Solid State Communications, Vol. 28, pp. 667-670.
© Pergamon Press Ltd. 1978. Printed in Great Britain.
0038-1098/78/1122-0667 $02.00•0
ABSENCE OF FERROMAGNETIC LONG RANGE ORDER IN RANDOM ISOTROPIC DIPOLAR MAGNETS
AND IN SIMILAR SYSTEMS*
A. Aharony
Department of Physics and Astronomy, Tel Aviv University, Ramat Aviv, Israel
(Received 29 August 1978 by I41.Low)
It is shown that random off-diagonal exchange interaction coefficients
destroy ferromagnetic long range order in isotropic systems with less than
four spatial dimensions. A special case is that of random isotropic systems
with dipole-dipole interactions. The proof breaks down when (cubic or
hexagonal) symmetry breaking terms are introduced. However, in some
cases such terms turn the transition into the ferromagnetic phase first
order, with a possible spin glass phase above it.
THE EFFECTS of random impurities on the critical
properties of magnets have recently drawn much
attention [1 ]. In some cases, the existence of impurities
leads to the drastic result that ferromagnetic long range
order, which exists in the absence of impurities, must
completely disappear once such impurities are
introduced. This has been shown to be the case when
random magnetic fields are introduced in systems with
rotational invariance (described by Heisenberg or XY
models) in d < 4 dimensions [2]. Very recently, the
same result was shown also to be true [3-5] for sys-
tems with random uniaxial anisotropy, i.e. with a single
ion term I~ i (ni • Si) 2 in the Hamiltonian, where n i is a
unit vector of random direction and Si is an n-component
spin vector at the site i in the system [6, 7]. The exper-
imental situations with such systems is far from clear,
since (a) no magnetic systems with random fields are
known and no detailed experiments on other such sys-
tems are available, (b) the experiments on amorphous
rare earth-iron alloys [6], to which the model of ran-
dom uniaxial anisotropy is assumed to apply, are not
very clear, and (c) the assumption of complete
rotational invariance, needed in the proofs of [2-4], is
usually not satisfied. The present paper contains two
new results, which are aimed at resolving these difficul-
ties.
First, it is shown that the random uniaxial aniso-
tropy is only a very special case of a large variety of
systems in which long range order is eliminated. Con-
sider a general spin system, with the Hamiltonian
n
1 Z E (1)
Jf = -2 ia ~,o=I
* Supported by a grant from the United States-Israel
Binational Science Foundation (BSF), Jerusalem,
Israel.
with random exchange coefficeints j~t3, such that the
configurational averages [J~t~]av maintain the n-
dimensional spin space isotropy. We show, that any ran-
dom off-diagonal exchange coefficient, J~ with ~ 4: f3,
will prevent the system from having ferromagnetic long
range order for d < 4.
A very important example, which is described by
(1) is that of a random isotropic dipolar magnet, with
[8]
-- (d-- 2)gigiu (8 2
= -- dxijxo/ro)/rli, (2)
where r o = r i -- ri and gi/sB is the magnetic moment of
the ion at site i. Both the moments gila~ and the
locations r i may be random, leading to random values of
Ji~#. For isotropic distributions of these random vari-
ables, the averages [JJ]av will describe the usual iso-
tropic dipolar system [8]. The effects of randomness on
this system were recently studied, in d = 4 -- e dimen-
sions [9]. It was found, that the isotropic pure dipolar
fixed point is unstable with respect to randomness. A
"random" stable dipolar fixed point was found, but it
was not clear whether this fixed point can be reached
from physical initial Hamiltonians. Our present results
show, that this fixed point cannot be reached, since no
ferromagnetic long range order is possible.
Many other examples exist. For example, quenched
random anisotropic strains, e7 n , may induce terms in
the Hamiltonian of the form [10]
Z g e?6s sf, (3)
agq6
leading to similar results. It is hoped that this large
variety of possible systems will make a detailed exper-
imental study of the general statements easier.
The second result relates to the assumption of
rotational invariance. Real crystals usually have sym-
metry breaking terms, e.g. of cubic or "hexagonal" sym-
metry [11,121,
667