Journal of Physics: Condensed Matter

You may also like

- Performance variation due to stiffness in a
Large-S expansion for quantum antiferromagnets tuna-inspired flexible foil model
Mariel-Luisa N Rosic, Patrick J M
on a triangular lattice Thornycroft, Kara L Feilich et al.

- Damage detection algorithm based on

identified system Markov parameters
To cite this article: A V Chubukov et al 1994 J. Phys.: Condens. Matter 6 8891 (DDA/ISMP) in building structures with
limited sensors
F Amini and K Karami

- Carrier-induced antisymmetric–symmetric
tendencies of spin stiffness in zigzag
View the article online for updates and enhancements. graphene nanoribbons
Teguh Budi Prayitno and Fumiyuki Ishii

This content was downloaded from IP address 84.236.58.101 on 15/02/2023 at 02:25

1. Phys.: Condens. Matter 6 (1994) 8891-8902. Printed in the UK

Large-S expansion for quantum antiferromagnets on a

triangular lattice
Andrey V Chubukovtxs, Subir Sachdevz and T Senthi18
7 Department of Physics. University of Wisconsin, Madison, WI, USA
Department of Physics. PO Box 208120. Yale University, New Haven, CT 06520-8120,
USA
g P L KapiUa Institute for Physical Problems, Moscow, Russia

Received 14 July 1994

Abstract. We repon spin-wave results for the sublattice magnetization (to order O(l/S2)) and
two spin stiffncsses and susceptibilities (to order O(l/S)) for the two-dimensional triangular
Heisenberg antiferromagnet. These stiffnesses and susceptibilities are used as input parameters
in scaling functions for various observables. The scaling results for uniform swptibility are
compared with recent numerical data.

1. Introduction

In recent years, there has been significant interest in the theory of the low-temperature
properties of frustrated quantum antiferromagnets in two dimensions (20). Among these
systems, antiferromagnets on a triangular lattice are perhaps the most popular objects, as
there are several experimental realizations such as VClz, VBr2, CsEu, Nan02 etc [l].
Theoretical studies of such antiferromagnets go back to 1974 when Fasekas and Anderson
[2] first suggested that for spin S = $, quantum fluctuations may be strong enough to
destroy the classical 120" ordering of Heisenberg spins. Since then, there have been a
number of numerical studies of the S = 1/2 antiferromagnet: some of them [3] do indicate
the presence of well defined long-range order reduced by quantum fluctuations by nearly the
same amount as predicted by non-interacting spin-wave theory [4], while others indicate a
substantially smaller (if any) ordered moment [5,6]. The latter, if true, would indicate that
triangular antiferromagnets are very close to a quantum disordering transition. Recently, we
considered [7] the low-temperature theory of frustrated antiferromagnets. similar in spirit
to the low-T theory of collinear antifemomagnets [8]. We described the general features of
such a theory,and assuming the existence of deconfined spinons in the quantum disordered
phase, derived a number of experimentally testable results for the uniform susceptibility,
correlation length, dynamical structure factor, spin-lattice relaxation rates etc. We found that
the behaviour of observables near the quantum transition is universal and depends only on
a few input parameters at T = 0. The triangular antiferromagnet is likely to be located on
the ordered side of the transition in which case input parameters are on-site magnetization
(NO),two spin stiffnesses, p l and pli, and two susceptibilities, XI and x ~ l .The fact that
one needs two stiffnesses is indeed a direct consequence of the well known fact that the
macroscopic order parameter in triangular (and any other frustrated) antiferromagnets is
given by a pair of mutually orthogonal unit vectors, nl and nz,which specify a plane. The
two stiffnesses then measure the energy cost of twists in the plane ( ~ 1 1 ) and perpendicular to

0953-8984/5'4/428891+12$19.50 0 1994 IOP Publishing Ltd 8891

8892 A V Chubukov et al

the plane of the order parameter ( p l ) . Two uniform susceptibilities, xll and X I , on the other
hand, define the response of the antiferromagnet with infinitesimal anisotropy to uniform
magnetic fields perpendicular and parallel to the plane of the order parameter, respectively.
Note the inversion in the order of ‘parallel’ and ‘perpendicular’-the susceptibility tensor
xu,o is defined with respect to a unit vector m = nl x Q?I perpendicular to the plane of
spin ordering [9,10]:

xu.p = xl&p + (XII - xdmamp. (1.1)

The two stiffnesses and susceptibilities are indeed related by spin-wave velocities
c: = PI/XI9 c; = PIdXII.
The goal of the present paper is to calculate the five input par,ameters for triangular
antiferromagnets in the expansion over the inverse spin S. This will allow us to make
quantitative predictions in the deconfined spinon scenario on the behaviour of observables,
and to compare scaling results with recent high-temperature series expansion studies of
hiangular antiferromagnets. We will also compare our result for sublattice magnetization,
NO,with recent series expansion 161 and other studies.
Although throughout the paper we will use a large-S approach, our chief interest is in
the case of S = 112 for which numerical studies have been performed. Indeed, for small S,
the expansion parameter for spin-wave calculations is not small. However, as we wilI see
below, the convergence of the perturbative series in 112s in hiangular antiferromagnets is
very good (as it is on the square lattice [11,12]), and the 1/S expansion is likely to give
quite accurate values of observables, even for S = 112.

2. 1/Sexpansion

2.1. Bose Handtonian

We now turn to a description of the calculations. We consider here the model with
interactions between nearest neighbours:

The procedure of doing the 1/S expansion is rather standard and involves several steps
which include (i) the transformation from spin operators to bosons via Holstein-Primakoff,
Dyson-Maleev, or some other transformation, (ii) the diagonalization of the quadratic form
in bosons, and (iii) the use of a standard perturbative technique for Bose liquids to treat
the interaction between spin waves. Non-interacting spin waves have energy which scales
as S, while the interaction vertex involving m bosons scales as S2-m’2; this gives rise to
an expansion in powers of 1/S for anharmonic contributions, similar to that in a weakly
interacting Bose gas.
Another important issue, related to the I j S expansion, is the number of Bose fields
which one has to introduce in order to keep imck of the whole spin-wave spectrum,
not just the low-energy modes. This is important because quantum fluctuations are not
divergent in ZD, and the 1/S expansion involves snms over the whole Brillouin zone. In the
general case, the number of different Bose fields is equivalent to the number of magnetic
sublattices. However, in several special cases, a multisublattice magnetic configuration can
be transformed into a onesublattice ferromagnetic one by applying a uniform twist on the
 Quantum nntiferromagnets on a triangular lattice 8893

coordinate frame. In this situation, the spin-wave spectrum has no gaps at the boundaries of
the reduced Brillouin zone and one can describe all excitations by a single bosonic field, as
in the case of a ferromagnet. Obviously, the triangular antiferromagnet in a zero magnetic
field is an example of such special behaviour: the 120" ordering becomes a ferromagnetic
one in the twisted coordinate frame with a pitch Q = (4n/3,4n/d). We will therefore use
a onesublattice description of triangular antiferromagnet whenever possible. This indeed
substantially simplifies the calculations.
We start with the transformation from spin operators to bosons. The choice of the
transformation is indeed only a matter of convenience, and the final results are independent
of the way in which bosons are introduced. Nevertheless, there are several possibilities
extensively discussed in the literature [13]. We found it most convenient to use here
the conventional Holstein-F'rimakoff transformation because it preserves the Hermitian
properties of the Hamiltonian. We therefore. use

S, = s -uta S+ = JZCX a S- = utJzs-;;i;;. (2.2)

Substituting this transformation into (2.1), expanding the radical, and restricting to only
cubic and quartic anharmonic terms, we obtain after some algebra

'H = "rio f3Js('Hz 3.'Ha -1- "ri4) (2.3)

where ?to= -9 J P N is the classical ground state energy, and other terms are

(2.4)

Here i kj, and

2
-!Jk = -2 sm
3
. kx(
2
- cos k-
2 2
")
, - cos Y . (2.5)

Finally, Ak and Bk E given by

Vk 3
&=I+- Bk = --&. (2.6)
2 2

At S = 00, anharmonic terms are absent and 'Hz can be diagonalized by a standard
Bogolubov transformation

ak = k(ck +xkc-k)
t
(2.7)
8894 A V Chubukov et a1

with

and

E&= (A: - @)I/’ = ((I - V&)(1 + 24))”’.

The diagonalization yields

(2. IO)

It follows from equation (2.9) that the excitation spectrum of the ideal gas of magnons
has three zero modes, as indeed it should. Two of these modes are at k = i Q where
Q = (4n/3,4.r/& is the ordering momentum in a triangular antiferromagnet, and the
third is at k = 0 and describes soft fluctuations of total magnetization. The expansion near
zero modes gives two spin-wave velocities

(2.1 1)

(2.12)

The ratio of the two at S = cc is q/cL = &. This waS also obtained in other
approaches [lo].
The infinite4 spin-wave results can be also used to get the fust quantum comection to
on-site magnetization [4]. Indeed, (uta) in (2.2) is nothing but the density of particles which
is finite due to the anomalous term in the quadratic form. From (2.7) and (2.8), we have
) (Ak - & ) / 2 4 , and therefore non-interacting spin waves reduce the sublattice
( a l u ~=
magnetization to

0.522
(2.13)

We next consider corrections to equations (2.11) and (2.13) due to the interactions
between spin waves. We will follow the same line of reasoning as for square-lattice
antiferromagnets. However, the presence of cubic terms makes the analysis considerably
more involved.
We start with the spin-wave velocity renormalization.
 Quantum antiferromagnets on a triangular lattice 8895

2.2. Spin-wave velocity

Our goal is to obtain the leading 1/Srenormalization of spin-wave excitations. For this we
consider first-order self-energy corrections due to quartic anharmonicities and second-order
corrections due to cubic anharmonicities (recall that cubic terms have the overall factor S
I
')
.
The corrections due to quartic terms are easy to compute, because of leading order in 1/S,
one can get away with simple one-loop diagrams. Equivalently, one can simply decouple
the fourfold term in equation (2.4) by making possible pair averaging. The quadratic form
allows for non-zero normal (u$zk) and anomalous (uku-k) pair products of Bose particles,
and the decoupling changes At and Bk to

(2.15)

A simple inspection then shows that the renormalized spectrum ( E k = (x: - still
keeps a zero mode at k = 0, but acquires a finite gap at k = iQ:

(2.16)

This finite gap is indeed an artifact of using only quartic terms, and cubic anharmonicities
should restore the correct structure of the spectrum, as we demonstrate below.
There are several ways to deal with the cubic terms: one can either calculate the
effective fouifold vertex produced by two triple vertices [ 14,151, and then use the decoupling
procedure, or one can transform to quasiparticles (i.e., diagonalize the quadratic form)
considering first only quartic corrections, and then calculate the renormalization of the
excitation spechum due to cubic terms in the second-order perturbation theory. Below we
use the second approach which is technically advantageous. We therefore first transform
from particle operators (ak) to quasiparticles (Ck) using equation (2.7), but with A k and B k
instead of An and Bk. The bare Hamiltonian then keeps the form of equation (2.10) with
& instead of Ek. On the other hand, the structure of cubic vertices becomes more involved
after the transformation to quasiparticles, and instead of equation (2.4) we obtain

The vertex functions @I and @2 are given by

8896 A V Chubukov et a1

and

*
f“’ - -( Xif&)I/’. (2.19)

The self-energy diagrams to order 1/S are shown in figure 1. We see that cubic terms give
rise to both normal and anomalous self-energy parts so that the dispersion relation again
has the form typical for a 2 x 2 problem:

(0 + L ( k 90))’= ( E k i-
Xdk, 0))’ - (X+.+(k,0))’ (2.20)

- -
where C,,a(k, o)= (Z+,-(k, o)f X+.-(-k, -a).However, it is not difficult to check
that C-,- E+.+ 1/S and therefore anomalous self-energy terms contribute to the
excitation energy only to order l/S2, while to order 1/S a solution of equation (2.20) is
simply 0 = Ek where

l?: = 8: 4- 2&X+,-(k, E k ) . (2.21)

We therefore need to evaluate here only the normal component of the self-energy. The
analytical expression for E+.-is

To leading order in 1/S we can indeed use non-renormalized values for Ax, Bk, E k in the
RHS of (2.22).

z + + = ~
Figure 1. Secondader self-energy corrections to magnon propagators due to cubic vertices.
Notice tM cubic terms always produce anomalous selfenergy terms.

We first demonstratethat & has a true zero mode at k = Q. For this we need to evaluate
Z+,-(Q, EQ). We found the following equality to be quite useful in the calculation

J?; %tQlz = (Aq+plz + BqiQf2) - (AqrQ/z - BqTQfZ). (2.23)

Substituting (2.23) into the expressions for the vertex functions and using AQ = BQ = $,
we obtain after simple algebra
 Quantum antiferromagnets on a triangular lattice 8897
Substituting, then, the vertex functions into the formula for the self-energy we obtain using
(2.19)

(2.25)

Finally, upon substituting this result into equation (2.21) and using (2.16) for &, we find
that the gap in the excitation spectrum disappears as it should.
Our next step is to expand E k and & near the zero modes, and obtain the corrections
to the spin-wave velocities to order 1/S. The expansion near k = 0 is quite straightforward
because &,(1,2;k ) and & 2 ( l r 2, k ) both scale as k at small k , and one can therefore safely
neglect Et in the denominators in (2.22). Doing the algebra, we obtain the renormalized
spin-wave velocity at k % 0 in the form

where

x +sin2 k
Q2, --sin. , k - + J5k, +sin2 kx -sky
2 4 4

Numerical integration then gives

= C,I (1 - !?$) (2.27)

The structure of the expansion near k = +Q is more involved and we refrain from presenting
the analytical expression for the spin-wave velocity. Numerically, we obtained

2 1 = Cl (1+ !g) (2.28)

Comparing (2.27) and (2.28). we observe that quantum fluctuations tend to diminish the
difference between the two spin-wave velocities. This is consistent with our 1 / N result in
[7] that the relative difference between tl and El, should disappear at the quantum-critical
point. We will use (2.27) and (2.28) below and now proceed with the calculations of
sublattice magnetization.

2.3. Sublattice magnetization

We have shown above that to leading order in 1/S, the correction to sublattice magnetization
comes already from non-interacting magnons (equation (2.13)). Here we obtain the next
term in the expansion in 1/S, which is also the leading 1/S correction to the density of
particles. We again have to consider both quartic and cubic terms, since they contribute at
8898 A V Chubukov et a1

the same order to x k ( & ) . AS before, quartic terms only renormalize the coefficients in
the quadratic form, and hence change the expression for the density of particles to

(2.29)

where i k and & are given by (2.14) and (2.15). In explicit form
---E”
E””--9 U2
k E k ’ k Ex 32s , E, E9 32s , E, 9 4 .
(2.30)

We see that the very last term behaves near q = Q as 1q - which makes the integral
over q divergent. The divergence is indeed an artificial one and should disappear when we
add the contributions of the cubic terms.
To see how cubic terms modify (2.29), we express the density of particles in terms of
the quasiparticle operators using (2.7) and (2.8):

(2.31)

The first two terms are just the renormalized spin-wave terms. The third correction is related
to the anomalous self-energy term in figure 1. Performing the frequency summation in this
term, we obtain

(2.32)

where

(2.33)

Finally, the last term in (2.31) contains the density of quasiparticles. This density is
finite to order 11s because among cubic non-linearities, there is the term which describes
simu~taneousemission of three spin waves. Evaluating the expectation value of &) by
the usual means, we obtain

(2.34)

where

(2.35)
 Quantum antiferromagnets on a trionguiar lattice 8899

We first show that the total expression for the density of particles is free from
divergencies. Simple inspection of equations (2.32)<2.35) shows that the divergent
contributions from the cubic terms (namely, l/E3 and l/EZ terms in (2.32) and 1/E2
terms in (2.34) ) come from the region k % Q, where \Y and T tend to constant values. For
these k , we again use (2.23), substitute it into the vertex functions, and after some simple
algebra obtain

(2.36)

Substituting further this expression into (2.32) and comparing the result with the divergent
piece in (2.30). we find that the l/E3 contributions from cubic and quartic terms, and the
1/E2 contributions from the two cubic terms cancel each other, so that the 1/S correction
to the density of particles is finite, as it of course should be. We then performed numerical
computation of the 1/S terms in (2.31) and obtained

S - 0.261 +- (2.37)

For S = 1/2, equation (2.37) yields (S) = 0.266, which is close to half the classical
value. A very similar result was obtained earlier by Miyake [16], who calculated the on-
site magnetization to order l/Sz by evaluating numerically the response to a staggered
magnetic field. His estimate for the l/p correction is however somewhat smaller than ours
(0.01 instead of 0.027). In any event, l/Sz terms are rather small and can hardly change
substantially the lowest-order spin-wave result for the magnetization [43. We therefore
found no support for the recent claim based on series expansion analysis [6] that the value
of magnetization is substantially lower than the spin-wave prediction. Note, in passing,
that for a squarelattice antiferromagnet, the first anharmonic correction to ( S ) is exactly
zero 1171. Indeed, cubic terms are absent from the square-lattice antiferromagnet, and 1/S
corrections due to quartic terms do not change the shape of the quasiparticle spectrum (that
is, & / E k = An/& ). The next-to-leading-order correction in the square-lattice case has
been calculated and found to be very small 1171.

2.4. Uniform susceptibilify

Now we calculate, to order l j S , the response of a triangular antiferromagnet to an external
magnetic field, i.e., transverse and longitudinal susceptibilities. For classical spins, the two
susceptibilities can easily be obtained by minimizing the ground state energy. This yields
XI = ,q = 2/91/5Ju2 where U is the interatomic spacing (az-.,%/2 is the unit cell volume).
As in the bulk of the paper, we define X I and X I I without the gyromagnetic ratio gpLg/Ft. We
see that the two susceptibilities are equal in the classical limit [181. This degeneracy in the
response to a magnetic field in a ZD hiangular antiferromagnet has attracted some attention
in the past as an example of the ‘order from disorder’ phenomenon [1&22]. For our present
purposes, it is sufficient to observe that the degeneracy is a purely classical effect. It is not
related to the symmetry properties of a quantum system and therefore should be broken by
quantum fluctuations.
Technically, the computations in a finite field are more involved because the transverse
field breaks the 120” ordering in the basal plane. In this case, a transformation to a twisted
coordinate frame is no longer advantageous because Umklapp processes also contribute to
8900 A V Chubukov~eta1

order 1f S. It is then more convenient to introduce a separate Bose field for each of three
sublattices. For the longitudinal response, the 120" ordering in the basal plane is preserved
and a onesublattice description with no Umklapp terms is still valid. However, one has to
be careful in this case as well, because in the presence of a field, the excitation spectrum is
no longer an even function of k . This is consistent with the fact that time reversal symmetry
in a magnetic field requires that in changing k + -k in the spectrum, one has to change
simultaneously the sign of H.
The corrections to the susceptibility tensor to order 11s were computed by Golosov
and one of us [21]. We refrain from presenting the details of the calculations and list here
only the results. To order Ifs, they are (notice that the definitions of XI and in [21] are
interchanged compared to ours):

(2.38)

where

(2.39)

and

where f") = f",) were defined in (2.19). Note that contrary to the situation in a stacked
3D triangular antiferromagnet where z XI, the transverse (in-plane) susceptibility
in the 2D case turns out to be larger than the longitudinal one; this gives rise to an
unconventional phase diagram in a magnetic field which has been discussed several times
in the literature [18-211.

2.5. Spin stiffness

With the values of the two spin-wave velocities and spin susceptibilities at hand, we are
now in a position to calculate the spin stiffnesses. To order 11.7 they are

where

Zf = I - 0.1I9f2.S Z{ = 1 - 0.67912.9. (2.42)

 Quantum antiferromagnets on a triangular lattice 8901
3. Conclusions

In this paper, we calculated in the 1,’sexpansion the T = 0 parameters of a ZD Heisenberg

antiferromagnet on a triangular lattice. The two spin stiffnesses and susceptibilities were
calculated to fist order in 1/S,while sublattice magnetization was calculated up to O(l/S2).
We found that numerically, the l/Sz correction to the magnetization is rather small even
for S = 112 so that one may hope that already the first order in the 1/S expansion
yields quite accurate results for all spin values. We expect the same to be true for
stiffnesses and susceptibilities. The application of our 1/S results to S = 112 then
yields XI = 0.09/Ja2, X I , = 0.07/Ja2, p~ = O.WJ, = 0.07J. We also found
NO = 0.266 which is about a half of the classical value. Both the on-site magnetization and
the spin stiffnesses are smaller than in a square-lattice antiferromagnet, where NO = 0.3
and ps = 0.18J 1231. However, the difference in magnetization is relatively small, and
for stiffness it is mostly associated with geometrical factors. We therefore cannot conclude
from our results that triangular antiferromagnets are much closer to quantum disordering
transition than squarelattice ones, as has been recently suggested [6].
Finally, for comparisons with experimental and numerical results at finite T , when long-
range order is absent, it is convenient to introduce the effective stiffness and susceptibility
averaged over all three spatial directions:

2 1 2 1
P = -PL
3 + ?PI1 x = ?XL + -xII.
3
(3.1)

Application of 1/S results yields

1 -0.399/2s
P=
1/5
J S tqsx = -
91/5Ja2
(1 - 0.339129 (3.2)

For S = 112, we then have p = 0.0875, x = 0.085/Ja2.

In a separate publication [7],we obtained the scaling expressions for various observables
near the critical point (in the quantum-critical region) with p and x as input parameters. In
particular, for the uniform susceptibility we obtained

x.(T)= (3.3)

where cz = PIX. Substituting the values of x and p into (3.3), we obtain for S = 112

xu(T)= (2) 2
[o.Wx + 0.14- J (3.4)

The temperature dependence of xu was recently studied in high-temperature series

expansions 1241. The data show that xu obeys a Curie-Weiss law at high T,passes
through a maximum at T 0.4J, and then falls. Numerical data are available only over a
relatively small temperature region below the maximum. Nevertheless, we fitted the data
by a dependence linear in T , as &equation (3.4), and found 0.13 iO.03 for the slope and
around 0.06 for the intercept-both results are in reasonable agreement with our theoretical
expression. The comparison with other observables is considered at some length in [7].
8902 A V Chubukov et a1

Acknowledgments

The research was supported by NSF grant No DMR-92 24290. We thank N Read for useful
communications and P Lecheminant for pointing out an insufficient precision of numerical
calculations in the original version.

References

[I] For a recent review see e.g.

Harrison A 1992 Ann. Rep. Pmg. Chem (R. Soc. ChemJ S7A 211, S8A 447
Plumer M L,Caille A, Mailhot A and Diep H 7 1994 M a p t i c System with Compering INernctions ed H
T Diep (New York World Scientific)
[2] Fasekas P and Anderson P W 1974 Phil. Mag. 30 432
[3] Miyashita S 1984 J. Phys. Soc.Japan 53 44
Bemu B, Lhuillier C and Pierre L 1992 Phys. Rev. Len. 69 2590
A h a P, Delamone B and Mouhanna D 1993 Phys. Rev. Len. 70 2483
[4] Jolicoeur T and Le Guillou J C 1989 Phys. Rev. B 40 2227
[SI See e.g.
Nishimon H and Nakanishi H 1988 1. Phys. Soc. Japan 57 626 and references therein
[6] Singh R R P and Huse D 1992 Phys. Rev. Lett. 68 1706
[7] Chubukov AV, Sachdev S and Senthil T 1994 Phys. Rev. Len. 72 2089
[SI Chubukov A, Sachdev S and linwu Ye 1994 Phys. Rev. B 49 I 1 919
Sachdev S and Jinwu Ye 1992 Phys. Rev. Lert. 69 2411
Chubukov A V and Sachdev S 1993 Phys. Rev. Len. 71 169
[9] Andreev A F and Marchenko V 1 1980 Sov. Phys.-Usp. 23 21
[IO] Dombre T and Read N 1989 Phys. Rev. B 39 6797
[Ill Canali C M,Girvin S M and Wallin M 1992 P h p . Rev. B 45 10131
[I21 IgarasM J 1992 Phys. Rev. B 46 10763
[131 K a w o v M I and Chubukov A V 1987 Usp. Fiz No& 153 537 CEngl. m l . 1987 So". Phys.-Usp. 30
1015)
Borovik-Romanov A S and Sinha S K (ed)1988 Spin Waves ondMagneric Dieleerrics alsevier)
[14] Rastelli E. Reatto L and Tarsi A 1985 J. Phys. C: Solid stare Phys. 18 353
1151 Chubukov A V 1991 Phys. Rev. 44 5362
[I61 Miyake S J 1992 J. Phys. Soc. Japan 61 983
[I71 Castillia G E and Chakravarty S 1991 Phys. Rev. B 43 13687
[IS] Kawamura H and Miyashita S 1985 J. Phys. Soc. Jopm 54 4530
[I91 Konhunov S E 1986 J. Phys. C: Solid.Stnte Phys. 19 5927
[20] Kleine B. Muller-Hartmann E, Frahm K and Fazekas P 1992 2 Pkys. B 87 103
[211 Chubukov A V and Golosov D 1991 1. Phys.: Codem. Muller 3 69
[U] Qing S h a g and Henley C L 1992 J. Phys.: Condens. Matter 4 2937
1231 Singh R R P 1989 Phys. Rev. B 39 9164
1241 Elmer N, Singh R R P and Young A P 1993 Phys. Rev. Lert. 71 1629