I LYSI| AI.

REVISE QOI UME f 26, gtJMBRR 2 APRIL, i 5, f962

Quantum Theory of the Dielectric Constant in Real Solids
STEPHEN L. ABLER*
Genera/ E/ectric Research Laboratory, Schenectady, Sew Fork
(Received October 12, 1961)
The quantum theory of the frequency- and wave-number-dependent dielectric constant in solids is ex-
tended in order to study the full dielectric constant tensor and to include local-field eGects. Within the
framework of the band theory, an explicit expression for the dielectric constant tensor, neglecting local-
field eGects, is derived. In addition to components which are the ordinary longitudinal and transverse di-
electric constants, there are components which couple transverse and longitudinal electromagnetic dis-
turbances. A formalism for calculating the local-field corrections to the dielectric constant is developed in
detail for the case of the longitudinal dielectric constant of a cubic insulating solid. 1n the coarsest (dipole)
approximation, the theory gives a Lorenz-Lorentz formula modified by self-polarization corrections arising
from the polarization of the charge in a unit cell by its own field.
~~~
UANTUM mechanical treatments of the fre-
quency-and wave-number-dependent dielectric
constant in solids have been given by Nozieres and
Pines' and
by
Ehrenreich and Cohen.
'
These authors
give explicit expressions for certain components of the
dielectric constant tensor, valid within the framework
of the random phase
approximation
(RPA). Expres-
sions are not given for the remaining components of
the dielectric constant tensor and local field sects are
neglected. This paper will generalize the treatment of
Ehrenreich and Cohen' so as to include additional
eGects of interest in real solids. In Sec. I an expression
for the full frequency- and wave-number-dependent
dielectric constant tensor in a solid of arbitrary sym-
metry will be derived, still neglecting local field e6ects.
The additional components obtained correspond to a
coupling between longitudinal and transverse electro-
magnetic disturbances. This coupling, which does not
appear in an isotropic free electron gas, is present in
solids of even cubic symmetry and vanishes only for
propagation along special directions of high symmetry.
In Sec. II, local Geld sects in insulators of cubic sym-
metry will be discussed. An integral equation will be
set up which determines the longitudinal dielectric
constant with local Geld corrections in the case of wave-
lengths large compared to the lattice constant but small
compared to the over-all crystal dimensions. The in-
tegral equation will be rewritten
by making a multipole
expansion of the potential in a given cell arising from
the charge density in all other cells. The solution, when
only dipole terms are retained, is the modified Lorenz-
Lorentz formula
duction in n due to polarization of the charge in a given
cell
by its own Geld.
The calculations of Secs. I and II are performed
within the framework of the one-electron (energy band)
approximation and use a linearized Liouville equation
to determine the single-particle density matrix. Since
in this context linearization is equivalent to the RPA,
'
the results obtained in Secs. I and II are still valid only
within the framework of the RPA.
I. DIELECTRIC CONSTANT TENSOR
We will Grst introduce a phenomenological dielectric
constant tensor. Let A(r, t) and
p(r, t) be the potentials
describing fields acting on a system of charged par-
ticles. In response to the fields, charge and current
densities p'"~(r,
t) and j'"~(r,
f),
which satisfy the equa-
tion of continuity &
j'"~+rip'"a/elf =0, will be induced
in the system. Let us immediately introduce Fourier
transforms
A(q, or), g(q, or),
j'"~(q,
or), etc.
, by
A(r, f)
=
dqdor
A(q, or) expi(q
r

orf), (1.1)
and similar equations.
In their treatments of the dielectric constant, Nozieres
and Pines' and Ehrenreich and Cohen,
2
following a
practice originated by Lindhard,
'
deGne longitudinal
and transverse dielectric constants
a~
and
e~
by
the
equations

iorLsin r&(q, or)

I]
E&n
r&(q, o)
=4~j'"'"
"(q~) (1 2)
4rr(rr

Ci)
C
1

(4rr/3) (o.

Ci)
The two constants describe respectively the longi-
tudinal current induced
by a purely-longitudinal electric
field and the transverse current induced by
a purely-
transverse electric field. In the case of a free-electron
gas a longitudinal (transverse) current cannot be in-
duced
by
a transverse (longitudinal) electric field.
I T
1 d cri i n f
where n is the polarizability of the solid calculated
without making local field corrections, and Cl is a re-
Consequently s and a give a comp ete es pt o o
*
Prese t address: Department of Physics Princeton Uni-
the linear dielectric properties. In solids, in general a
'P. Nozihres and D. Pines, Phys. Rev. 109, 741, 762, 1062
(1958); fbfd 111, 442 (1958); f.Md 115, 1254 (1959). . J. Lindhard, Kgl. Danske Videnskab. Selskab, Mat. -fys.
'
H. Ehrenreich and M. H Cohen, Phy.s. Rev. 115, 286 (1959). Medd. 28, 8 (1954).
413
STE PHEN L. ADLER
purely-transverse or a purely-longitudinal electric field
induces both transverse and longitudinal currents. In
this case the linear dielectric properties are fully de-
scribed by a dielectric-constant tensor defined
by

io
t e(q, u)

1] E(q,o~) =4~j'"'(q, (o). (1.3)

The longitudinal and transverse constants e~(q, a&) and
sr(q, or) can be simply related to
e(q, o~). Let
j
be a unit
vector parallel to the direction of propagation
q,
and
define
11=//
lr= 1

g
where 1 is the unit dyadic. Then letting
E(q,o~) be
purely longitudinal or purely transverse gives
e'(q, ~)
=1r,
e(,
M).
lr
sr(q, o~)=lr
e(q, ~) Ir.
The remaining components of the dielectric-constant
tensor are lr, e(q, o~) lr
and Iz, e(q, oi) 1J.
,
which vanish
for a free-electron gas but do not in general vanish for
a solid. They describe, respectively, the longitudinal
ttransverse) current induced
by
a transverse (longi-
(udinal) electric field.
An explicit expression for
e(q, oi) will be calculated in
the energy-band approximation. Consider the single-
particle Liouville equation
A(r, t) =A(q, a)) expi(q. r cut),

y(r, t) =y(q, oi) expi(q r

at),
(1.9)
and make the Ansatz that the time dependence of
(l'k+
q ~
p"'
~
lk) is exp(

i&et). The frequency o~ is taken

to have a small positive imaginary part, corresponding
to an adiabatic turning on of the perturbing potentials.
It is an easy calculation to show that
(l'k+q'~ep~lk)=b,
, ,ep(q, o~)(l'k+q~/k) (1.10)
and
-',
(/'k+q'~A
p+y A~/k&
=5,
, ,
(/'k+q~
p, +5k+A/2~/k) A(q, o). (1.11)
The abbreviation
(l'k+
q ~ f
(rp,) (
lk)
equation with respect to A and
P by setting p=
p&'i+p&'&
etc. The unperturbed density matrix, p&'), satisfies
p ~
k/)
=
f0(Eki) ~
k/) with
fo(Eki)
the Fermi-Dirac dis-
tribution function. The perturbation
p&" is linear in A
and P. Dropping quadratic terms gives
isa(/'k+q~ p&~ll
&/at
(EVk+g Eik)(l'k+I
p"'
Ilk&
+go(@k)

fo(Ei k+,)]
X(/'k+ql (e/2mc) (A p+ p A)+I &
Let us assume
i7iBp/Bt
=
)EI,p],
where
p
is the single-particle density matrix and
(1.6)

=
(1/~.)
ui k+,
*(r)
f(r,

its~, )uik(r)dr, (1.12)
H
=
(1/2m) Ly

(e/c) A(r, t)]'+ep(r, t)+U(r). (1.7)

Here U(r) is the periodic lattice potential. Let the state
functions for the unperturbed lattice be
~
k/)= V 'ukt
Xexp(ik r) with uki cell-periodic and V the volume of
the crystal. They satisfy the Schrodinger equation
Lp'/2m+U(r)]
~
k/&=Eki~ kl),
in which k is the wave
vector and l the band index. Linearize the Liouville
has been introduced, in which the integration extends
over a unit cell. Couplings of the wave vector
q
to wave
vectors
q+K,
where K is a reciprocal lattice vector,
have been neglected. These so-called Umklapp processes
give rise to the local field corrections and will be dis-
cussed in Sec. II. The solution of Eq. (1.8) is immedi-
ately obtained in the form
t f0(ELk) f0(EL'k+, ))L(/'k+q
/
lk)ep(qoi)

(l'k+q
f p, +ttk+Itq/2
[
lk) (e/mc)A(~os)]
(l'k+
q(
p&'i
[
lk)
=
(1.13)
Aco+Eik

Ei
k+,
The induced current and charge density may be
calculated from
j'"~(r,
t)
=
Trp&"j
&"
(r)+Trp&"j
&"
(r, t),
(1.14)
p'"
(r,t)
=
Trp"'p.
,
"'(r),
where
j
~'i
(r)
=
(A/2m) [(p,/m)5(r

r,)+8(r

r,
) (y,/m)],
j,
~&"
(r)
=
(e/mc) A(r, t)8(r

r,
),
(1.15)
p,
~&0'(r)
=
eb(r

r,
),
and r, and
p, are, respectively, the position and mo-
mentum operators. This gives
j'"'(q, o~)
=
e'A(q, o~)ll//mc V
+Pit. k(/k
~ p,
+/'tk+itq/2
~
l'k+
q)
X(/'k+q~
p
"i
~lk), (1.16)
.
'"'(q,
)
=
(/V) 2
'(/kl/'k+q)
X(/'k+q~poi
~/k&, (1.17)
with X the number of cells in the crystal.
Since j'"~(q,o~) and p'"~(q, or) are obtained by
linear-
ization of a gauge-invariant theory with respect to the
potentials, they are invariant under infinitesimal gauge
transformations, and thus also under arbitrary gauge
QUANTUM THEORY OF DIELECTRIC CONSTANT
transformations. This is verified explicitly, in the case
of the expression for j'
o(q, td), in the Appendix. Since
the theory is gauge invariant, we may transform to. a
j
t"'(q,
&o)
=
e'A(q, to) 1V/mc V
gauge in which /=0 without loss of generality. Note
that A will not in general be transverse in this gauge.
From Eqs. (1.13) and (1.16) we get
m'cU '&
(/k~ p, +/tk+fiq/2~/'k+q)Lfo(Etk)

fo(Et k+ )](/'k+q~ p,
+tttk+Aq/2~/k) A(q co)
(1 1g)
ttto+Etk

Et
kyo
Comparing with Eq. (1.3) and noting that
E(q,to) =iooA(q, co)/c gives
e(q, to)
=
(1

47re'X/mVoP)l
4xe'
m Vcr
(/k
~ p
+kk+ftq/2
~
/ k+
q)Lfo(Et k~o)

fo(Etk)]
(/'k+
q ~ p,
+5k+/'tq/2
(
/k)
(1.19)
Atd+Etk El'k+o
O=ptt
k~
(/k(/'k+q) )~go(Etk) fo(Et k+o)] (1 22)
Equations (1.20) and (1.21) are proved in the Ap-
pendix, while Eq. (1.22) is obtained
by
making the
change of variable k

+

k

q
and noting that
Nz
g EIlt+K

Zk and Nz~+K= Nz~,

where
K is a vector of the reciprocal lattice. Applying
the
identities to (1.19) gives
q (/k] p, +Itk+/tq/2)/'k+q)
=
(m/5) (Et k+o

Etk) (/k
~
/'k+
q),
(1.20)
0=

X+(1/trtt/')
Qtt k q (/k~ y, +Sk+tttq/2~/'k+q)
xI:fo(Etk)

fo(Et'+o)](/'k+ql/k), (1 21)
an explicit expression for the frequency- and wave- and
number-dependent dielectric constant tensor.
It is interesting to compute e~(q, to) =I&
e(q, co) lz,
directly from (1.19). This can be done
by using the
three identities
4~e'
=lr, 1+
l
r, e(q, to) 11,

11, 1 47re'1V/mVooo+
p q (/k~ y.+fik+Aq/2~/'k+q)
Aq'mUco'
'&
f1to
X(fo(Et'+, )

fo(Etk)](/'k+q f/k)
]-
(Aoo+Etk Et~k+o
I
(/kl/'k+q) ['Lfo(Et'k+ ) fo(Etk)]
AU
zz'x
/t&+Elk El'k+.
o
(1.23)
Equation (1.23) is just the longitudinal dielectric constant derived by
Ehrenreich and Cohen.
'
Vsing the same
identities employed to derive (1.23), it is easy to show that lz a(q, co) 11. and 11, e(q, td) 1r are given by
4tre lz (/k
~ p, +5k
~
/'k+q) (/'k+q
~
/k)t/Lfo(Et k+o) fo(Etk)]
lo e(q, to) 11,
=-
~qmV
'~
Aot+Etk Epkyo
4me'
t/(/k~/'k+q)(/'k+q~ p, +fgk~/k) lrLfo(Et k+o)

fo(Etk)]
lr, e(q)to) lr

coqmP' zz'l
~ot+Etk El'k+o
When the limit
q

+ 0 is taken, the dielectric constant tensor becomes
47re'N 4me'
(/k~ p,
~/'k)(/'k~
p. ~/k)Lfo(Et k)

fo(Etk)]
o(O, M)
=1-
+
mU~2 m'Uco' '&
~ot+Etk EVk
(1.24)
(1.25)
In a crystal of cubic symmetry, the sum over the
star of k, pk*(/kl y,
l/'k)(/'kl
pl/k),
is a multiple of
the unit dyadic, and consequently e(0,&o) is isotropic.
Thus, with the approximations made to get Eq. (1.19),
as
g

+0 the longitudinal and transverse dielectric

constants become equal and lr e(q, &o) I& and 11.
~
e(q, M) lr vanish. This is true for an arbitrary direction
of propagation in a cubic material, but will not in gen-
eral hold in the case of crystals of lower symmetry.
II. LOCAL-FIELD CORRECTIONS
In this section we will develop the theory of the
longitudinal dielectric constant with local-field correc-
tions, for a cubic insulating solid, in the case of wave-
STEPHEN L. ADLER
ct
' '=
P'
'(q, co) expi(q r cot)
is applied to the crystal. The total potential
p=p'"'
+P'"d
will in general contain rapidly varying terms
with wave vector
q+K,
where K is a vector of the
reciprocal lattice:
ct =+Kit
(q, K,co) expiI (q+K)
r

cot(. (2.2)
The potential
@
is the microscopic potential and deter-
mines how the charge in the crystal is polarized. The
macroscopic potential
(g),
is clearly given by
Q). =@(q,O, M) expi(q
r cot),

(2.3)
since exp(iq r) is nearly constant over the averaging
region while expI i(q+K). rf, (KNO), is very rapidly
lengths large relative to the lattice constant but small
relative to the over-all crystal dimensions. Local-field
effects arise in a real solid because the microscopic
electric field varies rapidly over the unit cell. Conse-
quently, the macroscopic fieM, which is the average of
the microscopic field over a region large compared with
the lattice constant but small compared with the wave-
length 2sr/ct, is not in general the same as the effective
or local field which polarizes the charge in the crystal
For example, suppose a slowly varying external potential
(2.1)
varying. The derivation in Sec. I assumed a potential
of form Eq. (2.1) instead of Eq. (2.2). In other words,
the distinction between the microscopic and macro-
scopic fields and potentials was neglected, with the
result that no local-field corrections were obtained. In
order to obtain the longitudinal dielectric constant with
local-field corrections, a total potential of the form Eq.
(2.2) must be assumed (with A=O), and the induced
potential,
find

PK
find(q K co) expiI (q+ K)
.
r cot
j (2 4)
must be calculated. The longitudinal dielectric constant
is obtained from the macroscopic total and induced
potentialsc according to an alternative form of Eq. (1.2),
i
"(q )
~= 1

(&'"')-(q,
)/(4-(q, ) (2 3)
Using Eq. (2.3),
this is
j
"(q~) (=1

4'"'(q,
O, ~)/4 (q, O, ~) (2 6)
The right-hand side of Eq. (2.6) is easily evaluated
in a formal manner. A calculation analogous to that of
Sec. I gives the relation between
P'"d
and
g
as
y'"'(q
K,~)
=
I q+
K
I

'
pre G(q+
K
q+
K'
co)
Xy (q,
K',
co), (2.7)
with
G(q+K, q+K', co)
=
4vre'
trco++ik +c'k+o
(tk
I
exp(

iK r,
) I
t'k+q)(t'k+qI exp(iK'. r,
) I
tk)Lfo(&ck)

fo(&l'k+o)]
(2.8)
P zzj
As before, the variable of integration in the matrix
element has been indicated
by
I', . Let us define
s(q+K,
q+K', co) and e

'(q+K, q+K', co) by

e(q+K, q+K', co)
=sic x

G(q+K, q+K', co)
I q+KI

', (2.9)
Px. e
(q+K,
q+K", oi)
Xs

'(q+
K",
q+
K',
co)
=
ciK rc
. (2.10)
t
The quantity s '(q+K, q+
K',
co) is just the dielectric
response function of Schwinger and Martin.
'
Equations
(2.9) and (2.10) have been given by Falk,
o
who treats
the nearly free electron case.
$
Rewrite Eq. (2.7) as
I
q+KI'Ly(q,
K co)

y'ad(q,
K,co)7
=Pic s(q+K, q+K', co)
I
q+K'I'@(q,
K',
co), (2.11)
and note that
g(q,
K co)
g'"
(q
K co)
=p'
'(q K co)
=p'""(q,
co)5K, o,
since the external potential (the poten-
tial due to charges located outside the crystal) is essen-
tially constant over a unit cell of the crystal. Using Eq.
(2.10) we find
$(q, K,co)=
Iq+KI
's
'(q+K
q
co)ct'p'"'(qco), (2.12)
4
L. Rosenfeld, Theory of Etectrorcs (North-Holland Publishing
Company, Amsterdam, 1.951), Chap. 2.
'
P. C. Martin and J. Schwinger, Phys. Rev. 115, 1342 (1959).
'
D. S. I'alk, Phys. Rev. 118, 105 (1960).
giving
i7 e(q co) cf= 1/ '(oq
q
co). (2.13)
Thus, the problem of finding the dielectric constant
with local field corrections reduces to that of solving
the integral equation
s

'(q+K, q+K', co)

=&x, x+Zx- G(q+K, q+K", ~)
I
q+K"
I
'
Xs

'(q+K" q+K', co), (2.14)

obtained by combining Eqs. (2.9) and (2.10).
The main purpose of this section is to develop a
systematic method of approximating the integral equa-
tion (2.14). This will be accomplished by
means of two
successive transformations. First, the integral equation
will be transformed from the K representation to an r
representation, where r is a continuous variable con-
fined to a unit cell of the real lattice centered about the
origin. In this representation, the kernel of the integral
equation will be split into two parts,
E~
and E~. These
describe the inhuence on a given cell of the field of the
polarized charge in all other cells (E~) and of the field
of the polarized charge in the same cell (Es), and are
connected, respectively, with the local field and self-
polarization corrections. A second transformation will
7
This equation has been independently obtained by N. Wiser.
I am indebted to M. H. Cohen and N. Wiser for communicating
their results prior to publication.
QUANTUM THEORY OF DIELECTRIC CONSTANT
then be made by expanding
E~
in a multipole series,
leading to an integral equation in what might be termed
a multipole representation. This equation can be solved
approximately by neglecting all but the first
I"
multi-
pole moments. The case when only dipole moments are
retained will be worked out explicitly, and leads to a
Iorenz-Lorentz formula modi6ed
by
self-polarization
corrections.
In the equations that follow, the ~ dependence of e
'
and G will no longer be indicated explicitly. To trans-
form to the r representation let us define
,

i(qr r) P
e,
x.re-, x.
r,

i(q+K q+K) (215)

1
c '(q,
q)
=
drdr'
e

'(q, r,
r').
&a O
2
(2.18)
The kernel of the integral equation,
1 'Va
E
(q, r, ri
)
=
dri
G(q, r, ri)
&a 0 4m
where R;
are the vectors of the real lattice and the
continuous variable r is confined to a unit cell centered
at R;=0. The inverse of the dielectric constant is ob-
tained from
G(q, r,
r') =
P
e'"'e

'x"'G(q+K
q+K').
The integral equation becomes
& '(q, r,
r')
1
=
v.
P
t'(r
r'
R,)+

dridr,
'
7 &a
2
(2.16)
expiq (R,
+ri'

ri)
&(P, (2.19)
I
R+ri'

riI
can be divided into two parts,
E(q, r,
ri') =E
(q, r, ri')+E
(q, r,
ri'). (2.20)
expiq (R;+ri'

ri)-
)&G(q, r, r&)

P
4m ~
~R;+r,
'

r,
~
&(e

'(q, r,
', r'), (2.17)
Since several kernels similar in structure will be intro-
duced in the course of the derivation, we specify them
all through the functional form
~[q,f,
gj=
U
ll'k
(L &l')
(If
Ii'k+q) (I'k+ql
g I
)[fo(k)

fo('+ )j
~~++Lk +L'
qk+
(2 21)
G,
K~ E8:
E.
L:
gS
~
f=V.
P;
8(r

r,

R;),
g= i.
P;
S(r,
'

r,

R;),
g= (v /4~) exp[iq
(ri'

r,
)j
XP exp(iq R,)/~
r,

ri'

R,
~,
(2.22)
g= (e./4~) exp[iq. (r
'

r.
)j/ ~
r.

r&'
~.
The prime on the sum defining
g
in
E~
means that the
term with R;=0 is to be omitted.
Since
r1'
and r, are restricted to lie within a unit cell
(No terms with
i=1'
appear in the summation in
the case of an insulator because all bands are either
empty or full.
)
The kernels
G(q, r, ri'), E~(q, r, r&'),
and E
(q, r,
ri') are obtained from this form
by
the
substitutions
centered. about the origin, ~r, (/)ri'

R, )(1 for all
R, /0 and the multipole expansion
expiq R;
=Q(r,)&
T(q,
ri') (2.23)
4m i (r,

r,
'

R,
(
i=o
is valid. Equation (2.23) serves as definition of the ex-
pansion coefficient T~. Substituting Eq. (2.17) into
Eq. (2.18), splitting the kernel according to Eq. (2.20),
making the multipole expansion of Eq. (2.23) and letting
q
-+
0 results in
e '(0,0)=1
i+j

K
r
B i+B,
e
q. (2.24)
m=1
The quantity B~~ is defined
by
g
B=lim

dr'dri'
exp(iq ri')T,
(q,
r&')e '(q, ri', r')
q~0
p
2
(2.25)
and Bie is the p= 1 case of (2.21) by
substituting
B e=lim
II~0
~
2
dr'dri'K e(r, ')e-'(q, ri', r') . (2.26)
The kernels
K
and K(ri') are obtained from Eq.
f=rg=(r,)"
Ee(r,
'):
f= (r,
)& exp(iq r,
),
(2.27)
g= (~ /4m) exp[iq
(ri'

r )j/Iri'
STEPHEN L. ADLER
~
(iB8++
Kz.
B z)
(2 29)
n=l
where the kernel Ki,
z is obtained from Eq. (2.21), by
the substitution
f=(.
)"
g=(.
)"
(2.30)
In the first term on the right-hand side of Eq. (2.29)
the evaluation
g'
lim-
q~0
zgi p
1i
dr
e'&'T
(q, r)
=0, p)1,
(2.31)
has been used.
'
Finally, an equation for the Bsmust
be derived. Let us write
D(q, r,
r') =6(r

r')

(1/v, )Es(q, r, r'), (2.32)

and define D '(q, r, r,
) by
(1/~,
)
dr, D '(q, r, r,)D(q, r,
,
r') =5(r r').

(2.33)
Then we can write
e '(q, r,
r')
=
D '(q, r,
r')+
(1/v,
') dridri'D '(q, r, ri)
XE
(q, ri,
ri') e '(q, ri', r').
(2.34)
Multiplying Eq. (2.34) by

iqKs(r)/p, ', integrating
and making a multipole expansion gives
In deriving Eq. (2.24), the relation
q
(i'k+
q l
r, exp(iq r,
) l
/k)
=
i(i'k+q ilk)+0(q'), (2.28)
valid when t/3', has been used.
The kernels Kiand K~ (ri') are known quantities.
In order to
complete
the set of equations, expressions
for B~ and Bs must be derived. Multiplying Eq.
(2.17) by
q
exp(iq
ri')
T(q,
ri') and integrating gives
k+p
B~=i
j8,,+P(

1)'
Dim Ti+(q,0)]
Ic=l
q~0
It should be noted that terms involving Bp~ have
been omitted from Eqs. (2.24), (2.29) and (2.35)
be-
cause they vanish in the limit
q
~0. For example, the
coe%cient of BP in Eq. (2.24) contains a factor
(i'k+ql exp(

iq r,
) l/k). Since
(i'k+ql exp(

iq r,
) l/k)
=
(i'k+ql/k)
i
q
(i'k+ql r, ilk)+0(q') =0(q')
alld
q
v exp(iq R,)
Bpz

dr
e'&'
P'
v. p 4m i lr+R,
l
q
exp(iK r) 1
dr+
v. o
K
lKPql'
q
the term drops out as
q
~0. Similarly, the terms in
Eq. (2.29) and Eq. (2.35) involving BP do not
contribute.
The preceding manipulations have replaced the inte-
gral equation (2.14) for e '(q+K, q+K')
by Eqs. (2.29)
and (2.35), which together constitute an integral equa-
tion for B~z(p=1,
, ~), and the integral Eq. (2.33)
for D
'.
Once the quantities B~z and Bis are known,
the dielectric constant with local-field corrections can
be calculated from Eq. (2.24). The point of this forma, l
rearrangement is that it is now possible to make an
approximation with a clear physical significance which
makes Eqs. (2.29) and (2.35) easily soluble. This is
simply to neglect all the B~with
p
greater than some
integer P. This means roughly that we are approxi-
mating the influence on a given cell of the charge in any
other cell by the first P multipole moments of this
charge. In many cases, we expect very good results to
be obtained for a small value of P. The most familiar
case is that of
I'=1
(dipole approximation). Utilizing
the fact that the only second-order tensor compatible
with cubic symmetry is the isotropic tensor, and noting
that the inhomogeneous term in the equation for B&~
is a vector parallel to
q,
Eqs. (2.24), (2.29), and (2.35)
become
e '(00)=1

iBiz
jq
K
~
j+B
s.
j
Bi j=i

2Dimj T2(q, 0)
jg
q~0
Z
drdri
n=l
0
XK(r)D '(O, r, ri)K(ri) B, (2.35)
where KP(ri) is de6ned
by
K~(ri): f=v,
P;
5(ri

r,

R,
),
g= (r,)". (2.36)
X(iBis
q+q
K,p jBiz
.
g),
B j=

(i/ri.
')
j
drdr,
0
XK, (r)D-'(O, r, r,)Ei (r,) yB,
y
i47rCiBiz

j
(2.37)
This is easily obtained by using
p
&
exp(iq. R;) expLi(q+K) (r,

r)]
4r; [r,

R;i K fq+Ki'
It is easy to show that
g
'
Kii
'g=

4ire, (2.38)
QUANTUM THEORY OF DIELECTRIC CONSTANT
e '(0,0)
4m (n

Cr)
1

(4~/3) (n

Cr)
(2.39)
This is the usual Lorenz-Lorentz formula, modified
by
the subtraction from 0, of
C,
=
(1/4m v.')q drdr,
&&Ki (r)(r~L1

(1/e )Esj '~r,

)K,z(rr) q. (2.40)
This self-polarization correction takes just the form that
is expected on the basis of a simple classical model. If
we compute the dielectric constant of a macroscopic
cubic lattice of uniform spheres composed of material
of polarization per unit volume n, we find that the di-
electric constant is determined
by
a Lorenz-Lorentz
formula, except that o. is replaced by
4s.n'/3
1+47m/3
(2.41)
The subtracted term is a self-polarization correction
arising from the inhuence on a given sphere of its sur-
face charge. To examine the qualitative form of
C&,
let us replace all the kernels Eappearing in Eq. (2.40)
by

Inn Lcf. Eq. (2.38)$. Then we see that the cor-
rection Ci also has the form An'/(1+Bn), with A, 8&0.
With the formalism developed, higher-order correc-
tions to the Lorenz-Lorentz formula can be obtained
by
where n is the polarizability calculated without making
local-field corrections. LThe second term on the right
of Eq. (1.21) is just 4~n.
J
The dipole sum
q Ts(q, O)
.
q
is not absolutely convergent. However, if we evaluate
it in a crystal of 6nite diameter L, letting L

+ ~ and
q

+ 0 while keeping qL&)1, it has the value

' inde-
pendent of the crystal shape.
'
This evaluation procedure
is the one that makes sense physically for wavelengths
in the infrared, visible and near ultraviolet. For such
wavelengths and for a typical crystal of dimension L
and lattice constant a, the inequality qL&)1 holds.
However, since qa((1 and since the matrix elements
and energies appearing in the kernels vary appreciably
only when
q
changes by
an amount of order 1/u, we
can still take the limit
q

+ 0 in the kernels.
Using these evaluations, Eqs. (2.37) may be readily
solved to yield
taking the cutoff integer
I'
larger than one. Note that
T~(0,0) is absolutely convergent for p&2, so no addi-
tional ambiguities regarding the method of sunnnation
appear when working to higher order. Although the
calculation has been carried out for crystals of cubic
symmetry in order to avoid a tensor dielectric constant,
its main features would be expected to carry over to
the case of arbitrary symmetry. If the restriction to
insulators is dropped, intraband terms (/=/')
appear in
Eq. (2.21). If these are treated in a free-electron ap-
proximation, which should be reasonable when motion
of the conduction electrons and holes is well described
by
an effective mass, the generalization of the above
derivation is straightforward and leads to
4~(n"

Ci)
=
1+4mn'+ (2.42)
e 1(00) 1

(4r/3)
(n"

C,
)
In Eq. (2.42)
ni
and
nii
are, respectively, the intraband
(/=/') and interband (///') parts of n, the polariza-
bility without local field corrections. The restriction
qL))1, necessary to evaluate the dipole wave sum
Ts(q, O),
cannot be relaxed without drastically altering
the derivation. In order to deal with wavelengths
comparable with the macroscopic dimensions of the
crystal it would be necessary to take into account sur-
face eRects, which of course has not been done in the
above derivation.
ACKNOWLEDGMENT
The author wishes to thank Dr. H. Ehrenreich for
suggesting the problem treated in this paper and for
many helpful discussions.
APPENDIX
In order to prove that the expression for j'"s(q,co) is
gauge invariant, we need the auxiliary identities:
P~
N~q(r)g~q*(r') =v,
P;
5(r
r'
R;), (A1)
q. (/k
~ p,
+hk+hq/2 ~/'k+
q)
=
(m/5) (Epg+s E~&) (/k ~/'k+q). (A2)
Expression (A1) is just the completeness relation for the
periodic parts of the Bloch functions. The identity
(A2) is obtained in a straightforward manner
by
writ-
ing k
p
Schrodinger equations for N&z* and I&
z+q,
multiplying the former by I& i,+~,
the latter by u&&~, and
subtracting. We may also regard (A2) as the result of
expanding
dr exp(

iq r)L& (/k~ 3,
&'&(r)
~/'k+q)+8(/k~ p&'&(r) ~/'k+q)/eI/j=O, (A3)
which shows that it is an expression of conservation of charge in the unperturbed theory,
'
M. H. Cohen and F. Keffer, Phys. Rev. 99, 1128 (1955),
420 STEPHEN L. ADLER
Let us now make the gauge transformation A ~
A+qf(q, &v), g

+
P+ (co/c)
f(q,
pp). Then
hj'"
/f(q
&o)
=
e'1Vq/mcV
g2
tn'cV
'~
(lk( p, +Ak+Aq/2~l'k+q)gp(Elm)

fp(Ei t+o)j(l'k+q~ p, +Ak+Aq/2)lk) q
A~+E
E
t ~+o
e'~
(lk
~ p, +Ak+Aq/2
(
l'k+q)L fo(E&~)

fo(E& k+ )g
Combining terms gives
yggt/
'R
A~+Elk Evk+p
d j'
~/f(q &o)
=
(e'/mcU)
{

XAq+P
z(lk
) p +Ak+Aq/2
(
l'k+
q)
[D'p(E&z)

fp(E~ q+o)
g(l'k+q
~
lk)
}
=
(e'/mcV)
[
XA
q+2 P&& fo(E&&) P& (lk
[p,
+Ak+Aq/2
(
l'k+q) (l'k+q [lk)
=
(e'/mcV)
t
1V
Aq+2 p~~ fo(Kx) (lk
j p.+Ak+Aq/2
~
lk) j=O (A4)
j n the last step, the fact that (lk
~ p, +Ak
~
lk) has odd parity under inversion of k has been used.