Calculation of critical indices for zero-gap semiconductors

A. A. Abrikosov
L. D. Landau Theoretical Physics Institute, USSR Academy of Sciences
(Submitted November 19, 1973)
Zh. Eksp. Teor. Fiz. 66, 1443-1460 (April 1974)

The indices defining the exponents in the power laws in the Green functions and physical
characteristics of zero-gap semiconductors of the second kind in the strong-coupling region (i.e., the
region of energies w<wo, where Wo= m *e 4/21<6 is the exciton binding energy, and momenta k <k 0,
where k 0= m *e 2/1<0 is the inverse Bohr radius of the exction) are calculated. Two methods are
applied: expansion in E=4-d (d is the dimensionality of space) and expansion in the "large
number of components," i.e., in l/II, where II is the polarization operator. In the first approximation
the two procedures give results that are numerically slightly different but qualitatively similar. It
follows from the results that there are several energy and momentum regions: (1) the "free region,"
(2) the region Wl<W<WO' kl<k <ko (wl~2.5XIO-4 Wo, kl~1.6XIO-2ko). in which the
"random-phase approximation" is applicable, i.e., the interaction of the carriers is altered but their
spectrum is conserved, and (3) the true strong-coupling region w < W1> k < k l' In the regions (2) and
(3) the results of Abrikosov and Benesiavskil are applicable, with a=v=2 in region (2) and a= 1.76,
V= 1.92 in region (3).

1. INTRODUCTION whence it can be seen that there is invariance with re-

spect to any rotations. The eigenvalues are doubly de-
It has been shown earlier (cf,[l,2]) that in zero-gap
generate and equal to
semiconductors of the second kind (e .g., a -Sn), a
strong-coupling regime arises in the vicinity of the 8",= (f±c) k'. (3)
point of contact of the branches of the spectrum (a point
Hence it can be seen that a zero-gap spectrum is ob-
of the order of ko = m*e 2/ Ko along the momentum axis
and Wo = m*e4/2K~ along the energy axis, where m* is tained for c > f,
the effective mass and Ko is the dielectric permittivity). It was shown in(1] that the zeroth-approximation term
In this region the Green functions and vertices acquire for the polarization operator is proportional to an inte-
the self-similar form: gral that converges like 11k at the lower limit (k is
G -, ( k ) -a [ rol roo ]
the external momentum). It is clear that in four-dimen-
-roo k: g (klk o) , ' sional space we would have a logarithmic integral. This
roo ( k ) ,_3 .[ ro/roo ]
corresponds precisely to the situation in phase transi-
f-- - d-- (1) tions. The application of Wilson's methodeS] is associ-
k o' ko (klk o), '
ated with formulation of the problem in a space with a
( k ) a-' [ rol roo ] noninteger number of dimensions d = 4 - €. In the
1- k: q (klk o)' .
present case, the problem is complicated by the fact
The indices a and v determine the behavior of phySical that operators JiJ. associated with the three-dimensional
quantities in the singular region, e.g., their dependence rotation group appear in the Hamiltonian and form ten-
on the temperature, magnetic field, frequency, etc. sor combinations with the momentum components. The
continuation to noninteger dimensions in such a situa-
In view of the fact that the problem of zero-gap semi- tion has been effected for the case J = 1 [5], but the
conductors is methodologically similar to the theory of
phase transitions, to determine the unknown indices a
r
method used is not applicable for J = 2. In view of this,
we shall proceed in the following way.
and v it is natural to attempt to apply the methods de-
veloped in recent years by Wilson and Fisher(3] and We write the second term in the Hamiltonian (2) in
Ferrell and Scalapino [4) and successfully used for the the form cH"
analysis of different types of phase transitions. The (4)
present article is devoted to this question.
and analyze the properties of the operators A v' Ac-
cording to (2), these operators possess the foilowing
2. CONTINUATION TO NONINTEGER DIMENSIONS properties:
The calculation of the indices should be performed
for a specific model. Although the indices do not depend -r.
....... A",,=O, {A ••, A,'/=-1 (v*/-1),
(5)
on the interaction constant, they can depend on the
SpA.,=O, A•.'=1, 1.;=3/, (v*/-1);
geometry (the little group and specific representation;
cf,UJ). We shall consider a very realistic situation,
namely, the case of a-Sn with a diamond-type lattice. { AiJ.II' ApO'} = 0 in all other cases. Here there is no
The bands touch at a point with little group Dh. We summation over repeated indices. All these properties
take a representation corresponding to J = "2; the cor- can be written in the form of a single formula:
responding Hamiltonian and spectrum are well known[I]. (6)
To simplify matters, we put 3 f3 = I' in the formula in (1];
the spectrum then possesses spherical symmetry. The Hence it follows that the anticommutator of two Hamil-
Hamiltonian in this case can be written in the form tonians H, at different points k is equal to
(2) {Hdk,), Hdk,)} =3(k,k,)'-k,'k,'. (7)

709 Sov. Phys.-JETP, Vol. 39, No.4, October 1974 Copyright © 1975 American Institute of Physics 709
In particular, if k1 = k2 = k we obtain 2c 2 k\ and aver- 1
Ad-td-'=- d_11,- ... +
[ 1 ] .. , 1- (d-d+2)'
1- (d-1)'
1] '" 1d-"'f, [
aging (7) over the directions of k2 gives zero. These
properties are carried over in an obvious way to a 1 [ 1 ] 'I. ... [ 4"3] 1d-to
Add=·-d_1 1,- ... - 1-(d_1)'
'I.
space with noninteger dimensions. For this it is suf-
ficient to require that
The condition (c) is fulfilled automatically. This is not
2
(H,(k,), H,(k')}=-d-[d(k,k,)'-k.'k,']. (8 ) surprising, as it already follows from (9) that {Apa,
-1
bAIIII} = 0 for any of the matrices Apa.
For this, we can take HI in the form (4), where the II

operator coefficients A1111 satisfy the condition If we construct such matrices as direct products of
- - d 2 (9)
Pauli matrices, we obtain the following cases:
{A." A p,}= d-1 (1),,,,1>,,+1>",1>,p) - d-1 1>.,1>p,.
for n = 1,
In the following we shall need to use the property (8)
more frequently than any other. However, in the calcu-
lation of the polarization operator it is necessary to for n = 2,
take the trace, and we need to know the rank of the
matrices Ailil . In a space of three dimensions this is
equal to 4. We shall determine the rank from the condi- for n = 3,
tion that it be sufficient to enable the conditions (9) to
be satisfied. We shall use the following method. For
the three-dimensional case, the matrices All II can be Thus, in a similar manner, we find that the number of
represented with the aid of direct products of two sets matrices equals 3 + 2(n - 1), where n is the number
of Pauli matrices: of types of Pauli matrices. This number should be
equal to (or greater than) (d - 1) (d + 2 )/2. The rank
of the matrices equals 2n. Hence we obtain
r(d) =2"=2 d(d+,)/,-,. (11 )
(10) In particular, for d = 3 we have r = 4; if d = 4, r = 16.
The rule (8) and the rank (11) of the matrices are suf-
ficient in practice for our purposes.

3. FIRST APPROXIMATION IN € = 4-d

A similar representation can also be used in other
cases. The operators All II are symmetric in 11 and II, According to the philosophy of Wilson's approach[3],
and ~AIlIl = O. If the dimensionality of s pace (i.e., the we can argue in the following way. We shall assume
number of values of 11) is d, then the number of such that d = 4 - E. If we put E = 0, we obtain logarithmic
matrices equals (d - 1) (d + 2 )/2. integrals. This gives us the possibility of choosing
principal diagrams in which the power of the interaction
The conditions (9) can be satisfied by means of linear constant corresponds to the power of a large logarithm.
combinations of matrices with zero trace, if these As a result of summation, certain functions of this
matrices anticommute, their squares are equal to unity, logarithm arise. Then, assuming E to be small, we
and there are (d - 1) (d + 2 )/2 of them. This is proved carry out the formal replacement:
as follows. We have at our disposal (d - 1) (d + 2 )/2
ko 1 '" d'k '" d'-'k k-'-k;'
matrices, which we denote by ('1, Y2, .•• . We take any In - = - - J - - .... constJ - - = const--
(d - 1)d/2 of them as the Ailil with Il;r. II. There remain k 2n'. k' • k' e'
d - 1 matrices, in terms of whic h we must express the For k « ko we are left with k- E/ Eo Putting E = 1, we
d matrices AlfJ.l... We must then have: a) Atll = 1, obtain the critical exponents for all the quantities.
b) {A IlIl , Allllj = -2/(d - 1) (11 '" II), c) ~AIlIl = O. We
11 We shall carry out the above program for our model.
denote All == Y 1. We choose A22 = P Y 1 + q Y2 in such a We put the external frequency equal to zero. Then ac-
way that the conditions (a) and (b) are satisfied. This cording toP], G ~ w;/(k/kor a . First of all we shall
gives p = -l/(d - 1), q = [1 - l/(d - ll]1/2. We next consider the polarization operator (Fig. 1). In the first
choose A3S = P1Y 1 + q1Y2 + rY 3 , again to satisfy the approximation, this is equal to
same conditions. This gives
1 1 [ 1 ] 'I. IT(') (0), k) =-i Sp JG'O) (O)t- ~ . k t- ~ )
Pt=- d-1' q,= - d-2 1- (d-1)' '
0)
( 0)+- k t +-
k) -d'k- - -dO). (12 )
[ 1] 'I. [ 1- (d-2)'
r= 1- (d-1)'
1] 'I. "G(O)
r. t 2' 2
t
(2n)d 2n
We put d = 4. According to the arguments expressed
Continuing in the same spirit, we obtain expressions for in(1], the integral over WI goes to zero for k = O. Con-
all the AIlIl (for integer d): sequently, we must expand in k under the integral. We
then obtain an integral of the type k 2 f dkjk1. The situ-
ation turns out to be very similar to that which was
found in an zero-gap semiconductors of the first kind
(with a linear spectrum). This is confirmed by the fact
that an estimate of the correction to Y represented in
Fig. 2 gives only unimportant terms. In view of this, we
can confine ourselves to summing the logarithmic terms
in the electron Green function G and Coulomb-quantum
Green function r, taking Y = 1.

710 SOy. Phys.·JETP, Vol. 39, No.4. October 1974 A. A. Abrikosov 710
 r(d) d
n(t)(w,k)=---z;;- d-1
(19 )
X (k+'+k_') [k+'k_'-(k+k-)')d'k,
S [(k+'+k_2_i6)-(w/e)')k/k_'(2n)' .

FIG. I FIG. 2 We shall put w '" 0 and confine ourselves to logarithmic

accuracy. Assuming that k « kh we find
We call attention to the following circumstance. The (')( ) __ r(d) _d_S kt'k'-(k,k)' ddk, (20)
Hamiltonian fk2 + cH 1 is a linear combination of two in- n O,k - 4e d-l k,' (2n)d .
variants of the rotation group. According to[l), we must We put d '" 4; in the integral,
assume that in the logarithmic approximation the con-
d'k " k'dk
stants f and c are replaced by functions of In (ko/k). __'_ .... 4
(2n)'
S i 'OdOS-'-'
n, s n (2n)' .
However, it is not difficult to see that logarithmic cor-
rections arise only in the term with H1. In fact, assum- We then obtain
ing that w « ck2, we have for the self-energy in the 1 4n·2n' e 2 (21)
zeroth approximation a=----4.
e (Ztt)' x,

~(I)(O,k)=i 4:~' S ~"G'O)(k+k"w,)d2:' (~:;,. (13) We turn now to the self-energy. It follows from sym-
metry considerations that ~(l) '" qcH1, where q can de-
The function G(O)( k 1 ) can be written in the form pend only on k2. Taking the anticommutator with H1( k),
G,O)(k )=(w -fk 2-cH )-'= w,-fk,'+cH, (14) according to (9) we obtain
, , , , (W,-/k,2)2_(ck,'-i6}'
2 ~= 4ne'd SJ.. { (k,k+k')' (k,k)'} d'k,
Substituting this expression into (13) and taking the in- q cxo(d-l) k,' (k.+k)' k,' (2n)d
tegral over w 1, we find
(here we have made use of the relation (k1' k)2
1;(')(0 k = 2ne2s~ H,(k+k,l d'k, . (15)
,) x, k,' (k+k,)2 (Ztt)' '" k~k2/d). Assuming that k « k1 and expanding in k,
we obtain
But it immediately follows from this that Tr ~ '" 0 2ne'd S d'k,
(since Tr H1 '" 0). Consequently, a logarithmic correc- q= (
exo d-l
) (1+4cos'O-Scos'O)-(2)'"
n k,
tion to the first invariant fk2 does not arise, and only
the correction to H 1 remains. This means that the Since q '" b In (ko/k) (cf. 17), we obtain from this (d '" 4)
diagonal term in G- 1 for w « Ck2 has the power 0!1, b=J.. 4n·Ztt' ~~ (22)
equal to 2 - O( E2). But if the term of the type H1 has a e (2n)' xo 6 .
lower power (of the type O! '" 2 - O( E)), and below we Substituting this expression into (18), we have
shall verify that this is precisely the case, then in the
approximation under consideration it is sufficient to a=v=2-1I2S. (23)
retain only this term. In view of this, we shall assume It is interesting to note that the indices O! and v for
below that f '" O. the electron G-function differ little from the "free
Earlier[l], the general case of a logarithmic situa- indices." As regards the function r( k), it becomes
tion with one invariant in the Hamiltonian was analyzed. close to the quantity ~k-1 obtained from the "random-
The result reduces to the following. It is necessary to phase approximation," Le", obtained on the basis of a
determine the quantities a and b from the relations calculation of the first-approximation diagram for the
po larization operator [6].
aln~=-r(O)n(t) bln~= {~(I)(k),H,(k)} . (16) Furthermore, in the approximation under considera-
k 'k ~~
tion it turns out that the term with fk2 in the Hamilton-
In the region where the logarithms are large, ian is not renormalized, and O! '" v. The next approxima-
tions would undoubtedly introduce corrections to these
G(O,k)-G(O) (Inik) -b/(aH}
,T(O,k)-flO)
( k)
Ink~
-a/(a+bl
. (17) statements. However, it is extremely difficult to calcu-
late the corresponding integrals and we shall apply
Replacing In (ko/k) - k-\ we obtain another approximate method, which, moreover, gives
an idea of how the transition to the strong-coupling re-
a=2 _ _b_ a_
'11-3=_2+_ (18) gime occurs.
a+b' a+b'
Le., (}I '" v, which, incidentally, was obviOUS, since
y ~ (k/k o '[X-v", 1 in the approximation under considera-
4. THE APPROXIMATION OF A LARGE NUMBER OF
tion. v can differ from O! in second order in E. But in COMPONENTS
the first approximation, according to (18), it is neces- In the preceding section, it was noted that indices
sary to determine the ratio bfa, and from it are obtained that are close to those given by taking into
b/a (18a) account only the first approximation for the polarization
a='II=2- l+b/a' operator. This circumstance is characteristic of the
situation in phase transitions when- the "order parame-
According to (12), (14) (with f '" 0) and (11), we have
ter" has a large number of components. Although this
n(t)(w, k) =-ir(d) S{w+w_+c 2[d(k+k_)'-k+2k_2]1(d-l)} case is rare in true physical models, we can neverthe-
, dw d'k less count on the fact that expansion in the number of
x[ (w_-ek_'+ill) (w_ +ek_ 2-i6) (w+-ck+'+iI\) (w+ +ek/-i6) )-' ~ (2n)' ' components gives a series that is not too slowly con-
vergent even for n '" 3. Such an approximation was pro-
where ~ '" k1 ± k/2 and w± '" w 1 ± w/2. Integrating over posed by Ferrell and Scalapino[4). The special role of
the frequencies, we obtain the first-order diagram for the polarization operator is

711 SOy. Phys.-JETP, Vol. 39, No.4, October 1974 A. A. Abrikosov 711
explained by the fact that, as a consequence of taking the f,(X)
3
trace, it is proportional to n. Since r ~ 1/1T appears
in the self-energy 2: and, at the same time, there is ~
r----"
no summation over the components, 2: will be of order
lin. In our case, the number of components is equal to a \
four and we can count on the good convergence of this
procedure.
1
"" ~
r-
- -
Although, as we shall see be low, the 11 n expansion
2 4 6 8 10 12 14 16 18 20
leads in first order to more complicated calculations x
than does the E expansion, it has important merits.
First of all, the calculations are performed in the real "Ix)
1,5
three-dimensional space and questions associated with
the nonuniqueness of the continuation to noninteger d
./
do not arise. It is then found that, even in the first ap-
proximation, Q ,c v. Having a real model, we shall per-
form the calculation only for this model, and not for ,
b

I "I'- ........
r--
-
arbitrary n.
The expression for the first approximation for the
polarization operator is obtained from (19), if we put
V 4 8 12 16 20 24
:r

d = 3. Introducing the variables x = cos e, z = 2kl/k FIG. 3

and s = 2wl ck2 , we obtain
able s, defined by the formula (27a), analytic in the
(1'= _ 3k
_ S' dx OOs dz z'(z'+t) (1-x'}dx . (24)
II 2en' , , [(z'+1-i6}'-s'] [(z'+1}'-4z'x'] plane of the complex variable s with cuts along the
real axis for 1 < I s I < "", and symmetric with respect
Integrating over x, we obtain to s - -so The formulas (27b) and (27c) correspond to
11 (1)- 3k SOO z'dz [ 1 (z'-1)' Z+1] taking the value on the upper side along the right cut
- --- z'+1-----In-- . (25)
8en' , (z'+1-i6}'-s' 2 z Iz-11 1 < s < 00 and on the lower side along the left cut
-""<s<-1.
The integral over z can be taken if we note that
co oo+'l~ On decrease of k and w in the approximation under
SdZ-++ S dz, consideration, we enter first of all a region in which
o -oo+iO IT (1) » r( 0). However, the self-energy correction to the
with electron Green function in this region is still small, and
z+1 z+1 only on further decrease of wand k does it begin to
In---+In-.
Iz-11 z-1 playa fundamental role. By assumption (cf.[ll), it is
precisely then that there arise the power dependences
We then obtain G ~ w-Q/v for w/wo» (k/ko)V or G ~ k-a for
. 3k wi Wo« (k/k o ) v (we recall that a < v and a < 2).
11("= ---cp(s), (26)
32en Matching between these laws and the free Green function
should occur in the region in which 2: is still small.
cp () (2+s)' ( ) 'I (2-s)' 'I 'I ( }'I
S =--arctg 1+s - '---lITctg(1-s)- '-(Hs) '- 1-s '
s s - This is possible only in the case when a/ v is close to
(27a) 1 and a is close to 2.
for 0 < s < 1, and In fact, for w/ Wo » (k/ko) v,
(2+s)' " n(2-s}' a
~
<0 ) -(I_a/.'
1- ~) In :,] .
[ (
cp(s} =---arctg(Hs}-"- (1+s) 1,_ - - - 8 (2-s) G-loo<oa/·oo<o ( ""<0 H
s ~

+i"[ (S-1),1,- I.
(2-s)' In H (S-'-1} 'I,
~ (s-1)"'-1
I] (27b)
On the other hand,

for s > 1;
Hence it follows that in the matching region, for wi Wo
cp(-s)=cp(s}; (27c) » (k/k o lV,
the asymptotic values are ~(<O}""-(1- ~) <OIn~. (30 )

cp(s} -+ 4 ( ; - 1 ), s-+O;
(28 ) In exactly the same way, if we assume that 2: = qcHl
8 for w/wo« (k/ko)v, we have
cp(s}-+-=(1+i}, s-+oo;
31's

the Singular values are

k )
G-'cokacok' ( To
-(2-.) [
""k't+(2-a} In ik ]

1 n - and
cp(1} =9 arctg-=.-- -1'2,
1'2 2
(29 ) q=(2-a} In (k,/k). (31 )
4n -
cp(2}=3-1'3+i. Thus in the approximation under consideration, de-
Graphs of the real and imaginary parts of the function termination of 2: with logarithmic accuracy is sufficient
to determine a and v. Of course, all these arguments
rp (rp = rp 1 + i rp 2) are given in Fig. 3.
are valid only in the case when 1 - a I v « 1 and 2 - a
It will be important for the following that rp (s) is the « 1. The self-energy is obtained from (13) by replacing
value on the real axis of a function of the complex vari- r(O) = 41Te 2/ Ko k 2 by r = -1/1T. In addition, it is necessary

712 SOy. Phys.·JETP, Vol. 39, No.4, October 1974 A. A. Abrikosov 712
to take into account that ~(O, 0) must be subtracted integrating over kb we obtain
from ~(w, k) (renormalization of the chemical poten- 16 S~ A (s) 000
tial). For finite W we have ~(oo)=--- ---dsooln- (37)
3n', (s+2)' 00 ,
d'k
~(oo,k)=-i Sn-l(ool,kl)G(oo-ool,k-k,)dool (2n;" (32) where Wo ~ ck~; ko is the upper limit of the integration
over k. But if w « Ck 2, we must, on the contrary, put
The function G( 0) can be written in the form w = 0 in the integral (36). We then obtain
G(O) (OO-OOh k-k,) = - 1 [ 1+ HI(k-kl)] [ 00-00,-8, (k-k,) +i6j-' 16 S~ k dk dQ A S H, (k-k, ) 1
2 8, (k-k , ) (33) ~(k)='-~ I ds , c I '4';" (S,) 81(k-k, ) -s-,+-2-:(:-k--k:-,"'"')':-:-lk::-:,"
+~ [1 H,(k-k, )] [00-00,+8, (k-k,)-i6j-l,
2 81(k-k,) Since its trace equals zero, this expression should be
proportional to HI; ~ = qCHI. Calculating the anticom-
where 101 (k) = ck 2. mutator with HI and making use of the rule (7), we have
For II we substitute the expression (26), (27). Owing 8 ~ dQ
to the fact that this is the value on the real axis of an q= - - - S dS I S k, dk,-A (S,)
3n'k' I 4n
analytic function with two cuts (with an appropriate
definition on the cut), the integral over WI can be x3~[~k~X~(~k_-~k~I)~]_'-~k7'~(k~-~k~'~)'__~~_1:-7~~
represented as a contour integral, where the contour C (k-k,)' s+2(k-kl)'lk,'
is depicted in Fig. 4a, in which the poles from
. G( 0) (w - WI) are also shown. Since the integral over Expanding in k/k l, we obtain
WI for ~(w, k) - ~(O, 0) converges in any case, the
__ 8 [ 2 S~ A(s) d 16S~ A(s) d + 64S~ A(s) d]
contour can be closed and deformed in such a way that q- 3n' 5. s+2 s - 5 I (s+2)' s 15 I (s+2)' s .
it takes the form C' (Fig. 4b). Since the function has
complex-conjugate values on the two sides of the cut, However, this expression is not completely accurate.
we obtain (we do not write the integral to be subtracted This can be seen from the fact that the first term in it
for ~(O, 0)) diverges as s ~"". Indeed, according to (28), A(s)
8 ~ dQ a: SI/2 as s _00. The inaccuracy has arisen in the use
~ = - - Cds , S ck, dk,--A (s,) of the transformation (35), which is valid only for the
3n'~ 4n
case when A(s) and B(s) fall off at infinity l). We can
H,(k-k,) ] [ 200 2(k-k,)' ]_' correct the situation by subtracting from B - iA a sym-
x {[. 1+I - - -S,- --'-,-:-"--
£, (k-k,) ck,' k ,' metric function of s with the same cuts and asymptotic
(34)
',+ [1- H,(k-k,)] [2;.-s,+ 2(k-,k , )' ]-I} value, in such a way that the result now be a decreas-
8 , (k-k,) ck, k, ing function. We note that the trouble arises entirely
8 S dQ [ H, (k-k,) from the terms of second order in k/kl, which originate
+ -- dklck,--B(slO)
from Hl/€1 (the zeroth approximation is taken from SlO
3,., 4n 8, (k-k,)
and the denominators of the first term in (34 ». There-
where SI = 2w/ck~, S10 = 2w/ck~ + 2(k - kl)2/k~ is the fore, we shall consider only these terms.
value of s 1 at the pole, A( s 1) = -1m [1/ rp( s d], and
B(sl) = Re[l/rp(sd] (we assume that k« kl and 2w Subtracting from 1/ rp (s) a function which we denote
« ck~; this is valid in the calculation of the logarithmic by 1/ rpoo( s) = Boo - iAoO, we obtain
part of the integrals--cf. below). This expression can 1 ~ 1 1
be transformed by making use of properties of the func- B(2)=B~ (2)--S
n,
[A(s)-A~($) j (--+--)
s+2 s-2
ds.
tion l/rp(s) = B(s) - iA(s), namely, the analytic proper-
ties and the symmetry in s. Making use of the Cauchy As a result, it turns out that in the expression for q we
theorem for the contour C" (Fig. 4c), we obtain must make the replacement

(35)
~S A (s) ds .... _ ~B~ (2) +
, s+2 2 I
j [ As+2
(s) _ A~ (s) _s-] ds.
s'-4

Substituting this relation into (37), we obtain For 1/ rp 00 (s) we can take the simplest functio n:
8 ~ dQ 11<p~ (s) ='/ .. [ (s+1) '''+ (1-s) 'I,];
~= __ f dsISckldk,--A(sl),
3n'~ 4n
then Boo = (3/16) (s + 1 )1/2 and Aoo = (rlS) (s _ 1)1/2.
2 (k-k,) ,
k,'
2(k-k,)'
r' (36) Substituting 1/ rpoo into the integrals, we find

S~ As+2
(s) ds .... _ 31'3n + S~ (A (s) _ ~ s(s-1) 'I, ) ds.
k ,' ]} I 32 I s+2 16 s'-4

But we can do still better. We add and subtract

In this form we need no longer imply the subtraction of 3/16..fS under the integral. We then obtain
~(O, 0). In fact, in view of the fact that f dnH I (kl) = 0,
the quantity ~(O, 0) = o. jA(s)dS .... _2.+j(A(S) -~)dS.
I s+2 8 I s+2 161's
If W » Ck2 we can put k = O. Expanding in wand
Carrying out these replacements in q and comparing
the expressions for q and for ~(w) with (31) and (30),
c' we obtain
-; I p== 2-<X=~-~(AI-8A,+ 32 A,) (38)
b 5n' 15n' 3'
<X 16
l--=-A, (39)
FIG. 4 'V 3n 2 '

713 SOy. Phys.-JETP, Vol. 39, No.4, October 1974 A. A. Abrikosov 713
 A,= f(
• A(s) 3.
8+2 -16l'8 )d8, A.>,= f
• A(s)
(8+2)' ds. (40)
the power of the term of the type cH 1 cannot be realized.
Finally, equality of the powers is also impossible, since,
if this were the case, it would be true for all f.
The function A( s) = -1m [11 <p (s)] = 1m <p (s)1 I <p (s W, One can also discuss all this in a different way.
and <p (s) is given by the formula (27b). Numerical cal- Matching of the power laws with perturbation theory is
culation gives: not always possible _ In particular, in matching with
A,= -0.6976; A,=0.1570; A,=0.01207; A/=0.00183. ordinary perturbation theory in the E-expansion
method [3] it is necessary to select the value of the in-
As a result, teraction constant for which this is possible. In our
2-a=0.238, 1-a!v=0.0847 (41a) case, for ckz» w, there can be terms of the type H,
and an isotropic term in G- ' . We can make three as-
or sumptions: a) the power of the term of the type H, is
a=1. 76, v=1,92. (41b) lower than the power of the isotropic term. Then
matching with the perturbation theory starting from
The result obtained is evidence that the approach f = 0 should exist, and has been effected in this Section;
used is highly successful. In fact, both 1 - alII and b) the power of the term of the type H, is higher than
(2 - 0::)/2 are of the order of 0.1. In view of this, the the power of the isotropic term. In this case, there
values obtained for a and 11 can be regarded as fairly should be matching of the strong-coupling regime with
reliable. But if we take the first approximation of the the perturbation theory starting from c = 0, and this is
E expansion, it turns out there that the electron Green not obtained; c) finally, by assuming the powers to be
function is weakly renormalized, but, in return, the equal we can take any value of the ratio fl c = 1] in the
Coulomb-quantum Green function is strongly renormal- initial Hamiltonian. The two powers should turn out to
ized: 1) = 3 - 11 "" 1, in place of 2 as in Coulomb's law. be equal for any 1]. But this is not corroborated for
This shows that, in the given problem, E = 4 - d is not small 1/.
as suitable an expansion parameter as l/.n, in contrast
to the theory of phase transitions. This is also indi- Thus, the possibility (a) considered above is the only
cated by the fact that, according to Sec. 3,1 - alII is a one.
quantity of order EZ, whereas 2 - a is of order E. In
reality, as can be seen from (41a), they are of the same 5. "ISOTROPIZATION" IN THE SINGULAR
order (nevertheless, it is true that there is a certain REGION
tendency for 1 - alII to be small). In view of this, we Up to now we have considered only the spherically
must conclude that, although the E-expansion method symmetric model with the Hamiltonian (2). It ~as found
can be applied in the given problem for certain qualita- that the power of the second term, which we called CH"
tive estimates (the absence of the term f, and argu- in the singular region is lower than the power of the
ments in favor of the lin expansion; cf. also Sec. 5), term fkz. This corresponds to the fact that the electron
for quantitative calculations it is necessary to use the and hole masses are, as it were, equalized in the
lin expansion. singular region Z ). We now show that "isotropization"
In the derivation performed we have assumed the of the spectrum occurs in the singular region, i.e.,
Hamiltonian to be equal to cH" on the basis of the re- even in the case when we do not assume isotropy at
sult from the preceding Section that the contribution large momenta, it arises automatically in the singular
fk z in the Hamiltonian is not renormalized. Actually, region.
this is somewhat inconsistent. In fact, this result was For this we note that an anisotropic model corre-
obtained in the first approximation of the E expansion, sponds to replacement of CH" the part of the Hamilton-
whereas here we are using a completely different ap- ian given by formula (4), by c,H , + czH z, where H, in-
proximation. Nevertheless, a certain argument can be cludes the terms with the "non-diagonal" matrices
adduced in favor of the validity of the procedure used. AIlII (Il;c 11) and Hz includes the terms with t~e matrices
Suppose that f;c O. In this case, the replacement A IIII . We call attention to the fact that, accordmg to the
s->-s-211ZX, rules (5) and (6), we can easily replace the normaliza-
tion of the matrices All 11 with Il ;c 11 without damaging
where 1] = fl c, takes place in the expression (24) for the the whole scheme. This gives the possibility of general-
po larization operator. Let 1] be small. In this case, izing the method proposed in Sec. 1 for continuing to
expanding in 1] gives a correction of order 1]z to n. noninteger dimensions for the case under consideration,
Substituting n into ~ leads to the appearance of correc- and for all integer d the number of matrices Yi and,
tions of order 1/z in all terms. In particular, a term consequently, their rank are not increased in compari-
proportional to 1] z k z ln(k o/k) can arise in ~. Combin- son with the isotropic model.
ing this with the term _fkz in G(O)-I, we arrive at the
conclusion that for small 1] the power 2 - 0:: in the iso- Without repeating all the arguments,we cite the
tropic term of the Hamiltonian will be proportioual to principal formulas. Writing
1], i.e., it will be considerably smaller than the constant
obtained for the term of the type cH" But this means
that, in this case, we can neglect the isotropic term in
..
c,H,+C,H,50C,1:, A,.k,k" (42 )

the asymptotic region w, k - O. Consequently, the case we have in place of (9)

we have considered is internally consistent. It is not djJ.' 2 2d,
difficult to see that the opposite assumption, namely, (A,,,A po )= d-1 (1l"ll vo +Il..IlVP)-T-1/l,./lpo+ d-1 (1-fl )/l,.po, (43)
that f;c 0, c = 0, leads to n(I) = 0, i.e., to the absence
of a strong-coupling situation. Consequently, the possi- where Il = c II cz, and ()Illlpa = 1 when all the indices
bility that the power of the isotropic term is lower than coincide and zero otherwise. It follows from (43) that

714 SOy. Phys.-JETP, Vol. 39, No.4, October 1974 A. A. Abrikosov 714
 <C,H, (k,) +c,H,(k,), c,H, (k,) +c,H,(k,)}= ~ [d", (k,k,)' (51)
d-l

-k,'k,'+d(l-,,') ~k,.'k2V']. (44) From Eqs. (49), (50) and (51) we can find the three func-
tions C1, C2, and d. Our purpose is not this, however,
In particular, for k1 = k2 = k the right-hand side be- but the determination of the laws
comes equal to 2 ( E/ C2 )2;
(52)
e=c,k' ["'+~("'-1)A22 ]'{. (45)
in the asymptotic region. Using the notation (3 = 2 X 21T2
d-l
x 41Te 21(21T) Ko, introducing JJ. = c 1/ C2 and differentiating
is the pole of the bare electron Green function;
the equations with respect to ~, we have
dIn" ~d(s){ A,,-'/, }
~= c.(s) [H'/,.4,,(,,'-i)]"'.: (53a)
With the same rule (11) for the rank of the matrices,
dlnc, ~d(s) { '/'-'/,.4,,} (53b)
we can solve the problem. To obtain the first approxi- ~= c,(s) [H'/,A,,(,,'-1)!'" ,:
mation in E = 4 - d it is sufficient to perform the cal-
culations with logarithmic accuracy for d = 4. Since in dIn d-' = ~d(s) { ,,'+'/.(,,'-1) (4"'-1)A"-'/'("'-1)'A",} . (53c)
the present case we have not one but two invariants, the ds c, [1 +'/,A,,(,,'-l) ]'1, ••
scheme we constructed earlier[ll is not directly applic-
First of all, we make the assumption that JJ. = const.
able. However, the basic line of the calculations is the For this it is necessary that the right-hand side of
same. We shall find II and :E in the logarithmic ap- (53a) vanish, or, in other words,
proximation, assuming c 1 and C2 (or C2 and JJ.) to be
n A 1/
functions of ~ = In (ko/k), and then obtain equations
f f f
11 211

from which it will be possible to determine them. 2:' sin' e de sin <p d<p dy [1 +,/,A::~"'~1) !'" =0. (54)

Substituting G = [w - C1H1 - C2H2r1 into (12), taking This equation is satisfied for JJ. = 1, since {A 22 } av = Y4.
the integral over wand transforming the expression This solution corresponds to the isotropic case. But in
obtained, we find principle (54) could also have other solutions.
II=-4k' ~ ~ f' { ,,'+'/. (,,'-i) (4,,'-1)A,,-'/, (,,'-i)' A22'} We note that the left-hand side of Eq. (54) has the
(2n)' . c,(~,) [1 +'/,(,,'-1)A,,],/, •• asymptotic forms 3 )
" (46)
{ A,,-'/, }
where (i-'/,A,,),I, ,; " ..... 0, (55a)
n 'lit

f sin' 6 d6 f sin <p d<p Jdy{ },

II

{ }•• = (2n') -, 1 { A,,-'/, }

~ (8/ 3An) 'I. av' J.t -+ 00.
(55b)
o 0 0

A22 was defined earlier,

The maximum value of A 22, corresponding to n~ = n~
A 222 = .E nfl.'ln v 2 n p\ = n~ = n~ = 1., equals ra,
and the minimum value equals
II*"*P
zero. Since A22 = 14, we should expect that values A22
and the projections nv of the unit vector are equal to: > Y4 will play the prinCipal role in the averaging of
n,=cos 6, n,=sin 6 cos <p, n,=sin e sin <p cos y, (55a), Le., the result will be positive, and that values
n,=sin e sin <p siny. A22 < 14 will play the main role in averaging of (55b),
In (46), JJ. depends on ~1. It follows from (46) that Le., the result will be negative.
4ne' 4ne' These arguments are reinforced by the following esti-
r= (k' - 4ne 'II/ Xo ) - -Xok ' d(~), (47)
Xo mate. We expand the integrand in (54) in A22 - 14 and
where d( ~) is a logarithmic function. confine ourselves to the first nonvanishing term. The
integral is then easily taken and we obtain
We turn now to the self-energy. We write, analog- 4 { ( 1 ) '} (,,'-1) 1 (,,'-1)
ously to formula (15), - 3" A,,- '4 .. [H'/, (1-\'-1) ]''' 120 [1+'/, (1-\'-1) J'" . (56)
f
~(I)(O,k)=-;;- ~
2ne'
e(k,)
d(~,)
c,H,(k,)+c,H,(k,) d'k,
(2n)'
This formula is accurate near JJ. 2 = 1 and somewhat
, ,(48) less accurate at the limits (the expansion parameter is
"" 8ne fd(s,)(k,k) c,H,(k,)+c,H,(k,) dk, = II (k)+ H ·(k). of the order of 13). But it is, however, sufficient for the
Xo k,' e(k,) (2n)d q" q" following conclusions to be drawn: 1) there is only one
value JJ. 2 = 1 at which the expression (54) vanishes;
We take the anticommutator with AJJ. v (JJ. "" 1/). Com- 2) for JJ. - 0 this expression is positive and for JJ. - 00
paring the two parts and applying the definition c 1 = C10 the coefficient of \ JJ. \-1 is negative.
+ q1, we obtain the equation
We return now to Eqs. (53) and make an assumption
2n' 4ne', S' { d(~')"(~')A"} ) about the powers in (52): 01> Q2' This means that for
c,(~)=clO+ (2n)' -;;- /,." (H'/,A"[,,'(W-ij)'" a; (49
~ _ 0 0 we have JJ. _ 0 0 , Le., the power of JJ. is positive.

By taking the antic om mutator of the formula (48) with But for this the right-hand side of (53a) must be posi-
A pp , we have, in the same way, ti ve for JJ. - 00 (d cannot become negative and in
reality C2 implies \ C2\)' But it is proportional to (55b),
2n' 4ne' Sl { d(~,) ('/,-'/,A,,) }
c,(~)=c20 + (2n)' -;,-2" . (H'/,A,,[,,' (S.)-1])'I' a: (50)
and, according to the analysis carried out above, the
coefficient of 1/\ JJ. \ is, in reality, negative. The same
From (46) and (47) we have contradiction arises if we assume that the powers in
1 4ne' 2n' (52) satisfy 0!1 < 0!2, Le., JJ. - 0 as ~ _ 0 0 .
--=H-----·
des) Xo (2n)' Thus, it follows from the arguments given that JJ. 2

715 Soy. Phys.-JETP, Vol. 39, No.4, October 1974 A. A. Abrikosov 715
= 1 is the only solution, i.e., "isotropization" arises
in the asymptotic region, irrespective of the initial These limits are somewhat different. In fact, ck~ ~ 2.5
conditions (c 10 and C20)' x 1O- 4 wo, whence it follows that, in prinCiple, there is
a certain region of energies in which :E is substantially
6. CONCLUSION renormalized and, at the same time, the dependence of
Although the calculations performed above pertained :E on W can be neglected. But this means in practice
to the singular region, in which all quantities are that a single-particle spectrum remains for energies
strongly renormalized because of the interaction, 2.5 x 10- 4 Wo > W > 10- 5 wo, although it differs from the
nevertheless, since we have used a matching method, "free" spectrum. For W < 10- 5 WOo the Single-particle
we can analyze qualitatively the transition to this region. spectrum is strongly damped.
Deviation from the "free" behavior begins for However, all these changes occur in the region of
k < ko, W < Woo So long as the momenta and energies extremely low energies. We shall call this the third
are not too small (we shall denote the lower boundary region. In the known zero-gap semiconductors, because
by k1, WI), the change in the electron Green function of the large dielectric permittivity Ko and the small
can be neglected. In the region k1 < k < ko, WI < W < Wo, effective mass, the second region is itself shifted into
which we shall call the second region (the first is the the region of energies of the order of 10- 3 - 10- 4 eV, so
"free" region), only the electron-electron interaction that even this region, in the purest samples, may be
describable by the function r changes. The inclusion difficult to "seek out." In principle, we can assume
of the zeroth-approximation polarization operator has that substances with significantly larger values of Wo
already been made by a number of authors [5]. Our cal- will be found. However, even in a case in which Wo
culation justifies this approximation in the second re- reaches 10 eV, the third region will remain extremely
gion. The Coulomb law in this region is replaced by an difficult to attain.
interaction proportional to II r2 (in the coordinates). In conclUSion, the author expresses his gratitude to
For W » ck 2, allowance for the polarization leads to A. A. Migdal and 1. E. Dzyaloshinskil for discussion of
the appearance of a dielectric permittivity proportional the work, to Yll. V. Petro v and V. 1. Fisher for the nu-
to 1Iw 1/ 2 • From formulas (26) and (28) it is not difficult merical calculations, and to S. 1. Anisimov for friendly
to obtain assistance in organizing these calculations.
_ + e'(1+isignlll)
X-Xo (Zllle)"" (57 ) I)The expression (34) contains no such divergence.
2)We recall that the single'particle spectrum may not exist in the singular
region (cL the Conclusion).
In this form, this formula is also true for W ~ Woo It can
3) A change of sign of J.i. corresponds only to a change of basis and, there-
be shown that formula (57) also remains valid if we take fore, has no physical significance.
an isotropic model, but with different masses, Le., 4)Our calculation had the purpose of matching with the singular region,
f,.. O. In this case, c = Y4 (lime + 1/mh). and we have therefore considered the isotropic model with f = O. Tn
calculations for the second region it is necessary to use the true "free"
Since the "free" Green function and the interaction
Green function.
have the form (1) with a = v = 2, all the temperature
and other dependences found earlier in[2] are operative 1 A. A. Abrikosov and S. D. Beneslavskil, Zh. Eksp.
in the second region. As regards estimating the coef- Teor. Fiz. 59, 1280 (1970) [SOy. Phys.-JETP 32, 699
ficients, the effect can be appreciably reduced in those (1971)].
cases in which interaction of the electrons is important. 2 A. A. Abrikosov and S. D. Beneslavski'i, J. Low Temp.

For example, the electron-electron scattering probabil- Phys. 5, 141 (1971).

ity will be proportional to the number (1 - alv) (41a), "K. G. Wilson and M. E. Fisher, Phys. Rev. Lett. 28,
which is approximately 0.08. In view of the fact that the 240 (1972); K. G. Wilson,Phys.Rev. Lett. 28, 548(lif72T
second region is very simple for calculations, all the 4R. A. Ferrell and D. J. Scalapino, Phys. Rev. Lett. 29,
expressions for the phYSical effects in this region can 413 (1972).
be found, without great complications, in an explicit SM. E. Fisher and A. Aharony, Phys. Rev. Lett. 30, 559
form with numerical coefficients (in any case, for the (1973).
isotropic model)4), and this has been done in a number 5 L . Liu and D. Brust, Phys. Rev. Lett. 20,651 (1968);
of papers (cf. the literature cited in[7]). Phys. Rev. 173,777 (1968); D. Sherrington and W. Kohn,
Phys. Rev. Lett. 21, 153 (1968); Rev. Mod. Phys. 40,
According to Sec. 4, change of the electron Green
767 (1968).
function begins either when 7 J. C. Broerman, p. 917 in Proc. Int. Conf. Physics of
Semiconductors 2, PWN, Warsaw, 1972.
or when Translated by P. J. Shepherd
150

716 SOY. Phys.·JETP, Vol. 39, No.4, October 1974 A. A. Abrikosov 716