Professional Documents
Culture Documents
Prog. Theor. Phys.-1981-Kasai-1561-73
Prog. Theor. Phys.-1981-Kasai-1561-73
§l. Introduction
Since Edwards and Anderson I) introduced the replica method into the in-
vestigation of the random magnetic mixture in the quenched state. a lot of works
with the use of the replica method 2HO ) have been done.
For the dilute system,5),7),I1),18) the replica method can be applied successfully.
For the system with competing interactions, on the contrary, some draw-backs 29 )
were pointed out, for example, by Thouless, Anderson and Palmer,3) in connection
with the negative entropy at the spin glass phase for the Sherrington-Kirkpatrick
modeL 2) Aharony8) estimated the effective interaction of the replicated system
(the ± ] model) at the ground state. He concluded that the order parameter of
the Mattis type 22 ) (the Edwards-Anderson order parameter) exists at the ground
state even for the quenched limit (n->O) and that the obtained state is frustra-
tionless. On the other hand, De Dominicis and Stephen lO ) pointed out that the
replicated system includes a frustration effect at the quenched limit if the limiting
procedure is taken as n->O and then T->O. Using the high temperature expan-
sion theory, Rapapore 3) indicated that there is no phase transition to the spin
24
glass for the ±] modeL By the Monte Carlo calculation, Bray, Moore and Reed )
showed that the Edwards-Anderson order parameter does not exist for the
frustrated system. With the recursive transfer matrix approach, Binder,
Kinzel, Sarbach and Stauffer l5 ) also conjectured that the spin glass phase does not
appear for the ±] modeL
Sherrington and Kirkpatrick's result 2) and Aharony's one 8 ) done by the
replica method seem to conflict with others 15 ),23),24) done by computational meth-
ods. Above discrepancy is harmful 13 ),15),25) for the deterministic statement, since
there exists no reason for the replica method to have fundamental errors. One of
1562 Y. Kasai, A. Okiji and 1. Syozi
the main purposes of this paper is to clarify a basic structure of the spin glass
state with the use of the replica method within the framework of the ±] model.
In a previous paper,20) the present authors proposed an analytic continuation of
Suzuki's real replica method 6) which is applied on the Syozi model. 19) The
asymptotic expression of the replicated system at the ground state was shown to
be equivalent to the 2n~1-state Potts model like Aharony's result. B) The expres-
sion of the effective interaction, however, is different from that of Aharony.19) It
can be considered that the essential behavior of the cooperative part in the
replicated system is described by this asymptotic Potts model. Using the asy-
mptotic Potts model, we can estimate the upper bound of the transition tempera-
(jij(8)=] or - j) (2'1)
where <ij> denotes a neighboring spin pair (or a bond), Si the i-th Ising spin
variable and (]= l/kT. The grand partition function En for the n-replica
method is given as
Ising Replicated System oj ±J Model 1563
(2·3)
where
=
l
exp[k~oKknrk~ l)]= xexp[Lnr1 ( l)]+exp[ - Lnr1 ( l)],
_ (2·4)
=
o+t)n-l(I_t)I= "i.:,nrk(l)t k (2·5)
k~O
and
=
"i.:, nrl(k)nrk (m)=2 n
k~O
o ,m,
1 (2·6)
Here, L = (3J, B denotes the number of the total bonds, x the fugacity, b( e) the
number of the ferromagnetic bond ( + j), mo( == n - "i.:,~~l mr), ml, "', mr, "', mM
mean the number of replicas contributed from the spin configurations Go, Gl, ''',
Gr, "', G M (where a spin configuration and the spin configuration obtained by
reversing all the spins of it are treated as the same) and <L denotes the number
of replicas where the pairs of spins on the i- and j-sites are anti parallel. It is to
be noted that the infinite series of the above expressions do not obey the absolute
convergence generally. It is necessary to introduce a concept of the quasi-
convergence like the Holder sum in this stage. The expression (2·3) for En is
specified 20 ) by Go in which spin pairs are all parallel. Thus the grand partition
function En can be expressed concretely by the partition function for the interact-
ing many-spin system with the homogeneous interactions {Kk}. The expression
(2·3) also shows the factorization of the replicated system into the spin inde-
pendent part, exp[B Ko] and the spin dependent part (the cooperative part),
ZM (Kl, ... ). In the following, we treat mainly the simple lattices like the linear
chain, square, cubic, etc., which could have Neel's antiferromagnetic ordering.
Jo
..
2
J,
3
J2
----- . .J.
J.- 2
N-2 N-1
B -,
Fig. 1. The linear chain and the exchange interaction h(j or - n, B= N-1.
(3' 1)
where N denotes the total number of sites. From the definition En in (2'3), we
can directly calculate it as
B
En=~BCt(2N coshBL)nxl=( 2N coshBL)n(1+x)B. (3'2)
In the following, we can also obtain the same result with the use of the polynomial
expression. To calculate the polynomial expression, we have to write down a
spin configuration Gr explicitly. The number r can be expressed by the diadic
system:
(3'3)
°
where rj = or 1. The j-th spin pair in Gr is defined as parallel if rj = and °
antiparallel if rj = 1. Thus the r-th configuration Gr is determined uniquely by
the number r itself:
r=O,l,"',M,
(3'4)
The number <L (== U in the expression (2'3) can be expressed simply as
where rj denotes the j-th diadic figure for the running index r. The grand
partition function En of the replicated system can be expressed as
= =
En=2n~ ... ~
ml=O mM =0
= =
=2n~ ... ~ (3'6)
ml=O mM=o
where the summation ~{Lj=±L} is carried out over the possible combination of L j
=+L or -L (j=O,"',B-1) and o(L) gives 1 for L=L and for L=-L. °
With the use of (3'6), En can be written as
Ising Replicated System of ±J Model 1565
B-1
=Zn ~ ITexp[Lm]x 8(Lj)
{Lj=±L}j=O
= = n! M B-1
X ~ ... ~
nM mr! exp[~ mT{ ~ (-ZLrj)}]
Since for the linear chain all the possible configurations r = 0, ... , M can be
°
obtained as the diadic expression by taking rj = or 1 independently for each j,
En can be rewritten as
B-1 B-1
En=Zn ~ IT exp[Lm]x 8(Lj){ IT ~ exp[-ZLrj]}n
{Lj=±L} j=O j=O Tj=O.l
B-1 B-1
=Zn ~ IT exp[Ln]x 8(Lj){ IT (I +exp[ -ZL])}n
{Lj=±L} j=O j=O
B-1 B-1
= Zn ~ IT X 8 (Lj){ IT (exp[L]+exp[ - L])}n
{Lj=±L} j=O j=O
B-1
=Zn ~ ITx 8(Lj)(ZcoshL)Bn
{Lj=±L} j=O
Accordingly,
En=Z(B+l)n(coshL)Bn(I+x)B, (3·9)
which agrees exactly with (3· Z).
In this section, we investigate the ground state of the n·replicated system for
x=l(p=l/Z, see Appendix A). For the integer n replicated system,19) an al-
lowed spin configuration at the ground state is composed of the same spin
configurations for all the n replicas. The bond distribution can be determined
frustrationlessly by the allowed spin configuration, i.e., a J-bond and a ( - f)-bond
are assigned by a parallel spin pair and an antiparallel spin pair, respectively. It
is shown that the spin configuration at the ground state can be described in a
similar way even for the noninteger n replicated system in the following. At first
1566 Y. Kasai, A. Okiji and I. Syozi
If one denotes the number of antiparallel spin pairs in Gr as Br, Eq. (4'3) can be
rewritten as
(4'4)
For L-->oo, the asymptotic expression of ~r is given as
and K2k+l'S are ineffective as a whole. If we replace K2k by Z-<n-l)nL for the
Ising Replicated System of ±J Model 1567
replicated system, after the analogy of the integer n replicated system the
effective Hamiltonian Hp is expressed as
Hp=-n]L:{o(O,
<0>
I )+o(n,
<0>
I)},
<0>
(4'8)
which means the 2n -I-state Potts model 20 ) with the interaction nf- It is con-
sidered that the thermodynamic property of the replicated system can essentially
be described by this asymptotic Potts model. Because, the asymptotic expres-
sion of the effective interaction Kk is obtained by neglecting the term
exp[ - Llnrl( m)l] compared to exp[Llnrl( m)l] in the formula (2'4)_ The neglected
On the other hand, for the Ko sub-system the internal energy Vo and the specific
heat Co can be calculated using (2'4) as
Vo/B = - B- 1 0 In exp[BKo]/o/3
=
= - ]2- n L: nCm( n-2m)tanh{L( n-2m)},
m=O
=
Co/B = k/3 2 P2- n L: nCm( n - 2m)2 /cosh 2 {L( n - 2m)} -= Co( n)/B , (5' 2)
m=O
1
eM
0.5 n=2
2 3
c'
n=0.8
0.5
n=2 -1
2 3 C'
-1.5
2 3 C'
§6. Discussion
In the preceding sections, it is shown that the asymptotic Potts model (0 < n
~ 2) represents the essential property of the spin dependent part ZM of the
replicated system and that the correspondence of the ZM sub-system to the Potts
model is regardless of the lattice structure.
Here, we estimate the transition temperature T M of the Mattis phase from the
asymptotic Potts model for the square lattice as an example. The transition
paine!) Ke of the 2n - I -state Potts model with the interaction K defined by (4' 7)
is given as
exp[2n-IKe]=1+~ . (6'1)
The transition point L.M -I( = kTM/J) of the Mattis phase is obtained as
n n
2n IKe (6'2)
This value of L.M - I is compared with the exact value of LM - I obtained for the
Table I.
n LM -I(exact) LM- 1
0 0 0
0.7593 1.442
2 1.3084 2.269
3 1.5186 2.730
00 2.2691 2.885
1570 Y Kasai, A. Okiji and 1. Syozi
c'
3
2
, Fig. 5. The n·dependence of the Mattis
I .... "' ...
: ,,/'
transition point L,,'( = kTM/J) and the upper
/"C(
/ : Mattis Phase bound L,,' estimated from the asymptotic
o/
o 4 n 5 Potts model.
(6·4)
where
(6·5)
The relation (6·4) means that the upper bound is obtained by the asymptotic
procedure (neglecting e- nL term). Further the consideration about the upper
bound of the transition temperature is expected to be valid for all other two-
dimensionallattices 32 ).33) and even for the three-dimensional lattice. There exists
formally the transition temperature for the case n = l( p = 1/2) which is the Syozi
model. l9 ) It is noted, however, that from the discussion given in §5, the coopera-
tive phenomena described by the ZM sub-system vanishes perfectly at n = 1, i.e.,
the amplitude of the cooperative part vanishes at n = 1. Further for n < 1, the
exponent of the specific heat for the asymptotic Potts model is predicted 34 ) as the
cusp type, that is, as n changes from 1 to 0, the exponent changes from - 2/3 to
-1.5645···. In this case the cusp must be downwards because the specific heat of
the ZM sub-system is negative.
Since the transition point Kc in (6·2) is expected to be finite for the 1/2-state
Potts model, the upper bound of the transition temperature, EM -\ always tends to
zero for n -> O. This is the important result obtained in this paper, that is, the
ordered phase characterized by the Edwards-Anderson order parameter, the
Ising Replicated System of ±J Model 1571
(A -5)
1572 Y Kasai, A. Okiji and I Syozi
+_ ~(o In
Ck - 1"20 oKI
Z)( oK
oK
k
1-
+
)_(
- -
l)k -
Ck. (A '6)
Thus
rn~Ck+~ nrm(k)+exp[Lnn(m)+]
k~O m~oexp[Lnrl( m)+]+exp[ - Lnn( m)+]
(A,?)
2-n~Ck+~ nrm(k)+exp[-Lnrl(m)+]
Accordingly,
p= 1/2. (A'9)
Appendix B
- - The Mattis Transition Temperature for n = 1 - -
However, it is expected that the expression (B, 6) contains the apparent interac-
tion K' 0 of the spin dependent part in the spin space with zero degree of freedom
(the I-state Potts model). In consideration of the condition (B'3) and the forms
(B·4), (B'5) and (B'6), we can propose the following form for K'o:
(B·n
where m is an undeterminate constant. With the use of the asymptotic condition
(4· 7) for n = 1, K' 0 -> L, the constant m can be determined as m = 0:
(B'8)
References
1) S. F. Edwards and P. W. Anderson, J. of Phys. F5 (1975), 965.
2) D. Sherrington and S. Kirkpatrick, Phys. Rev. Letters 35 (1975), 1792.
3) D. J. Thouless, P. W. Anderson and R. G. Palmer, Phil. Mag. 35 (1975), 593.
4) J. H. Chen and T. C. Lubensky, Phys. Rev. BI6 (1977), 2106.
5) M. J. Stephen and G. S. Grest, Phys. Rev. Letters 38 (1977), 567.
6) M. Suzuki, Prog. Theor. Phys. 58 (1977), ll5l.
7) E. Domany, J. of Phys. Cll (1978), L337.
8) A. Aharony, J. of Phys. Cll (1978), L457.
9) M. R. Giri and M. J. Stephen, J. of Phys. Cll (1978), L54l.
10) C. De Dominicis and M. J. Stephen, J. of Phys. Cll (1978), L969.
11) H. Nishimori, J. of Phys. CI2 (1979), L64, L905.
12) E. Domany, ]. of Phys. CI2 (1979), Ll19.
13) A. Aharony and P. Pfeuty, J. of Phys. CI2 (1979), Ll25.
14) Y. Kasai and T. Takano, Prog. Theor. Phys. 62 (1979), 566.
15) K. Binder, W. Kinzel, S. Sarbach and D. Stauffer, Farady Discussion 69 (1980), 26l.
16) S. Sarbach, J. of Phys. CI3 (1980), 5033, 5059.
17) S. Nambu and S. Naya, Prog. Theor. Phys. 63 (1980), 1098.
S. Nambu, Prog. Theor. Phys. 63 (1980), 1474.
18) Y. Kasai, T. Takano and I. Syozi, Prog. Theor. Phys. 63 (1980), 1917.
19) Y. Kasai, A. Okiji and I. Syozi, Prog. Theor. Phys. 65 (1981), 140.
20) Y. Kasai, A. Okiji and 1. Syozi, Prog. Theor. Phys. 65 (1981), 1439.
21) G. Toulouse, Comm. Phys. 2 (1977), ll5.
22) D. C. Mattis, Phys. Letters 56A (1976), 42l.
23) D. C. Rapaport, J. of Phys. CIO (1977), L543.
24) A. J. Bray, M. A. Moore and P. Reed, ]. of Phys. Cll (1978), 1187.
25) c. D. Dominicis, "Systems with Quenched Random Impurities" in Lecture Note in Physics,
ed. C. P. Enz (Springer Verlag, Berlin 1979), vol. 104, p. 252.
26) F. Matsubara and M. Sakata, Prog. Theor. Phys. 55 (1976), 672.
27) S. Katsura, Prog. Theor. Phys. 55 (1976), 1049.
28) Y. Ueno and T. Oguchi, J. Phys. Soc. Japan 40 (1976), 1513.
29) S. Katsura, S. Inawashiro and S. Fujiki, Physica 99A (1979), 193.
30) K. Honda and H. Nakano, Prog. Theor. Phys. 65 (981), 83, 95.
31) R. B. Potts, Proc. Cambridge Phil. Soc. 48 (1952), 106.
32) D. Kim and R. I. Joseph, J. of Phys. C7 (1974), Ll67.
33) F. Y. Wu, J. of Phys. CI2 (1979), L645.
34) M. P. M. den Nijs, J. of Phys. A12 (]979), 1857.