(2) ( F= 2E1 + e2

E.

Onsager,2 3 and Kirkwood4 have called this the cavity field. In Onsager's excellent treatment of the problem of dipolar liquids, account was taken of long-range interactions between the dipole in the central cavity and the surrounding medium. In Kirkwood's approach the short-range interactions between nearest neighbors also were considered. In particular, on introduction of the internal dielectric constant, n2, and the assumption of spherical isotropically polarizable molecules, Onsager obtained

e-n2) (2,E+ n2) ( _4__A V _2_212

V \3kT/\ 3 /

Here n is the refractive index appropriate to electronic and atomic polarizability, N is Avodagro's number, V is the molar volume of the liquid at temperature T, k is the Boltzmann constant, and M the value of the permanent dipole moment characteristic of the isolated molecule.

In his derivation of equation (2), Onsager neglected short-range order. He assumed all angles between the molecular dipole and the applied field to be equally likely aside from the orientation energy in the local field. This postulate makes equation (2), as it stands, inapplicable to ice or water. This assumption cannot be even approximately true for a perfect single crystal of ice I. The dipole of a water molecule points from the oxygen atom to the midpoint between its two hydrogen atoms. There are six ways of selecting which two of the four neighboring hydrogen atoms in ice I are to belong to the central oxygen atom, and these six ways correspond to the water dipole pointing in the plus and minus direction of each of the three axes of an appropriately chosen rectangular coordinate system. Now if the water dipole were free to take up each of these six positions, the mean value of cos2 0', where 0' is the angle between the direction of an applied field and the water dipole, would be 1/3 as it is for free rotation and one would still arrive at equation (2). However, if ice had this sixfold degeneracy, there would be a corresponding entropy per mole of R In 6 persisting to low temperatures instead of the 0.870 eu observed by Giauque,5 which, as Pauling6 points out, closely approximates an entropy of R In 6/4. Pauling's argument for 6/4 for the degeneracy proceeds as follows. Since neighboring oxygen atoms are equally likely to pre-empt the intermediate hydrogen atom, there is only a chance of (1/2)2 that a particular one of the six orientations of the water dipole be allowed. This gives the degeneracy 6/4 which is in close agreement with experiment. This degeneracy of 6/4 per water molecule allows some randomness of molecular orientation with respect to the direction of maximum polarization for a domain. The result is a decrease in the average component of the dipole moment, ,u cos 0, in the direction of maximum polarization of the domain.

Accordingly, the model discussed below differs from the Onsager model in one essential respect, namely, the method of obtaining the polarizability attributable to the permanent dipoles.

In Onsager's model this polarizability, obtained by space averaging ,u cos 0 over all possible values of the angle 0. This assumes that the relative population of dipole vectors is not constrained by shortrange directional interactions and leads to an incorrect value of this polarizability in cases where the dominant forces are directional short-range interactions. Kirkwood attempted to account for the short-range directional interactions by introduction of a correlation factor between a central molecule and nearest neighbors, but again it was assumed that the central dipole was free to orient in a local field, F.

The model suggested by the present authors is that the Bernal and Fowler7 and Bjerrum8 proposals for the structure of water (see Fig. 1), and by implication many .~~~.~~.~.~~.~~~~5~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~.... other hydrogen-bonding liquids and solids, be introduced immediately into the evaluation of the average dipole moment in the field direction attributable to permanent dipoles. This is done by assuming the following:

( 1 ) Solid or liquid water is made up of a mosaic of roughly bricklike domains with the dipoles in a particular domain having an average resultant moment /I cos 0 along the direction of maximum polarization for the domain while the direction of maximum polarization of neighboring domains is rotated through 1800 with respect to the first. For perfect tetrahedral bonding, cos 0 approaches its maximum value of 1, while for bent hydrogen bonds9 cos 0 is correspondingly smaller.

(2) When an electric field is applied to the system, the directions of maximum polarization of domains orient in the field until all domains are either lined up with the field or against it. Further, these domains with their directions of maximum polarization oriented in the direction of the field increase in volume at the expense of domains appositively oriented until equilibrium is attained. The resulting mean dipole moment for solidlike molecules is then scos OF

kT = .u Cos Oe p cos0 F

kT e -M cos Oe -coo F

kT + e -jA cos0 F kT _2 COS20 F

. provided cos OF k «1.

(3) Here F is the local field. Thus for ice and the solidlike part of water the factor

A2 JA~~~~c~~o~~s~22 0 3kT in equation (2) must be replaced by ' , while for the gaslike portion the factor -U-T is retained since such molecules oriented freely in the local field, F.10 (3) The interface between domains in the mosaic is made up of Bjerrum faults, vacancies, and related mismatch structures.

According to the significant structure theory of liquids," the fraction of molecules having solidlike character is V8/V, where V8 is the molar volume of the solidlike structure in the liquid and V is the volume of the liquid at the temperature of observation. The remaining fraction of the molecules, (V - V8/V), is gaslike.

Calculation of Dielectric Constants.-In the light of the foregoing discussion and the evidence from significant structure theory" that the fraction V,/V of the molecules of a liquid has solidlike properties and the remaining fraction (V-V-)IV has gaslike properties, we expect the dielectric constant of water to be given by (e - n2) (2E+ n2) = nN4(n + 2)2[V oAs 0 + V

VsT] For ice, one has (E- n2) (2,E+n2) =rNM/2 cos20(n2 +2)2

3e V's kT ( 3)

For convenience in actual calculations, equations (4) and (5) may be rearranged to give E e 2 + _ 2_n+ 2 _2 1 - + -y+ 67rN COS2O<(l )(77)Vs

1 (i-3os + co2](6) (4) (5) and e =2 E=2 +2e S'k2nosr0NA2)(n2+ '2I2V( 1TI (7) The value of cos2 0 can be determined as a parameter by evaluation at a single temperature using experimental quantities for the other factors in equation (7). The small deviation of cos2 0 from unity reflects ( 1 ) any approximations in our model, (2) approximations in the value used for ;A and n, and (3) deviations from tetrahedral structures due to bending of hydrogen bonds, and finally (4) some decrease in polarization in a domain because of the degeneracy of 6/4 per molecule.

As an application of equation (6), the calculated values of the dielectric constant of water are compared in Table 1 with observed values at several temperatures. In the calculation the one parameter, cos2 0, is taken as 0.964 and ,u as 1.84 debyes for both light and heavy water. The fact that for all forms of ice and water the values found for cos2 0 are nearly equal to unity is an extremely satisfying result.

For comparison, Oster and Kirkwood4b obtained, using the Kirkwood correlation parameter, the data shown in Table 2.

Judging from the data in Table 2 for the small temperature interval involved, the Kirkwood correlation parameter gives a poorer agreement with the temperature coefficient than that given by equation (6). Also, the Oster-Kirkwood model, which assumes free rotation about the hydrogen bond of the nearest neighbors of a central molecule, seems to be less appropriate for water than our model of water as a mosaic of domains.

As an application of equation (7), the calculated values of the dielectric constant of various forms of ice are compared in Table 3 with observed values at several temperatures. Unfortunately, we have not sufficient experimental data available for a good comp2a2ri3so~ne~xc~ep~t ~fo~ri~ce~I,~2w~hi~ch~s~ho2~ws~e~xce~ll~e1n~t~agreem0entwith experiment.

In the calculations, the following values of cos2 0 are used 0.977 for Ice I 0.949 for Ice V 0.90 for Ice III 1.138 for Ice VI Factors Affecting the Dipole Moment in Condensed Phases.-The factor (n2 ta2) corrects for the increase in dipole moment due to induction. The fact that for Ice VI the estimated value for cos2 0 is greater than 1 suggests that a slightly larger value of n2 Would probably be appropriate.

Many solids containing polar molecules show little polarization arising from molecular dipoles in the solid state. This is to be expected if the dipoles are arranged in the solid so that their mean component along the direction of maximum polarization of a domain is small or zero.

When the vapor pressure of a system gets high, the vacancies in the liquid tend to cluster like molecules in the vapor, and the gaslike part of the polarizability should be multiplied by the factor -1 Xn HAne\2raeiXsonthe mole fraction of molen \A/ cules in clusters of n molecules having a resultant dipole moment An.

Relaxation Times.-As Debye10 showed, the dielectric constant ea is a function of frequency and satisfies the equation where according to absolute rate theory,'4 we write 1 - = 2Tr , kT -exep h _AGt

RT.

(9)

Here eo and E. are the dielectric constants at low and high frequencies, and co is the circular frequency of the applied potential. The relaxation times for ice 13,250 = 7.7 X 10-1 e

-13,250 given by Auty and Cole'5 are TH,0 = 5.30 X 10-6 e 13,400 RT

for solid deuterium oxide. = 9.4 X 104 e for ordinary ice and kDo = 6.5 X 104 e rates of relaxation of light and heavy ice, respectively. ast

for ordinary ice and TDO

-13,400 for the specific

constant where the entropy of activation, ASt, is small is 1013. If we divide 9.4 X 104 by 101', we obtained 9.4 X 10-9 for the fraction, f, of molecules which are situated so they can rotate from one domain into a neighboring domain better oriented with respect to the field. If, for each domain, only one molecule at a time is so situated that it can migrate to a neighboring domain, the linear dimension of the domain would be about a tenth of a micron. If more molecules can migrate at a time, the volume of the domain would be proportionately larger, and if the entropy of activation, ASt, is negative, the domain is correspondingly smaller. From this point of view the model seems reasonable. If the reaction coordinate were rotation about an axis normal to the axis of the dipole, the frequency factor would be larger for H20 than for D20 by a factor V2. The reported frequency factors are at least not inconsistent with this requirement. Manson, Cagle, and Eyring'6 have found 13,800 calories for the activation energy for the rate of growth of microcrystals of ice from supersaturated vapor up to a size where they can be seen. The saturated vapor was seeded with AgCl to nucleate the liquid drops. The rate-determining step is the growth of of one ice crystal at the expense of another. This correspondence in activation energies for crystal growth and passage between domains supports our model of the mechanism of dielectric relaxation in ice.

Manson, Cagle, and Eyring'6 interpreted their results to show that a liquid domain of about 40 water molecules must freeze as a unit to form the bridge required in crystal growth. Jhon, Grosh, Ree, and Eyring"5 were able to explain the thermodynamic properties and viscosity of water on the assumption that it contained domains of about 46 molecules which changed as a unit from an "ice I-like" to an "ice III-like" state. These findings suggest the expected result that the domains which are hydrogen-bonded together should be much smaller in water than in ice, i.e., liquids lack long-range order.

Because the domains of long-range order in water are much smaller than in ice and because of an abundance of fluidized vacancies in water, we expect that the relaxation time for water will have an f factor much inearer unity and an activation energy about 10,000 calories less than for ice. This is in accord with the value of r = 0.9 X 10-11 reported by Collie, Hasted, and Ritson'7 for water. The relaxation time for D20 is about V2 times as long as for H20 in qualitative agreement with the findings in ice and with our expectations that the reaction coordinate will approximate a molecular rotation.

Conclusions.-In this paper there are several points to be emphasized. ( 1 ) Short-range order in many solids and in hydrogen-bonded liquids prevents the reorientation of a molecular dipole by rotating locally in response to an electric field. However, only under extraordinary circumstances will there be long-range order extending through a macroscopic sample. Consequently, the result of an applied field will be to make those domains grow which are favorably oriented with respect to the field at the expense of the less favorably oriented until a steady state is achieved. (2) Because relaxation involves only molecules at the interface between domains, the relaxation time measures the size of the domains. The lack of sufficient degeneracy, as shown by the entropy, to allow local relaxation supports this point of view. (3) The lack of long-range order in liquids allows one to predict much shorter relaxation times for this state. (4) Significant structure theory by introducing the concept of solidlike and gaslike molecules leads to an explanation of the observed temperature coefficient of the dielectric constant in liquids.

Debye's'9 calculation of the temperature dependence of the dielectric constant of water agrees well with experiment. It consists in assuming that each water molecule is restrained by a local potential field of about lOkT. The difficulty with this explanation is that the relaxation time should be the reciprocal of a vibrational frequency which is more than an order of magnitude shorter than the experimental value. This forces us to turn to domains as the explanation of relaxation. The observed orientational entropy indicates there is too much local restriction to obtain polarizations of ice and water of the magnitude observed by rotation in situ.

Domains with large electric moments would tend to line up parallel and antiparallel just as magnets juxtapose poles of opposite kinds. Our assumption that there are few domains with their direction of polarization oriented at angles intermediate between zero and 1800 is justified in this way. The presence of domains with intermediate orientation would lower the polarization and in the limit of complete randomness would lead to a factor of 1/3 multiplying 1s2/kT. This would require that the calculated Onsager field be low by just enough to compensate for this randomness.

Investigators have frequently looked, unsuccessfully, for the domain structure in water which our considerations suggest may be present. The low-angle X-ray scattering in water does not differ appreciably from that found for liquid argon. A fairly broad distribution in domain volume around a mean value of about 46 molecules may explain this discrepancy.

In subsequent publications, the dielectric constants of some of other well-known systems18 will be investigated using our proposed model in conjunction with the significant structure theory of liquids.

One of the authors (M. S. Jhon) wishes to thank the National Science Foundation for support of this work. Dean Eyring wishes to express his appreciation for the opportunity of discussing some of the above problems with Professor Peter Debye.

* Duke University. t University of Utah.