_J. Phys. Chem.

Solids

Pergamon Press 1962. Vol. 23, pp. 447450.

Printed in Great Britain.

DIELECTRIC

CONSTANTS
W. COCHRAN

AND

LATTICE

VIBRATIONS

and R. A. COWLEY

Crystallographic Laboratory, Cavendish Laboratory, Cambridge

(Received 7 September

1961)

relation is derived between the frequencies of lattice vibrations of long wavelength whose wave vector is in the a-direction in a crystal, the dielectric dispersion frequencies, and the cLQ of the dielectric tensor. In crystals ratio of the static and high-frequency limits of the component Q having orthorhombic symmetry or higher, the dielectric dispersion frequencies are equal to the frequencies of transverse optic modes for which the polarization is along a crystallographic axis, and the relation may then be expressed in a simpler form.
Abstract-A

THE static and high frequency dielectric constants of a crystal are related to the long wavelength limit of the frequencies of the lattice vibrations. LYDDANE et u1.u) showed, for diagonally cubic crystals with two atoms in the unit cell, that e(O) -=
&d

%5 2 -. ( UT >

(1)

Here e(s) is the static clamped dielectric constant and l @) is the high frequency dielectric constant-strictly speaking it is the dielectric constant at zero frequency when the contribution from the displacements of the atoms is suppressed. The frequencies WL and WT are the long wavelength limits for longitudinal and transverse vibrations respectively. Equation (1) can be obtained from the macroscopic theory of optical vibrations;(s) it is therefore independent of the model used to describe the crystal but does involve the electrostatic, adiabatic and harmonic approximations. The relation has been extended by one of US(~) so as to apply to diagonally cubic crystals with any number of atoms in the unit cell. A further extension to cubic crystals which are not diagonally cubic was made in the Appendix to another paper(d) but the proof was based on rather unsatisfactory assumptions. A modified form of equation (1) has been obtained by MERTEN(5) for uniaxial crystals which have two atoms in the unit cell. We show in this paper that these results are all special cases

of a more general result which applies to any crystal for which the adiabatic and harmonic approximations are valid. Equation (1) has in fact been found to hold within the limits of experimental error for potassium bromide.@) The existence of these relations is important for the theory of ferroelectric crystals.@) The equations which describe the vibrations of a crystal for long wavelengths can be obtained by postulating an energy density 4 at every point in the medium, which is a quadratic function of the macroscopic electric field E and the displacement U at the same point. E is a three-dimensional vector, but U is a 3n-dimensional vector whose components represent the displacements of the n different types of atoms, identified by k = 1,2, . . . z. In a wave of wave-vector q and frequency w the field at a point r will be E exp i(q r-d) and the displacement of the atoms U exp i(q r-d). The energy density can be written in matrix notation as
l l

+ = BU+LU-

U+QE-iE+XE

(2)

which may be compared with equation (33.1) of BORN and HuANG(~). Here U+ represents a transposed matrix, and L is a 3n x 3n matrix whose elements L,,(kk) depend on the force constants between atoms of type k and k. Q is a 3n x 3 matrix which may be referred to as the apparent charge density matrix, since its elements relate the polarization of the crystal to the displacement

447

448

W. COCHRAN

and R. A. COWLEY The 3n eigenvalues C$ of the equation waW = NW are the dielectric dispersion frequencies;(a) from equation (7) they are also the frequencies of modes of vibration for which p-1 is much greater than any dimension of the crystal but the macroscopic field is suppressed, for example, by having the crystal immersed in a conducting medium. Three of these frequencies correspond to acoustic modes for which !C$ = 0, say j = 1,2,3. This reflects the fact that the matrix N is singular, due to the relations (6) between its elements. If those rows and columns of N which involve the kth atom are deleted, then the minor w is of order its determinant is independent of the choice of k, and

of the atoms [see equation (4)]. The elements of the 3 x 3 matrix X are the components of the electronic susceptibility tensor. It is also useful to define a diagonal matrix p for which p&AK) = Pk for all a, where pa is the mass density of atoms of type k. The equations of motion are then &pU = LU-QE, is given by (3)

the polarization P = -&$/aE

P = Q+U+XE
and the macroscopic electric field by

(4)

E = +(+)($f)
where it is assumed throughout that 4 + 0. Equations (3) and (4) are given here as following from the potential function (2), as has been shown in more detail by BORN and HuANc~~). They may in fact also be derived from a wave-mechanical treatment of the theory of the lattice dynamics of crystals with polarizable atoms, which shows that equations (3) and (4) include the effect of all the electric multipoles of the electronic distribution around the atoms. (7) The condition that all results are invariant for a uniform translation of the crystal leads to the identities

3n-3,

3n

Det. N =

I-r j=4

Q23.

(9)

The matrix N is singular so its inverse cannot be defined in the usual way; however, an effective inverse N-1 may be formed by taking the inverse of N and adding zeros in those rows and columns which were eliminated from N to give #. When the crystal is in equilibrium in a static field, we have from equation (7) that NW = ZE. Using this result to eliminate W from equation (8) gives a relation between P and E, and leads to the following expression for the static dielectric constant :

p,b(kk)
and 1

= xL,,(kk) e

= 0 for all (a@

8;

= 6,+4r(Z+N-lZ+X),,,

(10)

Q&k)

= 0 for all (afi).

(6)

For our present purpose it is convenient to eliminate the matrix e by defining W,(k)
~a,dkk)

where the suffices a/3 indicate that the (CC/~) component of the matrix within the brackets should be taken. This result can also be obtained in another way.@) The high frequency dielectric constant is then

= d(pd
=

u,(k)

l r; =

8,,+4rr(X),,

(11)

Qz#W/l/(p~p~)

and -&B(k) = %&)/Z/pk. Equations (3) and (4) then become wsW = NW-ZE and

(7) (8)

in agreement with our earlier definition of X. So far the results are a recapitulation of results which have appeared in the literature@> 49 8) with some extensions. Now suppose axes are chosen so that the wave vector q is in the a-direction; this choice is made for simplicity in the formulae which follow and the direction a can be perfectly general. From equation (5), the macroscopic field associated with the wave is given by

P = Z+W+XE

E,=

-&Pa,

EB=

Ey=O,

DIELECTRIC

CONSTANTS

AND

LATTICE

VIBRATIONS

449

and therefore from equation (8), E,(1+47rXaoT) = -47T(z+IV)a. Substituting this result motion (6) leads to into 4Tr the equation (12) of

frequencies WJ(~) are the same as dielectric dispersion frequencies Sz,. A necessary and sufficient condition for this can be obtained by considering the jth eigenvector of N, n(j), and finding the condition for the frequency and eigenvector to be the same for the matrix

w2W =

N+-ZIZ+
%

W, 1

(13)

where I is a 3 x 3 matrix with unity for the (aa) component but with zeros elsewhere. This is the equation of motion for long wavelength vibrations with q in the u-direction, but for which q-1 is now much smaller than the length of the crystal in this direction. The matrix enclosed in brackets in equation (13) is singular and of rank 3n-3. The threemodesofzerofrequencycorrespondtoacoustic modes havingj = 1,2,3; and if the matrix is reduced to a non-singular matrix of order 3n-3 by eliminating rows and columns involving the Kth atom, then the determinant of this matrix gives the product of the squares of optical vibration frequencies W;(E). By an extension of a theorem given by COCHRAN[Ref. (4) Appendix] it follows that

as for the matrix N. The eigenvalue is obtained by multiplying the matrix N by n(j) and so the condition is that the part added to N must not contribute. This leads to

[z+n(j>L = 0,
which expresses the condition that there should be no polarization in the u-direction, due to the lattice vibration; from equation (12) this is also the condition that there should be no macroscopic electric field associated with the mode. Let us consider again the frequencies Q, which besides being the dielectric dispersion frequencies are also the frequencies of modes of vibration for which 4 is precisely zero but the crystal is shorted. From the results given by BORN and HUANG(~) and by COWLEY@) the ionic susceptibility at frequency W, much less than the electronic dispersion frequencies, is the difference between the susceptibility at frequency w and at high frequency, and can be expressed in terms of the modes n(j). In our notation the result is

1 +g

(Z+N-IZ,,)
I

and therefore from equations (8) and (10):

3n

g-1
Qf
fit ,-4

44

2=-

(0) Em

E(00) QG(

(14)
(16)

The dielectric dispersion frequencies Qj can be eliminated by considering a wave vector in the p-direction, when an analogous formula is obtained. Dividing one by the other

44
WJ(B)

2=

%aEj3j
p,(o)

(0) (4

(15)

ao! BB

Equations (14) and (15) are the most general equations of the Lyddane-Sachs-Teller type ; they are valid for any directions in a crystal of any symmetry. The formula (14) can be simplified if some of the
2G

where P(j) is the polarization associated with the mode whose frequency is Sz,. Now from the tensor property of the dielectric constant there exist mutually perpendicular axes, the principal axes of the ionic susceptibility tensor, for which X(~)(W) is diagonal. These axes are the crystallographic axes for crystals of symmetry class orthorhombic or higher, but for crystals of lower symmetry the axes are neither fixed with respect to the crystallographic axes, as frequency w varies, nor simply related to them. We see from equation (16) that the polarization of each of the modes whose frequencies are the Q

450

W.

COCHRAN

and

R.

A.

COWLEY

is along one of the crystallographic axes for crystals of orthorhombic or higher symmetry class. There will then be n- 1 modes associated with each axis. The difference between a mode for which q is precisely zero but the crystal is shorted and a mode for which q is small but q-1 is much smaller than a dimension of the crystal, which is not shorted, lies not in the value of q, but in the absence of a macroscopic field in the first case, and its possible presence in the second. From these considerations we conclude that as q becomes small the frequency WI of a mode tends to one of the values QJ provided only that the polarization of the mode is transverse and directed along a crystallographic axis. The direction of q may be anywhere in the plane perpendicular to this axis. Since we can now associate rz- 1 transverse and n- 1 longitudinal modes with each axis, equation (14) can be written

compared with a dimension of the crystal. The frequencies of transverse modes with which we have been concerned will therefore be the same as the infra-red absorption frequencies except possibly in the case of crystals of very small dimensions. The adiabatic approximations is valid provided that the electron transition frequencies are far removed from the phonon energies; this is undoubtedly satisfied for the electrons in a polar crystal, apart from the electrons associated with defects. The validity of the harmonic approximation has been discussed by SZIGETI@F lo),and the effect of higher order terms in the potential energy and dipole moment on the static dielectric constant is calculated for the alkali halides in the form: E(o) -E(W) = H+ G, where H is the harmonic contribution and G is due to the higher-order terms. SZIGETI estimates G from the derivates of the dielectric constants with respect to pressure and temperature as about 5 per cent of H+ G. The effect of higher-order terms on the frequencies of the lattice vibrations has not been worked out but we can estimate that the formulae given for the ratio ~rpk/eL$ are accurate to at least 5 per cent for real crystals.
Acknowledgement--R. A. COWLEY is indebted to the Department of Scientific and Industrial Research for the award of a maintenance grant.

where the u-direction is now a crystallographic axis, wL(a) is the frequency of one of the rz- 1 modes for which the polarization is longitudinal, with q parallel to the a-direction, and UT(E) is the frequency of one of the n- 1 modes for which the polarization is transverse and q is perpendicular to the u-direction. We may note that in crystals of high symmetry it is possible for WL to be equal to UT for a particular mode, which is symmetric and has no polarization associated with it. An example is provided by cubic barium titanate for which n- 1 = 4 but the product in equation (17) need only be taken over three values since one mode is symmetric-only two oxygen atoms move in this mode, and these in antiphase. These formulae have been obtained using the adiabatic, electrostatic and harmonic approximations. The electrostatic approximation leads to correct results for modes whose wavelength is small compared with the wavelength of electromagnetic radiation of the same frequency. In distinguishing modes with transverse polarization we have of course assumed that their wavelengths are small

REFERENCES
R. H., SACHS R. G. and TELLBR E., Phys. 1. LMDANE Rev. 59. 673 (1941). 2. BORN M.. and HUA~G K., Dynamical Theory of Crvstal Lattices. Oxford University Press (1954). W., Z. jzrisiallogs. 112,36 (1959). . 3. COCHRAN 4. COCHRANW., Adwanc. Phys. 9, 387 (1960). 5. MERTEN L., Z. Naturf. AlJ, 47 (1960). 6. WOODS A. D. B., Bre%I<Ho& B: N., COCHRAN w., SAKAMOTO M. and SINCLAIR R. N.. Bull. Amer. Phys. Sot. (Ser. II) 5, 462 (1960). 7. ROWLEY R. A. (to be published) (196l)a. 8. COWLEY R. A. (to be published) (196l)b. 9. SZIGETI B., Proc. Roy. Sot. A252, 217 (1959). 10. SZIGBTl B., Proc. Roy. Sm. A261, 274 (1961).