HAL Id: jpa-00246460
https://hal.archives-ouvertes.fr/jpa-00246460
Submitted on 1 Jan 1992
HAL is a multi-disciplinary open access archive for the deposit and dissemination of sci- entific research documents, whether they are pub- lished or not. The documents may come from teaching and research institutions in France or abroad, or from public or private research centers.
L’archive ouverte pluridisciplinaire HAL, est destinée au dépôt et à la diffusion de documents scientifiques de niveau recherche, publiés ou non, émanant des établissements d’enseignement et de recherche français ou étrangers, des laboratoires publics ou privés.
The Potts spin-glass in a transverse field
Yadin Goldschmidt
To cite this version:
Yadin Goldschmidt. The Potts spin-glass in a transverse field. Journal de Physique I, EDP Sciences,
1992, 2 (1), pp.31-39. �10.1051/jp1:1992116�. �jpa-00246460�
Classification Phydcs Abstracts
05.20 05.30 75.50L
The Potts spin-glass in
atransverse field
Yadin Y. Goldschmidt
Department of Physics and Astronomy, University ofPittsburgh,
Pittsburgh,
PA 15260, U-S-A-(Received
23 July 1991, accepted 20September 1991)
Abstract. ln this paper we derive an expression for the free energy of the infinite ranged quantum Potts spin-glass in
a transverse field using the replica method and the Suzuky-Trotter
formula. Results are obtained for general p
(number
of Pottsstates)
and explicit calculationsare done for p
= 3
using
exact spin summations for M < 8 where M is the size of the lattice in the Trotter direction. The phase diagram is obtained and is also compared with results for thestatic approximation which is non trivial even in the paramagnetic phase.
I Introduction.
The
investigation
of the quantumspin-glass
model attracted much attentionrecently [1-17].
It has been found that quantum
tunneling
has adisordering
ef§ect similar to the temperature.In the case of the
Ising
model in a transversefield,
the transverse field simulates the effect oftunneling
in protonglasses
[3]. It has been found that thisfield,
when strongenough, destroyes
the
spin- glass phase
even at T= 0~ but the nature of the
spin-glass phase,
when itexists,
for the case of the infinite range
model,
is very similar to the classical case(r
=0)
and is characterizedby
a Parisi order paramaterq(z) (or alternatively by
a non trivial distribution ofoverlaps P(q)) [18]. Experimental
realizations of protonglass
occur in random mixtures of ferroelectric and antiferroelectric materials [3], and it has beenproposed recently
that thequantum Heisenberg spin-glass
mayplay
a role for some range ofdoping
in thenewly
discoveredhigh-n
materials(in
theinsulating phase)
[19].The classical infinite
ranged
Pottsspin glass
has someinteresting
features which have beeninvestigated
in recent years [20,21].
Inparticular
for p > 4(p
is the number of Pottsstates)
the
spin-glass
transition has some first orderfeatures,
and for any p > 3 the order parameterq(z)
has aparticularly simple
form for some range of temperatures below Tg [21].In this paper we
generalize
theIsing
model in a transverse field to the case of the Potts model withgeneral
p and use thereplica
method and theSuzuky-Trotter
formula[22,
23] to map the model into anequivalent
classical system. The Potts model in a transverse field which representtunneling
between the pequivalent
Potts states isquite
non trivial. Even the staticapproximation
in theparamagnetic phase
is difficult toimplement
forgeneral
p.32 JOURNAL DE PHYSIQUE I N°1
After
deriving
somegeneral
results we concenrate on the case p= 3 where the transition is
expected
to be of second order for any value of the transverse field r(we verify
thisexplicitly
for M =2).
We first determine the location of thephase boundary using
the staticapproximation.
We then carry out exact
spin
summations up to M < 8 andextrapolate
the results to M = ocusing
theI/M~
law [24]. The result obtainedby
this method for therc(T
=0)
is found to be lower than staticapproximation
result which is known from theIsing
case to over-estimate the value of the transverse field necessary todestroy
the transition at T= 0.
Finally,
Wepoint
out
questions
and directions for future research.2. The free energy and order paranJeters for
general
p.Following
the discussion in reference [25] we consider thefollowing
Hamiltonian for the Potts model in a transverse field:p-I p-I
~f =
~
j,,£(u±u,)m r£ ~
vp(i)
iJ i J i
(ij) m=0 I m=0
Here
(ij)
is a summation over all distinctpairs
ofsites,
and it is a summation over all sites.V and U are site variables
(operators)
which on agiven
sitesatisfy
theZp algebra
VU =
e~~~'/PUV, V+U
=
e~~'/PUV+,
VP= UP
=
(2)
In a
particular representation
in which U isdiagonal
one hasU
nj
=e~'+ nj,
n = 0,, p
(3j
V
n) =I
n +ii
andPI
= 0)(4)
Jij
are random variables which forsimplicity
can be chosen tosatisfy
a Gaussianprobability
distribution of widthJ/@
and mean JoIN
where N is the total number of sites.We now make use of the Suzuki-Trotter formula to map the system into a classical system
plus
an extra '~Trotter" dimension. The effective classical Hamiltonian for the Mthapproxi-
mant is
given by
~
M N
~~~~~~ ~
M
( ( ~"~~'l~~>~J(~)
~ ~~ $ ~n,(k),n,(k+i)
+CNM, (5)
with
~ ~
B =
In(
~~'~j~~),
C =In(
~ ~'(6)
e M I P
and 7
=
fly.
Inderiving equation (5)
we have used the identitiesexp(7 f
V~)
= 1+ ~~~f V~' (7)
m=0 P
m=0
(n exp(7 f
V~) n')
=
bn,n,
+~~~
(8)
m=0 P
where the
eigenstates n)
of U where defined inequation (3).
In
equation (5)
the classical variablesn,(k)
take the values0,
p- I on sites(I, k)
where k =I,
,
M labels the Trotter direction and
satisfy
theperiodic boundary
condition in this"time"
direction. Ultimately
the limit M- oc must be taken. The
quenched averaging
overthe random bonds is done
by
the use of thereplica
trick[26].
After somealgebra
we obtain:"
"4flln 11 £iQ17'(k> k')12
+t ~o ~o ~imz<~,j~
no , ~~ ~ ~ ~
-(In~hexp(G) (9)
with
p2 j2
~ "
~ £(~~na
,(k),na(k+i) +
tl)
+~ 2 Jf2
~ ~ ~ Qll
(~>k')(P~n"(k),~ l)(P~n«'(k'),s
~)a i P
««, ;1, »
+
jjj ii LMI(k)(p&n«~i~,~ i) (lo)
~ ~ i
where the linfit
n - 0 has to be taken. We now define
j~a
(~ ~l) Qaa(~ ~l) (~~)
~8 ~8 '
and reserve the notation
Qff'
for thecase where a
#
o'. Based on the Potts symmetry of the model we look forstationary
solutionssatisfying
Q(f'
=
Q°°'(pbm
1)(12)
Rls
=R°(pb~s
11(13)
We will further
simplify
theexpression
for the free energyby making
thefollowing
assump-tions and arguments:
First since we are interested
only
in thespin-glass phase
and not in aferromagnetic phase (or
a mixedphase)
we chooseJo
to benegative
such thatMf(k)
= 0. This choice has alsobeen used in the classical case
[20, 21].
Next we assume that
Q°°'(k, k')
does notdepend
on the Trotter index. This has also beenused in the
Ising
case and follows from the nature ofQ
as thespin-glass
order parameter, and it is found to be a consistant solution of thestationarity
conditions. ForR°(k, k')
we will assumethat it does not
depend
on a since there is noprefered
direction inreplica
space and that thedependence
onk,
k' comesthrough
the difference k k'(because
of translation invariance in the Trotterdirection).
Tosimplify
matters we will use a first stepreplica
symmetrybreaking
scheme [27] in the
spin glass phase
which for the classical case(for
p >2)
is known to be exact in a range of temperature belowTg.
[21]Thus one puts a =
(L7)
with L = I,,
n/m;
7 " 1, m andQ°°'
"
Q2
L"
L'>'f # 'f' (141
Q°°'
=
Qii
L#
L'(15)
In the classical case for p > 2 one has
Qii
" 0 in a range of temperatures belowTg.
Since it is notguaranteed
to hold for r#
0 we willkeep
the mostgeneral
case(15).
First step RSB34 JOURNAL DE PHYSIQUE I N°1
even if not exact is a
good
firstapproximation
to thespin glass
case as has been found for theIsing spin glass
in a transverse field [14](p
=2).
We will now use the Wallace-Zia [28]
representation
of the Potts variables as p I dimen- sional vectorsS(n),
n =0,
,p I.
We have
pbnn,
I =S(n) S(n') (16)
The free energy
can now be written in the form
p2
j2~ i
~~
~~ ~~
~~~~~~2 ~~l) ~(
+j~2
~~~~
~ ~' ~ii'
jDP~~z
injDP~~y Z'$ (17)
m
~~~~~
~p-i, jj
~~'~)exp(- £ z?/2) (~~)
@
~n>=i I
and
similarly
forDP~~y. ZH
isgiven by
Ttexp(H)
whereH =
j L(R(I
k k'
1)
Q2)S(n(k)) S(n(k'))+
+
~(
~S(n(k)) S(n(k
+1))
+ ~ +C)
~ P P
+$( Qii
y +
/& z) ~ s(n(k)) (19)
from
equations (17)-(19)
we obtain thestationarity equations
for the R's andQ's
q
_ -l~
ji~
(~ij
~~
p- I
I ~ ~yZg
where
(A)H
"lh[A exp(H)]/lh exp(H) (23)
In the
pararnagnetic phase equation (17)
becomesPi
=(jj~ (p
11~ R(
k k'1)~ in zH
(24)
~ ~
kk,
with
H =
~ £
R(I
k k'I>S(n(k» S(n(k'l>
+L()S(n(k» S(n(k
+ii>
+)
+
Cl (251
and
R~i
k k' i> ")iS<n<k» S<n<k'»iH
(26>In the case that the transition is continuous
<which
for q < 4 is known to be true when r=
0),
then thephase boundary
is determiendby
the condition This is foundby expanding
thefree energy in powers of
Q
andidentifying
the coefficient of thequadratic term>:
ji £
~<
k k' I> "TC/J
<27>k,k'
For the case when the transition is of first order
equation
<27> is nolonger
valid since the location of the transition is nolonger
determinedby vanishing
of the coefficient of theQ~
term in theexpansion
of the free energy in power ofQ.
In thatcase the location of the transition
can be found
by matching
the freeenergies
in theparamagnetic
and in thespin-glass phases.
3.
Explicit
calculations for p= 3.
In this section we will carry out
explicit
calculations for the three state Potts model. We willassume that as in the classical case with r =
0,
the transition is continuous and hence we canuse the condition
(27>.
We have verified that this is the case for M = 2(see below>
where wewere able to calculate the order parameters
Q2
andQii
in thespin-glass phase.
Before we
proceed
with the exact calculations we willbriefly
describe the results of the staticapproximation.
In the static
approximation R(k,
k'> is taken to beindependent
ofk,
k'. We then find thatequation
(24> taken the formfl f
=~~~~
R~
7 lim in G <28)with
G
=
j
Dz 'hexPl~ £(S(n(kl S(n(k+ Ill 2>+ $4
z
.~lS<n<k»
+S<n(k+ I»11
(29>s(o>
=<v5,o>, s<i>
=
<-(, (>, s<2)
=
<- (, (>
<30)
The trace in
equation (29)
can be obtained from theeigenvalues
of a 3 x 3 transfer matrixrepresenting
a I dimensional Potts model. Because of the Mdependence
of B B--
In(M/r)
for
large M)
and of themagnetic
field one needs to consider a non- trivial M- oc limit.
This means that one has to calculate the
I/M
correction to theeigenvalues li
of the transfermatrix.
Pitting
li = I +flui /M
+O(I /M~
we findG =
j
Dz£efl~'
<31)
I
l
with u,
being
solutions of theequation
u~ 3<(h( I )h(
+r~)u 2r~ #h(
+~#hi hi
= 0 <32>
36 JOURNAL DE PHYSIQUE I N°1
h;
=J4
z;(33)
Equation (33)
can be solvednumerically
andtogether
withequations (28) (31)
and the condition R =I/flcJ
can be used to find thephase boundary
between thespin-glass
andparamagnetic phases.
Here we consider thelimiting
cases of r= 0 and T = 0.
In the case r = 0 one
easily
find~~~
"~
~~~'
"~
~~~
~~~~'
~~~~~ ~~~
~~~~p j
=1p2 j2
j~2p2 j2
j~ f~ ~(~~j
~ from which R
= I and also Tc = J. This agrees with known results for the classical case.
In the case Tc - 0 we would like to find r
(Tc)
in that limit. We have to consider the limit flc - oc and R- 0
(since
Rc=
I/flcJ
at thetransition)
and thus h; inequation (32)
are also small. Under these conditions thelargest
solution forequation (32)
isum = 2r +
(h(
+h()
+= 2r +
J~Rz~
+(36)
3r 3r
Thus the free energy is
given
in the small T limitby Pi
=(~~
R2
3Pr
+in(1 ~((~ (37)
From this
equation
and the condition ll~ = I/flcJ
we findrc(T
"0)
"~J (38)
In case of the
Ising
model(p
=2)
it is known that the staticapproximation
tends to overestimate the value of rc at T =0.(The
exact value is aboutrc(0)
-- 1.51J whereas the static
approximation yields
rc (0> =2J>.
We now
proceed
to an exact evaluation of the location of thephase boundary using
exactspin
summations. We have evaluated the free energy(24)
and solved for R (( k k' for lattices with the size of the Trotter dimensions up to M = 8(with periodic b-c-).
For M = 2we have also solved for the solution in the
spin- glass phase.
In thespin-glass
one has to carry out the two dimensionalintegrals
d~z andd~y
inequations <20)-(22)
and the solution isquite
time
consuming.
The results of thespin-glass phase
calculationindicates,
at least for M= 2,
that:
(a)
The transition is of second order for p= 3 even for r
#
0;(b)
The first stagereplica breaking
withQii
" 0 is a solution of the
stationarity
conditionfor a range of temperatures below
Tc<r)
as it is for r= 0.
More work has to be done to establish these results for M = oc
although
the results seemsplausible. Assuming
the transition to be continuous we derived the location of thephase boundary
from the solution ofequations (26)
and(27).
For temperatures 0.95, 0.9, 0.8, 0.6 we have calculatedrc(T>
for M= 2, 4, 8. For T
= 0A we evaluated
rc<T>
for M =4,8.
The values obtained forrc(T)
as a function ofI/M~
lie onnearly straight
lines.(This
is the knownI/M~
law[24])
andby
this method weextrapolated
to M= cxJ
(allowing
for some small errorbars)).
The results are
displayed
infigures
I andfigure
2.Figure
2 shows the location of thephase boundary
and we see thatrc(T
=
0)
-- 1.05- 1-1This is
compared
with the value of4/3
obtained from the static
approximation
which is known to overestimate the value ofrc(T)
for small T.2.0
Legend 1.8
. T=0.g5
x T=O.g
6 a T=O.8
. T=0.6
1_4 ~"~'~
l.2
~ i.o
0.8
O.6
O.4
O.2
0 O.025 O.OSO 0.075 0.100 O.125 0,150 O,175 0.200 0.225 O.250
1/M62
Fig. 1. rc ~3.
1/M~
foT various values of T for the p= 3 Potts SG.
1.2
1-o
0.8
~
O.6
O.4
, '
,
O.2 ,
0
0 0.2 0.4 O.6 0.8 1.0 .2
r
Fig. 2. Phase boundary between the
Spin-glass
and paramagnetic phases for thecase p = 3. The
dotted line is a possible extrapolation to T
= 0.
38 JOURNAL DE PHYSIQUE I N°1
4. Conclusions.
In this paper we extended the
Ising spin-glass
in a transverse field to a p-component Potts model with a transverse field that inducestunneling
among the p-states. This model may haveexperimental
realizations in ferroelectric antiferroelectric mixtures and is alsointeresting
from the theoreticalpoint
view. More work is needed toinvestigate higher
pvalues, especially
p >4, where the transition is known to have some first order features for r = 0
[21].
A solution for p- oc may also be found. These are
projects
for future research.Acknowledgements.
This work has been
supported by
the National Science Foundation under grant#
DMR- 90l6907.Note added in proofs: After the
completion
of this w6rk we became aware of apreprint by
K. Walasek and K. Lukierska-Walasek who consider the 3 state Potts
glass
in a transverse field.They
do not usereplicas,
but rather an uncontrolled effective fieldapproximation
thatneglects
correlations among different components of thecavity
field. Their result ofrc(0)
= 2/3
underestimates the value of the critical field at T
= 0.
References
[I]
Bray A. J. and Moore M. A., J. Phys. G13 (1980) L655.[2] Sommers, J. Ma9n. Mater. 22
(1981)
267.[3] Pirc R., Tadic B. and Blinc R., Z. Phys. B 61
(1985)
69.[4] Ishii H. and Yamamoto T., J. Phys. G. 18
(1985)
6225; J. Phys. C 20(1987)
6053.[5] Usadel K. D., Solid State Gommun. 58
(1986)
629; Nucl. Phys. B(Proc. Suppl.)
5A(1988)
91.[6] Walasek K, and Lukierska-Walasek K., Phys. Rev. 34
(1986)
4962.[7] Usadel I<. D, and Schmidtz B., Solid State Commun. 64
(1987)
975.[8] Dobroslavievic V. and Stratt R., Phys. Rev. B 36
(1987)
8484.[9] Yokota T., Phys. Lent. A 125
(1987)
482.[10] Ray P., ChakrabaTti B. K, and Chakrabarti A., Phys. Rev. B 39
(1989)
11828.[11] Thirumalai D., Li Q. and Kirkpatrick T. R., J. Phys. A 22
(1989)
3339.[12] Buttner G. and Usadel K. D., Phys. Rev. B 41
(1990)
428; ibid 41(1990)
9221.[13] Goldschmidt Y. Y., Phys. Rev. B 41
(1990)
4858.[14] Goldschmidt Y. Y. and Lai P-Y., Phys. Rev. Lett. 64
(1990)
2467.[15] Lai P-Y. and Goldschmidt Y. Y., Europhys. Lent. 14
(1991)
165.[16] Goldschmidt Y. Y, and Lai P-Y., Phys. Rev. B 43
(1991)
11434.[17] Goldschmidt Y. Y. and Lai P-Y., Physica A, to be published.
[18] see e.g. Binder I<. and
Young
A. P., Rev. Mod. Phys. 58(1986)
801.[19] Aharony A, et. al., Phys. Rev. Lent. 60
(1988)
1330.[20] Elderfield D. and Sherrington D., J. Phys. C16
(1983)
L497, L971, L1169.[21] Gross D-J-, I<anter I. and Sompolinsky H., Phys. Rev. Lent. 55,
(1985)
304.[22] Trotter H. F., Proc. Am. Jfath. Sac, lo
(1959)
545.[23] Suzuki M., Prog. Theor. Phys. 56
(1976)
1454.[24] Suzuki M., Quantum Monte Carlo Methods, Springer Series in Solid-State Physics 74, M. Suzuki Ed.
(Springer-Verlag, 1987).
[25] Goldschmidt Y. Y, and Shigemitsu J., Nucl. Phys. B 200
(1982)
149.[26] Sherrington D, and Kirkpatrick S., Phys. Rev. Lent., 35,
(1975)
1792.[27] Parisi G., J. Phys. A
13(1980)
L155, L1101, L1887.[28] Zia R. K. P. and Wallace D. J., J. Phys. A 8