , the total moas e'+ n2 and to unity for zero frequency as e' ment of the chain molecule, we then have approaches the static dielectric constant. The inflection slope is '/zfor the case of a single time of P = C P b k relaxation (a = 1) and decreases to zero as the limit of an infinitely wide distribution (a = 0) is and for reached. For CY 6 0.3, the Jx curve is practically linear, over a fairly wide range of frequency near x = 0, which is in agreement with the ap where n proximate linearity of the €'log f curves reported for polyvinyl chloridetricresyl phosphate systems." V. The Average Moment of a Polymer and is the moment of a CX group in free space. Molecule.Let us consider a long chain polymer Combining with (65), we find of the type (CH,CHCl),, containing n monoZc=POdl+S9l (68) meric units. Assume that all polymeric mole By a method similar to the one suggested by Eycules in the system have the same chain length. ring12we find Neglect interaction of neighboring chains, by aslim S,,= S = 2@/(l  8) (69) n+ m suming the polymer present in dilute solution in a nonpolar solvent. Then we know from the work where /3 = */g, the square of the cosine of the of Mark and Kuhn4ssthat a long chain molecule tetrahedral angle.I3 is curled up, rather than extended, with an average (12) Eyring, Phys. Rev., 39, 746 (1932). (13) The details of the calculation are as follows. Neglecting the length proportional to the square root of n. Let effect of end groups, there are 2n CC bonds and n CCI bonds. p be the dipole moment of the chain molecule in a Let &sl be a unit vector in the direction of the 1th CC1 bond, &ZI+I particular configuration, and p be the vector sum a unit vector from that carbon to its nearest neighbor, and so on un& ~ k + lis the unit vector between the carbon next to the Zkth one, of p and the moment induced by the molecule in til holding the kth C1atom, and let ask be a unit vector in the direction its environment. Let ( p  l . ~be ) ~ ~a mean value of the kth CC1 bond. Then PC= ~ p a l k and taken over all configurations of the chain per= mi mitted by free rotation a t each CC bond. This The value of cos YM is thenPZgiven by average will include certain configurations which cos YM =  a 2 ~ a 2 1 require an (impossible) interpenetration of the and in order to obtain its average, we must average over rotation at between k and 1. chain. We shall also neglect dipoledipole inter allIfo fbjtheandbonds c i are mutually perpendicular unit vectors perpendicular The following table of formulas gives the slopes of the J  x curves at the midpoints, where x = 0.
(11) Fuoss, Taxa JOWWAL, 60, 461 (1938).
to at, we have a s

cos
70
a2jl f
sin
Yo
cos Qtb bnbi
+ sin
DIPOLE MOMENTS IN POLYVINYL CHLORIDEDIPHENYL SYSTEMS
Feb., 1941
Substituting (69) in (68), we obtain i; =
4 P0/2
(70)
The significance of (70) is that the average moment per monomer unit is