PHYSICAL REVIEW B 71, 134424 s2005d
Hysteresis and avalanches in the T = 0 random-field Ising model with two-spin-flip dynamics
Eduard Vives*
Departamento d’Estructura i Constituents de la Matèria, Universitat de Barcelona Diagonal 647, Facultat de Física,
08028 Barcelona, Catalonia
Martin Luc Rosinberg and Gilles Tarjus
Laboratoire de Physique Théorique des Liquides, Université Pierre et Marie Curie, 4 Place Jussieu, 75252 Paris, France
sReceived 12 November 2004; published 29 April 2005d
We study the nonequilibrium behavior of the three-dimensional Gaussian random-field Ising model at
T = 0 in the presence of a uniform external field using a two-spin-flip dynamics. The deterministic, historydependent evolution of the system is compared with the one obtained with the standard one-spin-flip dynamics
used in previous studies of the model. The change in the dynamics yields a significant suppression of coercivity, but the distribution of avalanches sin number and sized stays remarkably similar, except for the largest
ones that are responsible for the jump in the saturation magnetization curve at low disorder in the thermodynamic limit. By performing a finite-size scaling study, we find strong evidence that the change in the dynamics
does not modify the universality class of the disorder-induced phase transition.
DOI: 10.1103/PhysRevB.71.134424
PACS numberssd: 75.60.Ej, 05.70.Jk, 75.40.Mg, 75.50.Lk
I. INTRODUCTION
The nonequilibrium random field Ising model sRFIMd
was introduced by Sethna et al.1 as a model for the
Barkhausen effect in ferromagnets and more generally as a
prototype for many experimental systems that show hysteretic and jerky behavior when driven by an external force.
Because of the presence of disorder, these systems have a
“complex free energy landscape” with a multitude of local
minima sor metastable statesd separated by sizeable barriers,
which makes thermally activated processes essentially irrelevant at low enough temperature sthe lifetime of metastable
states may then be considered as infinited. As a consequence,
these systems remain far from equilibrium on the experimental time scales seven when the driving rate goes to zerod and
their response to the external force is made of a series of
jumps savalanchesd between neighboring metastable states.
This type of behavior is very well modeled by the ferromagnetic RFIM with a zero-temperature single-spin-flip dynamics in which a spin flips only if this lowers its energy. The
local character of the energy minimization is then at the origin of irreversibility. With this dynamics, the RFIM satisfies
the property of return-point memory sor “wiping out” effectd
which is a feature observed in several experimental systems
with good approximation. Moreover, in dimension d ù 3, the
model is known to exhibit an out-of-equilibrium phase transition between a strong-disorder regime where the magnetization hysteresis loop is smooth on the macroscopic scale
and a weak-disorder one where it has a discontinuous jump.
Such a transition has been observed in thin Co/ CoO films2
and Cu-Al-Mn alloys,3 and it has been recently suggested4
that it may also be associated to the change in the adsorption
behavior of 4He in dilute silica aerogels.5 The two regimes,
strong and weak disorder, are separated by a critical point
characterized by universal exponents and scaling laws which
have been extensively studied by analytical and numerical
methods.6,7 In particular, much effort has been recently devoted to analyze the number of avalanches, their size, and
1098-0121/2005/71s13d/134424s8d/$23.00
their geometrical properties above, below, and at
criticality.8,9 These results provide a comprehensive, though
rather complex scenario for the phase diagram of the nonequilibrium RFIM with a metastable dynamics in the thermodynamic limit. A recent discussion of the relevance of the
model to the description of the Barkhausen effect in real
magnets can be found in Refs. 10 and 11.
An issue that so far has not been studied is the robustness
of this theoretical description with respect to a change in the
dynamics sin the literature on the RFIM, it is implicitly taken
for granted that this should not matterd. The single-spin-flip
dynamics, however, is not the unique sand may be not the
bestd way of simulating hysteretic dynamical processes in
actual systems. It is clear for instance that the hysteresis loop
will shrink if the dynamics allows for a better equilibration
of the system by employing multiple-spin flips. Then, what
will be the avalanche properties? Will there still be a phase
transition? If so, will the critical behavior be the same as
with the single-spin-flip dynamics? There is in fact the intriguing possibility, supported by numerical simulations and
analytical arguments,6,9,12,13 that the nonequilibrium and
equilibrium transitions of the T = 0 RFIM belong to the same
universality class, even if criticality occurs in zero external
field at equilibrium and at a nonzero coercive field in the
irreversible evolution.14 Since the ground state is stable with
respect to the flip of an arbitrary sfinited number of spins, this
may indicate that the disorder-induced transition has a universal character at criticality which does not depend on the
specific choice of the dynamics.9
In order to shed some light on this issue and check the
robustness of the transition, we study here the nonequilibrium T = 0 RFIM with a two-spin-flip dynamics. We compare
the results with those obtained with the standard one-spinflip dynamics, in particular those concerning the number and
size of the avalanches. We first show in Sec. II that one can
indeed define a two-spin-flip algorithm that yields a deterministic evolution of the system with the external field. In
particular, the dynamics satisfies the “abelian” property
134424-1
©2005 The American Physical Society
PHYSICAL REVIEW B 71, 134424 s2005d
VIVES, ROSINBERG, AND TARJUS
which guarantees that the same two-spin-flip stable configuration is attained, whatever the order in which the unstable
spins are relaxed during an avalanche. It also has the property of return-point memory. In Sec. III we present the results
of our numerical simulations on three-dimensional s3Dd lattices and compare them to the behavior of the same system
with a one-spin-flip dynamics. The hysteresis loops are significantly reduced when allowing two-spin flips, but the main
features, including the presence of a disorder-induced transition with an associated critical point, are not significantly
altered. We then perform in Sec. IV a finite-size scaling
analysis, which allows for a determination of the critical
properties. We find that, within statistical uncertainty, the exponents and scaling functions are identical to those obtained
with the standard one-spin-flip dynamics. The main conclusions of the study are reported in Sec. V.
II. MODEL AND DYNAMICS
The model is defined on a cubic lattice of linear size L
with periodic boundary conditions. On each site si = 1,…,N
= L3d there is an Ising spin variable sSi = ± 1d. The Hamiltonian is
H=−J
o SiS j − oi hiSi − H oi Si ,
s1d
ki,jl
where the first sum extends over all distinct pairs of nearestneighbor snnd , H is the external applied field, and hi are
quenched random fields drawn independently from a Gaussian distribution with zero mean and standard deviation s. We
are interested in studying the sequence of states along irreversible paths at T = 0 when the system is driven by the external field H sin the adiabatic limit corresponding to a vanishingly small rate of change of the external fieldd. For this
purpose, the Hamiltonian must be supplemented by some
dynamical rules.
The standard one-spin-flip dynamics used in previous
studies consists in minimizing the energy of each spin
Hi = − Si f i ,
s2d
where
fi = J
S j + hi + H
o
jsid
FIG. 1. Stability diagrams showing the state with minimum local energy according to the values of the fields created by the neighborhood sdefined in the textd: sad corresponds to a single spin i and
sbd to a pair of neighboring spins i , j.
come unstable and thus initiate an avalanche. The avalanche
stops when a new metastable state is reached. The external
field is then changed again, and so on. When the spins that
become unstable during the avalanche are sequentially reversed se.g., by increasing i from 1 to Nd, it is of course
crucial that the final state does not depend on the sequential
order. Thanks to the ferromagnetic nature of the couplings,
this is indeed the case as a result of the so-called “nopassing” and abelian properties of this dynamics.1,16 Moreover, the same state is also reached when all unstable spins
are flipped in parallel, which allows to measure the “time” it
takes an avalanche to occur.7
Setting the rules for a two-spin-flip dynamics is rather
straightforward. By definition, two-spin-flip stable states are
spin configurations whose energy sdefined solely by the
Hamiltoniand cannot be lowered by the flip of one or two
spins sclearly, new features are only introduced when these
two spins are nnd. The slocald energy to be minimized is thus
the one associated with a pair ij of nn spins
Hij = − Si f 8i − S j f 8j − JSiS j ,
s3d
s4d
where
is the net field at site i fthe summation in Eq. s3d is over the
z nn of ig. From the above expression, it is clear that the
minimization of Hi is obtained by aligning each spin with its
local field, Si = signsf id, as represented schematically in Fig.
1sad. This provides a stability criterion for any state with
respect to this one-spin-flip dynamics. This dynamical rule
may be implemented by an algorithm that propagates one
avalanche at a time.15 Starting from a stable configuration,
one increases sor decreasesd the external field until the local
field f i at some site i becomes zero sthis corresponds to the
vanishing of the local minimum in which the system was
trappedd. The spin Si swhich is uniquely defined because the
distribution of the random fields is continuousd is then
flipped, which in turn may cause neighboring spins to be-
f i8 = J
o
Sk + hi + H
s5d
ksidÞj
is the field experienced by Si without the influence of
the neighbor S j fthe summation in Eq. s5d is over the nn
of i excepting jg. One can then think of the dynamics as
made of single-spin flips and “irreducibly cooperative” twospin flips. As pictured in Fig. 1sbd, a single-spin flip occurs
whenever the net field on Si, f i = f i8 + JS j, or on S j, f j = f 8j
+ JSi, changes sign. This corresponds to the changes
↓ ↓ ↔ ↑ ↓, ↓ ↓ ↔ ↓ ↑, ↑ ↑ ↔ ↑ ↓, ↑ ↑ ↔ ↓ ↑ in the diagram. An
irreducibly cooperative two-spin flip involves a nn pair of
spins with the same sign s↑↑ or ↓↓d that cannot flip individually si.e., without the simultaneous flip of the neighbord. This
134424-2
PHYSICAL REVIEW B 71, 134424 s2005d
HYSTERESIS AND AVALANCHES IN THE T = 0…
occurs whenever the net field on the pair of aligned spins,
f ij = f i8 + f 8j , changes sign in the region of the diagram where
−J ø f 8i ø J and −J ø f 8j ø J. These two last conditions come
from the fact that the spins cannot individually flip shence
neither f i nor f j changes sign before the cooperative flip of
the paird and the condition that once a pair has flipped, none
of the spins can flip back individually.
The corresponding algorithm is a simple extension of the
one described above for the one-spin-flip dynamics. Starting
from a two-spin-flip stable configuration, the external field is
varied until one finds a pair of spins that becomes marginally
stable: the representative point of this pair in the diagram of
Fig. 1sbd leaves the region where it was originally sassociated with ↑ ↑ , ↑ ↓ , ↓ ↑, or ↓↓d and, depending on the border
which is first attained, only a single spin flips or the two
spins flip simultaneously.17
It is not hard to show that this dynamics obeys the same
properties as the one-spin-flip dynamics, in particular the
crucial abelian property. This is again a consequence of the
ferromagnetic nature of the interactions. One only needs to
note that the state of the system can be represented by a set
of zN / 2 points scorresponding to all the distinct nn pairsd in
the diagram of Fig. 1sbd. As the external field is monotonously increased sresp. decreasedd, the local fields can only
increase sresp. decreased, the spins can only flip up sresp.
downd, and the points can only move up and right sresp.
down and leftd in the diagram. By slightly modifying the
arguments of Ref. 1, one then can prove the no-passing rule,
the abelian property and the existence of return-point
memory.
Instead of paraphrasing the demonstrations given in Ref.
1, we choose here to illustrate these properties by a numerical example. sWe have also performed numerical tests in
many situations and found no violations of these properties.d
The evolution of a system with size L = 30 and s = 2.5 is
shown in Fig. 2 where the energy per spin, e = H / N, is plotted as a function of the magnetization m = S Si / N sstrictly
speaking, e is the enthalpyd. The external field H is varied
from a very large initial value where all spins are up to the
final value H = −2 shere and after, J is taken as the energy
unitd. Two of the curves display the sequence of unstable
states that are obtained after a sudden change of the external
field using either a sequential or a parallel updating algorithm. We also show the metastable evolution corresponding
to the adiabatic driving swith sequential updatingd along the
hysteresis loop. In all cases the final state is the same. This is
true even when the intermediate states are distinct, for instance when the spins are chosen sequentially in a different
order.
The property of return-point memory property is illustrated in Fig. 3, again for a system with size L = 30 and
s = 2.5. The minor loop is obtained by reversing the evolution of the external field first in the decreasing branch at
H = −0.85 and then at H = + 0.80. As can be seen from the
inset, the internal loop closes before the return point, so that
the evolution follows that of the major loop for a small region of H * −0.85.
III. NUMERICAL SIMULATIONS
We now present the results of numerical simulations performed on 3-dimensional cubic lattices using either the one-
FIG. 2. sColor onlined Evolution of the enthalpy and the magnetization sper spind for a system of size L = 30 and s = 2.5 as the
external field is changed from H = + ` to H = −2. The trajectories
corresponding to sequential and parallel updating algorithms are
also compared with the adiabatic one. Points indicate the intermediate unstable states. The inset shows a closeup of the region around
the end point. Notice that the end point is the same in all cases.
spin-flip or two-spin-flip dynamics with sequential updating.
Averaged quantities were obtained with statistics over
103 – 105 different realizations of the random field distribution and system sizes ranging from L = 8 up to L = 48. As
emphasized in Ref. 8, in order to describe properly avalanche
properties sespecially the “spanning” avalanchesd, it is more
important to perform averages over many disorder realizations than to simulate very large system sizes.
A. Hysteresis loops
Figure 4 shows the hysteresis loops obtained in a single
sample for two different values of the disorder s. For com-
FIG. 3. sColor onlined Major hysteresis loop and internal loop
s−0.85ø H ø 0.8d obtained with the two-spin-flip dynamics for a
system of size L = 30 and s = 2.5. The inset shows the details around
the point at H = −0.85 revealing the property of return-point
memory.
134424-3
PHYSICAL REVIEW B 71, 134424 s2005d
VIVES, ROSINBERG, AND TARJUS
FIG. 4. sColor onlined Magnetization curves obtained with the
one-spin and two-spin-flip dynamics in a sample of size L = 30 for
sad s = 2 and sbd s = 3: dotted-dashed lines correspond to the onespin-flip dynamics and dashed lines to the two-spin-flip dynamics.
In addition, the ground-state magnetization is shown as continuous
lines.
parison, we also display the magnetization curves obtained
with the one-spin-flip dynamics and with the algorithm of
Ref. 18 which gives the exact ground-state sequilibriumd
magnetization. The corresponding behavior of the enthalpy
per spin along the ascending branches of the loops is reported in Fig. 5 snote in passing that the ground-state enthalpy does not show any discontinuity as the external field
is varied18d. As could be expected, the main effect of the new
dynamics is to reduce the size of the hysteresis loops. Specifically, the coercivity si.e., the magnitude of the external
FIG. 5. sColor onlined Enthalpy per spin as a function of H
corresponding to the loops in Fig. 5 sfor clarity, only the ascending
branches are shownd. The continuous lines represent the groundstate behavior.
field for which the magnetization is equal to zerod is decreased by more than 30% when allowing pairs of spins to
flip together. Accordingly, the enthalpy difference between
the ground state and the metastable states which are visited
along the loops is also reduced. Nevertheless, the loops display the same key feature, that is a change from a discontinuous to a continuous behavior as the disorder is increased.
This suggests that there is also an out-of-equilibrium
disorder-induced phase transition under the two-spin-flip dynamics, with a critical value of s at which the discontinuity
appears in the thermodynamic limit.
It is worth pointing out that the new dynamical rules allow the system to effectively overcome energy barriers of
magnitude up to DE = 2J. Indeed, the difference between the
two dynamics shows up when a pair of nn spins with same
sign can cooperatively flip, say from ↓↓ to ↑↑ when the external field is increased, whereas each of its spin cannot individually flip. This means that along one-spin-flip paths, the
system has now been able to bypass the higher-energy states,
either ↑↓ or ↓↑. By using Eq. s4d, it is easy to show that the
relevant barrier height associated with this process is at most
2J. Since cooperative flips of more than two spins do not
occur with the chosen dynamics and the system’s trajectory
otherwise go through states of decreasing energy, one concludes that DE = 2J is the maximum barrier height that the
system may overcome when passing from the one-spin-flip
to the two-spin-flip dynamics.
B. Avalanches
As shown in recent studies,8,9 a good characterization of
the disorder-induced critical point can be reached by analyzing the number and size distribution of the magnetization
jumps savalanchesd that compose the hysteresis loops in finite systems. For that purpose, it is necessary to classify the
avalanches in several categories, according to their behavior
as the system size L is increased. One first has to distinguish
whether or not an avalanche spans the system from one side
to the other, in one, two, or three spatial directions sindicated
in the following by the index ad. For each individual avalanche, this is a property that can be easily detected during
the simulation. Avalanches are thus classified as being nonspanning sa = nsd, one-dimensional s1Dd spanning sa = 1d,
two-dimensional s2Dd spanning sa = 2d, or three-dimensional
s3Dd spanning sa = 3d.
Figure 6 shows the number of 1D, 2D, and 3D spanning
avalanches recorded along the descending branch of the hysteresis loops as a function of s. The data, averaged over
disorder, correspond to a system of size L = 24. It can be seen
that the behavior of the three quantities is completely equivalent under the two dynamics. The only difference is a shift
toward larger values of s when the two-spin-flip dynamics is
used. The same shift is also found for all studied system
s1d
sizes. This is a first indication that ss2d
c . sc , as will be
confirmed by the finite-size scaling analysis presented in the
next section. In the case of the one-spin-flip dynamics, a
detailed analysis was performed in Refs. 8 and 9, revealing
the scenario that occurs in the thermodynamic limit and that
is already suggested by the data shown in Fig. 6: when
134424-4
PHYSICAL REVIEW B 71, 134424 s2005d
HYSTERESIS AND AVALANCHES IN THE T = 0…
FIG. 6. sColor onlined Average numbers of 1D, 2D, and 3Dspanning avalanches as a function of s in a system of size L = 24.
Continuous and dashed lines are guides for the eye and correspond
to the one-spin-flip and two-spin-flip dynamics, respectively.
L → `, N1ssd and N2ssd are expected to display a d singularity at sc and N3ssd a steplike behavior. It was shown,
moreover, that there are two types of 3D spanning avalanches, subcritical and critical, which scale with different
exponents. The former are responsible for the discontinuity
in the magnetization curve in the thermodynamic limit sthere
is only one, compact, subcritical avalanche for s , scd
whereas the latter only exist at sc shence, the additional d
singularity at the edge of the step function whose signature is
already visible in Fig. 6d. In a finite system, however, all
kinds of avalanches may exist close enough to the critical
point, and it is quite difficult to discriminate subcritical from
critical avalanches. In Ref. 9, an elaborate analysis was
needed to show that these avalanches have different fractal
dimensions at criticality. This study is impossible here be-
cause of the complexity of the two-spin-flip algorithm that
forbids the use of large systems with good enough statistics.
Therefore, in the following, we shall not distinguish between
these subcritical and critical 3D-spanning avalanches.
Figure 7 shows the snormalizedd avalanche size distributions Dass ; s , Ld obtained along one branch of the hysteresis
loop for three different values of s sfor clarity, the results
obtained with the one-spin-flip dynamics are represented by
continuous linesd. Surprisingly, one can see that the distribution of non-spanning avalanches in Figs. 7sad–7scd is almost
unaltered by the change in the dynamics. The only small
difference sbarely visible on the figured induced by the twospin-flip dynamics is that there are a little less avalanches of
size s = 1 and a little more avalanches of size s = 2, but the
rest of the distribution is almost the same. In particular, with
both dynamics, the expected power-law behavior of the distribution will be characterized by the same exponent teff
< 2.0 in the thermodynamic limit.9
The size distributions D1 and D2 of the 1D and 2Dspanning avalanches shown in Figs. 7sdd–7sgd also appear to
be identical with the two dynamics, at least within statistical
error bars. sNote that these avalanches do not exist for s
= 1.6 because this value is much lower than sc for both dynamics.d The only visible differences between the two dynamics occur in D3, the size distribution of the 3D-spanning
avalanches. Specifically, for s = 1.60 and 2.25 si.e., below sc
and very close to sc, respectivelyd, the large 3D-spanning
avalanches tend to be shifted to even larger sizes. According
to Refs. 8 and 9, these avalanches are probably subcritical
spanning avalanches and their average size is thus a measure
of the order parameter. Therefore, this result is another indis1d
cation that ss2d
c . sc . In contrast, for s = 2.80 swhich is
clearly above scd, the distribution D3 is not affected by the
dynamics fFig. 7sjdg: in this case, one expects to detect only
critical 3D-spanning avalanches in a finite system.
FIG. 7. sColor onlined Avalanche size distributions for s
= 1.60 sa,hd, 2.25 sb,d,f,id, and
2.80 sc,e,g,jd. The open symbols
and the continuous lines correspond to the two-spin-flip and
one-spin-flip dynamics, respectively. The first row shows the distribution Dns of non-spanning avalanches and the other rows show
the distributions D1 , D2, and D3
of the 1D, 2D, and 3D-spanning
avalanches, respectively. All data
have been obtained in a system of
size L = 24. In sbd, the dashed line
indicates the expected power-law
behavior of Dns at criticality with
exponent teff = 2.0.
134424-5
PHYSICAL REVIEW B 71, 134424 s2005d
VIVES, ROSINBERG, AND TARJUS
FIG. 8. sColor onlined Number of avalanches
spanning in 1D sad and 2D sbd obtained with the
two-spin-flip dynamics for different system sizes.
The corresponding scaling plots in scd and sdd
have been obtained according to Eqs. s6d–s8d
s2d
with sc = 2.25, A = −0.2, n = 1.2, and u = 0.1.
IV. FINITE-SIZE SCALING ANALYSIS
ss2d
c
and
As already mentioned, a precise determination of
of some of the critical exponents requires a detailed finitesize scaling analysis which, unfortunately, is not possible
with the present algorithm. What can be done, however, is to
check whether the set of exponents found with the one-spinflip dynamics can also be used to scale the two-spin-flip data.
We first consider N1 and N2, the numbers of 1D and 2Dspanning avalanches, which are quantities that have a simple
scaling behavior. Following previous work,8,9 we assume the
forms
N1ss,Ld = LuÑ1su2L1/nd
s6d
N2ss,Ld = LuÑ2su2L1/nd,
s7d
thus indicate that the two-spin-flip dynamics only induces a
shift of the critical value of the disorder but does not change
the universality class of the transition.
A second check of this result may be obtained by measuring the average field kH3ssdl at which the 3D-spanning avalanches occur. This quantity was studied with the one-spinflip dynamics in Ref. 9, which allowed to map out the firstorder line scorresponding to the macroscopic jump in the
magnetizationd in the diagram H − s. The dependence of
kH3ssdl with the system size is shown in Fig. 9sad. Note that
the transition line extends above sc in a finite system. However, the end point, beyond which no 3D-spanning avalanches are found swith the present sampling of disorder realizationsd, becomes closer and closer to the critical point
sHc , scd as L is increased.
where u and n are critical exponents, Ñ1 and Ñ2 are scaling
functions, and u2 is a scaling variable that measures the distance to the critical point. It is defined as
u2 =
s − ss2d
c
ss2d
c
+A
F
s − ss2d
c
ss2d
c
G
2
,
s8d
where the parameter A accounts for a second-order correction that plays a role when the studied systems are not very
large, as is the case here. In Ref. 8 the best choice for the
collapse of the scaling plots was obtained with n = 1.2, u
= 0.1, and A = −0.2, values that we keep here. The only free
parameter is thus ss2d
c .
As shown in Fig. 8, a very good collapse of the new data
can be obtained with ss2d
c = 2.25. Taking into account the
quality of the plots, there is an uncertainty of ±0.01 on this
value, but, clearly, the value ss1d
c = 2.21 obtained with the
one-spin-flip dynamics8 can be discarded. We have also
checked that this conclusion is not modified when the nonuniversal parameter A is allowed to vary. These scaling plots
FIG. 9. sColor onlined sad Average field kH3ss , Ldl at which the
3D-spanning avalanches occur as a function of s for different syss2d
tem sizes. sbd Scaling plot of the data below sc according to Eq.
s2d
s9d with Hc = −0.885, B8 = 0.25, and m = 1.5.
134424-6
PHYSICAL REVIEW B 71, 134424 s2005d
HYSTERESIS AND AVALANCHES IN THE T = 0…
similar analysis for Ñ2ss , Ld shows the same agreement. This
is again a strong indication for universality, that goes beyond
the equality of the critical exponents.19
V. CONCLUSION
FIG. 10. sColor onlined Comparison of the scaling plots of the
number of 1D-spanning avalanches obtained with the one-spin-flip
ssymbols with error barsd and two-spin-flip ssymbols without error
barsd dynamics. The two collapses only differ by the value of the
s1d
s2d
critical disorder fsc = 2.21 and sc = 2.25g. The continuous line is
the Gaussian fit proposed in Ref. 8.
According to Ref. 9, this set of curves should scale as
F
1 − B8
kH3lss,Ld = Hs2d
c
s − ss2d
c
ss2d
c
G
− L−1/mĥ3su2L1/nd, s9d
where Hs2d
c is the critical field, B8 is a nonuniversal tilting
constant, m is a critical exponent, and ĥ3 is the corresponding
scaling function. Strictly speaking, the scaling should be
done separately for the average fields kH3−l and kH3cl at
which the subcritical and critical 3D-spanning avalanches
occur.9 Indeed, the number of these avalanches scales differently with L. However we expect the lack of scaling to have
a ssmalld effect only in the region u2L1/n . 1 where the two
kinds of avalanches coexist.9
In Ref. 9, the best collapse of the one-spin-flip data was
obtained with m = 1.5, B8 = 0.25, and Hs1d
c = −1.425. Here, we
keep the same values for m and B8, set ss2d
c = 2.25, and consider the critical field as the only free parameter. As shown in
Fig. 9sbd, a very good collapse can be obtained with Hs2d
c
= −0.885 ffor clarity, we only present the collapse for s
, ss2d
c g. Again, this result is consistent with the assumption
that the new dynamics does not change the universality class
of the transition. On the other hand, the significant decrease
in the critical field sin absolute valued is in line with the
decrease in coercivity illustrated by Fig. 4.
Finally, it is interesting to study the influence of the dynamics on the finite-size scaling functions. In Fig. 10, we
compare the scaling collapses of the number of 1D-spanning
avalanches, Ñ1ss , Ld, obtained with the two dynamics sfor
the two-spin-flip dynamics, this is the same curve as in Fig.
8d. As can be seen, the agreement between the two curves is
quite remarkable. Even the deviations saround u2L1/n < 0d
from the Gaussian fit proposed in Ref. 8 are the same. A
We have shown in this paper that the nonequilibrium behavior of the 3D RFIM at zero temperature is not qualitatively altered when going from the standard one-spin-flip
metastable dynamics to a two-spin-flip metastable dynamics.
The coercivity sthat is essentially the width of the hysteresis
loop associated with the evolution of the magnetization with
the driving fieldd is significantly reduced by allowing twospin flips, but the main features of the hysteretic behavior,
including the presence of a disorder-induced nonequilibrium
transition and the distribution of avalanches, remain similar.
By using a finite-size scaling analysis, focused on the number and size of the avalanches, we have furthermore provided
strong evidence that the critical behavior sexponents and
scaling functionsd obtained with the two-spin-flip dynamics
is in the same universality class as that obtained with the
one-spin-flip dynamics.
Changing the dynamical rules used to study the evolution
of the model, as done here, helps address several important
questions. The first one is the robustness of the hysteretic
scenario provided by the T = 0 RFIM with the standard onespin-flip dynamics. As discussed in the introduction, a basic
assumption underlying the theoretical description is that the
system gets trapped in metastable states on the experimental
time scale and can only escape when a change in the external
field makes the relevant state looses its stability. However,
real systems are not at T = 0 and some partial, local equilibration, due for instance to thermally activated processes,
may take place even though the system remains far from
equilibrium on the experimental time scale. Introducing cooperative two-spin flips is a simple way to check the effect of
partial equilibration processes. Our results clearly point towards the robustness of the whole theoretical picture drawn
from previous studies of the model using the T = 0 one-spinflip dynamics.1,6–9
A second question concerns the relation between the
disorder-induced critical properties observed in the nonequilibrium behavior of the RFIM at T = 0 and the equilibrium
critical behavior associated with the paramagnetic to ferromagnetic transition. There is numerical evidence that, at least
to a good approximation, critical exponents and scaling functions associated with the two kinds of criticality are the
same.6,9,13 sAdditional, but less conclusive evidence is provided by exact, but mean-field-like results on the Bethe
lattice13 and by perturbation theory near the upper critical
dimension d = 6.6d Our present findings suggest that a whole
series of T = 0 metastable dynamics involving k-spin flips
lead to the same nonequilibrium criticality, with the critical
disorder strength increasing with k and the critical coercive
field decreasing with k. Equilibrium behavior at T = 0 as a
function of the external field involves the system’s ground
state, i.e., a state stable to flips of any arbitrary finite number
134424-7
PHYSICAL REVIEW B 71, 134424 s2005d
VIVES, ROSINBERG, AND TARJUS
of spins. Based on the numerical closeness of the critical
exponents and scaling functions, on the fact that the critical
s2d
s1d
eq
disorder strength satisfies seq
c . sc . sc fwhereas Hc = 0
s2d
s1d
, Hc , Hc g, and on the similarity of the underlying physics at T = 0, it is tempting to speculate that the equilibrium
behavior can be obtained as the limit of a series of k-spin-flip
metastable dynamics with increasing k, with the critical
properties of the whole series belonging to the same universality class and governed by the same fixed point.9 In the
opposite case, one would expect to see for large enough k a
crossover between a behavior dominated by the equilibrium
fixed point and the asymptotic behavior controlled by the
nonequilibrium fixed point. No such crossover is seen in the
present study ssee, e.g., the distribution of avalanches in Fig.
*Electronic address: eduard@ecm.ub.es
1
J. P. Sethna, K. Dahmen, S. Kartha, J. A. Krumhansl, B. W.
Roberts, and J. D. Shore, Phys. Rev. Lett. 70, 3347 s1993d.
2 A. Berger, A. Inomata, J. S. Jiang, J. E. Pearson, and S. D. Bader,
Phys. Rev. Lett. 85, 4176 s2000d.
3 J. Marcos, E. Vives, Ll. Mañosa, M. Acet, E. Duman, M. Morin,
V. Novak, and A. Planes, Phys. Rev. B 67, 224406 s2003d.
4 F. Detcheverry, E. Kierlik, M. L. Rosinberg, and G. Tarjus, Phys.
Rev. E 68, 061504 s2003d; Langmuir 20, 8006 s2004d.
5 D. J. Tulimieri, J. Yoon, and M. H. W. Chan, Phys. Rev. Lett. 82,
121 s1999d.
6 K. Dahmen and J. P. Sethna, Phys. Rev. B 53, 14 872 s1996d.
7 O. Perković, K. A. Dahmen, and J. P. Sethna, Phys. Rev. B 59,
6106 s1999d.
8 F. J. Pérez-Reche and E. Vives, Phys. Rev. B 67, 134421 s2003d.
9 F. J. Pérez-Reche and E. Vives, Phys. Rev. B 70, 214422 s2004d.
10 G. Durin and S. Zapperi, in The Science of Hysteresis, edited by
G. Bertotti and I. Mayergoyz sElsevier, New York, 2005d.
11 J. P. Sethna, K. A. Dahmen, and O. Perković, in The Science of
Hysteresis, edited by G. Bertotti and I. Mayergoyz sElsevier,
New York, in pressd.
12 A. Maritan, M. Cieplak, M. R. Swift, and J. R. Banavar, Phys.
7d, but the values of k considered here, k = 1 and 2, are of
course too small to detect such effect, if present.
ACKNOWLEDGMENTS
We acknowledge fruitful discussions with Francisco-José
Perez-Reche and Xavier Illa. E.V. also acknowledges the
hospitality of the Laboratoire de Physique Théorique des
Liquides sUPMC, Parisd during his stay as an invited professor in July 2004. This work has received financial support
from CICyT sSpaind, Project No. MAT2004-01291, and
CIRIT sCataloniad, Project No. 2001SGR00066. The Laboratoire de Physique Théorique des Liquides is the UMR 7600
of the CNRS.
Rev. Lett. 72, 946 s1994d.
Colaiori, M. J. Alava, G. Durin, A. Magni, and S. Zapperi,
Phys. Rev. Lett. 92, 257203 s2004d; M. J. Alava, V. Basso, F.
Colaiori, L. Dante, G. Durin, A. Magni, and S. Zapperi, Phys.
Rev. B 71, 064423 s2005d.
14 This objection, however, may be circumvented by considering the
remanent magnetization obtained in zero external field after a
demagnetization procedure as the order parameter of the transition ssee Ref. 13d.
15 M. C. Kuntz, O. Perković, K. A. Dahmen, B. W. Roberts, and J.
P. Sethna, Comput. Sci. Eng. 1, 73 s1999d.
16 D. Dhar, P. Shukla, and J. P. Sethna, J. Phys. A 30, 5259 s1997d.
17
Unfortunately, because of the additional complexity introduced
by the cooperative flip of nn pairs, one cannot use anymore the
so-called sorted-list algorithm proposed in Ref. 15. This in turn
implies that very large system sizes cannot be simulated.
18 C. Frontera, J. Goicoechea, J. Ortin, and E. Vives, J. Comput.
Phys. 160, 117 s2000d.
19 Note also that for the scaling collapses it has been possible to use
the same values of the a priori nonuniversal constants A and B8
in Eqs. s8d and s9d.
13 F.
134424-8