You are on page 1of 13

PHYSICAL REVIEW B VOLUME 21, NUMBER 11 1 JUNE 1980

Ising model with solitons, phasons, and "the devil's staircase"


Research
Per Bak
IBM Tharrra J. H a(soir Ceirrei, err h(aiiir H'eights, New Yark /05')8

J. von Boehm
NORD/ TA, BID'&Iarrrsi'ej I 7, Crrperrhaperr, Derrirrarh
(Received 24 A ugust 1979)

We have analyzed the modulated phase ot' an Ising model with competing interactions in an
et'fort to increase the understanding of the spatially modulated phases t'ound in many physical
systems. The analysis has three stages. First, the mean-field phase diagram is calculated nu-
merically. A large, possibly infinite, number of phases where the periodicity ot' the ordered
structure is cornrnensurate with the lattice is f'ound. The resulting periodicity-versus-
temperature curve thus probably has an inf'inity of'steps; i. e. , it exhibits "the devil's staircase"
behavior. Then the mean-field theory is analyzed analytically, and it is shown that the stability
of' the commensurate phases can be understood within a domain-wall or "soliton" theory. The
solitons from a regular lattice near the transitions to the commensurate phases. The elementary
excitations in the soliton lattice are the phasons. Third, the effects of' temperature-induced fluc-
tuations, ignored in the mean-field theory, are estimated by calculating the entropy contribution
to the free energy f'rom the phasons. It is found that the stability ranges of the commensurate
phases are reduced, but the staircase survives at finite temperatures. On the basis of our calcu-
lations a phase diagram is constructed.

I. INTRODUCTION rate phases the periodicity is strictly incommensurate.


Aubry' has analyzed a simple model, originally intro-
Spatially modulated systems are very common in duced by Frank and Van der Merwe, ' consisting of a
nature. The periodic phase may be an ordered mag- one-dimensional array of atoms connected with
netic structure' (spin-density wave), a periodic lattice springs and interacting with an external periodic po-
distortion, ' a charge-density wave, ' or a periodic ar- tential. The periodicity is varied by applying pressure
rangement of atoms in a host lattice. ' In this paper to the array. Aubry found that the periodicity locks
we analyze a simple three-dimensional Ising model at any commensurate value, and that there are no in-

lated phase."
with competing interactions which exhibits a modu-
The model in its present form was in-
vented by Elliot' to describe the sinusoidally modu-
commensurate phases. The resulting wave-vector-
versus-pressure curve is continuous, but nonanalytic.
The mathernaticians have invented the very pic-
lated phase of erbium. Despite the simplicity of the turesque name "the devil's staircase" for this peculiar
model, the phase diagram turns out to be extremely behavior. "
Unfortunately, there is no hope of distin-
rich, including a number of transitions between guishing between the various types of nonanalytical
periodic phases, including a floating unpinned phase. behavior in a numerical calculation, or in an experi-
The theory is quite complicated, involving esoteric ment. In an experiment, one should expect to see at
concepts such as solitons, phasons, and "the devil' s most a finite number of steps. It is therefore impera-
staircase. " tive to do analytical work on simple model systems.
In general, the distinction is made between "com- In addition to the above mentioned global
mensurate" phases, where the periodicity of the phenomenological theories, Frank and Van der
modulated structure is a (simple) multiple of the lat- Merwe, ' McMillan, ' Luban et g/. , ' and Bak and
tice constant and incommensurate phases where this Emery' have developed a domain-wall, or "soliton""
is not the case. In other words, the wave vector theory for the transition from an incommensurate
characterizing the commensurate phase is a rational phase to a specific commensurate phase, the CI tran-
fraction of a reciprocal-lattice vector. There exist a sition. The analysis to be presented here makes ex-
few phenomenological theories for modulated sys- tensive use of the soliton picture.
tems. Using a simple Landau-type argument, Dzy- Our original motivation for studying a magnetic
aloshinsky' argued that the periodic phase is always model of a modulated structure was the results of
commensurate. Pokrovsky' also suggests that the some interesting experiments' "
on the rare-earth
wave-vector-versus-temperature curve "locks-in" at magnet CeSb. As the temperature is lowered, CeSb
every rational value, but in between the commensu- displays a series of phase transitions between com-

21 5297 1980 The American Physical Society


5298 PER BAK AND J. YON BOEHM 21

mensurate phases. To investigate the change of "experimentally" by a Monte Carlo method, and they
periodicity versus temperature, we have recently' have been able to confirm some of our findings.
studied a simple S = , model with ternperature- Our analysis has three stages. First the mean-field
independent competing nearest- and next-nearest- phase diagram is calculated numerically (Sec. III).
neighbor interactions. Our findings, based on a nu- Indeed, the MF phase diagram includes a large, pos-
sibly infinite, number of commensurate phases, as
merica1 solution to the mean-field (MF) theory, were
the I'ollowing: (i) The periodicity changes in a step- could be expected from our previous analysis, " and
wise manner very similar to the behavior of CeSb. in agreement with the various phenomenological
Over large temperature intervals, the periodicity theories. '' A short account of this numerical work
"locks-in" at a few commensurate values, and (ii) by was given at the APS March meeting, 1979.
refining the calculation, we found more and more Second, in Sec. IV, the mean-field theory is
phases. These additional phases were stable in only analyzed analytically, and it is shown that the stability
extremely narrow temperature intervals and would of the commensurate phases can be understood
probably not be observable. This is consistent with within a "soliton" picture, The solitons are domain
"the devil's staircase. " walls between essentially commensurate domains.
There is no reason why this behavior should be The results of the soliton theory agree quantitatively
unique for an S = , model. The staircase is a conse- with the numerical results. The theory explains the
5

temperature dependence of the wave vector near the


quence of the fact that the spins are confined to a lat-
main commensurate phases, and the corresponding
tice, and has nothing to do with the particular choice
magnetic structures. The solitons form a regular lat-
of spin value. For the maximum possible insight it is
tice near the transitions to the commensurate steps.
important to study the simplest possible model exhi-
The excitations in the soliton lattice are the phasons,
biting a modulated phase. The Elliott' model is a
1 in complete analogy with the phonons in a regular
simple spin- , model with exactly the desired quali-
atomic lattice.
ties (Fig. I). It displays a Lifshitz point (Fig. 2) Third, in Sec, V, the finite-temperature effects,
separating a uniformly ordered ferromagnetic phase which were ignored in the MF theory, are estimated
(q =0), a modulated ordered phase (q WO), and a by calculating the phason contribution to the free en-
paramagnetic disordered phase. The model has been ergy. The phasons or "murons" reduce the domain-
studied in some detail near the phase boundary to the wall energy and favor the incommensurate phases.
disordered phase, and near the Lifshitz point. ' In Ho~ever, the staircase seems to exist all the way up
this paper we shall be concerned with the properties to T, . In the limit T T, the widths of the com-
of the modulated phase at lower temperatures away mensurate phases vanish.
from the transition to the disordered phase. Selke
and Fisher' are currently studying the same model

0.5 I.O

FIG. 2. Mean-field phase diagram. FM: ferromagnetic


phase. PM: paramagnetic phase, and S: nsinusoidally"
FICi. 1. Ising model with competing interactions. modulated phase.
21 ISING MODEL WITH SOLITONS, PHASONS, AND. .. 5299

II ~ MODEL T=O -J2 /J 1


&0.5 (a)

1 Q) li d il d l~ ~ n li n u h li ll ll h II
'The model that we shall study is a simple spin- ,

Ising model (S =+ I) (Fig. I). The spins within the


xy planes interact through a ferromagnetic nearest- 0 Z
)
neighbor coupling, J1 0; the nearest-neighbor in-
teraction between planes is also J1, but the interac-
-J2 /J =05 (b)
I I

tion between spins in next-nearest-neighbor layers is


(
antiferromagnetic, J~ 0. The competing interac-
V'

tions stabilize the periodic phase. 0 5 Z


The mean-field phase diagram is shown in Fig. 2.
It consists of three phases': (i) a disordered -J2iJl &O. s
paramagnetic phase, (ii) a ferromagnetic phase, and
n I ,
V'
(iii) the sinusoidally modulated phase characterized
by a wave vector q =2m(0, 0, q). As we shall see
later, this picture is greatly oversimplified. The MF
transition line between the paramagnetic phase and
the modulated phase can be calculated as follows. FIG. 3. Ground-state spin configurations ( T =0). (a)
The single-spin susceptibility Xo is given by the Curie Ferromagnetic phase. (b) Ferrom &gnetic phase with dom &in

law wall. (c) "Two-up-two-down" phase, q = .


4

xo
1/ T (2. 1)
ground state is formed by introducing the maximum
We have chosen a unit system with the Boltzmann
constant equal to unity. The susceptibility in the
number of domain walls (with spacing I). This
gives a phase with a periodicity of 4(qo= , ). The
) 1

paramagnetic phase is
ordered structure is formed by two layers of "up"
spins followed by two layers of "down" spins, and so
(2.2)
on. Since the periodicity is different at T =0 and at
T the periodicity of the modulation changes as the
where J(q) is the Fourier transform of the interac- temperature is varied, in contrast to previous belief. '
tions J,
II
between the spins at sites i and i
'
The problem to be investigated here is how this tem-
perature variation takes place.
J(q)=QJ. . exp(iq r. . ) (2.3) Note that the stability of the q = 4 phase could be
1

discussed within a domain-wall picture. We shall see


The transition temperature T, and the corresponding that at higher temperatures the stability of' the vari-
periodicity 1/q, are given by the values for which the ous commensurate phases can also be described
susceptibility (2.2) first diverges as the temperature is within a domain-wall picture, but in this case the
lowered: solution is far from trivial.

T, =4J1+2J2cos2~q, +2J2cos4~q, (2.4)


III. MEAN-FIELD CALCULATION OF THE SPATIALLY
J1 MODULATED PHASE
277q~ = cos
The simplest way of obtaining information on the
phase diagram corresponding to a given microscopic
The periodicity at T, thus depends on Jt/J~. The model is usually by means of mean-field theory.
Lifshitz point P separating the three phases has the Whereas mean-field theory is certainly insufficient
coordinates J&/Jt = , , T~ = S.S Jt. The line
1
near phase transitions where the correlation length
separating the ferromagnetic and the sinusoidal phase diverges, it is usually successful in describing the or-
is shown approximately. dered phase away from the transitions, and in
At T =0 the structure can be found exactly. When describing the topology ot phase diagrams. In Secs.
Jt/Jt is small the ground state is ferromagnetic. IV Vl it will be shown that even if the mean field in
The cost of creating a "domain wall" as shown in Fig. our case may not give an accurate description of the
3 is 2J2+ Jt. When J2= , J1, the ferromagnetic
1
modulated phase, it forms an important first step in
state becomes marginally stable with respect to our analysis.
domain-wall formation. The ground state is infinitely The Hamiltonian to be studied is
degenerate corresponding to all the possible ways to 3C= $ J. SS . (3.1)
introduce domain walls. When J2 & , J1, the
1

(ii )
5300 PER BAK AND J. VON BOEHM 21

'
The interactions between spins at sites i and i are The average free energy per spin,
given in Fig. l. In the MF approximation this Hamil-
tonian is replaced by the simpler single-spin Hamil- F(N, T) =
N g
N 1

T ln Trexp MF
tonian ()

BCMF = QHS; + 2 $H;(S;) (3.2) H,


Tln 2cosh + (S~)H,
J J

where (S;) is the thermal average of the spin at site i


and the effective field H; arises from the interactions (3.8)
with the surrounding spins is then calculated. This free energy is minimized
H;=QJ (S. ) (3.3) when self-consistency is obtained. At each tempera-
ture, the stable wave vector, or periodicity, is the one
that gives the lowest free energy. In principle, calcu-
If we know the fields H;, the average spins can be lations should be performed for N going to infinity to
calculated from the equation allow for any incommensurate structure. In practice
we have carried out calculations for N up to 17. As
(S, ) =tanh(H/T) (3.4)
we shall see, the large-N periodicities are stable only
The problem is to solve Eqs. (3.3) and (3.4) self- for extremely narrow temperature intervals, so most
consistently. For a ferromagnet, or an antiferrromag- of the phase diagram (away from T, where our nu-
net, this is almost trivial. All the (S;) ~'s are identi-
~
merical approach breaks down) is filled by simple
cal so there is only one pair of equations to solve. commensurate wave vectors, which justifies our pro-
Here, we have in principle an infinity of coupled cedure. We believe that no additional significant in-
equations. We must allow for any structure which is sight can be achieved by extending the numerical cal-
spatially modulated in the z direction, and we cannot culation to higher values of N.
assume a purely sinusoidal arrangement of spins Near T, the MF equations can be solved analytical-
since higher harmonics will certainly develop. We ly (Sec. II). The stable periodicity is here given by
can make the ansatz that the spins are identical only the value for which J(q) attains its maximum value,
within xy planes, but vary from layer to layer. The and q may thus have any incommensurate value.
MF equations then become This is the reason that our approach which assumes
commensurability does not work near T, . The nu-
(3.5) merical calculations pick the commensurate value
that is nearest to the correct "incommensurate" value.
Figure 4(a) shows the resulting wave vector versus
(SJ) =tanh (H, /T) (3.6) temperature for J2/J, =0.6. Disregarding the in-

where (S, ) is the average spin in the J' plane and


terval T/T, ) 0.9, more than 95% of the phase di-
agram is exhausted by the simple commensurate
J(j j') are effective interlayer and intralayer in-
values q = 4, , , and 6. However, other wave vec-
1 1 1

teractions
tors are indeed stable, but for extremely narrow tem-
J(0) =4Ji, J(I) = Ji, J(2) = J2 (3.7)
=2.
perature intervals. The wave vector q = 9 is stable
for a temperature range hT =0.0004Tq= , is
3

Ho~ever, there is in principle still an infinity of equa- stable for b, T =0.0004Tand q = ,6 is stable for
3

tions to solve.
b, T =0.03T, . An important point is that some
A very similar model was analyzed in Ref. 18, and
periodicities (even within our limited set) do not be-
the calculations here proceed in essentially the same
way. At each temperature it is assumed that the spin
corne stable. For example, q goes from
2 1
to 6
.

structure repeats itself after N layers. For each N the without assuming the intermediate value q = 17 '
mean-field equations are solved numerically by This is consistent with the "devil's staircase" as de-
means of an iteration procedure. As the starting fined, for example, by Aubry, but certainly not with
average spin configuration either a sinusoidal struc- a smoothly varying wave vector nor with a constant
ture or the self-consistent solution at a nearby tem- incommensurate wave vector.
perature (for the same values of Ji and J2) is used. Note also that the curve is not monotonic. At first
This configuration is used to generate the fields H, , the wave vector q decreases as the temperature is
which again are used to calculate a new spin config- lowered, then it increases until it locks at 4 at the
uration, Usually, self-consistency is obtained after lowest temperatures. The maximum periodicity is 6.
only a few iterations, but for certain values of N and A similar behavior has been found for the suscepti-
T the convergence is slower and some hundred itera-
tions were needed. model. "
bility maximum for the one-dimensional Ising
Most interesting, the periodicity of the or-
21 ISING MODEL WITH SOLITONS, PHASONS, AND . . ~ 5301

I I t 1 I I
I
IO
I/4 P I/ I I I/9 2/17 PM
0.25
(a)
4 I/I5
2/9-
3/l4-
0.20 I/5
3/l6 2/I I

2/I I 3/IT
J 2 = 06
0 I

J = I
0.2 0 25 0.3 0.35 0.4 0.45 0.5 0.55 0.6 0.65 0.7
I

Q. l 5 J~ /Jl

FIG. 6. Phase diagram constructed on the basis of' the nu-


0.0
merical MF calculations. The broken lines result f'rom the
soliton theory.

I/4
0.25 ed in the following section.
(b)
4/IT ~ Figure 5 (and Figs. 7 and 9) show some typical
stable spin arrangements. Note that near T, the con-
2/9 figurations are almost sinusoidal whereas at lower
3/I4
temperatures they contain a significant amount of
0.20 I/5 higher harmonics. For example, the q = structure
I

3/(6 approaches the "squared-up" structure where all the


JC =-OT
spins are + I at low temperatures.
J -I I On the basis of calculations for various values of
Jt/J~, the phase diagram in Fig. 6 has been con-
O. I 5 T-
C structed. At the point ( T, Jt/Jt) = (0, 0. 5) there is
I

0.0 an infinite degeneracy with respect to periodicity so


all the periodic phases must merge at this point. The
bulging of the q = 6 phase is associated with the
FIG. 4. Wave vector vs temperature as calculated nurner-
icallyf'or
(a) J2/JI = 0.6 and (b) J2jJI = 0.7. The f'ull nonmonotonic temperature dependence of the wave
curve is the result of' the theoretical calculation in Sec. 1V. vector. Near Twhere our approximations break
down, a refined calculation would produce a finer
mesh of transition lines as T, is approached. %e
dered structure in erbium exhibits the same type of
shall see that the finer details of the staircase will
behavior. This behavior can also be explained within ef-
probably not be smeared by finite-temperature
the domain-wall, or soliton, theory as we shall see.
fects, But first let us try to understand analytically
Figure 4(b) shows the temperature dependence for
the complicated structure of the MF phase diagram.
J2/Jt = 0.7. Again the stepwise behavior is ob-
served, but the curve seems monotonic. The full
curve is the result of the soliton theory to be present- IV. SOLITON THEORY OF THE MODULATED PHASE

In this section it will be sho~n that the main


features of the MF phase diagram found numerically
can be understood in terms of a soliton theory. The
solitons are static domain walls between commensu-
rate domains. In this respect they differ fundamen-
tally from the solitons appearing for example in the
q =I/5 JP /JI= -0.6 T=3.06 JI theory of the magnetic chain in a field. 23 The theory
I- makes contact with more phenomenological theories
A and provides the basis for the calculation of the na-
ture of the phase diagram at finite temperatures.

A. Stability of q &
"up-up-down-down" phase

FIG. 5. Typical spin arrangements as calculated numeri- One of the most striking features of the phase di-
cally. The solid curves indicate the main harmonic. agrarn is the stability over large regimes of the phase
5302 PER BAK AND J. VON BOEHM

diagram of the q = 1
phase. It will be shown here
4
So
that the stability is related to a positive soliton-
forrnation energy; the transition to other wave vec-
tors is brought about by spontaneous formation of in-
H, ,= g J (S ) + T tanh '(S;) (4.3)

teracting solitons, or domain walls.


As the starting point, the MF free energy and, from Eq. (4. 1), the free energy per spin be-
F ( (S;) ) is calculated near T, in order to generate a comes (apart from a constant)
Landau functional. Note that F is not the same free
energy as the one given by Eq. (3.8). Here F is the
$J... (S, ) (S., )+ $J
(s,. )
free energy in an external field H, expressed in
terms of (S;), the average spin at site i
, II
r
tanh '(r do

,
=H,CX( ((S, )) . (4. 1) (4 4)

The relation H, ,
( (S;) ) is given by the equation
where N is the number of spins. Minimization of Eq.
H;+H, ,=tanh {4.4) with respect to (S;) yields the MF equations
(S ) =tanh $ J, (S, ) + H, ,/T
.I
(3.3) and (3.4), as it should. Expanding in power
S-
I i series, and introducing Fourier transforms and
(4.2) J(q) of (S;) and J respectively, we find

F = $ [J{q ) T)S-S -+ , T X q(+ qq+ q3+ q4 7)


X S- S- S- S-5{
1 4

+ 30 T S S q 5( q + 1
'
+ q6 7 )+ (4.5)
1

Here, vis a reciprocal-lattice vector. The expansion includes two types of terms. First, there are translationally
invariant terms where the wave vectors add up to zero. The second type are "umklapp" terms where the wave
vectors add up to a reciprocal-lattice vector. These terms, which reflect the fact that the spins are situated on a
lattice are responsible for the stability of the commensurate phases. In principle, to get the complete phase di-
agram, we should retain all regular and umklapp terms. Here, the expansion will be truncated after the fourth-
order term since we are interested in the q = 4 phase,
1

Sufficiently close to T, it is sufficient to consider only one wave vector q0 and small fluctuations around it.
Near q0 =2'(0, 0, ) the expansion takes the form

F = ${r +aq, +cq,'+c'qg )S, + S


q

V4

(4.6)

where I = T 4J1 +2Jqa=2J1, c = 4Jq, c'=4J1, and arrive at the following effective Landau-
and q, and qq are the wave-vector deviations from q0 Ginzburg free-energy functional:
parallel and perpendicular to the z axis, respectively.
We have retained urnklapp terms corresponding to 7 F= dr F r)
equal to a unit vector in the z direction.
We now introduce the continuous order parameters

Si4(r)=42
(
3

dqexp(iq r)S -+-


F{ r) = ~ rllS(i, l
+ 'TIS(i41
)

2m V0+V
e
+ c )7 Sti4 + c') ')7gS(i4)
(4.7)
&
2c
3

s, i( )=42 Jdee p( e )s +
8 T(S)/4
+S 1/4 ) (4.8)
21 ISING MODEL %ITH SOLITONS, PHASONS, AND. . . 5303

The order parameter Sti4( r ) describes a spin density portional to the distance L between the domain stalls
wave
q
-== 2m (4. 16)
S( r ) = (2m i (
1 1
Sti4( r ) exp z) ] (4.9)
2
i.e. , q is the soliton density. This wave vector is
The lattice now manifests itself through the last term, given by the equation'4
which arises from the umklapp terms. To obtain the
q' =

2
stable spin configurations, Eq. (4.8) should be
n.
Z(&) E(&), (4. 17)
minimized with respect to S~/4(r ). Clearly, S]/4 5
depends upon z only.
where K and E are the elliptic integrals of the first
Following Refs. 12 and 14 we now assume that
and second kind, and q is defined by the equation
near the commensurate phase ( q = ) the amplitude
1

of the order parameter is constant whereas the phase (u/u, )'" = ,' 7rq/E(zi) (4. 18)
is allowed to be space dependent; i.e. ,
where v, is a critical value of v given by
S+,i4(z) =A exp+i&(z) (4. io)
v =m5 (4. 19)
The commensurate phase is given by Eqs. (4.9) and
(4. 10) with $ constant. The amplitude A is found by Near the commensurate phase the free energy takes
minimizing Eq. (4.8) in the commensurate phase. By the asymptotic form"
choosing P(z) =5z, we describe a modulated phase 1

with E 4
Jv 16tu 2mJv
isi
q
S(r) = ( z)]
1
(4. )
1
A exp {i gz) exp [2rri
2 (4.2o)
This is an incommensurate spin-density wave with The ai~tisolitons are stable because 5 is negative. The
wave vector q = +8/2e.
1
The free-energy function- first term, proportional to the soliton density, is a sol-
al per unit area perpendicular to z becomes iton formation energy; the second term is a weak
repulsion between solitons. The commensurate
F= dz cA
1 d@
5 +v 1+cosp$, 412) phase is stable as long as the first term is positive.
2 dz hen the coefficient becomes negative, for small v
or large ~5] the commensurate q = phase becomes
1

with
unstable with respect to soliton formation. This hap-
Ji TA' pens at a temperature given by Eqs. (4. 19) and
4 ~ '
4J 96J (4. i3)
and T'c, =4Jt 2Jz+n Jt /8Jz (4.21)
Az =3{4Jt 2Jz T)/T (4. 13) This curve has been plotted in Fig. 6 (lower broken
line). Figures 7 and 9 show the spin structures corre-
The first term favors the incommensurate wave vec-
sponding to the soliton solutions for Jz/Jt = 2 and
tor q = 4 +g/2e. The second term favors the com-
1

0.7, respectively. In terms of the soliton picture,


phase, $ = , m. A transition between the
1
mensurate the narrow stable phases found numerically near Tcf
commensurate phase and the incommensurate phase have a simple interpretation. They are "pinned" soli-
tons. The q = state can be formed by including
4
takes place because of the competition between these
two terms. The phase function which minimizes Eq.
one domain wall every four periods in the q. = 4
(4. 12) satisfies the sine-Gordon equation
The q = structure corresponds to one
3
structure.
$"(z) +4v sin 4$ =0 (4. 14) soliton every three periods, and so on. The q = ,
].

structure can be interpreted as having one soliton for


One of the solutions is the (anti) soliton each period. The positions of the solitons character-
' izing these higher-order commensurate phases are
4 (z) = , vr tan 'exp (4tuz) (4. i S)
pinned by higher-order umklapp terms ignored in the
The soliton describes a domain wall which separates present approximation, where the phase functional
two almost commensurate regions (Figs. 7 and 9). In (4. 12) is invariant under a uniform shift along z.
general, the solutions are regularly spaced solitons, a The broken arrows indicate the structure that was cal-
culated numerically (q = ). One period of the soli-
4
"soliton lattice". The deviation of the average wave
vector from q 0 in the z direction, q, is inversely pro- ton lattice is shown. Almost perfect agreement with
5304 PER BAK AND J. VON BOEHM 21

I ]
-5 5 Z
2

0.5 I
j ji
I
I I
I I )
I I
I I
I I

FIG. 7. Soliton solution and the corresponding spin struc- FIG. 9. Soliton solution and the corresponding spin struc-
ture, calculated for J&/J] = 2 at the temperature at which ture calculated for J&/J[ = 0. 7 at Tc]. The broken arrows
the commensurate phase becomes unstable ( TC1). The bro- indicate the numerical solution.
ken arrows indicate the numerical solution obtained using
4
the formalism in Sec. III. (q = ,7 ). For clarity, the arrows
are shifted slightly away from the lattice positions. B. Stability of the "three-up-three-down" phase (q )
6

Maybe the most surprising numerical result is the


the numerical solution is obtained as long as Jq is not "bulging" of the phase diagram (Fig. 6) corresponding
too close to , J1. Figure 8 shows the commensu- to the stability of the q = , phase over a large frac-
1 1

rate structure (with no domain walls) which is stable tion of the phase diagram for J~ near 0.5J]. For a
just below Tct. given value of Jq the periodicity first increases to
Figure 4(b) shows the temperature dependence of N =6 as the temperature is lowered, then it decreases
the wave vector for Jz/Jt = 0.7. The calculated to N =4. In this section it will be seen that this
curve does not "lock-in" at other commensurate behavior can readily be understood within the soliton
values. This is due to the fact that we have ignored theory.
higher-order umklapp terms. For example, the lock- Our starting point will again be the Landau expan-
in at q = 5 could, in principle, be accounted for by sion (4.5). This time, however, we chose qp =2m'
1

including tenth-order terms. Near Tct the wave vec- x (0, 0, 61 ) and keep the sixth-order umklapp terms.
tor assumes the asymptotic form Repeating all the steps leading to Eq. (4. 12), we find
the following free-energy functional
q
in-'( T Tct) . (4.22) 1

F dz cA ,
1 d@
5 +v(1+cos pP) p =6
which explains the very steep behavior of the q vs T 2 dz
curve, and the difficulty of finding numerical solu-
tions in this regime. (4. 24)
The soliton has a characteristic width where A ance $ now describe a spin-density wave
4J~
(4.23) S(r) = (A /~2) exp [i P(z) ] exp [2rri ( 6
z) ] (4.2S)
4tu,
and
Hence the width is large when J~ is large (Figs. 7 c = Jt
1
2J) (4.26)
and 9). We expect that our continuum approxima-
tion is best when 5 is large. In this light, it may be A4T
(4.27)
surprising that the calculation gives a meaningful 60( J1 4 JP)
result for Jz/Jt = 0.7, where 5 =2.8.
J3(J, +2J, )
J, +4J, (4.28)

0.5 jI I
(4.29)
I I I
I ) I
I I
)

fu
-5 I
I
I
I 5Z Again, we find soliton solutions

@(z) = rr + tanh ' ex


p (6tuz) (4.30)
FIG. 8. Commensurate structure which is stable at a tern-
perature just below TC1 for J&/Jt = 0.7. The broken ar- Defining q =2rr/6L the wave vector versus T curve
rows indicate the numerical solution. is given by exactly the same equations as for the
ISING MODEL WITH SOLITONS, PHASONS, AND. . . 5305

V. FINITE-TEMPERATURE EFFECTS.
PHASON INSTABILITIES

Superficially, it seems that any commensurate


phase will become stable when the corresponding "6"
Z--2 I

in Eq. (4. 12) becomes small enough. The energy


cost of forming a domain wall is infinite since the
wall extends to infinity in two directions and thus is
always large compared with the thermal energy T.
z--2
However, the walls are not completely rigid, and
there is an entropy associated with the possibility of
FIG. 10. Soliton solution and the corresponding
spin having elastic deformations of the walls. We shall
near the q = 6 phase calculated for J2/J1= 0.6.
1
structure see that for most commensurate phases the entropy
The broken arrows indicate the numerical solution contribution to the free energy overpowers the ener-
(& = ).
3
17
gy term and destabilizes the commensurate phase.
The key to this behavior is that both the energy and
the entropy are proportional to the soliton density q.
q = case, Eqs. (4. 17), (4. 18), and (4.20). The The incommensurate phase near the CI transition
can be thought of as a soliton lattice. Allowing for $
transition temperature T~1 becomes
' 1/2
fluctuations in the xy plane, we get an additional term
5(5 J1 J2)
1

3 in our "free-energy" functional (4. 12)


TC1 =5J1 J2+ 4 m(J1+2Jq)
I 2
d@ d&
h, = J, W' dx
+ (5.1)
(4.31) dy

which has also been plotted in Fig. 6 (upper broken The excitations in the soliton lattice are the phasons
lines). At J~ = 0.55J1 and 0.6J1 the agreement is in complete analogy with the phonons in an atomic
very good (within 1'/0), whereas, when J2 is closer to lattice. The phason modes describe shifts 0(x,y, z) of
,1 J1, there is some discrepancy. This is due to the the soliton lattice in the z direction (Fig. 12}
insufficiency of our numerical procedure near T, . y (z) @0(z e(x, y, z)) (5.2}
The calculation explains the stability of the q = ,
1

When Eq. (5.2) is inserted into Eq. (5.1}, Hq takes


phase and the associated bulging of the phase di-
the form
agram. Figure 10 shows the soliton solution for
J2/J~ = 0.6 compared with the numerical solution.
1
I
2 1

d8 dg 0
Again, it is evident that the soliton gives a good dx dy dz
description. Figure 11 shows the commensurate
structure just below TC1. The small disagreement
with the numerical solution is due to the fact that we J &28tu
1 dr
&
dH
+ d8 (5.3)
pL dx dy
have ignored the coupling to higher harmonics by
truncating our Landau expansion.
In principle, it can be argued that because of the x{y)
higher-order terms any commensurate phase becomes 8 (x, y)
stable in a finite temperature interval. This is, in
fact, the argument used by Dzyaloshinsky. ' The
stepwise behavior can thus be understood as a conse-
quence of umklapp terms in the Landau expansion.
In Sec. V the effects of fluctuations will be con-
sidered, and it will be argued that the mean-field pic-
ture is qualitatively correct at nonzero temperatures.

tt l i tt tt i l tt alt l tt
'l Z-
2
t l

FIG. 11. Commensurate q = structure


1
which is stable
6
just below TC1 for J2/J1=
0.6. The broken arrows indicate FIG. 12. Nonrigid solitons (domain walls). H(x, y) indi-
the numerical solution. cates the local shift of the domain wall in the = direction.
5306 PER BAK AND J. VON BOEHM 21

%'e have ignored the exponentially weak interactions mensurate phase is


between the solitons so there is no energy associated d

with 8 gradients in the z direction. The energy per Fr ='


='iccA2 8
domain wall is

dd, 8d
p
fd d
dd
dx
d dd
dy
(5.4) +,Sm
b In,vr2+b +u'In,
b++
7r2
q (5 II)
Transforming to q space,
1 for q 0
8(x, y) = dqdq exp [i (qx+ qy) ] q The phason entropy thus gives a negative contribu-
2n.
tion to the domain-wall free energy. The commensu-
(5.5) rate phase is unstable when the quantity in the curly
brackets is negative. The CI transition takes place
the deformation energy can be expressed in terms of when the domain-wall energy is zero. This happens
the phason modes at a lower temperature than the one given by the
mean-field theory. For a given commensurate phase
JtA'8tu the variables in Eq. (5. 11) are all known functions of
Eg= dqdq (q2+q2) t)2 . (5.6)
277 p JI, J2, and T.
At first, let us consider the case p =4 to study the
The energy of a single phason mode is thus
stability of the q = 4 phase. Since we know all the
I

8JtA tu ~ 2) quantities in Eq. (5. 11), the transition temperature


(5.7)
p TcI can be found from Eq, (5. 11) by iteration. We
tabulate Tct for a few values of J2lJt in Table I. The
In the commensurate phase the phason energy has a
effect of the fluctuations is to suppress TcI, but the
gap at q =0 corresponding to phase oscillations commensurate phase remains stable at low enough
around the minimum of the vcos p potential, and for
temperatures. Note that TcI does not approach T,
q ~0 (as TcI does) as J2 increases. Near T, the phason
'
Ep(q)= Ep(q)+ , cA'up' (5.8) entropy destroys the tendency towards commensurate
ordering.
At this point, a word about the formalism is impor- The transition line indicating the temperature
where the q = 4 phase becomes unstable is shown in
I
tant. Strictly speaking, what we need to calculate the
effects of fluctuations is not a Landau expansion of
the free energy, but a Hamiltonian giving the energy
Fig. 13. For J2/Jt )
0.6, Tc~ was calculated from
Eq. (5.11), for smaller values the transition tempera-
of any phase fluctuation. A way to derive such a ture was estimated by adding the entropy term to the
Hamiltonian from our original spin Hamiltonian is
through the Hubbard-Stratonovitch transformation. " numerical solution for the ground-state energy. The
corresponding transition temperatures are very low,
Application of this transformation yields a Hamiltoni- and the effect of the phasons is insignificant.
an which is essentially isomorphous to Etl. (4.5). The Selke and Fisher have calculated the transition
grourrd state of this Hamiltonian is approximately the line by means of a Monte Carlo method. Their cal-
soliton lattice with energy Eo= F given by Eq. (4.20) culation is in quantitative agreement with our result
near the commensurate phase. The total free-energy based on the soliton-phason picture. By calculating
density difference including the phason modes is the phason entropy to higher order in q, one finds
r terms of the form
F = Eo+
T
dqdq~ ln F3= Tq (5. 12)

Eo+
L 2m ~ de dq~ ln 2,
q2+b
(5.9) TABLE I. Values of TCI for several J2/JI.

~here J2/JI Tc
2
cp 8
(5. 10)
2JI 07 5.4 3.64 2. 57
Replacing the integration area with a circle with ra- I 6 4. 77 3.36
dius n we find that the total free energy of the in-
2 8 7.38 5.24
commensurate phase relative to that of the com-
21 ISING MODEL WITH SOLITONS, PHASONS, AND. . . 5307

ton picture. The most important result is that the


mean-field picture will hold even near T, , and "the
devil's staircase" will survive. The width of the
high-order commensurate phase (in J2/Jl space) goes
as Jv~ (T, T)'v "i near the critical temperature.
For p =6, the (mean-field) width is linear in T, T,
whereas for p ) 6 the width decreases more rapidly.
The phase diagram must take the form shown in
Fig. 13. Near T =0 an infinite number of phases ex-
ists. At not too high temperatures the computer
results have been used to construct the diagram, and
0.2 0.3 05 06 07
near T, the qualitative arguments discussed in this
section have been applied. The shaded areas indicate
FIG. 13. Resulting phase diagram for the Ising model regimes where higher-order commensurate phases are
with competing interactions. stable. Since the widths decrease at ( T, T)'
the commensurate phases will not fill up the whole

in two dimensions. "


The addition of this term to Eq.
phase diagram despite the fact that the commensurate
phases are infinitely close. It is thus possible to have
(5.1'7) gives a different temperature dependence of a truly incommensurate phase. The possibility of
the wave vector near Tct such behavior has been noted by Pokrovsky' and Au-
bry in different contexts. In the case p =6 we have
also assumed "mean-field" behavior near T, although
However, there is no reason to expect corrections to our calculations indicate that this may not be correct.
Eq. (4.22) for the three-dimensional system con-
sidered here. '
Now, we consider the q = 6 ("three-up-three- VI. CONCLUDING REMARKS
down") phase, which was found to be stable for J2
In this paper we have analyzed analytically and nu-
near ,Jt in a rather large region of the phase di-
t
merically the spatially modulated phase of an Ising
agram extending to T = T, , giving rise to the bulging model with competing interactions. The phase di-
of the phase diagram. By inserting Eqs. agram turns out to be much more complicated than
(4.26) (4.29) into Eq. (5.11) with p =6 we can previously anticipated: There seems to be an infinity
again determine Tcl as a function of & (or J2/Jl). of commensurate phases. Experimentally, it has
The highest temperature at which this phase can pos- been found that some modulated magnetic systems
sibly be stable is obtained for 5=0 (J2= Jl). 1
We have several commensurate phases. ' " The widths
find that above a critical temperature, T=2.
7 Jt, the of the high-order commensurate phases must neces-
commensurate phase is not stable for any value of J2/Jl sarily be very small, and there is no possibility of
since Fq in Fq. (5. 11) is then always negative. The positively identifying the "devil's staircase" experi-
bulging of the phase diagram is conserved, but takes mentally. Near T, all commensurate phases are nar-
place at a much lower temperature. For row and the system is indistinguishable from an in-
J2/Jl = 0 55 Tcl = 2.6Jl. commensurate unpinned one. For a model including
The stability of higher-order commensurate phases, quantum effects it has been argued that quantum
q = n/p, is governed by higher-order terms. The fluctuations will destroy the staircase even at T =0.
domain-wall energy is proportional to
3' jv A'3 'v "i
( T, T) "+v't, whereas the
domain-wall entropy term is proportional to ACKNOWLEDGMENTS
v lnv ( T, T)' "
. If critical fluctuations are tak-
en into account the exponents will be modified. For %e would like to thank D. Mukamel for discus-
the high-order commensurate phases (p
tropy term is unimportant compared to the energy
)
6) the en- sions on his work on finite temperature effects on
modulated systems and W. Selke for keeping us in-
term as T T, . The cases p = 4 and p = 6 (where formed on his Monte Carlo work with M. E. Fisher.
the lnv factor destabilizes the commensurate phase) %'e are grateful to T. Schultz and S. Kirkpatrick for a
are thus special. The most probable reason is that for careful reading of the manuscript and for constructive

T T, amplitude fluctuations are destroying the soli- comments.
5308 PER BAK AND J. VON BOEHM

'See, for example, W. C. Koehler, in Magetic Properties o/' ' P. Bak, in Solitos ad Codesed MatteI Physics. edited by
Rare Earth Metals, edited by R. J. Elliott (Plenum, New A. R. Bishop and T. Schneider (Springer, New York, 1978).
York, 1972), p. 81. ' P. Fisher, B. Lebech, G. Meier, B. D. Rainford, and O.
~R. Comes, S. M. Shapiro, G. Shirane, A. F. Garito, and A. Vogt, J. Phys. C 11, 345 (1978).
J. Heeger, Phys. Rev. Lett. 35, 1518 (1975). ' J. Rossat Mignod, P. Burlet, J. Villain, H. Bartholin, Wang
~J. M. Hastings, J. P. Pouget, G. Shirane, A. J. Heeger, N. Tseng-Si, D. Florence, and O. Vogt, Phys. Rev. B 16, 440
D. Miro, and A. G. MacDiarmid, Phys. Rev. Lett. 39, (1977).
1484 (1977). ' J. von Boehm and P. Bak, Phys. Rev. Lett. 42, 122
4U. Enz, Physica (Utrecht) 26, 698 (1960). (1978).
R. J. Elliott, Phys. Rev. 124, 346 (1961). ' W. Selke, Z. Phys. B 29, 133 (1978); R. M. Hornreich, R.
6M. Habenschuss, C. Stassis, S. K. Sinha, H. W. Deckmann, Liebermann, H. G. Schuster, and W. Selke (unpublished).
and F. H. Spedding, Phys. Rev. B 10, 1021 (1974), and W. Selke and M. E. Fisher, Phys. Rev. B 20, 257 (1979);
references cited therein. Int. J. Magn. Magn. Mater. (to be published).
71.E. Dzyaloshinsky, Collective Properties of'Phvsical Svstets ~'P. Bak and J. von Boehm, Bull. Am. Phys. Soc. 24, 429
(Academic, New York, 1974), p. 143. (1978); J. Appl. Phys. (to be published).
V. L. Pokrovsky, Solid State Commun. 26, 77 (1978). W. Selke and H. Schuster (unpublished).
9S. Aubry, in Solitos ad Codesed Matter Physics, edited H. J. Mikeska, J. Phys. C 11, L29 (1978); J. K. Kjems and
by A. R. Bishop and T. Schneider (Springer, New York, M. Steiner, Phys. Rev. Lett. 41, 1137 (1978).
1978). ~4P. G. de Gennes, Solid State Commun. 6, 163 (1968).
' and J. H. Van der Merwe, Proc. R. Soc. Lon-
F. C. Frank ~5R. L. Stratonovich, Sov. Phys. Dokl. 2, 416 (1958); J.
don Sect. A 198, 205, 216 (1949); 200, 125 (1949). Hubbard, Phys. Rev. Lett. 3, 77 (1959).
"B. Mandelbrot, FIactals: Fon, Chattge ad Ditesio D. Mukamel (unpublished).
(Freeman, San Francisco, 1977). V. L. Pokrovsky and A. L. Talapov, Phys. Rev. Lett. 42,
' W. L. McMillan, Phys. Rev. B 14, 1496 (1976); 16, 4655 65 (1979). For an excellent review, see J. Villain, in
(1977). Proceedings of the NATO Advanced Study Institute,
' M. Luban, D. Mukamel, and S. Shtrikman, Phys. Rev. B Geilo, Norway, April, 1979 (unpublished).
10, 360 (1974). 8P. Bak and H. Fukuyama, Phys. Rev. B (to be published).
' P. Bak and V. J. Emery, Phys. Rev. Lett. 36, 978 (1976).

You might also like