HAL Id: jpa-00246356
https://hal.archives-ouvertes.fr/jpa-00246356
Submitted on 1 Jan 1991
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.
On the behavior of Ising spin glasses in a uniform magnetic field
David Huse, Daniel Fisher
To cite this version:
David Huse, Daniel Fisher. On the behavior of Ising spin glasses in a uniform magnetic field. Journal
de Physique I, EDP Sciences, 1991, 1 (5), pp.621-625. �10.1051/jp1:1991157�. �jpa-00246356�
L
Phys.
I 1(1991)
621-625 MAi 1991, PAGE 621Classification
P%ysksAbsmwts
75.50-75.10
On the behavior of Ising spin glasses in
auniform magnetic
field
David A~ Huse
(I)
and Daniel S. Fhher (2,*(1)
AT&T BellLaboratories, Murray Hill,
NJ07974,
U.s.A.(2) Physics Department,
PrincetonUniversity,
Princeton, NJ08544,
U.s.A.(Received
30 November 199(accepted
14 March1991)
Abstract. We show how the recent computer simulation data
reported by
Caracciolo ei al for the three-dimensional + J[sing spin glass
in a uniformmagnetic
field are, in contrast to theirclaims, quite
consistent with thesimple scaling
ordroplet theory developed by
Mcmillan, Flsher and Huse, andBray
and Moore, and do notprovide
evidence for a de AJmeida-Thouless transition.Caracciolo et al
[1, 2]
haverecently reported
results ofcomputer
simulations of the + J[sing spin glass
model in a uniformmagnetic
field on asimple
cubic lattice.They
claim that their re-sults are in
good agreement
with the exotic behavior of theinfinite-range Sherrigton-Kirkpatrick (SK)
model[3,
4] and areincompatible
with thesimpler scaling
ordroplet theory [5-8]
of finite-dimensional
short-range spin glasses.
Here wedhpute
thatclaim, showing
how the results re-ported
in references[1, 2]
arequite
conshtent with thescaling
ordroplet theory presented
in references[5-8].
Infact,
we argue that in references[1, 2]
thetemperatures
studied were toohigh
and the sizes studied were too small to address the issues on which the two theoretical
approaches differ,
I.e. the nature and extent of the lowtemperature
orderedphase.
We alsobriefly
discussother recent numerical studies and what
regimes
need to be accessed toreally
address these hsues.The + J
Ising spin glass
model withonly nearest-neighbor
interactions on asimple
cubic lattice has beenextensively
simulated[9, 10]
for zeromagnetic
field. These studies indicate aphase
transition, although
the evidence b notindisputable. However,
T there is indeed aphase
transition itappears quite
certain that it occurs at atemperature
less than 1.3(in
units where J= ~l and
kB
=I). Ogielski
[9] estimatesTc(h
=0)
= 1.175 +0.025; however,
we feel his error bars areoverly optimistic, especially
on the lowertemperature
side. The simulations of references 11, 2] are allperformed
atapplied
field h= 0.2 and thus
h/Tc(0) /
0.16. At such anapplied
field
(scaled by 7~(0))
the reduction of7~(h)
fromTc(0)
in theSherrigton-Kirkpatrick
model is(Tc(0) Tc(h)) /Tc(0)
> 0.25. Thus T there were a de AJmeida-Thoulessphase
transition line[3,4]
in thesimple
cubic +J model one wouldexpect 7~(h
=
0.2) ~
0.9 if onesimply
scales the SKphase diagram [11].
This indicates that even if thesystem
did behave as in mean fieldtheory,
(* l+esent address:
Physics Department,
HarvardUniversity, Cambridge,
MA02138,
U-S-A-622 JOURNALDEPHYSIQUEI N°5
the lowest
temperature
simulated in references[1, 2],
T ci0.833,
would be very near to or above We de AJmeida-Thouless line. Thus it should come as nosurprise
that their data are all consistent withbeing
in We disorderedphase.
We have
argued
elsewhere[6,
8] that the orderedphase
of anIsing spin glass
withfinite-range
interactions b unstable in
equilibrium
to a smallmagnetic
field for all T <Tc(0)
and all dimen- sionalities d. Theresulting
disorderedparamagnetic phase
has a finite correlationlength (
thatdiverges
as a power of the field h for h - 0 and T <Tc(0).
The relaxation time for T <Tc(0)
b
expected [6, 8]
todiverge
asexp (A(~ /T)
for h - 0 and thus(
- oo, where#
is apositive exponent
and A is theamplitude
for the free energy barriers[6, 8].
Thus we conclude[6, 8]
that there is no de Nmeida-Thouless line in theequilibrium phase diagram. (Mcmillan
[5j andBray
and Moore [7j have made similar
arguments
for d =3,
butthey
did notattempt
to rule out thepossibility
of a de AJmeida-Thouless line forhigh
d.)
For small field and T$ Tc (0)
the correlationlength
and correlation time arequite large [6, 8],
so it would be very difficult to determine whether or notthey
are finite in a simulationstudy.
Much of datareported
in references[1, 2]
are from this difficult low fieldregime.
We will now examine the data and
interpretation given
in reference[I]
in the same sequenceas
the?
arepresented
there:a)
MAGNEnzATION vs. TEMPERATURE.Figure
la of reference[I]
shows that themagneti-
zation at h
= 0.2 b constant within error bars over the
temperature
range 1.0$
T$
1.5. Asargued above,
even if there h a de AJmeida-Thouless line thesetemperatures
are above it. Thus themagnetization
appears to benearly temperature independent
in thispart
of theparamagnetic phase.
Thisappears
to be indisagreement
with We behavior of the SKmodel,
where the magne- tization becomestemperature independent only
belowTc(h).
Of course, as shown infigure
16 ofreference
[I],
the averageexchange
energy continues todrop
after themagnetization
saturates,as it must since the
specilic
heat ispositive.
Witllin thescaling
ordroplet theory [6, 8],
the actual numerical values of themagnetization
and energy are to alarge
extent determinedby
the details of smalllength
scale fluctuations(small droplets);
these are not universal so it is not clear howone could make a detailed
prediction
of thetemperature-dependence
of themagnetization
and We energy.b)
CORRELATION BETWEEN CONFIGURATIONALAND ENERGY OVERLAPS. In referencesIi, 2j they
simulatedmulIiple replicas
of the samesample (same Ji;'s)
and measured theconfigurational overlap
ofreplicas
a andfl
~
q +
j £ alai, (I)
I=i
where each
replica
contains N[sing spins al.
The"energy overlap"
qe +
[ L a;alar al (21
lo)
was also
measured,
where the sum runs over all 3Nnearest-neighbor pairs
ofspins. They
noted that in order for both q and qe to differsubstantially
fromunity,
thedroplets
that constitute thedifference between the instantaneous
spin configurations
of the tworeplicas
must have a non-zero average surface-to-volume ratio.Indeed,
we have shown[6,
8] that the difference of q fromunity
is
predominantly
due to the smalldroplets;
similararguments apply
to qe. These smalldroplets,
N°5 ISING SPIN GLASSES IN A MAGNETIC FIELD 623
which contain
only
a fewspins,
do have alarge
surface-to-volume ratio and it b theirpresence
thatproduces
most of the deviation of q and qe fromunity.
Note that thesedroplets
do not allspend approximately
one-half of the timeflipped,
as isimplied
in the discussion offigure
13 of reference[2];
rather the averagepolarizations
of thedroplets
willgenerally
becontinuously
distributed[8].
The
strong
correlation observed between q and qe is inevitable: when q is near either I or-I,
there are fewdroplets present
so qe must be nearI,
while when the twoconfigurations
arequite
different so q b near zero, qe must be at a minimum. For a
paramagnetic
state one may constrain q in thethermodynamic
limitby
theconjugate field,
e, and measure qe. Theresulting
functionqe(q) should,
because of the aboveconstraints,
be wellapproximated by
an even function of q.c)
NONTRIVIALITY oFP(q ).
For ageneral
finite-sizesystem P(q),
theprobability
distribution of q, isexpected
to have nontrivial structure aslong
as the linearsample
size L h less than or oforder the correlation
length (.
It honly
for L »f
thatP(q)
becomes verysharply peaked
in aparamagnetic phase.
Thus theP(q J's
for L < 14 and T < I shown infigures
2 and 3 of referenceill
are notsurprising,
since within anytheory
the correlationlength
isexpected
to belarge
for T <Tc(0)
and low field. lb demonstrate thatP(q)
is indeed nontrivial(I.e.,
not adelta-function)
for L - oo one must show that
(q2) (q)2
e xso/L~
does not vanish for L - oo.However,
the data shown infigure
5 of referenceill
show that xso/L~ steadily
decreases withincreasing L,
andare thus
quite
consistent withP(q collapsing
to asingle
delta-functionpeak
in thethermodynamic limit,
as weexpect
occurs.d)
ABSENCE oF SELF-AVERAGING oFP(q).
One may consider theprobability
distributionPj (q)
of q for agiven
finite-sizesample
J andcompare
it toP(q),
which is the average ofPj(q)
over all
possible samples
of that size. As a measure of thetypical
difference betweenPj(q)
andP(q)
thequantity
S +
/ dq ~ L (PJ
(qJ (qJJ~)
~~~(3J
was
proposed
in reference[ii,
whereNJ
is the number ofsamples
summed over.However,
thismeasure
ofself-averaging
isseverely
flawed because of theexpectation (even
in the SKmodel)
thatP(q
contains a delta-functioncomponent
for L - oo. Forexample,
if thesystem
isparamagnetic
then for L »
(
oneexpects Pj(q )for
agiven sample
J to bepeaked
at qj, with qjvarying by
of orderL-~/2
fromsample
tosample
when h#
0. The width of thbpeak
inPj(q)
should also be ofordqr L~~/2.
Thus thesample-to-sample
fluctuations inPj (q)
at fixed q should be of orderP(q)
itself. Thin means S will not vanish withdiverging
L for L >(
even ifP(q) collapses
to a delta-function as isexpected
in a(self-averaging) paramagnetic phase.
Theproblem
is that S is a useful measure ofself-averaging only
whenP(q)
contains no delta-functioncomponent
for L - oo. Asimpler
andprobably
more reliable measure ofself-averaging
isgiven by
the ratio ofthe
sample-to-sample
variation of xsG to its mean.d)
ULTRAMETRICITY. Here wejust
note that aspace
of statesconsbting
of one state istrivhlly
ultrametric. Aparamagnetic phase clearly
mustsatisfy
the test forultrametricity
done in reference[ii.
Thus the data shown infigure
4 arequite
conshtent with what should occur in aparamagnetic phase.
624 JOURNAL DE PHYSIQUE I N°5
EmSTENCE oF A PHASE TRANsiTioN. In a
paramagnetic phase
with alarge
correlationlength, (,
if one measures xsG for various linearsample
si2esL,
one should find that for L <2f
the finite-she effects are
strong
and xso increases withincreasing
L. For L »(,
on the(her
hand,
xso should saturate andapproach
the finite value it attains in the L- oo limit. In
figure
5 of reference
[ii, xso(L)
at h= 0.2 and T ci 0.833 and T
= I is shown to increase with L as L varies from 6 to 14. Given the she of the error bars and the small range of L it is no
surprise
that the data arequite
consistent with atemperature-dependent
powerlaw,
xso+w
L~(~).
The factthat xso h
increasing fairly rapidly
with L means that the correlationlength
isgreater
than or oforder,
say, 7 latticeunits,
which is to beexpected, considering
the low field andtemperatures.
At the zero-field criticaltemperature
xso +wL2~Q,
with[9, 10]
q ci -0.3.Discussion.
Unfortunately,
it is not clearhow,
for three-dimensionalspin-glass models,
one cannumerically
access the low
temperature, large L, long
timeregime
in the orderedphase
where there are realqualitative
differences between the behavior of the SK model[3],
on the onehand,
and thescaling
or
droplet theory [5-8],
on the other.Perhaps
a carefulstudy
ofnonequilibrium
domaingrowth [12]
for T <7~ might
shed somelight
on thbquestion.
Infact,
no one hasyet produced
any solidevidence for true
long-range spin-glass
order atpositive temperature
in thesemodels,
even in zeromagnetic
field. Ml thelow-temperature
behavior observed bquite
consistent with thesystem being
at or near a criticalpoint
withspatial
correlations thatdecay
as apower
of the dbtance.The ordered
phase
is more accessiblenumerically
for four-dimensionalmodels,
as demon- stratedby Reger,
Bhatt andYoung [13]. They
measuredP(q)
for theIsing spin glass
with Gaussian- distributedJ;;'s (with
varianceunity)
on four-dimensionalhypercubic
lattices withperiodic
bound- ary conditions for sizes up to L= 6 and
temperatures
down to T = 1.2. The zero-field transitionTc(0)
appears to occur near T= 1.75. At T
= 1.2 one can see
(Fig.
I of Ref.[13])
that the widthof the
peak
inP(q)
normalizedby
itsposition
decreasesconsiderably
ongoing
from L= 4 to
L =
6,
as isexpected
in the orderedphase. However,
the si2es arequite
small and one cannot tell from their datawhether,
for L - oo,P(q)
isapproaching simply
a delta-function as weexpect
oris instead
approaching
a distribution with a continuouspart,
as does the SK model. It isimportant
to
emphasize
that ingeneral P(q)
is not agood
indicator of thepresence
or absence of many states[14].
Another
approach
to thequestion
of the existence of the de AJmeida-Thouless line is to ex-amine contours of constant xsG in the
(h, T) plane [15].
In the SK model such contours go tolarge
h for T-
0,
while in thedroplet theory they
go to h = 0 for T - 0 for asystem
withcontinuously-distributed J,,'s,
which iS a realqualitative
difference. In reference[lsj
such astudy
was
attempted,
but since iI wasmostly
based onhigh-T
seriesexpansions
which failt6
converge well aboveTc(0)
and was restricted to the + J model with its extensiveground
stateentropy,
it doesnot
help
to resolve thequestion
for d > 3.Perhaps
athorough
simulationstudy
ofxsG(h, T)
fora model with
continuously-d
istributedJ;;'s
would be useful inaddressing
thisquestion.
Yet an-other
promising approach
is advancedalgorithms
such as thatproposed by Swendson
andWang [16].
Thepotential
of thesealgorithms
in anapplied
field has notyet
beenexplored.
We conclude with
general
remarks on the difficulties to be facedby
numerical simulations. In order toprobe
the behavior ofspin glasses
in amagnetic
field(or
inzero-field)
atlong enough
dis- tance scales to have ahope
fordistinguishing
between the theoreticalpredictions
for the orderedphase,
thesample
sizes must be considerablelarger
than the critical zero-field correlationlength
[8]
(- (T)
whichdiverges
as the zero-field criticalpoint
isapproached
from below. Otherwise oneN°5 [SING SPIN GLASSES IN A MAGNETIC FIELD 625
just
sees the critical behavior. Thus data onrelatively
smallsamples
in small fields nearTc(0)
are not
adequate.
With a fixed amount of availablecomputer time,
one should instead choose thetemperature
to maximize the ratio of L to the correlationlength (- (T)
for which one can still achieveequilibrium.
Thisoptimum temperature
isprobably roughly I/2
ofTc(0).
One must thentry
todhtinguish between,
forexample,
a xso whichdiverges
at zero field as h~~H and onewhich
diverges
at a critical field as[h hc(T)]~~H.
In either case, the relaxation time isexpected
to
diverge exponentially rapidly
as xsodiverges [8, 12],
so thatequilibration
ofsufficiently large systems
isextremely
difficult.Acknowledgements.
Daniel S. Fisher is
partially supported by
theNSfI
DMR 8719523 and the A~P Sloan Foundation.References
[1] CARAccioLo
S.,
PARisiG.,
PATERNELLO S. and SouRLAsN., Eumphys.
Loin. ii(1990)
783.[2] CARACCIOLC S., PARist G., PATERNELLO S. and SouRLAs
N.,
JPhys.
Frvnce Sl(1990)
1877.[3] MtzARD M., PARisi G. and VtRAsoRo
M-A-, Spin
GlassTheory
andBeyond (World
Scientific, Sin ga- pore,1987).
[4] For a reveiw of
spin glasses:
BINDER K. and YOUNG A-P, Rev MhLPhys.
58(1986)
801.[5j MCMILLAN
WL.,
J P/ij>s. c17(1984)
3179.[fl
FISHER D-S- and HusE D-A-,P%ys. Rev Lett. 56(1986)
1601.[7jBRAY AJ. and MOORE M-A-, in
Glassy Dynamics
andOptimization,
J.L. van Hemmen and I.Morgen-
stern Eds.
(Springer, Berlin, 1987).
[8] FtSHER D.s. and HusE D.A.,
Phys.
Ret; 838(1988)
386.[9] OGIELSXi
A-T, P/iys.
Rev 832(1985)
7384.[10] BHATr R-N- and YOUNG A-P,
Phys.
Rev 837(1988)
5606.[ll]
Note that for d < 6, the de Almeida-Thouless line, if itexists,
isexpected
to scale for small h as[Tc(0) Tc(h)]
~h2/(P+~),
see FISHER D-S- and SOMPOLINSKYH., Phys.
Rev Leit. s4(1985)
1063. Estimates of
(fl
+7)
for three dimensions range from 3 to5,
see, e-g-, Refs. [9, 10] and GESCHWINDs.,
HusE D.A. and DEVLING-E-,Phys.
Rev B41(1990) 2650).
the mean-field estimate ofTc(h) /Tc(0)
is thuslikely
to be an overestimate.[12] FISHER D.s. and HusE D-A-,
Phys.
Rev 838(1988) 373;
HusE D-A-Phys.
Rev 843(Apr.
1,1991)
in press.[13] REGER
J.D.,
BHAaT R.N. and YOUNGA-P, P/iys.
Rev Lett. 64(1990)
1859.[14] HusE D-A- and FISHER D-s-, L