You are on page 1of 18

PHYSICAL REVIEW B VOLUME 55, NUMBER 14 1 APRIL 1997-II

Boundary-condition effects in anharmonic lattice dynamics: Existence criteria


for intrinsic localized modes from extended-mode properties
D. Bonart,* T. Rössler, and J. B. Page
Department of Physics and Astronomy, Arizona State University, Tempe, Arizona 85287-1504
~Received 20 August 1996; revised manuscript received 26 November 1996!
Recent theoretical studies of periodic one-dimensional anharmonic lattices with standard periodic boundary
conditions established fundamental connections between the existence of intrinsic localized modes ~ILM’s! and
the stability of the extended lattice modes into which they evolve with decreasing amplitude. While the
odd-order anharmonicity drops out of the equations of motion for the extended modes within these boundary
conditions, it nevertheless produces an amplitude-dependent period-averaged ‘‘dynamical stress’’ across the
supercell boundaries. Here, we allow the supercell length to adjust so as to eliminate this stress and find that
the frequency vs amplitude curves for the anharmonic extended modes are markedly changed for a variety of
realistic nearest-neighbor interactions, whereas highly localized ILM’s are little affected. Nevertheless, in all
cases ILM existence remains intimately connected to an instability of the associated extended mode. Further-
more, the use of zero-stress periodic boundary conditions now allows one to predict the spatial extent of an
ILM from extended-mode stability properties in lattices with odd-order anharmonicity. Most importantly, for
the zero-stress periodic boundary conditions we obtain an additional ILM existence criterion, based on simple
dynamical properties of the unperturbed related extended mode. Since well-localized ILMs are independent of
the specific choice of boundary conditions, our results yield promising tools for ILM predictions in real
systems. @S0163-1829~97!01014-X#

I. INTRODUCTION The studies of Ref. 6 used standard periodic boundary


conditions, in which the supercell length Na is fixed. Here
In striking contrast to the familiar plane-wave vibrational N is the number of particles and a is their equilibrium spac-
excitations in harmonic periodic lattices, it is known theoreti- ing. These boundary conditions are ubiquitous in solid-state
cally that lattices with nonlinear interactions can support physics and are convenient for ‘‘bulk’’ properties, i.e., those
highly localized stationary vibrations.1,2 These so-called ‘‘in- which are independent of the detailed nature of the bound-
trinsic localized modes’’ ~ILM’s! can be centered on any site aries. This should hold for ILM dynamics, provided the
and are optic-mode-like, in that adjacent particles move p modes are highly localized relative to Na. However, the as-
out of phase. In the low-amplitude limit, ILM’s broaden spa- sociated anharmonic ExM’s will be shown here to be af-
tially and for some cases reduce to stationary envelope soli- fected by the boundary conditions in a nontrivial manner, in
tons of the sort described more than 20 years ago for weakly contrast to the situation in harmonic lattices, raising the
anharmonic one-dimensional lattices,3,4 whereas such a de- question of the role of boundary conditions in the ExM-ILM
scription does not apply for large-amplitude highly localized interrelations detailed in Ref. 6. In the following we use
ILM’s.5 These unusual excitations have elicited numerous more general periodic boundary conditions to study these
studies over the last several years, most of which are focused connections, and we will show that the earlier results are
on one-dimensional ~1D! lattices and employ a mixture of generalized in interesting and often unexpected ways. More-
analytic techniques and direct numerical simulations. Vari- over, we will obtain a second ILM existence criterion based
ous aspects of ILM’s are surveyed in Ref. 2. on properties of the ILM’s associated ExM. This criterion is
For the experimental establishment of ILM’s, it is impor- not based on stability properties and is complementary to the
tant to develop practicable existence criteria. Reference 6 first. It appears promising for the a priori prediction of ILM
obtained a fundamental connection between the existence of existence in real systems.
ILM’s and the stability of the extended lattice modes To better motivate the generalized periodic boundary con-
~ExM’s! into which they spatially broaden with decreasing ditions, we next outline some additional quantitative aspects
amplitude. Monatomic lattices were studied, for interactions of the results of Ref. 6. Within specific models, it was shown
ranging from simple quadratic plus quartic nearest-neighbor that the ExM stability analysis can go beyond ILM existence
potentials (k 2 ,k 4 ) to realistic full potentials V(r), e.g., Born- prediction and yield quantitative information about ILM lo-
Mayer plus Coulomb, Lennard-Jones, etc. It was concluded calization. For the monatomic lattices studied there, the rel-
that for a given mode amplitude, ILM’s only exist when their evant ExM is the zone boundary mode ~ZBM!, in which
associated ExM’s are dynamically unstable with purely real adjacent particles have equal and opposite displacements.
instability growth rates. The practical significance of this re- This mode’s spatial periodicity allows a Fourier decomposi-
sult is that the stability analysis is a linear problem and is tion of the instability perturbations, with the result that the
therefore simple to implement as a useful prelude to the wave vector of the fastest-growing perturbation introduces a
more difficult problem of direct numerical searches for characteristic length scale. For the simple case of nearest-
ILM’s. neighbor (k 2 ,k 4 ) interactions, the ZBM is unstable and

0163-1829/97/55~14!/8829~18!/$10.00 55 8829 © 1997 The American Physical Society


8830 D. BONART. T. RÖSSLER, AND J. B. PAGE 55

ILM’s exist at all amplitudes, and the ZBM instability length We will discuss 1D diatomic lattices with nearest-
scale matches the spatial width of the ILM having the same neighbor potentials. Our numerical studies will be detailed
frequency as the ZBM. Moreover, finite-time molecular dy- for the case of (k 2 ,k 3 ,k 4 ) interactions, with parameter val-
namics ~MD! simulations showed that the unstable ZBM ues appropriate to realistic potentials. Results for other cases,
evolves with time into a periodic array of localized excita- such as full potentials V(r), or the monatomic limit, will be
tions having the same frequency as the original ZBM, but given at appropriate points. Several studies of ILM’s in 1D
larger amplitudes. Note that the finite-time MD simulations diatomic lattices have been carried out ~see Refs. 8–10!.
justify making the above length-scale comparison at the Here, our focus is on the effects of boundary conditions on
same ZBM and ILM frequency. This comparison is facili- ExM-ILM interrelations, and it turns out that the
tated by the fact that the ZBM instability and the ILM exist (k 2 ,k 3 ,k 4 ) diatomic lattice offers a particularly fruitful
over identical frequency ranges. model system, exhibiting a richer variety of ExM-ILM be-
Real interatomic potentials are dominated by cubic anhar- havior than does a model using full potentials. Indeed, the
monicities, and it was also shown in Ref. 6 that the addition results for full potentials are readily discussed within the
of nonzero k 3 changes the detailed manner in which the context of the results obtained here. In addition, we will see
above existence criterion is realized: in monatomic that the (k 2 ,k 3 ,k 4 ) model studies reveal an interesting and
(k 2 ,k 3 ,k 4 ) lattices with realistic parameter values, ZBM in- unexpected ILM spatial broadening behavior as the mode
stability and ILM existence occur above a finite threshold amplitude is changed.
amplitude, below which the ZBM is stable and ILM’s are not It is customary to classify ILM’s according to whether
found. The ZBM instability perturbation again introduces a their frequencies are above the maximum harmonic fre-
characteristic length scale, but quantitative comparisons with quency of the lattice or are within the gap between the acous-
the corresponding ILM length scales are now difficult, for tic and optic harmonic phonon bands. However, we will see
two reasons: ~1! for the standard periodic boundary condition that for either of the periodic boundary conditions to be used
lattices studied in Ref. 6, the frequency ranges of the un- here, this classification is not clear cut, e.g., ILM’s can occur
stable ZBM’s and the corresponding ILM’s no longer fully within the harmonic phonon bands. We feel that the more
coincide and ~2! finite-time MD runs show that the unstable useful classification is in terms of the ILM’s associated ExM.
ZBM’s again break up into periodic arrays of localized ex- Thus we will discuss ‘‘optical zone center mode’’ ILM’s,
citations, but the frequencies as well as amplitudes change in which broaden into an anharmonic ExM having the pattern
the process. The underlying cause of these differences from of the harmonic k50 optic phonon mode, and ‘‘optical zone
the (k 2 ,k 4 ) case is the amplitude-dependent local static dis- boundary mode’’ ILM’s, which broaden into an anharmonic
tortion which accompanies ILM’s in lattices with nonzero version of the harmonic k5 p /2a optical zone boundary
k 3 ,7 and the effect is to prevent the simple length scale com- mode. These two classes exhibit quite different behaviors,
parison outlined above. and it is important to consider them both. The results of Ref.
Within the standard periodic boundary conditions of Ref. 6 are recovered from the StdPBC results for the first class, in
6, the symmetry of the ZBM displacement pattern renders the monatomic limit.
the mode’s equation of motion independent of odd-order an- The following section describes our theoretical frame-
harmonicity. Nevertheless, such anharmonicity strongly af- work, after which we discuss boundary conditions and define
fects the mode’s instability properties, owing to the symme- dynamical stress. Our numerical results are detailed in Sec.
try breaking introduced by the instability perturbations. IV, which is split into two parts according to the ExM in-
Moreover, even without an instability perturbation, the ZBM volved. We first consider the anharmonic optical zone center
in a lattice with odd-order anharmonicity sets up a ‘‘dynami- mode, discussing its dynamics, stability and related ILM be-
cal stress,’’ i.e., a nonzero period-average force across any havior, after which the same topics are discussed for the
imaginary plane perpendicular to the chain. Its presence in a anharmonic optical zone boundary mode. Section V com-
real lattice would drive a uniform expansion, dependent on prises a detailed discussion and synthesis, including the ad-
the ZBM amplitude, such that the dynamical stress vanishes. ditional existence criterion mentioned earlier. The paper is
The consequences of such a condition will be studied here, concluded in Sec. VI, and two appendices provide additional
by allowing the supercell length to change with mode ampli- details.
tude such that the dynamical stress remains zero. We will
call these boundary conditions ‘‘zero-stress periodic bound- II. THEORETICAL BASIS
ary conditions’’ ~ZSPBC’s!, in contrast to the standard peri-
odic boundary conditions ~StdPBC’s! used earlier. Employ- We discuss longitudinal motion in a 1D lattice of particles
ing ZSPBC’s rather than free-end boundary conditions having anharmonic interactions V(r) between nearest neigh-
preserves the simplicity of the ILM-related anharmonic bors. For an infinite lattice, the Hamiltonian is
ExM’s, permitting both a meaningful comparison with the
StdPBC case and, more importantly, an extension of the
ExM instability/ILM length-scale comparisons outlined
above to the case of nonzero k 3 . Since the properties of
H5 (n F p 2n
2m n
G
1V ~ r n 2r n21 ! , ~1!

well-localized ILM’s are independent of the boundary con-


ditions, the use of numerical model studies of ExM’s within where $ m n % gives the masses and $ p n (t) % and $ r n (t) % are the
the ZSPBC’s to predict ILM existence and properties tran- instantaneous momenta and positions. We will mainly con-
scends these boundary conditions and is thus applicable to sider diatomic lattices of alternating masses m and M .m,
real systems. and V(r) is either a full potential such as Born-Mayer plus
55 BOUNDARY-CONDITION EFFECTS IN ANHARMONIC . . . 8831

Coulomb ~BMC!: V(r)5le 2r/ r 2q 2 /r, or the corresponding change, which vanishes for any periodic motion. Moreover,
fourth-order Taylor series expansion: since F̄ n,n11 52F̄ n11,n , we see from Eq. ~5! that the time-
average force on a particle due to either one of its neighbors
1 1 1 is independent of the lattice site, i.e., F̄ n,n21 [F̄.
V ~ r ! 5V ~ a ! 1 k 2 ~ r2a ! 2 1 k 3 ~ r2a ! 3 1 k 4 ~ r2a ! 4 . The preceding discussion applies both to localized and
2 3 4
~2! extended stationary modes. We now focus on ExM’s which
have the same dynamic displacement patterns with and with-
Here a is the static-lattice nearest-neighbor equilibrium sepa- out anharmonicity present. In particular, for a diatomic lat-
ration, obtained by minimizing the total potential energy. For tice we study the anharmonic versions of the harmonic
our assumption of nearest-neighbor interactions, a occurs at k50 optical zone-center mode ~OZCM! and the
the minimum of V(r). k5 p /2a optical zone-boundary mode ~OZBM!. The
To treat a finite lattice of N particles, we will use periodic respective dynamic displacement patterns $ c n % are
boundary conditions r n1N 5r n 1L, where L is the repeat A( . . . ,1,2m/M ,1,2m/M , . . . ) and A( . . . ,1,0,21,0,1,0,
length; these are realized in Eq. ~1! by summing n from 1 to 21,0, . . . ), with A being the amplitude of the light particles
N and defining r 0 [r N 2L. A major focus of this paper is on in each case. In some instances we will also give our OZCM
the effects of dynamically induced changes of L in anhar- results for the case M 5m of a monatomic lattice; the static
monic lattices. Thus L will generally not be equal to its static equilibrium lattice constant is then a rather than 2a and the
equilibrium value Na. In passing, we note that free-end mode is the monatomic lattice ZBM, with the dynamic dis-
boundary conditions would correspond to r 0 [r 1 2a. placement pattern A( . . . ,1,21,1,21, . . . ).
As discussed in previous work,2,6 the ‘‘rotating wave ap- With anharmonicity present, each of the above three ExM
proximation’’ ~RWA! has proven convenient for giving ac- dynamic displacement patterns can produce static displace-
curate stationary solutions to the equations of motion derived ments consisting of just a simple uniform expansion of the
from Eq. ~1!, for both spatially extended and localized an- lattice, for a broad class of realistic potentials and a large
harmonic modes. In the RWA, the motion of particle n is range of dynamic displacements. Indeed, we note that for all
assumed to be of the form three modes, the relative dynamic displacements
Dc n [c n 2c n21 between adjacent particles can be written as
r n ~ t ! 5c n cos~ v t ! 1b n 1na. ~3! Dc n [(21) n 2Ã, where à is an effective mode amplitude,
given by (11m/M )(A/2), A/2, and A, for the diatomic
After inserting this ansatz into the equations of motion, we
OZCM, the diatomic OZBM, and the monatomic ZBM, re-
multiply the resulting equations by either cos(vt) or unity
spectively. Within the RWA we then obtain
and average over a single period. This yields a system of
2N coupled nonlinear equations for the dynamic displace-
ments $ c n % and the static displacements $ b n % :
F̄ ~ Ã ! 52
1
2p
E 2p
d f V 8 ~ 2Ãcosf 1Db n 1a ! , ~6!

E
0
1 2p
05m n v 2 c n 2 d f cosf @ V 8 ~ r n 2r n21 !
p 0 where Db n [b n 2b n21 gives the relative static displace-
ments. Since we have seen that F̄ is independent of n, the
2V 8 ~ r n11 2r n !# , ~4a! right-hand side of Eq. ~6! cannot depend on n. A simple
uniform expansion Db n [D satisfies this constraint. To
052
1
2p
E0
2p
d f @ V 8 ~ r n 2r n21 ! 2V 8 ~ r n11 2r n !# ,
check for other possible static displacement patterns, we as-
sume that Db n is a continuous function of à and that the
~4b! interparticle potential V(r) has only a single minimum ~at
r5a). It can then be shown that for standard realistic poten-
where f 5 v t and V 8 (r n 2r n21 )[(dV/dr) u r5r n 2r n21 . Once tials ~e.g., BMC, Lennard-Jones!, the relative static displace-
the boundary conditions are specified, these equations can be ments Db n cannot be site dependent over a large range of
solved using standard numerical routines; the solutions in dynamic displacements; specifically, they are site indepen-
conjunction with Eq. ~3! constitute the RWA. Wherever pos- dent for 2Ã1D(Ã)<0.30a ~see Ref. 12!. Accordingly, we
sible we will compare our RWA predictions with results ob-
will only consider uniform expansions a→â5a1D when
tained by direct MD simulations of the full equations of mo-
discussing the three ExM’s described above. The specific
tion for the Hamiltonian ~1!, using a fifth-order Gear
predictor-corrector method.11 functional dependence D(Ã) depends on the choice of
For future reference we note that Eq. ~4b! can be rewritten boundary conditions, as discussed in the following section.

05F̄ n,n21 1F̄ n,n11 , ~5! III. BOUNDARY CONDITIONS


AND DYNAMICAL STRESS
where F n,m is the force on particle n due to particle m and
the overbar indicates a time average over a mode period. For bulk properties of lattices, it is traditional and conve-
This equation is actually independent of the RWA, as is eas- nient to employ Born-von Karman periodic boundary condi-
ily seen: the right-hand side of Eq. ~5! is proportional to the tions. As noted above, we do this by requiring
time-integral of the total force on a particle over one period, r n1N 5r n 1L for an N-particle lattice, where L is the repeat
and is thus proportional to the corresponding momentum length. For purely harmonic vibrations, the repeat length is
8832 D. BONART. T. RÖSSLER, AND J. B. PAGE 55

just the static lattice value Na, but with anharmonicity function of the mode amplitude: â(Ã)5a1D(Ã), with the
present, we will allow L to vary, contrasting the results so functional form depending on the particular mode. Such
obtained with those for L5Na. changes in the lattice constant for the case of small-
amplitude ExM’s in monatomic (k 2 ,k 3 ,k 4 ) lattices have
A. Standard periodic boundary conditions been included in Refs. 4 and 13. Here, we will show that the
ZSPBC’s alter the interrelations between ILM’s and ExM’s,
Standard periodic boundary conditions with L fixed at compared with the StdPBC results obtained in Ref. 6.
Na are denoted as StdPBC’s, whether they are applied to In Appendix B, it is shown that the ZSPBC’s follow from
harmonic or anharmonic lattices. For the solution of Eqs. ~4! minimizing the period-average supercell potential energy.
within the RWA, these are realized by setting c n1N 5c n and Thus they are a dynamical analog to the use of periodic
b n1N 5b n . StdPBC’s were employed throughout our earlier boundary conditions in computing the equilibrium lattice
study of the interrelations between ILM’s and ExM’s in Ref. constant, where one would minimize the static potential en-
6. ergy of the supercell.
As discussed in Sec. II, for an anharmonic lattice vibrat- Similar to a lattice with ZSPBC’s, a lattice having free
ing in any of the three ExM’s introduced there, the relative ends would allow for a relaxation of the dynamical stress by
static displacements Db n 5D are site independent. Hence for static distortion. However, the presence of free ends breaks
StdPBC’s, D vanishes identically, and the period-average the periodicity, so that the simplicity of the ExM mode pat-
separations of adjacent particles remain equal to the static terns is lost, preventing a straightforward comparison of the
equilibrium value a, even with one of these ExM’s present. ExM-ILM interrelations for zero and nonzero dynamical
An additional consequence of the StdPBC’s is that the stress. Moreover, we find that the lack of a common ampli-
odd-order terms of the Taylor series of the interaction poten- tude range for the free-end ExM’s and ILM’s prevents us
tial do not occur in the equations of motion for any of the from obtaining useful ExM-ILM results of the sort obtained
three ExM’s. This follows directly from Eq. ~4a! and the here using ZSPBC’s. Nevertheless, in the limit of highly
symmetry of the mode patterns, as shown explicitly in Ap- localized ILM’s, the ZSPBC’s and free-end boundary condi-
pendix A. tions become equivalent. Hence the latter can be used in MD
simulations of these modes, allowing one to avoid the diffi-
cult task of implementing the zero-stress condition in MD.
B. Dynamical stress and zero-stress periodic
boundary conditions
IV. RESULTS
With vibrations present in our 1D system, and for the case
of general two-body interactions, we define the dynamical We have studied the interrelations between ILM’s and the
stress across an imaginary plane to be the period-average anharmonic ExM’s introduced in Sec. II, as a function of the
force on the particles on one side due to the particles on the periodic boundary conditions and for a wide variety of inter-
other side. For the present case of interactions restricted to actions in diatomic lattices. Many of the results are qualita-
nearest neighbors, this quantity is just the period-average tively similar for realistic full potentials and their
force on a particle due to one of its nearest neighbors. Note (k 2 ,k 3 ,k 4 ) series expansions. Moreover, the latter case often
that in addition to the forces between particles that move, our allows one to proceed further analytically, and it yields a
definition of dynamical stress includes the forces between more diverse range of ExM-ILM behavior. Since the results
particles which are essentially at rest, e.g., between adjacent for full potentials are easily described within the context of
particles far from the center of a highly localized ILM. If the the (k 2 ,k 3 ,k 4 ) results, we will detail the latter case, pointing
interaction potential contains only positive terms of even or- out important differences and similarities for full potentials.
der, there are no static displacements and no dynamical Unless explicitly stated otherwise, we take k 3 a/k 2 5210,
stresses, and the StdPBC’s are appropriate for ILM’s and the k 4 a 2 /k 2 550, and m/M 51/4. These force constant ratios are
ExM’s we are treating. However, for realistic full potentials, appropriate for realistic potentials and were used in our ear-
e.g., BMC, Lennard-Jones, or Morse, the odd-order terms are lier studies for StdPBC’s in monatomic systems.6 Further-
important, and they result in nonzero dynamical stress across more, the corresponding (k 2 ,k 3 ,k 4 ) potential function has
the supercell boundary for StdPBC’s. For the case of highly only a single minimum, as is the case for realistic full poten-
localized ILM’s, this stress vanishes in the N→` limit of a tials. Frequencies will be given in terms of the k50 har-
large number of particles, but for the ExM’s, it does not. monic frequency v max 5 A2k 2 mM /(m1M ).
Accordingly, with odd-order anharmonicity present, we
modify the StdPBC’s by allowing the periodicity length L to
A. Optical zone center mode and related ILM’s
vary such that the dynamical stress vanishes across the su-
percell boundary. As a result L is changed from its static 1. Dynamics of the OZCM
equilibrium value Na to a value which depends on the am-
a. Standard periodic boundary conditions. As shown in
plitude and nature of the mode ~ILM or ExM!. As noted in
Appendix A, k 3 plays no role in the dynamics of the OZCM
the introduction, these zero-stress periodic boundary condi-
within StdPBC’s, owing to the symmetry of the dynamic
tions are abbreviated ZSPBC’s. For the solution of Eqs. ~4!
displacement pattern. The RWA frequency is determined by
within the RWA, they are implemented by setting

S D
c n1N 5c n and b n1N 5b n 1L2Na.
With one of our three ExM’s present, the ZSPBC’s deter- v StdPBC 2 k4
5113 Ã 2 , ~7!
mine D(Ã) and hence the nearest-neighbor distance as a v max k2
55 BOUNDARY-CONDITION EFFECTS IN ANHARMONIC . . . 8833

FIG. 2. Fractional lattice constant change D/a vs normalized


FIG. 1. Frequency vs amplitude curves for the optical zone- amplitude A/a of the OZCM for the 40-particle diatomic lattice of
center mode ~OZCM! and the related intrinsic localized mode Fig. 1, with ZSPBC’s. The dashed line corresponds to the large-
~OZCM-ILM! in a 40-particle diatomic (k 2 ,k 3 ,k 4 ) lattice, for the amplitude asymptotic value D ` 52k 3 /3k 4 . This graph also applies
parameter values given in the text. The frequency and amplitude are to the optical zone boundary mode ~OZBM!, provided the abscissa
scaled to the maximum harmonic lattice frequency v max and the values are multiplied by (11m/M ); the value of D ` is the same for
static lattice nearest-neighbor distance a, respectively. The solid both modes.
curves ~a! and ~a 8 ) are for the OZCM with standard periodic
boundary conditions ~StdPBC’s! and zero-stress periodic boundary
conditions ~ZSPBC’s!, respectively. The dashed curves ~b! and
unique. For the k 3 ,0 case considered here, D increases
~b 8 ) are for the OZCM-ILM’s under the same two boundary con- monotonically, eventually saturating at an asymptotic value
ditions. For these ILM curves the plotted amplitudes are those of given by D ` 52k 3 /3k 4 . This behavior is shown in Fig. 2.
the mode’s central particle, which is a light mass. The thin horizon- Relative to the new equilibrium positions, the force constants
tal lines locate the top and bottom of the harmonic optical phonon are renormalized according to k̂ 2 [k 2 12k 3 D13k 4 D 2 ,
band, indicated by the vertical bar. The top of the harmonic acoustic k̂ 3 [k 3 13k 4 D and k̂ 4 [k 4 . Note that k̂ 3 goes to zero as D
band is at v 50.45v max , below the range of the graph. The crosses approaches its saturation value. An earlier study14 showed
and open circles are the results of MD measurements of mode fre- that the application of static external stress of the right mag-
quencies for StdPBC’s and free-end boundary conditions, nitude to a finite diatomic (k 2 ,k 3 ,k 4 ) lattice can produce a
respectively—for well-localized ILM’s, free ends should give the
static strain which eliminates cubic anharmonicity. The lat-
same results as ZSPBC’s, and our MD measurements and RWA
tice constant change predicted there is just D ` , which is
predictions differ by 2% at most. The two diamonds give the pre-
produced here by the presence of a large-amplitude OZCM.
dicted thresholds above which the OZCM is unstable for each of the
two periodic boundary conditions. Note that these thresholds are
Equation ~7! for the OZCM also applies for ZSPBC’s,
also the points where the two ILM curves join the OZCM curves. provided the force constants are replaced by their renormal-
ized versions given above. Thus the RWA squared frequency
for ZSPBC’s can be expressed as
where Ã5(11m/M )(A/2) and A is the amplitude of a light

S D
mass. Hence for StdPBC’s, the OZCM frequency increases
monotonically with amplitude starting from v max , as shown v ZSPBC 2 k̂ 2 k̂ 4 k3 k4
5 13 Ã 2 5112 D13 ~ D 2 1Ã 2 ! ,
by the solid line ~a! in Fig. 1. This curve is the diatomic v max k2 k2 k2 k2
lattice version of the solid curve in Fig. 8 of Ref. 6, which is ~9!
for a monatomic (k 2 ,k 3 ,k 4 ) lattice—recall that in the mon-
atomic limit the diatomic OZCM becomes the monatomic which, of course, can also be derived directly. From Eq. ~8!,
ZBM. For the case of realistic full potentials, the OZCM we find that D522Ã 2 k 3 /k 2 for small amplitudes. The cor-
v (A) curve exhibits the same qualitative behavior. responding limit for the squared frequency is
b. Zero-stress periodic boundary conditions. In contrast, ( v ZSPBC / v max) 2 511(3k 4 /k 2 ) @ 124k 23 /(3k 2 k 4 ) # Ã 2 ~see
when ZSPBC’s are used, the OZCM produces a changed Ref. 15!. We can thus distinguish two kinds of small-
period-average nearest-neighbor distance â(Ã)5a1D(Ã). amplitude behavior: for 4k 23 /(3k 2 k 4 ),1 the frequency in-
The requirement of vanishing dynamical stress yields an creases with amplitude, while for 4k 23 /(3k 2 k 4 ).1 it de-
equation for D(Ã): creases. For our force constants, this quantity is equal to
8/3, and we will thus focus on the second case, although

D1
k3 2 k4 3
k2
D 1 D 12Ã 2
k2
k3
k2
k4
S
13 D 50.
k2 D ~8!
some pertinent results for the first case will be summarized
in Sec. V. For small amplitudes, then, the frequency of the
OZCM moves downward into the harmonic optical phonon
Since we do not consider double-well potentials, the only band, as shown by curve ~a 8 ) in Fig. 1. With increasing
solution of this equation for Ã50 is D50. One can show amplitude, the positive term 3(k 4 /k 2 )(D 2 1Ã 2 ) in Eq. ~9!
analytically that with increasing amplitude, D remains gains importance, and at A50.063a the frequency reaches a
8834 D. BONART. T. RÖSSLER, AND J. B. PAGE 55

minimum value of v 50.91v max , beyond which it increases


monotonically. In contrast, for realistic full potentials V(r),
the presence of higher-order terms is found to prevent this
bottoming out, and the OZCM frequency descends through
the harmonic optical phonon band into the phonon gap.
Comparing the solid curves ~a! and ~a 8 ) in Fig. 1, we see
that the relaxation of the dynamical stress lowers the OZCM
frequency for a given amplitude. Furthermore, the zero-stress
OZCM has the lower energy at a given amplitude, as ex-
pected since the ZSPBC’s follow from minimizing the
period-average potential energy at a fixed amplitude ~Appen-
dix B!.

2. Stability of the OZCM


We have examined the stability properties of the
(k 2 ,k 3 ,k 4 ) OZCM using the RWA stability analysis de-
scribed in Refs. 6 and 16, with the result that the qualitative
stability properties are the same for the two types of periodic
boundary conditions. The stability analysis assumes infini-
tesimal displacement and phase perturbations having expo-
nential time dependence exp(lt), leading to an eigenvalue FIG. 3. Upper panel: The solid curve gives the predicted RWA
problem for the growth rates $ l % . Instabilities correspond to maximum instability growth rate of the OZCM as a function of the
l’s with positive real parts, and whenever these are found, normalized amplitude, for a diatomic (k 2 ,k 3 ,k 4 ) lattice with
we consider only the fastest-growing instability, associated StdPBC’s. The parameter values are given in the text. The dia-
with the eigenvalue l max having the maximum positive real monds give growth rate measurements obtained from MD simula-
part. In Ref. 6, it was shown that for the ZBM in a mon- tions for a 40-particle lattice. Lower panel: Wave vector of the
fastest-growing Fourier component of the instability perturbation as
atomic (k 2 ,k 4 ) or (k 2 ,k 3 ,k 4 ) lattice with StdPBC’s, the oc-
a function of the normalized amplitude. To achieve good resolution,
currence of instabilities with growth rates l max which are
the stability analysis for both panels was based on a wave vector
purely real is intimately connected with the existence of grid appropriate to a 2000-particle lattice.
ILM’s. Following Ref. 6, we term such instabilities ‘‘ILM
related,’’ and in the remainder of this paper we will only leads to a breakup of the mode into a periodic array of lo-
discuss ExM instabilities of this type.17 calized excitations whose spacing is very close to the pre-
For the small-amplitude OZCM with either StdPBC’s or ferred instability length.
ZSPBC’s, we do not find purely real l max’s, implying by For ZSPBC’s, the qualitative features of the curves of Fig.
analogy with the monatomic StdPBC studies of Ref. 6 that 3 are retained, but the ILM-related instability threshold am-
related ILM’s do not exist at small amplitudes. However, for plitude is different for the two periodic boundary conditions.
both periodic boundary conditions, we find that a finite In the limit of an infinite lattice and StdPBC’s, A thresh can be
threshold amplitude A thresh exists, above which the RWA sta- obtained analytically, following an argument analogous to
bility analysis predicts a purely real l max . The upper panel that given in Ref. 6 for the case of a ZBM in a monatomic
of Fig. 3 plots our predicted maximum growth rate as a func- (k 2 ,k 3 ,k 4 ) lattice. For the OZCM in an infinite diatomic
tion of the OZCM amplitude for StdPBC’s. The diamonds (k 2 ,k 3 ,k 4 ) lattice with StdPBC’s, we find that the instability
give the results of MD measurements based on the ‘‘projec- threshold criterion is
tion’’ method of Ref. 6, for a 40-particle lattice. The agree-
ment between the predicted and measured values is within
7%. By decomposing the instability perturbation into its
plane-wave Fourier components, we can obtain the wave
3L 4 1
9
2S D
11
m
M
2
L 24 >4L 23 , ~10!

vector (k p ) max of the fastest-growing component, as dis- where L 3 [k 3 A/k 2 and L 4 [k 4 A 2 /k 2 . The use of the renor-
cussed in Ref. 6 for the monatomic lattice ZBM. The lower malized force constants in place of (k 2 ,k 3 ,k 4 ) in Eq. ~10!
panel of Fig. 3 plots (k p ) max as a function of the OZCM yields the corresponding instability threshold criterion for the
amplitude for StdPBC’s. At the instability threshold ampli- ZSPBC version of the OZCM. However, due to the ampli-
tude A thresh50.120a, (k p ) max vanishes, and it increases to its tude dependence of the renormalized force constants, this
maximum allowed value of p /2a over a restricted range of criterion is not straightforward to solve analytically for
amplitudes. The corresponding wavelength 2 p /(k p ) max in- A thresh , in contrast to the StdPBC case. Nevertheless, a rather
troduces a preferred instability length scale at each ampli- lengthy analytic calculation shows that A thresh for the OZCM
tude, similar to that for the monatomic ZBM in Ref. 6. As in the infinite lattice with ZSPBC’s coincides exactly with
the amplitude of the OZCM increases from A thresh , the insta- the minimum of this mode’s frequency vs amplitude curve.
bility length scale decreases from infinity, attaining a maxi- The instability thresholds computed numerically for a 40-
mum value of 4a, after which it remains constant for in- particle lattice with both boundary conditions are shown by
creasing amplitude. Finite-time MD simulations of unstable the diamonds on curves ~a! and ~a 8 ) in Fig. 1; for
OZCM’s in lattices with StdPBC’s reveal that the instability StdPBC’s the threshold is at (A thresh /a, v / v max)
55 BOUNDARY-CONDITION EFFECTS IN ANHARMONIC . . . 8835

for both boundary conditions, i.e., the particles away from


the mode center are essentially at rest, as shown in the upper
panel of Fig. 4 for the ZSPBC case. The slight difference
between the two ILM curves of Fig. 1 in this range arises
from the different static displacements away from the mode
center for the two types of periodic boundary conditions. For
StdPBC’s, the total length L of the lattice is held fixed at its
static equilibrium value, and a highly localized ILM pro-
duces dynamical stress, resulting in static displacements
$ b n % which decrease uniformly to zero at the supercell
boundaries; thus there is a constant static strain away from
the ILM center. The magnitude of this strain and the corre-
sponding dynamical stress decrease with increasing number
of particles N and vanish only in the infinite lattice limit. On
the other hand, for a finite lattice with ZSPBC’s, the dynami-
cal stress set up by the ILM is relieved by allowing the
supercell length to increase. As a result, a highly localized
ILM in a ZSPBC lattice produces zero strain away from the
mode center, even for a finite lattice ~see top panel of Fig.
FIG. 4. Dynamic ~circles! and static ~squares! displacements for 4!—the lattice on each side of the ILM simply moves out
an odd-parity OZCM-ILM in the 40-particle diatomic lattice of Fig.
uniformly. Thus when the ILM localization is much smaller
1, with ZSPBC’s. The small ~large! symbols represent
than the total number of particles, as in the upper panel of
the light ~heavy! masses, respectively. Upper panel: At
Fig. 4, the ILM properties should be very similar for
(A/a, v / v max)5(0.194,1.40) the mode is highly localized, with the
particles away from the mode center being essentially at rest and
StdPBC’s and ZSPBC’s. This is the case in Fig. 1, for ILM
having constant static displacements. Lower panel: At amplitudes greater than A'0.15a.
(A/a, v / v max)5(0.099,0.95) the dynamic displacement pattern ex- However, the situation is very different for these ILM’s at
hibits an OZCM-like extended background. Note the accompanying smaller amplitudes. As the amplitude is decreased from the
lattice constant change away from the ILM center. highly localized case discussed above, the ILM frequencies
for the two periodic boundary conditions decrease together
in Fig. 1, approaching v max from above. Further decrease of
5(0.120,1.36), while for ZSPBC’s the values are
the amplitude leads to a separation of the two ILM curves:
(0.063,0.91). The differences between the 40-particle thresh-
the ZSPBC ILM curve continues down into the harmonic
old amplitudes and their infinite lattice counterparts are un-
optical phonon band until it merges with the curve for the
observable on the scale of this figure.
ZSPBC OZCM at (A/a, v / v max)5(0.063,0.91). This point
No MD runs showing the finite-time evolution of either
corresponds almost exactly to the minimum of the OZCM
the unstable OZCM or the unstable OZBM for ZSPBC’s are
curve. On the other hand, with StdPBC’s, the ILM frequency
reported here, since the implementation of the zero-stress
never enters the harmonic optical phonon band. Instead,
condition in MD is difficult, particularly when dealing with
shortly before v max is reached, the frequency vs amplitude
the spatial inhomogeneities introduced by the ExM instabili-
curve has a sharp minimum ~‘‘knee’’! and the frequency
ties or the presence of ILM’s.
rises over a very small amplitude range, until the ILM curve
merges with the StdPBC OZCM curve at
3. OZCM-related ILM’s (A/a, v / v max)5(0.120,1.36).
Next, we consider ILM’s that are related to the OZCM in This clearly distinct behavior with decreasing amplitudes
our diatomic lattice, in a sense which will become clear be- can be understood from the nature of the dynamic displace-
low. Since the dynamic displacement pattern of the OZCM is ment patterns. When the frequency is above v max , the ILM
odd under reflection in a site, we restrict our attention to remains localized and the particles away from the mode cen-
ILM’s with this parity. These ILM’s are just the diatomic ter behave like particles in a harmonic lattice which is driven
versions of the known odd-parity ILM’s in monatomic above its maximum frequency v max , i.e., they vibrate with
(k 2 ,k 3 ,k 4 ) lattices. The upper panel of Fig. 4 shows the very small amplitudes that die away rapidly with distance
dynamic and static displacements for such an ILM in a 40- from the mode center. As the ILM frequency decreases to
particle lattice with ZSPBC’s. For this ILM, v 51.40v max . become resonant with v max , the entire lattice responds, and
In Fig. 1, the frequency of this odd-parity ILM in a 40- the ILM dynamic displacement pattern acquires a small-
particle diatomic lattice is shown as a function of the ampli- amplitude extended background away from the mode center.
tude of the mode’s central particle. The dashed curves ~b! For StdPBC’s, this occurs slightly above v max , whereas for
and ~b 8 ) are for ILM’s in lattices with StdPBC’s and ZSPBC’s, it occurs at v max . In each case, the background
ZSPBC’s, respectively. It can be seen that for amplitudes pattern is just that of the corresponding small-amplitude
greater than A'0.15a, the ILM curves for the two different OZCM having the same frequency as the ILM.18 Hence, for
periodic boundary conditions are nearly coincident. Inspec- StdPBC’s, the extended background amplitude and fre-
tion of the corresponding displacement patterns in this range quency correspond to a point in Fig. 1 on the OZCM curve
shows that the dynamic displacements are likewise almost ~a! just above v max , whereas for ZSPBC’s, they correspond
identical. At such amplitudes, the modes are highly localized to a small-amplitude point just below v max on the OZCM
8836 D. BONART. T. RÖSSLER, AND J. B. PAGE 55

curve ~a 8 ). For a further decrease of the ILM amplitude, the is illustrated by the solid curve ~a! in Fig. 5. When realistic
extended background amplitude grows relative to the local- full potentials are used, the OZBM v (A) curve exhibits the
ized portion of the ILM; comparing curves ~a! and ~a 8 ), we same qualitative behavior.
see that for StdPBC’s, the background OZCM amplitude can b. Zero-stress periodic boundary conditions. For these
grow only if the mode frequency increases, whereas for boundary conditions, the OZBM produces a lattice constant
ZSPBC’s, an increase in the background OZCM amplitude change D(Ã) which is given by Eq. ~8!, but with Ã5A/2.
corresponds to a frequency decrease. Thus the different be- Thus Fig. 2 applies to the OZBM, provided the abscissa
havior of the two ILM curves in Fig. 1 results from the values are multiplied by (11m/M ). The expressions for the
different frequency vs amplitude behavior of the corre- renormalized force constants in terms of D are the same as
sponding extended background portions. In either case, the those given below Eq. ~8!, and they lead to the RWA squared
extended background amplitude grows as the ILM amplitude frequency
decreases, until the ILM becomes the OZCM for the particu-
lar periodic boundary condition. We will refer to these
ILM’s as ‘‘OZCM-ILM’s,’’ to distinguish them from the
OZBM-related ILM’s to be discussed in the next section.
S D S DS
v max
2
v ZSPBC
5 11
m
M
21
k̂ 2
k2
k̂ 4
13 Ã 2
k2
D
The lower panel of Fig. 4 illustrates an OZCM-ILM in a
ZSPBC 40-particle lattice for v 50.95v max , well inside the
harmonic optical phonon band. The mode’s extended back-
S DF
5 11
m
M
21

112
k3
k2
k4
G
D13 ~ D 2 1Ã 2 ! ,
k2
ground is readily apparent. ILM’s with an extended back- ~12!
ground were discussed briefly in Ref. 6, for a monatomic
(k 2 ,k 3 ,k 4 ) lattice with StdPBC’s. from which we compute the solid curve ~a 8 ) of Fig. 5. Com-
Finally, we emphasize that for each periodic boundary paring Eqs. ~11! and ~12! with their OZCM counterparts Eqs.
condition, the point in Fig. 1 where the OZCM-ILM fre- ~7! and ~9!, we observe that the only differences are the
quency vs amplitude curve merges with the corresponding constant prefactor (11m/M ) 21 in Eqs. ~11! and ~12! and
OZCM curve and where we have seen that the OZCM-ILM the definitions of the effective mode amplitude Ã. Therefore,
actually becomes the OZCM, agrees exactly with the OZCM the OZBM frequency vs amplitude curves for the two peri-
instability threshold amplitude in our 40-particle lattice, odic boundary conditions exhibit the same qualitative behav-
shown in the figure as a diamond. Furthermore, for ampli- ior as those for the OZCM, and the discussion given in Sec.
tudes below A thresh , we find no OZCM-ILM’s in either case. IV A 1 below Eq. ~9! applies here as well. Note, however,
It is interesting to note from Fig. 1 that with ZSPBC’s, the that the minimum of our OZBM ZSPBC frequency vs am-
frequency range of the unstable OZCM and the OZCM-ILM plitude curve occurs within the harmonic phonon gap, at
are the same, whereas this is seen not to be the case for (A/a, v / v max)5(0.078,0.81), rather than within the har-
StdPBC’s. This mismatch of the frequency ranges of the un- monic optical phonon band as for the OZCM. For realistic
stable ExM and its related ILM for StdPBC’s was briefly full potentials the OZBM frequency has no minimum, de-
discussed in the introduction in the context of the ZBM in a scending instead through the phonon gap.
monatomic (k 2 ,k 3 ,k 4 ) lattice, and we will return to this
point later. 2. Stability of the OZBM
For either periodic boundary condition, changing the par- Although the frequency vs amplitude curves for the
ticle number N does not affect our qualitative results, such as OZBM and the OZCM are very similar, we find that these
the junction between the OZCM and OZCM-ILM frequency modes’ ILM-related stability properties differ markedly and
vs amplitude curves and the nature of the evolution of the are, in fact, complementary: whereas the OZCM is stable for
OZCM-ILM displacement patterns along the curves. amplitudes below an instability threshold, the OZBM is
stable above a critical amplitude. The two types of periodic
B. Optical zone boundary mode and related ILM’s boundary conditions only affect the details of this behavior,
analogous to the situation for the OZCM. More specifically,
1. Dynamics of the OZBM for both StdPBC’s and ZSPBC’s, the RWA stability analysis
a. Standard periodic boundary conditions. For these predicts the OZBM to have an ILM-related instability, i.e.,
boundary conditions, the OZBM dynamics are not affected an instability with a purely real growth rate l max , only for
by cubic anharmonicity, just as for the OZCM ~see Appendix amplitudes from zero up to a critical threshold value
A!. The RWA frequency of the OZBM is determined by A thresh . The upper panel of Fig. 6 shows our predicted
growth rate as a function of the OZBM amplitude for

S D S D S D
21 StdPBC’s. Measured growth rates for MD simulations of the
v StdPBC 2 m k4 2
5 11 113 Ã , ~11! OZBM in a 40-particle lattice are indicated by diamonds,
v max M k2 and the agreement with the RWA prediction is to within less
than 3% except for the point at the largest amplitude, where
where à is now A/2 rather than the OZCM value the deviation is 10%. For our force constants the RWA
(11m/M )(A/2), with A again being the amplitude of a light instability threshold occurs at (A thresh /a, v / v max)
mass. The frequency of the OZBM thus increases monotoni- 5(0.149,1.21).
cally with increasing amplitude, starting from the corre- In the lower panel of Fig. 6 we plot the wave vector
harm 5 v max / A11m/M and
sponding harmonic frequency v OZBM (k p ) max of the fastest-growing Fourier component of the
moving up through the harmonic optical phonon band. This RWA instability perturbation as a function of the OZBM
55 BOUNDARY-CONDITION EFFECTS IN ANHARMONIC . . . 8837

FIG. 5. Same as Fig. 1, but for the OZBM and its associated ILM. The inset magnifies a portion of the figure near the minimum of the
curve for the ZSPBC OZBM. The solid circle, square and upward-pointing triangle correspond to specific mode patterns discussed in the
text.

amplitude for StdPBC’s. We find that (k p ) max decreases from OZBM Dk5 u p /2a2(k p ) maxu . Hence the minimum in the
p /2a at small amplitudes to a minimum of 0.25p /2a at (k p ) max vs amplitude curve of Fig. 6 corresponds to a maxi-
A50.084a and then increases back to p /2a at A thresh . With mum of Dk and hence to a minimum preferred instability
an instability perturbation of wave vector (k p ) max present, length scale. Finite-time MD simulations of unstable
the spatial modulation of an ExM of wave vector k is deter- OZBM’s for StdPBC’s reveal that the instability causes the
mined by Dk5 u k2(k p ) maxu , so that the ExM’s preferred in- mode to break up into a periodic array of localized excita-
stability length scale is 2 p /Dk. Note that for the OZCM tions whose spacing is close to the preferred instability
discussed earlier, k50 and Dk5(k p ) max , whereas for the length.
Replacing the StdPBC’s by ZSPBC’s, we obtain qualita-
tively similar results. For these boundary conditions, the
RWA instability threshold is at (A thresh /a, v / v max)
5(0.078,0.81), which coincides exactly with the minimum
in the ZSPBC OZBM frequency vs amplitude curve ~a 8 ) of
Fig. 5. Furthermore, (k p ) max attains its minimum value
0.57p /2a at A50.054a. In Fig. 5, the computed OZBM in-
stability thresholds for both periodic boundary conditions in
a 40-particle lattice are shown by diamonds.

3. OZBM-related ILM’s
Finally, we discuss the ILM’s which are related to the
OZBM in our diatomic lattice in the same sense that the
OZCM-ILM’s discussed in Sec. IV A 3 are related to the
OZCM. Provided the amplitude is not too high, the frequen-
cies of these ‘‘OZBM-ILM’s’’ turn out to be in the harmonic
phonon gap. Regarding the symmetry of these ILM’s, we
first note that the dynamic displacement pattern of the
OZBM is odd under reflection in a light mass site and even
under reflection in a heavy mass site. Gap ILM’s having one
or the other of these symmetry properties have been consid-
ered in Refs. 9 and 10. Here, we will focus on gap ILM’s that
FIG. 6. Same as Fig. 3, but for the OZBM. are odd under reflection in a light mass site. These are the
8838 D. BONART. T. RÖSSLER, AND J. B. PAGE 55

To discuss the results of Fig. 5 in more detail, we now


focus on the ZSPBC case and will later simply point out the
differences for StdPBC’s. As we follow the OZBM-ILM
curve ~b 8 ) from small amplitudes to its large-amplitude
junction with curve ~a 8 ), the mode’s spatial width changes
from delocalized ~small-amplitude OZBM! to highly local-
ized and then again to delocalized ~large-amplitude OZBM!.
The mode shown in the middle panel of Fig. 7 is, in fact, the
minimum-width OZBM-ILM for our 40-particle lattice, and
the upper and lower panels of this figure show the mode for
smaller and larger amplitudes, respectively. To quantify the
broadening, we introduce a convenient measure of the ILM’s
spatial width as
N
u c nu
s5 (
n51 A
, ~13!

where N is the total number of particles in the lattice. With


this definition, more highly localized modes have smaller
values of s . For the OZBM-ILM, the maximum value of s
is obtained when the mode broadens to become the OZBM;
this gives s max5N/2, since the heavy masses are at rest and
all of the light masses have the same amplitude. For the three
FIG. 7. Dynamic ~circles! and static ~squares! displacements for OZBM-ILM patterns of Fig. 7, the values of s / s max are 0.47
an odd-parity OZBM-ILM in the 40-particle diatomic lattice of Fig. for the upper and lower panels and 0.12 for the middle panel.
1, with ZSPBC’s. The small ~large! symbols represent the light The two panels of Fig. 8 give our results for s / s max as a
~heavy! masses, respectively. Upper panel: Displacements for function of the OZBM-ILM amplitude or frequency, respec-
(A/a, v / v max)5(0.017,0.89). Middle panel: At (A/a, v / v max) tively, for our 40-particle lattice with ZSPBC’s. For an infi-
5(0.067,0.85), the mode has its smallest spatial width. Lower nite lattice, the small-amplitude OZBM-ILM exists down to
panel: At (A/a, v / v max)5(0.097,0.83), the ILM has spatially A50, corresponding to the frequency v 5 v OZBM harm , the bot-
broadened by acquiring a central ‘‘core’’ of 17 particles having an tom of the harmonic optical phonon band. However, because
OZBM-like dynamic displacement pattern. our lattice is finite, there is a very small threshold amplitude
A50.008a for OZBM-ILM existence, as seen in the left-
most localized gap ILM’s for a given amplitude, and the hand panel of Fig. 8. The corresponding threshold frequency
maximum dynamic displacement occurs for the light mass at is seen in the right-hand panel to be slightly below v OZBM harm ,
the mode center. The central panel of Fig. 7 shows the dy- shown by the thin vertical line. For amplitudes just above the
namic and static displacements of such a gap ILM in a 40- threshold, the dynamic displacement pattern is essentially
particle lattice with ZSPBC’s. The frequency is that of a small-amplitude OZBM, but with the dynamic dis-
v 50.85v max , and the ILM is highly localized. placements gradually rising a small amount as one moves
from the ends of the lattice towards the center. For this de-
In Fig. 5, the dashed curves ~b! and ~b 8 ) give the fre-
localized pattern s / s max'1. With increasing amplitude, the
quency of the odd-parity gap ILM as a function of the
frequency decreases within the harmonic gap ~Fig. 5!, and
central-particle amplitude for StdPBC’s and ZSPBC’s, re-
the mode’s spatial width decreases smoothly, as does
spectively, in our 40-particle lattice. For StdPBC’s the ILM s / s max . The smallest width is reached at (A/a, v / v max)
frequency eventually moves up into the harmonic optical 5(0.069,0.85), where s / s max has its minimum value of
phonon band, while for ZSPBC’s it remains in the gap. For 0.12, giving the mode pattern shown in the middle panel of
both boundary conditions, however, the ILM curves merge Fig. 7. This most localized OZBM-ILM in our lattice is seen
with the corresponding OZBM curves at both small and large to be strongly dominated by the motion of the central par-
amplitudes, and we will thus henceforth refer to these ILM’s ticle. The solid circles in Fig. 5 and in both panels of Fig. 8
as OZBM-ILM’s, rather than as gap ILM’s. correspond to this mode.
The amplitude vs frequency behavior of the OZBM- Increasing the ILM amplitude beyond this point, we find
ILM’s in Fig. 5 is more complex than the previously dis- that the spatial width of the ILM begins to increase. How-
cussed OZCM-ILM behavior shown in Fig. 1. A striking ever, as seen in Fig. 8, this increase is now accompanied by
feature of the ILM curves of Fig. 5 is their oscillatory nature ‘‘oscillations’’ of both the ILM amplitude and frequency; as
for amplitudes greater than A'0.08a. For ZSPBC’s, this is a function of s / s max , their period is close to 0.1. The oscil-
shown more clearly in the inset. The oscillations are inti- lations indicate that the ILM’s width is now changing by a
mately related to the manner in which the spatial width of the less continuous process than at small amplitudes, where no
OZBM-ILM changes along the curves, as will now be de- oscillations are seen. Detailed inspection of the mode pattern
tailed. In Sec. V we will discuss this behavior in terms of a reveals that after the first of these oscillations, the amplitudes
simplified model and will give conditions under which it is of the two light masses on either side of the center of the
likely to occur. highly localized mode shown in the middle panel of Fig. 7
55 BOUNDARY-CONDITION EFFECTS IN ANHARMONIC . . . 8839

‘‘hole’’ in the full-lattice OZBM. Indeed, if the site labels are


all shifted by N/2, the pattern appears as if one had simply
subtracted from the OZBM a highly localized OZBM-ILM,
centered in the lattice. One can develop this ‘‘anti’’-ILM
point of view to give a completely equivalent description of
the OZBM-ILM, although we will not do so here.
The final amplitude and frequency oscillations for our 40-
particle lattice are marked in Figs. 5 and 8 by upward-
pointing triangles. Further decrease in the OZBM-ILM am-
plitude is accompanied by a decrease in the frequency, which
then tracks very closely the OZBM frequency vs amplitude
curve ~a 8 ) in Fig. 5 downward, until the minimum is reached
at the diamond, marking the OZBM instability threshold. At
this point the OZBM-core fills the entire lattice, i.e., the
OZBM-ILM has become the OZBM.
FIG. 8. Normalized spatial width parameter s / s max @Eq. ~13!# Turning now to the StdPBC case, we find the same quali-
for the OZBM-ILM as a function of the normalized amplitude ~left tative behavior for the OZBM-ILM dynamic displacement
panel! and frequency ~right panel!, for the 40-particle diatomic lat- pattern as we follow curve ~b! in Fig. 5, starting from small
tice of Fig. 1, with ZSPBC’s. The downward-pointing triangle, the amplitudes and ending at its large-amplitude junction with
circle, and the square correspond to the modes whose patterns are the OZBM curve ~a!: the mode first changes smoothly from
given in the top, middle and lower panels of Fig. 7, respectively. delocalized ~small-amplitude OZBM! to highly localized, af-
The upward-pointing triangle corresponds to the similarly marked ter which it delocalizes with accompanying amplitude and
point in Fig. 5. Crosses represent MD measurements of the OZBM- frequency oscillations. The mode becomes a large-amplitude
ILM frequency for the case of free-end boundary conditions. The
OZBM at the upper diamond, corresponding to the OZBM
MD measurements and RWA predictions differ by 1% at most. The
thin vertical line denotes the bottom of the harmonic optical phonon
instability threshold for our 40-particle StdPBC lattice. The
band. differences between the OZBM-ILM curves ~b! and ~b 8 ) re-
sult from the presence of nonzero dynamical stress produced
have grown to become almost equal with that of the central by the ILM in the StdPBC case; this effect is necessarily
particle, whereas the other amplitudes have changed only absent for ZSPBC’s. Note that the presence of dynamical
slightly. The additional two large amplitudes increase s by stress renders the mode frequency at a given amplitude
approximately 2, so that s / s max grows by about 0.1 for our higher for StdPBC’s than for ZSPBC’s in our 40-particle
40-particle lattice. During the next oscillation period of lattice. Even so, we see that for amplitudes from zero to
A/a and v / v max , the amplitudes of two additional light A'0.08a, the effect is not large and curve ~b! approximately
masses increase until they closely match that of the central follows curve ~b 8 ). At this point the mode for either bound-
particle, and so on. The lower panel of Fig. 7 shows the ary condition is beyond its maximum localization ~see the
dynamic and static displacements for (A/a, v / v max) middle panel of Fig. 7! and has begun to broaden by devel-
5(0.097,0.83) and s / s max50.47. It is seen that a ‘‘core’’ of oping an OZBM-like central core, as detailed above for the
17 particles exhibits an OZBM-like displacement pattern, ZSPBC case. This broadening continues, with its attendant
while the rest of the lattice barely vibrates. Note that for this frequency and amplitude oscillations, until the size of the
type of broadening, N s / s max gives an approximate measure OZBM-like core is such as to produce enough dynamical
of the number of particles in the core. The locations of this stress in the StdPBC case to markedly affect the mode dy-
ILM on the curves of Figs. 5 and 8 are shown by squares.
namics. For our 40-particle lattice this occurs above
As the width of the OZBM-like core region grows by the
A'0.08a, and the curves ~b! and ~b 8 ) in Fig. 5 become very
addition of pairs of light masses, the dynamical behavior
near the mode center approaches that of the full lattice different. The difference just reflects the strong influence of
OZBM having the same amplitude as the OZBM-ILM’s cen- nonzero dynamical stress on the pure OZBM curve for
tral particle. This is seen in the inset of Fig. 5, where the StdPBC’s, as is evidenced by comparison of curves ~a!
oscillations in the OZBM-ILM amplitude vs frequency curve and ~a 8 ). As the large-amplitude OZBM-ILM for each peri-
~b 8 ) eventually parallel a portion of the OZBM amplitude vs odic boundary condition broadens to become the correspond-
frequency curve ~a 8 ). Indeed, the later oscillations cannot be ing OZBM, the accompanying oscillations in its frequency
resolved from each other or from the OZBM curve, and the vs amplitude curve parallel and approach the OZBM curve to
square denoting the location of the mode shown in the lower which it eventually attaches.
panel of Fig. 7 is actually removed from the OZBM curve Finally, for either of our periodic boundary conditions,
~a 8 ) by approximately five oscillations. changing the particle number N does not affect our qualita-
Clearly, in a finite lattice only a finite number of the os- tive results, such as the existence of two junctions between
cillations can occur, since the number of particles in the the OZBM and OZBM-ILM frequency vs amplitude curves,
mode’s OZBM-like core cannot exceed the total number N. the appearance of an OZBM-like ‘‘core’’ in the OZBM-ILM
The limiting situation is approached when the core encom- with its accompanying frequency and amplitude oscillations,
passes almost the entire lattice, and since we are using peri- and the nature of the evolution of the ILM displacement
odic boundary conditions, the mode pattern then looks like a patterns along the frequency vs amplitude curves.
8840 D. BONART. T. RÖSSLER, AND J. B. PAGE 55

TABLE I. Results relating ExM properties and ILM existence for specific model lattices: ~a! diatomic
(k 2 ,k 3 ,k 4 ) lattice with 4k 23 /(3k 4 k 2 ).1, ~b! diatomic (k 2 ,k 3 ,k 4 ) lattice with 0<4k 23 /(3k 4 k 2 ),1, ~c! di-
atomic lattice with full BMC potential. Column 3 gives a range for the mode amplitude A of both the ExM
specified in column 2 and the ExM’s related ILM’s of column 7. The quantity A thresh denotes the threshold
amplitude for the ILM-related instability of the ExM. The ExM stability behavior of column 4 as well as the
signs of the ExM quantities v 9 (k) and v ZSPBC 8 (A) defined in Sec. V A 2 apply to the entire amplitude range
of column 3. Column 7 indicates whether ILM’s related to the ExM of column 2 exist with amplitudes within
this range, and if so, of which type they are. The various ILM types are described in Sec. V. Entries in
columns 4 and 7 are valid for both StdPBC’s and ZSPBC’s, although the corresponding A thresh may differ.

ExM ~also gives Amplitude ILM-related ExM-related


Model m→M limit! range ExM stability v 9 (k) 8
v ZSPBC (A) ILM’s

OZCM~→ZBM! A,A thresh stable ,0 ,0 no ILM’s

A.A thresh unstable ,0 .0 type ~1!


~a!
A,A thresh unstable .0 ,0 types ~2!, ~3!
OZBM
A.A thresh stable .0 .0 type ~3!

OZCM ~→ZBM! A.0 unstable ,0 .0 type ~2!


~b!
OZBM A.0 stable .0 .0 no ILM’s

OZCM (→ZBM! A.0 stable ,0 ,0 no ILM’s


~c!
OZBM A.0 unstable .0 ,0 type ~2!

V. DISCUSSION ~2! behavior turns out to encompass all amplitudes, i.e., type-
~3! broadening is absent. Moreover, type ~2! is the only one
In this section, we use our diatomic (k 2 ,k 3 ,k 4 ) lattice found to exist with realistic full potentials V(r), where it
results of Sec. IV, together with findings for other model occurs for OZBM-ILM’s in diatomic lattices at all ampli-
lattices, to obtain basic interrelations between ILM’s and tudes; we thus regard type ~2! as the ‘‘generic’’ ILM behav-
their associated ExM’s. In Sec. IV, we encountered three ior. All three types will play roles in the remainder of this
qualitatively different ways for the dynamic displacement section.
pattern of an ILM to change as one moves along its fre-
quency vs amplitude curve: ~1! For the OZCM-ILM’s dis-
cussed in Sec. IV A 3, the pattern begins as a localized re- A. Existence criteria for type-„1… and type-„2… ILM’s
gion of increased displacements ‘‘on top’’ of a finite- Table I collects some pertinent ExM-ILM interrelations
amplitude background OZCM ~e.g., lower panel of Fig. 4! that we have obtained for several model lattices, for both
and evolves with increasing amplitude into a highly localized StdPBC’s and ZSPBC’s. The models are: ~a! the diatomic
mode without background ~e.g., upper panel of Fig. 4!. ~2! (k 2 ,k 3 ,k 4 ) lattice for 4k 23 /(3k 4 k 2 ).1, which is discussed in
For the OZBM-ILM discussed in Sec. IV B 3, the mode detail in this paper, ~b! the same lattice, but with
evolves from a very small-amplitude OZBM with dynamic 0<4k 23 /(3k 4 k 2 ),1, and ~c! a diatomic lattice with full
displacements gradually rising towards the mode center ~e.g., BMC potential. Moreover, the table includes the results for
upper panel of Fig. 7! to a highly localized mode ~e.g., the monatomic limit of these lattices. For each model, it lists
middle panel of Fig. 7!. Further increase in the amplitude the relevant ExM’s, a studied amplitude range, whether or
leads to a very different behavior, as follows. ~3! The not the ExM exhibits an ILM-related instability in this range,
OZBM-ILM of ~2! evolves from being highly localized to a and dynamical properties of the ExM to be discussed in de-
mode having a broad and sharply delineated OZBM-like tail in Sec. V A 2. Most importantly, the last column de-
core ~e.g., lower panel of Fig. 7!. The width of the core scribes whether or not ILM’s related to the ExM exist with
increases with increasing amplitude by the addition of pairs amplitudes within the studied range ~but not necessarily cov-
of particles to its ends until it becomes the OZBM, and this ering the entire range!, and if they exist, which of the three
process is accompanied by unusual frequency and amplitude ILM types they are.
‘‘oscillations,’’ seen in Figs. 5 and 8. Our results show that if
one of these three ILM behaviors occurs for the ZSPBC’s, it
1. Criterion based on extended mode stability
also occurs for StdPBC’s and vice versa, although the de-
tailed amplitude ranges may differ. The results of columns 4 and 7 in Table I provide com-
Various examples of these three types of ILM spatial be- prehensive numerical evidence that, independent of the
havior are encountered in monatomic and diatomic lattices choice of periodic boundary conditions, ILM’s of types ~1!
for diverse interparticle potentials. For some systems, type- and ~2! exist only for ILM amplitudes where the related ExM
55 BOUNDARY-CONDITION EFFECTS IN ANHARMONIC . . . 8841

for the ZBM in a monatomic (k 2 ,k 3 ,k 4 ) lattice reviewed in


the introduction.
For all the other model lattices that exhibit ILM’s of types
~1! or ~2! as listed in Table I, we find similar results for the
match between the preferred instability length scale of an
ExM and the spatial extent of its related ILM at the same
frequency with ZSPBC’s ~after subtracting any ExM-like
background which exists, e.g., that of the lower panel of Fig.
4!. Previously this connection was only shown for the much
simpler case of (k 2 ,k 4 ) lattices with StdPBC’s. In that case,
there is no dynamical stress, and the StdPBC’s are equivalent
to the ZSPBC’s. With odd-order potentials, however, the use
of StdPBC’s generally results in a mismatch of the frequency
ranges of the unstable ExM’s and their related ILM’s, ren-
dering impossible the length scale comparisons at a fixed
frequency. The use of ZSPBC’s now allows a confirmation
FIG. 9. Dashed curve: Preferred instability length scale vs fre- of this important ExM-ILM connection for lattices with odd-
quency for the OZBM in a 2000-particle diatomic (k 2 ,k 3 ,k 4 ) lat- order anharmonicity, including those with full BMC nearest-
tice, for the parameter values given in the text and ZSPBC’s. This neighbor potentials.
quantity is the wavelength corresponding to the difference Studies of anharmonic ExM stability can thus provide
u p /2a2(k p ) maxu between the wave vector of the OZBM and the
valuable information about the range over which ILM’s exist
wave vector (k p ) max of the fastest-growing Fourier component of
in model lattices and, for ZSPBC’s, can also yield informa-
the instability perturbation. Diamonds: Spatial extent of the OZBM-
tion about their spatial extent. Such studies form a useful
ILM vs frequency in the corresponding 40-particle lattice. This
quantity is 2N s / s max .
prelude to direct searches for ILM’s—the stability analysis is
a linear problem and hence is much simpler than searching
for the ILM itself.
is unstable. This generalizes the findings of Ref. 6, which
were for StdPBC’s, to a wider variety of model systems and 2. Criterion based on extended mode dynamics
to the ZSPBC’s. We now propose a practical, additional existence criterion
In addition, we now present another quantitative connec- for ILM’s of types ~1! and ~2!, which involves both the an-
tion between an ILM and stability properties of its associated harmonic and harmonic dynamical properties of the ILM-
ExM, for the case of ZSPBC’s. As discussed in Secs. IV A 2 related ExM’s. The criterion is based on our numerical stud-
and IV B 2 for the OZCM and OZBM with ies for the ZSPBC’s, and it is as follows: for ZSPBC’s, type-
StdPBCs, the wavelength corresponding to the difference be- ~1! or type-~2! ILM’s related to a given ExM only exist for
tween the wave vector of the ExM and that of the fastest- amplitudes A where
growing Fourier component of its instability perturbation in-
troduces a preferred instability length scale. Following the 8
v ZSPBC~ A ! v 9 ~ k ! ,0 ~14!
analysis of Ref. 6 @which was for the ZBM in a monatomic
(k 2 ,k 4 ) lattice with StdPBC’s#, we now compare the ZSPBC 8
for that ExM. Here, v ZSPBC (A)[d v ZSPBC /dA is the slope of
OZBM instability length scale with the spatial extent of the the frequency vs amplitude curve of the anharmonic ExM for
type-~2! OZBM-ILM having the same frequency in our di- ZSPBC’s, and v 9 (k)[d 2 v /dk 2 is the curvature of the har-
atomic (k 2 ,k 3 ,k 4 ) lattice with 4k 23 /(3k 4 k 2 ).1. The motiva- monic dispersion curve at the wave vector of the ExM. Table
tion for comparing at equal frequencies is discussed in Ref. I gives the signs of these quantities for the models listed.
6: there, MD simulations of the ZBM in simple (k 2 ,k 4 ) Note from Figs. 1 and 5 that for our diatomic nearest-
monatomic lattices, where the StdPBC’s and ZSPBC’s are neighbor (k 2 ,k 3 ,k 4 ) model lattice of Sec. IV, for both the
identical since there is no dynamical stress, revealed that the OCZM and the OZBM, v ZSPBC 8 (A) is negative for small am-
unstable ZBM decays into a periodic array of localized ex- plitudes and then becomes positive at the mode’s instability
citations having larger amplitudes than the original ZBM, but threshold A thresh . Moreover, the curvature v 9 (k) of the har-
having the same frequency. Here, as a measure for the spatial monic dispersion is negative for the OZCM (k50) and posi-
extent of the OZBM-ILM we use 2N s / s max , where s and tive for the OZBM (k5 p /2a). Hence for the OZCM, the
s max are defined in Sec. IV B 3. The comparison is made in criterion is met only for amplitudes from the ZSPBC insta-
Fig. 9, and the good agreement seen there strongly implies bility value A thresh50.063a onward, precisely the same range
that the preferred instability length scale is just the spatial where we have seen that type-~1! OZCM-ILM’s exist for
extent of the ILM, in complete analogy to the (k 2 ,k 4 ) case ZSPBC’s. On the other hand, for the OZBM in this lattice,
considered in Ref. 6. Note that a similar comparison is not the criterion is satisfied only over the amplitude range from
possible here for the StdPBC case, since the corresponding zero to the ZSPBC instability point for the OZBM, at
unstable OZBM and type-~2! OZBM-ILM do not share a A thresh50.078a. This is the amplitude range within which we
common frequency range. Furthermore, finite-time MD have seen that type-~2! OZBM-ILM’s exist for ZSPBC’s.
simulations reveal that the unstable OZBM for StdPBC’s Although the criterion applies specifically to ZSPBC’s,
decays into a periodic array of localized excitations having a we have already remarked that if ILM’s related to an ExM of
frequency different than that of the original OZBM, just as a given type ~1!, ~2!, or ~3! exist in a given lattice for one of
8842 D. BONART. T. RÖSSLER, AND J. B. PAGE 55

our periodic boundary conditions, they also exist for the (k 2 ,k 4 ) lattice, the dynamics of the envelope c (x,t) are de-
other, albeit over a possibly different range of amplitudes. scribed by the nonlinear Schrödinger equation ~NLSE!
Moreover, when the ILM’s are highly localized, we have
seen that their properties become essentially independent of ]c ] 2c
i 1 P 2 2Q u c u 2 c 50, ~15!
the boundary conditions, as is obvious physically. Accord- ]t ]j
ingly, the existence criterion given here can give useful
where t [ e 2 t and j [ e (x2 v g t) are the slowly varying time
information on ILM existence beyond the context of the
and space variables describing the variation of the envelope
ZSPBC’s.
and v g [d v /dk is the group velocity of the carrier wave.23
This criterion has been found to hold for every model
The coefficients P and Q are, up to positive numerical fac-
system we have studied, as is seen by inspection of columns tors, equal to v 9 (k) and limA→0 @ v 8 (A)/A # for the carrier
5, 6, and 7 of Table I. Moreover, previous ILM studies for wave, respectively. The sign of their product determines both
full potentials V(r) other than the BMC interaction included the type of solutions of Eq. ~15!, and the stability properties
in the table are also consistent with the criterion. For a di- of the associated anharmonic extended modes, i.e., the car-
atomic lattice with nearest-neighbor full potentials of the rier wave. For PQ.0, Eq. ~15! does not admit spatially
Toda, Morse, or Lennard-Jones forms, v ZSPBC 8 (A),0 for localized solutions having vanishing displacements far from
both the OZCM and the OZBM at all nonzero amplitudes, the mode center, and the corresponding extended carrier
and the criterion predicts ILM’s for the latter but not for the mode is stable against infinitesimal amplitude and phase per-
former. Consistent with this, OZBM-ILM’s are found in the turbations. For PQ,0, however, localized solutions exist
harmonic phonon gap for all amplitudes9 ~until the harmonic and the extended carrier mode is unstable.23
acoustic phonon band is approached19!, whereas OZCM- For a lattice with cubic anharmonicity, a static distortion
ILM’s have not been reported. In the monatomic limit, the corresponding to the $ b n % in our RWA solutions accompa-
OZCM becomes the ZBM, and no ZBM-ILM’s were found.6 nies envelope solitons.4 This leads to a pair of coupled equa-
Note that our criterion argues against a speculation offered in tions for the envelope functions which describe the static and
Ref. 20 that ILM’s ‘‘on top’’ of an extended ZBM back- dynamic components of the particle displacements. These
ground mode @i.e., type-~1! ILM’s# might exist for a Toda can be combined to yield an equation of the form of Eq. ~15!,
chain. but with different coefficients P̃ and Q̃. Although these
The specific lattice models discussed so far have been quantities still determine the properties of localized and ex-
restricted to nearest-neighbor interactions. Since the inclu- tended modes, Q̃ is in general no longer simply proportional
sion of interactions beyond nearest neighbors can markedly to limA→0 @ v 8 (A)/A # . We suspect that unawareness of this
affect the harmonic dispersion relation, it is interesting and fact may be the reason that Ref. 4 gives what we believe to
important to study their effect on the above ILM criterion. be an incorrect frequency v (A,k) for small-amplitude anhar-
Accordingly, we added a second-neighbor harmonic force monic ExM’s of general wave vector in monatomic
constant k (2nn)
2 /k 2 50.2 to our diatomic (k 2 ,k 3 ,k 4 ) lattice. (k 2 ,k 3 ,k 4 ) lattices.15 At any rate, it can be shown for these
The effect is to increase the harmonic OZBM frequency be- lattices that the simple relation Q̃}limA→0 @ v 8 (A)/A # holds
yond the harmonic OZCM frequency and to change the signs only for the ZBM, and then only under ZSPBC’s.
of v 9 (k) for these two modes. On the other hand, the quali- Consequently, the existence criterion obtained from the
tative features of their frequency vs amplitude curves are not NLSE approach for the ZBM in (k 2 ,k 4 ) and (k 2 ,k 3 ,k 4 )
altered by this added second-neighbor harmonic interaction. monatomic lattices is that for limA→0 @ v 8 (A,k5 p /a)/A #
Equation ~14! thus predicts the opposite behavior from the 3 v 9 (k5 p /a),0 the ILM exists and the ZBM is unstable,
case without second-neighbor interactions: OZCM-ILM’s whereas if this quantity is positive the ILM does not exist
should now exist below a finite amplitude threshold, whereas and the ZBM is stable, all for ZSPBC’s. For the case of the
OZBM-ILM’s should exist above a threshold. We have in- (k 2 ,k 4 ) lattice the ZBM introduces no dynamical stress, so
deed verified these predictions. the ZSPBC’s and StdPBC’s are equivalent. These small-
The above ILM existence criterion is not limited to small amplitude existence results from the NLSE agree with our
amplitudes. However, in the limit of vanishingly small am- more general RWA-based existence criterion of Eq. ~14!,
plitudes we can put it on a rigorous basis in terms of the which applies for arbitrary amplitudes.
mathematical properties of continuous nonlinear dynamical We re-emphasize that the NLSE analysis is based on the
systems. It is well known3,4,21 that small-amplitude localized assumption of slow spatial and temporal variations of the
modes in anharmonic 1D lattices, so-called envelope soli- envelope as well as small amplitudes, so that its range of
tons, can be described in terms of an extended carrier mode validity is much more restricted than that of our RWA ap-
having a modulated envelope whose spatial and temporal proach. This is reflected by the fact that the detailed dynami-
variations are slow compared with those of the carrier. In cal behavior within these two approaches deviates markedly
general, these are propagating solutions for any wave vector, for large amplitudes. For instance, the above NLSE analysis
but for particular wave vectors, e.g., at the zone boundary, cannot predict the switchover between ILM nonexistence
they become stationary envelope solitons and connect with and ILM existence at a finite threshold amplitude, as we have
the small-amplitude versions of the ILM’s studied in this encountered in Sec. IV A 3. Furthermore, the unusual large-
paper. Moreover, this approach has proven very useful be- amplitude frequency and amplitude oscillations which ac-
yond the context of 1D lattices, e.g., as a starting point for company type-~3! broadening and are discussed further in
theoretical studies of localized anharmonic edge and surface the following section, cannot be described within the NLSE
modes in crystals.22 For the simplest case of a monatomic approach. Nevertheless, the results of our studies strongly
55 BOUNDARY-CONDITION EFFECTS IN ANHARMONIC . . . 8843

imply that our ILM existence and ExM instability criterion 5# with type-~3! broadening and associated amplitude and
8
based on the sign of v ZSPBC (A) v 9 (k) is a general property frequency oscillations. On the other hand, it is at first sight
which applies well beyond the range of validity of the NLSE not clear that the frequency vs amplitude behavior of the
approach. OZBM-ILM’s core for StdPBC’s would exhibit a minimum,
as needed for the above simple model to apply. However, we
B. Unusual large-amplitude behavior: type-„3… ILM’s have already stressed that the differences between the type-
Finally, we return briefly to the unusual amplitude and ~3! OZBM-ILM frequency vs amplitude curves for StdPBC’s
frequency oscillations observed in our model studies for the and ZSPBC’s in Fig. 5 arise from the dynamical stress
type-~3! broadening of the OZBM-ILM beyond its point of present with StdPBC’s. The dynamical stress depends on the
highest localization. The oscillations are seen in Figs. 5 and finite size of our lattice; for an OZBM-ILM having a fixed-
8, and they occur at large amplitudes. It will be recalled that width core, it decreases as the size of the lattice is increased,
the type-~3! broadening is characterized by the presence of vanishing in the infinite lattice limit. In this limit, the lack of
an OZBM-like ‘‘core’’ ~e.g., lower panel of Fig. 7!, whose dynamical stress would result in the core behaving dynami-
width increases by the addition of particles to its ends, two at cally exactly like the OZBM-like core of the same spatial
a time. width for ZSPBC’s in a finite lattice and thus having a mini-
We find similar oscillations for ILM’s in a monatomic mum in its frequency vs amplitude behavior. Hence, the ba-
(k 2 ,k 4 ,k 6 ) lattice with k 2 and k 4 positive and negative k 6 ; sic physics underlying type-~3! ILM frequency and ampli-
hence the essential aspect of their origin is not the presence tude oscillations is the same for ZSPBC’s and StdPBC’s, but
of odd-order anharmonicity or a harmonic phonon gap. In- its details are significantly complicated by the presence of
stead, we feel that the relevant aspects of the model for this dynamical stress and finite-size effects in the latter.
behavior are that small-amplitude ILM’s exist, and that the
frequency vs amplitude behavior of the ExM core which ap- VI. SUMMARY AND CONCLUSION
pears in type-~3! broadening possesses an extremum, so that
the ExM core at a given frequency can occur with two dif- In this paper we have studied boundary condition effects
ferent amplitudes. These features hold for both the mon- on the properties of extended lattice modes and intrinsic lo-
atomic (k 2 ,k 4 ,k 6 ) lattice mentioned above and for our di- calized modes and their interrelations in periodic 1D anhar-
atomic (k 2 ,k 3 ,k 4 ) lattice with ZSPBC’s. For the latter case, monic lattices. In contrast to the more familiar situation of
this is evident from curves ~a 8 ) and ~b 8 ) in Fig. 5. The harmonic or weakly anharmonic lattices, where the boundary
situation for StdPBC’s, where the oscillations also occur in conditions play a passive role in determining bulk dynamical
our lattice, is more complicated, and its discussion will be properties, we have seen that strongly anharmonic dynamical
deferred until the ZSPBC case is treated. For our model, the behavior is markedly affected by the boundary conditions.
situation is simplified by the rather large mass discrepancy, Specifically, for lattices with odd-order interparticle anhar-
which results in the heavy masses having essentially no dy- monicity, the use of standard periodic boundary conditions,
namic displacement. This means that, to a good approxima- which keep the length of the supercell constant, generally
tion, each light particle added to the core has an individual results in nonzero ‘‘dynamical stress’’ whenever an ExM or
frequency vs amplitude behavior which mimics that of the ILM is present. Accordingly, we have studied the effects of
core itself, i.e., possesses a minimum. We incorporate these ‘‘zero-stress periodic boundary conditions,’’ which allow the
features into a highly simplified model consisting of two supercell length to adjust such that the dynamical stress van-
anharmonic oscillators having identical frequency vs ampli- ishes.
tude curves which possess a minimum. The oscillators are Within a rotating wave approximation, we have investi-
then weakly coupled. We make the RWA and start from a gated a variety of models. Our results have been illustrated in
displacement pattern in which one oscillator ~representing detail via calculations for a diatomic lattice with harmonic,
the OZBM-like core! has a large amplitude, while the second cubic and quartic (k 2 ,k 3 ,k 4 ) nearest-neighbor interactions,
oscillator ~representing a light mass to be added to the core! this model serving to illustrate a wide range of interesting
has a small amplitude, at the same frequency for each oscil- phenomena. For the anharmonic ExM’s studied, we have
lator. A continuum of solutions is then found, such that a seen that the frequency at a given amplitude is reduced when
smooth transition can occur to a displacement pattern where the StdPBC’s are replaced by ZSPBC’s. For both the optical
the amplitude of both oscillators is the same as the initial zone center mode and the optical zone boundary mode in our
large amplitude of the first oscillator. In terms of this model, diatomic (k 2 ,k 3 ,k 4 ) lattice, the frequency increases mono-
the core has added one light particle. During this transition, tonically with the mode amplitude for StdPBC’s, while it
the amplitudes of the two oscillators approach each other by exhibits a minimum when ZSPBC’s are used. A stability
moving towards the minimum of the frequency vs amplitude analysis reveals that the qualitative stability properties of
curve until they match, and the oscillators’ frequencies shift these ExM’s are the same for both periodic boundary condi-
together. From the minimum point, the amplitudes and fre- tions: the OZCM is unstable above a critical amplitude,
quencies increase together, until both oscillators attain the while the OZBM is unstable below a critical amplitude, with
initial large amplitude of the ‘‘core’’ oscillator. In this tran- the critical amplitude differing for the two types of boundary
sition, the core oscillator has gone through a single period of conditions. Remarkably, for ZSPBC’s the critical amplitudes
the frequency and amplitude oscillations observed for the for both ExM’s occur exactly at the minimum of their fre-
type-~3! OZBM-ILM’s in our diatomic lattice. quency vs amplitude curves.
For the case of StdPBC’s, our diatomic (k 2 ,k 3 ,k 4 ) lattice Within each type of periodic boundary condition, the
also yields small-amplitude OZBM-ILM’s @curve ~b! in Fig. properties of the ExM-related ILM’s were explored; these
8844 D. BONART. T. RÖSSLER, AND J. B. PAGE 55

are the ILM’s which spatially broaden to become the ExM as monic lattice dynamics have yielded clarifying insights into
the amplitude is reduced. Just as for the ExM’s, the ILM the fundamental connections between ILM’s and ExM’s.
frequency at a given amplitude is reduced when ZSPBC’s Moreover, we have obtained promising tools for future ILM
are used rather than StdPBC’s. For amplitude ranges where investigations. The ILM existence criterion @Eq. ~14!# re-
the ILM’s are highly localized compared to the extent of the quires only knowledge of two basic ExM properties: the cur-
lattice, the ILM frequency vs amplitude curves are rather vature of the ExM harmonic dispersion relation, which is
insensitive to the choice of periodic boundary conditions, as readily available for most real crystals, and the slope
expected. However, the portions of these curves which de- v 8 (A) for the anharmonic ExM under ZSPBC’s, which can
scribe the ILM’s as they spatially broaden to become the be computed via nonlinear potential models or first-
associated ExM are strongly affected. principles techniques.24 The most attractive situation would
For both periodic boundary conditions, we have found be for materials whose relevant ExM’s can be treated for-
three qualitatively different types of ILM’s, characterized by mally within a 1D analysis such as that used here. With a
the way their dynamic displacement patterns change as one prediction of ILM existence in hand, one can then study the
follows their v (A) curves. These types are described in the stability of the ExM within the ZSPBC’s to obtain a priori
first paragraph of Sec. V and are represented by the mode information on the spatial extent of the ILM displacement
patterns shown in Fig. 4 @type ~1!#, the upper two panels of patterns.
Fig. 7 @type ~2!#, and the lower panel of Fig. 7 @type ~3!#. Realistic interparticle potentials have strong odd-order an-
ILM’s of type ~3! are unusual, in that they exhibit a central harmonicity, and we have seen that with such anharmonicity
‘‘core’’ which is essentially just a finite-length OZBM. With present, both of the above modeling steps require the use of
increasing amplitude, the core broadens until it becomes the ZSPBC’s. Of course, once such ExM studies have been used
OZBM, and this process is accompanied by frequency and to predict ILM’s in a given lattice, the ILM properties for the
amplitude oscillations. Insight into these oscillations was ob- highly localized cases will be independent of the specific
tained via a simple model of two coupled anharmonic oscil- boundary conditions, as long as the ILM’s spatial extent is
lators. Type-~2! ILM’s exist from zero amplitude, where they small compared with the length of the supercell. Thus, be-
are broad. With increasing amplitude, they become localized sides yielding fundamental insights into the nature of
but may convert to type-~3! broadening at a finite amplitude, strongly anharmonic lattice dynamics, the use of ZSPBC’s
as in our diatomic (k 2 ,k 3 ,k 4 ) lattice, or they can remain provides an appealing practical framework for predicting
localized and of type ~2!. The ILM’s for realistic full poten- ILM’s in real crystals.
tials V(r) exhibit the latter behavior, remaining of type ~2!
for all amplitudes. ACKNOWLEDGMENTS
Connecting the ExM and ILM results, we found that for
both periodic boundary conditions, ILM’s of types ~1! and This work was supported by NSF Grant No. DMR-
~2! exist only at amplitudes for which their related ExM’s are 9510182. D. Bonart acknowledges the support of Deutsche
unstable, generalizing earlier qualitatively similar results for Forschungsgemeinschaft Grant No. Ma 1074/5-1 during a
monatomic (k 2 ,k 3 ,k 4 ) lattices with StdPBC’s.6 Further- portion of this work.
more, for ZSPBC’s we have seen that the unstable ExM’s
introduce preferred length scales which match the spatial ex- APPENDIX A: BOUNDARY CONDITIONS AND THE
tent of the ExM’s related ILM at the same frequency. For DEPENDENCE OF THE EXTENDED MODE DYNAMICS
lattices with odd-order anharmonicity, such an analysis is not ON ODD-ORDER POTENTIAL TERMS
generally possible with StdPBC’s, because the frequency
ranges of the unstable ExM and its related ILM differ due to As discussed in Sec. II, the relative dynamic displace-
the nonzero dynamical stress. However, the analysis is ments for the three ExM’s introduced there are given by
straightforward when ZSPBC’s are used. Most importantly, Dc n 5(21) n 2Ã, and the corresponding relative static dis-
the ZSPBC’s have led to a useful second criterion @Eq. ~14!# placements are Db n 5D. Within the StdPBC’s, D50, and
for the existence of ILM’s of types ~1! and ~2!, complement- we can write Eq. ~4a! for the RWA dynamic displacements
ing the above criterion based on ExM instability. This addi- as
tional criterion involves just two relatively simple properties
of the ILM’s associated ExM, namely the curvature v 9 (k) of
the harmonic dispersion at the ExM wave vector and the
m nv 2c n5
1
p
E
0
2p
d f cosf $ V 8 @~ 21 ! n 2Ãcosf 1a #
8
slope v ZSPBC (A) of the anharmonic ExM frequency vs am-
plitude curve. In the limit of vanishing amplitudes for type- 2V 8 @~ 21 ! n11 2Ãcosf 1a # % ~A1!
~2! ILM’s, this criterion reduces to earlier results for lattice
envelope solitons obtained by means of a nonlinear Schrö- or
dinger equation ~NLSE! approach. However, our criterion

E
applies for finite amplitudes, well beyond the range where 2 ~ 21 ! n 2p
the NLSE approximation breaks down, and it includes type- m nv 2c n5 d f cosf V 8 ~ 2Ãcosf 1a ! ,
~1! ILM’s. p 0

Although the use of periodic boundary conditions which ~A2!


require the dynamical stress to vanish might seem esoteric
from the point of view of harmonic or weakly anharmonic where the latter form is obtained by shifting the integration
lattice dynamics, our studies of its effects on strongly anhar- variable by p . Expanding V 8 about a, we get
55 BOUNDARY-CONDITION EFFECTS IN ANHARMONIC . . . 8845

m nv c n5
2
2 ~ 21 ! n
p m50
`
~ 2Ã ! m V ~ m11 ! ~ a !
( m!
E
0
2p
d f cos m11
f, Ū5
1
T
E
0
T
dt
N L int

( ( V l k50
n51 l51
H l21

( @ Dc n2k f ~ t ! 1Db n2k 1a # J .


~A3! ~B2!

where V (m11) (a)[d m11 V(r)/dr m11 u r5a . The integral van- Imposing periodic boundary conditions by defining
ishes for m11 odd, so that only the even-order terms in the Dc n1N [Dc n and Db n1N [Db n , we ensure that all terms in
expansion of the potential energy affect the dynamics of our this sum are well defined.
ExM’s. Now we assume an arbitrary given relative dynamic dis-
Note that an analogous result also follows for the placement pattern $ Dc n % and require ] Ū/ ] Db m 50 for all
ZSPBC’s, provided the potential is expanded about the mP $ 1, . . . ,N % , which is a necessary condition for the mini-
period-average nearest-neighbor distance â(Ã)5a1D(Ã). mization of Ū with respect to the set $ Db n % . Note that the
In this case, the harmonic and anharmonic coefficients are $ Db n % constitute a set of N independent variables as long as
renormalized from their static equilibrium values, and only we do not constrain the total length L of the supercell. We
the even-order renormalized coefficients $ V (m) (â) % are in- then have
volved in the dynamics of the ExM’s. However, to avoid

E
N L int l21
confusion, we always use the unrenormalized coefficients 1 T ] V l ~ r n 2r n2l !
$ V (m) (a) % appropriate to the static equilibrium lattice when T 0
dt ((
n51 l51 ]rn ( k50
d n2k,m
expanding the potential. Then both the even and the odd
L int l21
potential terms contribute to our ExM’s within the
ZSPBC’s, but not for the StdPBC’s, as we have shown. 5 ((
l51 k50
F̄ m1k,m1k2l 50, ~B3!

APPENDIX B: MINIMIZATION OF THE


PERIOD-AVERAGE POTENTIAL ENERGY where F̄ i, j is the period-average force on particle i due to
AND VANISHING DYNAMICAL STRESS particle j. Inspection of this equation reveals that the double
sum involving F̄ is just the sum over all the period-average
We show that the condition of zero dynamical stress interparticle forces acting from one side across an imaginary
across the supercell boundary follows from requiring that the plane separating particles m and m21. This is exactly the
period-averaged potential energy of the periodic boundary definition of the period-average dynamical stress across that
condition supercell is minimized. We consider an plane as given in the main text. Thus we have shown that the
N-particle 1D lattice, where particles n and n2l interact via minimization of the period-average potential energy in the
an assumed central potential V l (r n 2r n2l ), with supercell leads to vanishing period-average dynamical stress
l51, . . . ,L int . Here, L int,N/2 restricts the range of the in- across an imaginary plane separating any pair of adjacent
teraction. This is more general than the nearest-neighbor in- particles in the supercell, including the supercell boundary.
teractions case considered in the main text, reducing to it for Note, however, that our argument does not require the sec-
L int51. In addition, we generalize the assumed time depen- ond derivative of Ū with respect to the set $ Db n % to be
dence for the particle motion, given by Eq. ~3!, to positive and is therefore not reversible; it applies whenever
r n (t)5c n f (t)1b n 1na, where f (0)51 and f (t) is periodic Ū is an extremum. Although we have not proven in general
with period T, but has no zero-frequency Fourier component. that the zero dynamical stress condition is equivalent to a
Within the RWA of Eq. ~3! we would have f (t)5cos(vt). minimization, we have evidence that this is indeed the case:
The instantaneous potential energy is given by for the parameter values of the diatomic (k 2 ,k 3 ,k 4 ) lattice
N L int N L int studied in Sec. IV, it is easy to show analytically that Ū is a
U~ t !5 (( V l ~ r n 2r n2l ! 5 ( ( V l @~ c n 2c n2l ! f ~ t ! minimum for any mode pattern. Moreover, for any physi-
n51 l51 n51 l51 cally reasonable values of (k 2 ,k 3 ,k 4 ) ~i.e., k 2 .0 and
1 ~ b n 2b n2l ! 1la # . ~B1! k 4 .0), it can be shown analytically within the RWA that
this quantity is a minimum for both types of diatomic ExM’s
Since the interaction only depends on relative positions, we studied in this paper. Finally, numerical checks for ILM’s in
can rewrite Eq. ~B1! in terms of relative dynamic and static such lattices also yielded a minimum.
displacements between adjacent particles Dc n [c n 2c n21 Furthermore, we can show that if we require the period-
and Db n [b n 2b n21 . Thus c n 2c n2l 5c n 2c n21 average dynamical stress to vanish across an imaginary plane
1c n21 2•••2c n2l11 1c n2l11 2c n2l 5 ( l21 k50 Dc n2k and separating any particular pair of adjacent particles in the su-
similarly for b n 2b n2l . The period-average total potential percell, it will vanish for all other pairs of adjacent particles,
energy is irrespective of the range of the interaction.

*Also at Institute for Theoretical Physics, University of Regens- 2


A.J. Sievers and J.B. Page, in Dynamical Properties of Solids,
burg, D-93040 Regensburg, Germany. edited by G.K. Horton and A.A. Maradudin ~North Holland,
1
A. S. Dolgov, Fiz. Tverd. Tela 28, 1641 ~1986! @Sov. Phys. Solid Amsterdam, 1995!, Vol. 7, p. 137.
State 28, 907 ~1986!#; A. J. Sievers and S. Takeno, Phys. Rev. 3
A. M. Kosevich and A. S. Kovalev, Zh. Éksp. Teor. Fiz. 67, 1793
Lett. 61, 970 ~1988!; J. B. Page, Phys. Rev. B 41, 7835 ~1990!. ~1974! @Sov. Phys. JETP 40, 891 ~1974!#.
8846 D. BONART. T. RÖSSLER, AND J. B. PAGE 55

4
A. Tsurui, Prog. Theor. Phys. 48, 1196 ~1972!. the ZBM. However, Eq. ~5.2! of Ref. 4 and the corrected Eq.
5
Y. A. Kosevich, Phys. Lett. A 173, 257 ~1993!. ~44! of Ref. 13 disagree with each other for kÞ p /a. We have
6
K. W. Sandusky and J. B. Page, Phys. Rev. B 50, 866 ~1994!. therefore rederived Eq. ~5.2! of Ref. 4 using Eqs. ~4.1!, ~4.2!,
7
and ~5.1! of that paper and obtain V5 v $ 11 @ 2 q v 2 1 p 2 (4
3
S. R. Bickham, S. A. Kiselev, and A. J. Sievers, Phys. Rev. B 47,
14 206 ~1993!. 23 v ) # u f 61 u e % , in complete agreement with the corrected
2 2 2
8
M. Aoki, J. Phys. Soc. Jpn. 61, 3024 ~1992!; O. A. Chubykalo Eq. ~44! of Ref. 13.
16
and Y. S. Kivshar, Phys. Rev. E 48, 4128 ~1993!; M. Aoki, S. K. W. Sandusky, J. B. Page, and K. E. Schmidt, Phys. Rev. B 46,
Takeno, and A. J. Sievers, J. Phys. Soc. Jpn. 62, 4295 ~1993!; 6161 ~1992!.
17
M. Aoki and S. Takeno, ibid. 64, 809 ~1995!; A. Franchini, V. As in Ref. 6, we also find complex instability growth rates
Bortolani, and R. F. Wallis, Phys. Rev. B 53, 5420 ~1996!. l max , for both StdPBC’s and ZSPBC’s, over a range of OZCM
9
S. A. Kiselev, S. R. Bickham, and A. J. Sievers, Phys. Rev. B 48, amplitudes. The imaginary parts of these growth rates are com-
13 508 ~1993!. parable with the mode frequencies, so that these instabilities are
10
S. A. Kiselev, S. R. Bickham, and A. J. Sievers, Phys. Rev. B 50, not well described within the approximations of the RWA sta-
9135 ~1994!. bility analysis. In Ref. 6, such instabilities were investigated via
11 an exact ‘‘Floquet analysis,’’ and were found to be unrelated to
See, for instance, M. P. Allen and D. J. Tildesley, Computer
the question of ILM existence for the systems studied there.
Simulations of Liquids ~Clarendon, Oxford, 1987!. 18
12 More precisely, in our finite lattice with StdPBC’s, the ILM’s
The assumption of two distinct relative static displacements
background pattern away from the mode center corresponds to a
Db n yielding the same time-average force F̄(Ã) at a given à small-amplitude OZCM having the frequency of the ILM, but
leads to imaginary RWA frequencies, as long as the particle differing from the pure OZCM by the presence of a constant
motion is restricted to relative displacements r at which the strain. This ILM-induced strain renormalizes the force constants
fourth derivative of V(r) is positive. This encompasses relative such that the frequency of the background OZCM as a function
displacements deviating from the static equilibrium nearest- of its amplitude is raised slightly compared with the pure OZCM
neighbor separation a by as much as 30% for a Lennard-Jones curve ~a! of Fig. 1, while retaining the same qualitative behav-
potential and 50% for the BMC potentials considered in this ior. The effect of this strain is small in our 40-particle lattice and
work; moreover, it obviously imposes no restriction on the mo- vanishes in the infinite lattice limit. With ZSPBC’s, such strain
tion for a (k 2 ,k 3 ,k 4 ) potential with k 4 .0, such as used here. is completely removed by the supercell length change, for any
13
S. C. Lowell, Proc. R. Soc. London, Ser. A 318, 93 ~1970!. size lattice.
14
S. Wang, Phys. Lett. A 191, 261 ~1994!. 19
For completeness, we note that in diatomic lattices with realistic
15
In the monatomic limit the OZCM becomes the ZBM with potentials, at large amplitudes the frequency of the OZBM-ILM
Ã5A. We can compare this case with the results obtained for eventually enters the acoustic band and the mode changes its
small-amplitude anharmonic ExM’s in monatomic (k 2 ,k 3 ,k 4 ) character, exhibiting an acoustic ZBM-like extended mode back-
lattices in Refs. 13 and 4. Setting k5 p /a in Eq. ~43! of Ref. 13 ground, as mentioned in Ref. 10.
yields v (A) for the ZBM in agreement with our small-amplitude 20
S. Flach, Physica D 91, 223 ~1996!.
limit. However, when Eqs. ~44! and ~45! of Ref. 13 are com- 21
N. Flytzanis, St. Pnevmatikos, and M. Remoissenet, J. Phys. C
bined, the resulting v (A) disagrees with our result. We have 18, 4603 ~1985!; St. Pnevmatikos, N. Flytzanis, and M. Remois-
therefore rederived Eq. ~44! from Eq. ~43! and find that the senet, Phys. Rev. B 33, 2308 ~1986!; C. Tchawoua, T. C. Ko-
quantity f c appearing in the former equation is given by fane, and A. S. Bokosah, ibid. 50, 4189 ~1994!.
@ 4 (11 2 ba 22 ) # 2 (1/2), rather than the expression given by Eq.
3 1 22
D. Bonart, A. P. Mayer, and U. Schröder, Phys. Rev. Lett. 75,
~45! in Ref. 13. The corrected version of Eq. ~44! for k5 p /a 870 ~1995!; U. Schröder, D. Bonart, and A. P. Mayer, Physica B
then agrees with our small-amplitude limit for v (A) of the 219, 390 ~1996!.
ZBM. In Ref. 4, Eq. ~5.2! gives the v (A) behavior of small- 23
M. Remoissenet, Waves Called Solitons ~Springer, Berlin, 1994!.
amplitude anharmonic ExM’s, but it is not derived. For 24
R. Car and M. Parrinello, Phys. Rev. Lett. 55, 2471 ~1985!; O. F.
k5 p /a we find agreement with our small-amplitude limit for Sankey and D. Niklewski, Phys. Rev. B 40, 3979 ~1989!.

You might also like