Home Search Collections Journals About Contact us My IOPscience

Decay of a fundamental soliton in a periodically modulated nonlinear waveguide

This content has been downloaded from IOPscience. Please scroll down to see the full text.

1996 Phys. Scr. 53 385

(http://iopscience.iop.org/1402-4896/53/4/001)

View the table of contents for this issue, or go to the journal homepage for more

Download details:

IP Address: 93.180.53.211
This content was downloaded on 24/12/2013 at 09:07

Please note that terms and conditions apply.

Physica Scripta. Vol. 53, 385-393, 1996

Decay of a Fundamental Soliton in a Periodically Modulated

Non1inear Waveguide
Roger Grimshaw,’ Jianming He and Boris A. Malomed’
Department of Mathematics, Monash University, Clayton, Victoria 3168, Australia

Received July 26, I995 : accepted August 21, I995

Abstract after this, it was proposed to use the fibers with a variable
dispersion for the improvement of reshaping ultrashort
We report results of systematic numerical simulations of the nonlinear
Schrodinger equation whose dispersion coefficient is a periodic function of
optical solitons in long lossy optical fibers [SI. The basic
the propagation distance (formally, time). This is a model of propagation of idea was that, by adjusting the variable dispersion coeffi-
a subpicosecond optical soliton in silica fibers with a variable cross section, cient to periodic distribution of the gain along the fiber, one
which are now available for experiments. We demonstrate that, while very would be able to considerably suppress noise generation by
broad solitons are practically stable in this model, only slowly decaying the amplifiers. Independently, it was proposed to use the
into radiation, the fundamental solitons whose dispersion length is compa-
rable to the modulation period suffer a sudden splitting into a pair of sec-
nonlinear optical fibers with a similar profile of the inhomo-
ondary solitons, accompanied by a burst of radiation, after an initial period geneous dispersion coefficient in application to another
of steady evolution. The splitting takes place provided that the modulation practically important problem, viz., high-repetition-rate gen-
amplitude exceeds a certain critical value. This remarkable fact, as well as eration of a soliton array from a continuous wave [6].
the magnitude of the critical value, qualitatively agrees with predictions It is relevant to mention that nonlinear waveguides sup-
made recently for the same model on the basis of a variational approx-
imation. Drawing a separatrix between the nonsplitting and splitting soli- porting solitons is a rather general concept, which extends
tons in a parametric plane, we find a complicated structure. A set of beyond the framework of nonlinear optics. One example of
stability islands is found inside the splitting regime. Furthermore, we a solitonic waveguide in an altogether different physical
demonstrate that, at large values of the modulation amplitude, there exists context is the natural guide for internal waves in the ocean
a stability “isthmus” between two large splitting areas, in which the soliton (see [7] and [8]). The guided internal waves may easily
retrieves its stability. Unexpectedly, the soliton’s stability is again restored
at fairly large values of the modulation amplitude. In most cases, the become nonlinear, and, simultaneously, the natural oceanic
(quasi-)stable soliton is actually a breather, which demonstrates persistent waveguides may be strongly inhomogeneous [SI. In the
long-period shape oscillations. lowest approximation, the corresponding model again
amounts to the NLS equation with a variable dispersion
coefficient (see, e.g., [SI).
1. Introduction Thus, analysis of soliton propagation in nonlinear inho-
mogeneous waveguides is a relevant topic, with possibilities
The current activity in the field of nonlinear light propaga- for real applications. Among problems of this type, con-
tion in optical fibers and planar systems [l] has invoked a sideration of the evolution of a fundamental soliton in a
renewed interest in soliton dynamics in nonlinear wave- waveguide with a periodically modulated dispersion coeffi-
guides. In particular, propagation of optical solitons in cient is especially interesting, as it should allow the testing
inhomogeneous fibers was theoretically considered some of hidden dynamical properties of the soliton (in particular,
time ago with an intention to use a natural compression of its feasible internal quasi-resonances). On the other hand,
the soliton in a waveguide with decreasing dispersion for fabrication of a long optical fiber with a periodically modu-
obtaining ultrashort pulses [2]. This problem has gained lated dispersion is quite possible, and, in combination with
practical importance with the arrival of technology allowing the availability of powerful laser sources of subpicosecond
the fabrication (by drawing from the melt) of sufficiently solitons, theoretical analysis of this problem can suggest
long variable-cross-section silica fibers with a well- ideas for new experiments. The objective of the present
controlled local dispersion [3]. Experiments with sub- paper is to report results of systematic numerical simula-
picosecond solitons have demonstrated that high-quality tions of the propagation of the fundamental soliton in a
compression can indeed be achieved in these fibers, which model of a nonlinear optical fiber with a periodically modu-
have been given the name dispersion-decreasing fibers lated dispersion coefficient. It will be demonstrated that,
(DDF) [4]. It has also been demonstrated that the DDF can while a broad soliton with small energy evolves in this
effectively operate in conjugation with the famous Erbium- model rather trivially, very slowly decaying into radiation
doped optical amplifiers, which is an important advantage and keeping an almost unperturbed shape, evolution of a
for using these inhomogeneous fibers in photonics. Soon shorter soliton with a larger energy, whose characteristic
dispersion length (the so-called soliton period, see below) is
comparable to or smaller than the modulation period, may
e-mail: rhjg@wave.maths.monash.edu.au.
be quite nontrivial, giving rise to an abrupt destruction of
Permanent address: Department of Applied Mathematics, School of
Mathematical Sciences, Raymond and Beverly Sackler Faculty of Exact the soliton after a steady initial period. This seems to be
Sciences, Tel Aviv University, Ramat Aviv 69978, Israel; e-mail: qualitatively the same effect which has recently been predict-
malomed@leo.math.tau.ac.il. ed semi-analytically in [lo]. The nontrivial behavior sets in
Physica Scripta 5 3
386 R. Grimshaw, J . He and B. A. Malomed

above a certain critical value (threshold) of the modulation 2. Formulation of tbe model and comparison with the
amplitude (which weakly depends upon the soliton’s variational approximation
energy), in accord with the prediction in [lo], where the 2.1. The model
abrupt destruction of the soliton was predicted in the form
of its rapid “spreading out” into radiation. In this work, we Recently, evolution of a soliton in a model with a harmo-
demonstrate that the actual destruction mode is different; nically modulated dispersion was considered in a semi-
the soliton splits into two secondary solitons, which is analytical approximation in [lo]. The model was taken in
accompanied by emission of radiation. This particular mode the form of the perturbed nonlinear Schrodinger (NLS)
could not be predicted in [lo], as the simplest variational equation,
approximation employed in that work did not have enough iu, + +(I + E sin ( k z ) ) ~ ,+, 1 U 1% = 0. (1)
degrees of freedom. An essential feature of the splitting is
that the secondary solitons, having an essentially smaller In terms of optical fibers, z is the propagation distance,
energy than the primary one, are much more stable. z = t - z<. is the so-called reduced time (6,is the group
After a proper rescaling, the model considered in this velocity of the carrier wave), u(z, z) is the complex envelope
work keeps only two nontrivial parameters mentioned of the electromagnetic field in the fiber (all the variables z , z
above, viz., the modulation amplitude and the initial energy and u are assumed properly nondimensionalized), the cubic
of the soliton. Our simulations demonstrate that the bound- term accounts for the Kerr effect, the mean value of the dis-
ary between splitting and nonsplitting solitons in the corre- persion coefficient is normalized to be 1, and the amplitude
E measures the depth of modulation of the local dispersion
sponding parametric plane (at fixed energy, this border
corresponds to the above-mentioned threshold) has a com- coefficient, the modulation period being 24k. The obvious
plicated shape. Moreover, inside the splitting region we have scale invariance of the NLS eq. (1) allows us to fix dimen-
discovered a set of small stability islands. We make a con- sionalities of the quantities by setting k 1, which will be
jecture that the exact shape of the border is “fuzzy”, possibly implied below.
having a fractal dimension, and that distribution of the sta- As concerns applicability of the relatively simple model
bility islands is also fractal. Of course, more extensive simu- (l), which ignores modulation of all the parameters of the
lations are necessary to better substantiate this hypothesis. nonlinear waveguide excepting the dispersion coefficient,
However, despite the complexity of the border between the and excitation of all other modes but the one governed by
two parametric regions, it demonstrates a simple gross (l),it is necessary to stress that the effective dispersion of a
feature; above a certain minimal energy, the mean position silica optical fiber is rather small (to allow the weak Kerr
of the border almost does not depend upon the energy. This nonlinearity to manifest itself), which is achieved by a
fact was predicted in [lo], and, moreover, the nearly con- partial mutual compensation of two different contributions ;
stant threshold value of the modulation amplitude corre- the material dispersion of the silica, and the dispersion
sponding to the mean position of the border is quite close to induced by the waveguide geometry [l]. Therefore, this is
that predicted in [101. just the dispersion coefficient which is especially sensitive
Our simulations reveal another fairly novel and unex- even to a weak inhomogeneity of the fiber. As for excitation
pected gross feature of the soliton’s evolution; with further of other modes, it is known that a microscopic periodic
increase of the modulation amplitude, we find a region in modulation of the fiber’s refraction index (usually called
which the soliton effectively retrieves its stability, and then grating) may generate a counterpropagating wave through
we encounter another threshold beyond which the soliton the resonant Bragg scattering (see, e.g., [ll]), but in the
gets destroyed once again (in a more violent way). Thus, present case this process is ruled out by the disparity
there is a stability isthmus between two large ranges of the between the assumed period of the dispersion modulation
splitting (see Fig. 5 below). Nonetheless, as we keep increas- (dozens or hundreds of meters, see below) and the light
ing the modulation amplitude, we reach a new border wavelength.
(actually, it proves to be the ultimate one), beyond which the
soliton is again relatively stable. It is noteworthy that, close 2.2. Comparison with the variational approximation
to the first splitting threshold, then inside the above-
In [lo], emphasis was made on the search for a mechanism
mentioned stability isthmus, and, finally, beyond the utter-
of decay of the soliton in the periodically modulated wave-
most stability border, the soliton is essentially different from
the unperturbed one; it seems more like a breather, demon- guide based on resonant excitation of the soliton’s internal
strating persistent long-period shape oscillations. The period oscillations. Indeed, it is known, first of all from numerical
of the oscillations is typically almost ten times as large as simulations [l2], that decay of a pulse consisting of a
the underlying modulation period. soliton and some amount of radiation into the pure soliton
The rest of the paper is organized as follows. In Section 2, and free radiation waves takes a sufficiently long time,
we give a detailed formulation of the model. For the sake of demonstrating long-lived shape oscillations of the pulse. A
comparison, in this section we also recapitulate the varia- fully analytical approximate description of the pulse’s oscil-
tional approximation to the same problem developed in lations, completely neglecting the slow radiative attenuation,
[lo]. At the end of Section 2, we briefly discuss some recent was put forward on the basis of a variational approximation
related works. In Section 3, we display basic numerical in [13]. This approximation allows one to predict the fre-
results and discuss them in detail. In Section 4,we compare quency of the shape oscillations as a function of initial
our results with somewhat similar results obtained pre- amplitude and width of the pulse. It is assumed that initially
viously in terms of different models. Concluding remarks are the pulse has the form
collected in Section 5 . uo(z) = A sech (.t./a), (2)
Physica Scripta 53
 Decay of a Fundamental Soliton in a Periodically Modulated Nonlinear Waveguide 387

but with a “wrong” relation between the amplitude A and remedy this shortcoming). Anyway, comparison with
width a, different from that for the exact fundamental numerical simulations demonstrates that the variational
soliton, Aa = 1. The variational approximation is based on approximation is qualitatively correct, and, on the other
postulating that, at any value of z, the pulse has the shape hand, its unique asset is its ability to be implemented in a
Ccf. (211 fully analytical form.
Returning to [lo], the same variational approximation
+
u(z, t ) = A sech (z/a) exp (i$ ibz2), (3) was used there to analyze the soliton’s dynamics in the
in which the functional dependence upon z is fixed a priori, periodically modulated fiber described by (1). This form of
but the free parameters, A , a, 4 and b are allowed to be the perturbed NLS equation still admits a Lagrangian
arbitrary functions of z, b being the famous (though it may representation, and the energy (5) remains an exact integral
also be called notorious) chirp coefficient. Further analysis of motion. The analysis, based on the same ansatz (3), yields
assumes substitution of the ansatz (3) into the Lagrangian of the final evolution equation in the form
the NLS equation, and integrating out the explicit z-
dependence, thus arriving at an effective Lagrangian, which (7)
is a function of the aforementioned free parameters and
their first derivatives in z (actually, only d$/dz and dbldz where D(z) 3 1 + E sin z is the local dispersion coefficient in
are present [13]). Finally, one can deduce from this (l), which now plays the role of the inverse effective z-
Lagrangian a system of variational ordinary differential dependent mass in the equation of motion (7), and the effec-
equations which govern evolution of all these parameters in tive potential is [cf. (4)]
z. Further, all the parameters can be excluded from the
equations in favour of a single one, a, which measures the L
U(a, Z) = - [D(z)u-~- ~ E u - ’ ] .
instantaneous width of the pulse according to (3). In partic- n2
ular, the relation between the chirp b and a is b =
f a - ‘(daldz) (i.e., appearance of the chirp is inevitable once The basic idea of the analysis developed in [lo] was that
the pulse’s width is allowed to vary [13]). The remaining a possible resonance between the natural oscillations in the
evolutional equation for a(z) takes the form of the Newton’s above-mentioned potential well and the periodic modula-
equation of motion for a unit-mass particle in the effective tion of the local dispersion might give rise to a strong dya-
potential (obtained in [13]) mica1 response of the soliton to the modulation. The
resonantly driven oscillations were analyzed in detail for
2 small modulation amplitude E and small oscillations near
U(a) = - ( a - 2 - ~ E U - ’ ) , (4) the well’s bottom. However, the main emphasis was made
n2
on numerical simulations of the ordinary differential eq. (7)
where E A2a is an integral of motion of the variational without assuming E or the oscillations’ amplitude small, with
equations. Actually, E coincides with the exact integral of
the objective of detecting a transition from the oscillations
motion of the underlying NLS equation, which is usually
of the particle trapped in the well to another type of motion,
called (in optics) energy of the pulse:
viz., progressive motion of the free particle to infinity. In the
framework of the approximation adopted, this transition
clearly implies a decay of the soliton into radiation.

-
Numerical simulations were done in [lo] for the solitons
If the point corresponding to the initial pulse (2) gets into with initial energies 1. It was found that for each value of
the potential well described by (4), the pulse is predicted to E in this range, there was a critical value E,, , at which the
vibrate indefinitely long. Of course, in reality its vibrations transition took place. At E slightly below the critical value,
will be subject to radiative damping, which is ignored in this the observed oscillations were clearly chaotic. The decay
approximation. If the initial point is outside the potential observed when E slightly exceeded E,, can be described as
well, the pulse is expected to spread out unlimitedly, which follows : the particle would perform a number of seemingly
actually implies decay of the pulse into radiation. It imme- chaotic oscillations, with rather large amplitudes, inside the
diately follows from the form of the potential (4) and from well, and then it would be suddenly kicked out from the
the definition of E that the decay takes place provided that well, finding itself moving to infinity. In all the range of the
~131 energies considered (which was not very broad), E,, was
found to be nearly constant, taking values between 0.20 and
0.25.
As concerns the physical realizability of this phenomenon,
However, for the initial pulse (2) an exact solution in terms it is relevant to emphasize that the resonant conditions
of the inverse scattering transform [14] is known [12], imply that the modulation period must be of the same order
which predicts that the exact threshold value of the product of magnitude as the so-called soliton period (the soliton’s
Aa for the decay is lower, (Aa),,, = 1/2. Thus, some initial dispersion length), which, in the notation adopted here, is
pulses which are predicted by the variational approximation
to completely decay, will in reality produce a fundamental T = 4 7 ~ E - ~ (9)
soliton after shedding off a substantial amount of radiation, for the ideal soliton [note that the natural period of small
although the gap between the approximate and exact oscillations near the bottom of the potential well (4) is
thresholds is not broad (in [13], some improvements to the Z ~ E - ~For ] . usual optical solitons with widths of the order
variational approximation were set forward in order to of dozens of picoseconds, which are intended for use in the
Physica Scripta 53
388 R . Grimshaw, J . H e and B. A. Malomed

long communication lines, the soliton period is several in the presence of the periodic modulation of the birefrin-
hundreds of kilometers [l], so these broad solitons certainly gence [unlike (7), this was a system of equations for two
cannot be used to realize the effect. However, for ultrashort degrees of freedom, viz., the separation between the centers
subpicosecond solitons, which are now also available for of the two components and their common width]. Further
experiments, the soliton period, being proportional to the numerical simulations of these ordinary differential equa-
squared soliton’s width, can be made as small as several tions have demonstrated that a splitting of the vector
dozens meters, which should open the way to real experi- soliton into two single-component solitons is possible under
ments. the action of the modulation, and this splitting has a well-
Of course, the variational approximation, which leaves pronounced resonant character ; a threshold value of the
the soliton only two degrees of freedom [amplitude and modulation amplitude, necessary to induce the splitting, has
width, which further collapse into the single-degree-of deep minima at the values of the soliton’s energy at which
freedom system (7), taking into account the integral of the modulation period is close to resonances with the eigen-
motion (5)], must be thoroughly checked against direct frequencies of the free internal vibrations of the vector
numerical simulations of (l), which is the main objective of soliton.
the present work. In particular, as was mentioned above, the It is well known that, aside from the fundamental soli-
exact solution of the NLS equation for the initial pulse (2) tons, the unperturbed NLS equation also has the so-called
yielded a decay threshold which was lower than that pre- N-soliton solutions, which may be formally regarded as
dicted by the variational approximation; thus, one may bound states of N fundamental solitons. However, in the
expect that the solitons are actually more stable than was absence of perturbations the binding energy of any N-
predicted in [ 101. Results of systematic numerical simula- soliton is exactly equal to zero [14], that is why a pertur-
tions of (1) with the solitonic initial conditions will be set bation may easily split it into fundamental solitons.
forth in the next section. It will be demonstrated that a criti- Recently, splitting of a 2-soliton was simulated numerically
cal value of E (the threshold) does exist, such that the soliton in the model (1) [19]. It was found that the splitting had a
very slowly decays into radiation below the threshold, but well-pronounced resonant character. Still earlier, splitting of
above it the soliton suddenly gets destroyed after a steady the 2-soliton was simulated in a model describing propaga-
initial period of evolution. Moreover, the critical value tion of a broad soliton in a long communication line with
proves to be quite close to that predicted in [lo], and, unex- losses and periodic reshaping [20] (although the pertur-
pectedly enough, it is somewhat smaller than was predicted bation in that model is dissipative, it is well known that it
by the variational approximation. Another salient feature of can be cast into a Lagrangian form [21]). In that work, a
the results presented below is that beyond the destruction complex resonant structure was found too. It is necessary,
threshold the soliton does not fully decay into radiation, but however, to emphasize a principal difference of the problem
instead splits into two new solitons and some amount of considered in the present work: we are dealing with splitting
radiation. This feature, of course, could not be predicted of the fundamental soliton, which is obviously a “deeper”
within the framework of the variational approximation used dynamical process than the splitting of the two-soliton state
in [lo]. with the zero binding energy.
To conclude this section, it is pertinent to mention that
2.3. Comparison with other works the problem to be considered below can be formulated
Very recently, some results of simulations of the same equally well in terms of planar nonlinear light guides [22] if
model, as well as results of a perturbative calculation of one replaces the temporal variable z in (1) by the transverse
radiative corrections to the solitonic wave field induced by coordinate x. In this case, one will be dealing with the
the periodic modulation of the dispersion coefficient, were planar system modulated in the z-direction but homoge-
published in [lS]. However, only an early stage of the neous in x ; accordingly, the soliton will be a spatial soliton
soliton’s evolution was actually simulated in that work, instead of one in a temporal domain.
when a weak emission of radiation took place, but no
sudden splitting or other remarkable events happened. 3. Numerical results
Another effect, qualitatively similar to the one considered
3.1. Stability of the broad solitons
in [lo] and in the present work, was recently considered in
[16] ; resonant splitting of a vector (two-component) soliton In all the simulations, we solved (1) numerically with the
in a system of two coupled equations governing nonlinear initial condition [cf. (2)]
light propagation in a birefringent optical fiber, the two
uo(z) = E sech (ET), (10)
wave variables corresponding to two independent polariza-
tions of the light. It was also assumed that the fiber was which should give rise to an exact soliton with the energy E
periodically modulated, but this time the only modulated in the absence of the modulation (in applications to non-
parameter of the model was the birefringence coefficient (the linear fiber optics, a commonly used physical characteristic
group-velocity difference between the two polarizations). In of the soliton is its peak power E’). Equation (1) was solved
the homogeneous (unmodulated) two-component model, the with periodic boundary conditions in T; however, an artifi-
vector soliton has some internal vibrational (quasi-)modes, cial strongly localized dissipation was installed at the edges
which can be effectively studied by means of the variational of the integration domain, in order to absorb the emitted
approximation [17] (see also [18]). The most important radiation and prevent it from re-entering the integration
mode is related to mutual oscillations of centers of two com- domain.
ponents of the vector soliton. In [16], a system of effective In this problem, we are dealing with two control param-
variational equations of motion was derived for this model eters, viz., the modulation amplitude E and the initial
Physica Scripta 53
 Decay of a Fundamental Soliton in a Periodically Modulated Nonlinear Waveguide 389

soliton’s energy E [recall that the modulation wavenumber 3.2. TheJirst splitting threshold
k in (1) was set equal to 11. First of all, we Will mention Now, we proceed to a more interesting case of solitons with
relatively trivial results obtained at small values of the a larger initial energy, so that the soliton period (9) becomes
modulation amplitude E. A typical picture is shown in Fig. 1. comparable to or smaller than the modulation period 271.
AS one sees, the weak periodic modulation gives rise to a Although in an experiment it would be natural to vary the
weak emission of small-amplitude radiation waves by the initial soliton’s energy at a fixed value of the modulation
Soliton. Due to the radiative losses, the soliton gradually amplitude E , for a better understanding of the theoretical
becomes broader and lower, but no dramatic events results it is more convenient to gradually increase E at a
happen. We always observed similar behavior when E was fixed value of E. The structure of the soliton becomes qualit-
essentially larger, but the initial energy E was small enough, atively different from the simple, nearly unperturbed, struc-
so that the soliton was broad from the very beginning. ture observed at small E ; the soliton gradually turns into a
The reason for the relative stability of the soliton in the breather, demonstrating long-period persistent oscillations
case Of Small E is trivial: a weak perturbation gives rise to a of its width and height, as it is shown in Fig. 2. Notice that
weak effect. It is not difficult either to find a qualitative the period of the shape oscillations, approximately equal to
explanation for the effective stability Of the broad SOlitOn, 40 units in the case shown in Fig. 2, is much larger than the
even if E is not small. Indeed, when the Soliton period given modulation period 2x. It is also noticeable that the soliton
by (9)is much longer than the modulation period 2% the sheds a conspicuous amount of radiation at the initial stage
variable Part of the dispersion COefficient in (1) can be of the evolution, which we interpret as rearrangement of the
simply averaged Out, SO that the soliton, in the first aPProx- soliton into a breather adapted to the modulation in the
imation, feels only the constant mean value of the disper- system.
sion, while the variable Part gives rise to a weak However, the most important fact revealed by the simula-
higher-order effect in the form of emission of radiation. tions following this pattern (increasing E at fixed E) is that,
A remarkable fact is that the stability of the broad soli- Once E has attained a certain critical value &,,, which is a
tons Persists UP to quite large values of the ~odulation function of the energy E, the soliton suddenly splits after a
amplitude E. Notice that when E becomes larger than 0% steady initial period of evolution. Two typical examples of
the local dispersion coefficient in (1) can change sign. Evi- the splitting at values of E which are slightly larger than the
dently, in this case the approximation based on (7) is irrele- corresponding cfitical values are shown in Fig. 3. The
vant, as singular points periodically appear in the equation example shown in Fig. 3b is especially interesting, as, firstly,
(actually, passage of a localized pulse through a point where it is Seen that a weak “vestige” of the original soliton
the dispersion changes its sign can be considered in the remains in its place, and, secondly, it Seems that the two
framework of the variational approximation, but this con- solitons into which the original soliton splits recombine for
sideration requires a more accurate analysis [23]). Never- a short while, and then they eventually separate.
theless, our simulations demonstrate that the evolution Of ~ ~that ~the most
~ important
l l qualitative prediction of
broad solitons remains steady even in this case. the variational approximation developed in [lo] is that the
Of course, in a real Physical system higher-order disPer- soliton must suddenly start to spread out after an initial
sions are present, so that one should add to eq. (1) the third period of steady shape oscillations, provided that E exceeds
derivative in a vicinity of the point where the coefficient in a critical value. T E ~result was obtained for energies at
front of the second derivative vanishes [I]. The variational which the soliton period was not too different from the
approximation can be extended to this case as well [241. modulation period, so that the sudden decay of the soliton
However, in this work we do not consider effects Produced could be realized, in loose terms, as a resonant effect. As we
by the higher-order dispersion. said above. the variational approximation (at least, the
simple version employed in [lo]) underestimated the ability
of a solitary pulse to produce a genuine soliton. Therefore,

Fig. 1. A typical example of the steady evolution of a soliton governed by

(1) for small e: E = 0.1, E’ = 2.5, where E’ is the peak power of the soliton
( E is the soliton’s energy) defined according to (10). In this and all sub-
sequent plots, we show the z - T profiles of I U [. Hereafter, the modulation Fig. 2. A typical example of the stable breather-like soliton close to the
wavenumber kin (1) is set equal to 1. splitting threshold: E = 0.3, E’ = 2.5.

Physica Scripta 53
390 R. Grimshaw, J . He and B. A. Malomed

Fig. 4. Rapid decay of the soliton deep inside the splitting range: E = 0.8,
E’ = 2.5.

(Fig. 3) when E - E,, is small. It is not surprising that qualit-

atively the same dependence of duration of the pre-decay
stage of the soliton’s evolution upon the overcriticality was
observed in the framework of the variational approximation
in [lo].
As is seen in Figs 3 and 4, the soliton splits into a sym-
metric pair of secondary solitons with considerably smaller
energies, which is accompanied by emission of a conspicu-
ous amount of radiation. The appearance of the secondary
solitons is an essentially novel fact, which could not be pre-
dicted by means of the approximation employed in [lo]. It
is, of course, interesting to trace further evolution of a sec-
Fig. 3. Typical examples of the soliton splitting at E slightly exceeding eSr: ondary so1iton* Since it appears with a sma11er energyy it
(a) E = 0.3; E* = 2.9; (b) E = 0.4; E’ = 2.5. may be expected to be more stable than the primary soliton.
Our preliminary simulations aimed at isolating a secondary
soliton agree with this expectation, but this issue will be
one would expect the exact value E,,, which is to be found considered in detail elsewhere.
here from the numerical simulations, to be larger than the Now, we aim to follow changes which happen with
value predicted by the variational approximation for the further increase of E. The results to be described below are
same soliton’s energy. A surprising fact is that, in reality, the summarized in the phase diagram shown in Fig. 5. The first
exact critical value proves to be somewhat smaller than the remarkable fact is the existence of a whole set of small sta-
one predicted by the variational approximation. Actually, bility islands inside the splitting region. Resolution of a fine
for all the soliton’s peak power E’ > 2.5, the values E,, was structure which we believe should exist in this phase
found to vary between 0.15 and 0.20, which should be com- diagram would imply a huge computational job, which was
pared to the above-mentioned critical value produced by beyond our practical possibilities. Nevertheless, we set
the variational approximation, which was varying between forward a hypothesis that the genuine border between the
0.20 and 0.25. splitting and nonsplitting regions may have a fractal struc-
Anyway, it is relevant to emphasize that the variational ture, and that the sit of the stability islands adjacent to the
approximation, crude as it is, has correctly predicted the border is fractal too.
“catastrophic” decay of the soliton after a period of a rela- It is also noteworthy that the left splitting region in Fig. 5
tively steady initial evolution, and, moreover, E,, produced (the right one will be discussed below) is almost exactly
by this approximation is not too far away from the exact bounded from below by the value E 2 = 2. On the other
value. This is evidence in favor for using the variational hand, (9) yields, at this value of E’, the soliton period equal
approximation to obtain at least qualitative information 2n, i.e., the one coinciding with the underlying modulation
about basic properties of complicated dynamical models. period. This gives an additional argument in favor of inter-
The duration of the initial steady stage of the soliton’s preting the splitting as a certain resonant effect, in accord-
evolution, at which only a weak emission of radiation takes ance with the philosophy of [lo].
place, depends upon the size of the overcriticality E - E,, . In
principle, the steady stage must be infinitely long when E is 3.3. Higher sP1itting-unsP1ittingthresho1ds
exactly equal to the critical value. With increase of the over- When E is essentially larger than the above-mentioned E,, ,
criticality, this stage is expected to become shorter. In Fig. 4, we encounter a new parametric border separating different
we display a typical example of the splitting taking place far types of evolution of the soliton. Beyond this border, the
beyond the threshold. As can be seen from this figure, the solitons retrieve their stability, but at still larger E there is
splitting indeed becomes quick in comparison with the cases another threshold, beyond which the solitons split again.
Physica Scripta 53
 Decay of a Fundamental Soliton in a Periodically Modulated Nonlinear Waveguide 391

Fig. 7. A typical example of the soliton splitting in the domain to the right
of the “stability isthmus”: E = 1.4,E’ = 2.5.

in (1) changes its sign periodically. Moreover, we could not

find any end of this isthmus from the side of the large E and
Fig. 5. The phase diagram on the parametric plane (E, E’). The filled and
E . Thus it SeemS thaty at Some va1ues Of its energy9 the
unfilled rectangles correspond, respectively, to the stable and splitting soli- soliton finds an effective way to stabilize itself in a system
tons. where the dispersion changes sign.
In Fig. 6, we display typical pictures of quasi-stable
Thus, we have found a “stability isthmus” of nearly constant soliton propagation at two parametric points which are
close, respectively, to the left and right edges of the isthmus.
width, sandwiched between two large splitting areas (Fig. 5).
It is noteworthy that almost all of the “isthmus” lies at Qualitatively, these pictures are not very different from Fig.
2, which corresponded to a point near the first splitting
E > 1, so that the corresponding local dispersion coefficient
threshold. As well as in Fig. 2, the most salient feature of
Fig. 6 is persistent long-period shape oscillations of the
soliton, lending it a breather-like structure.
A typical example of the splitting in the area lying to the
right of the stability isthmus is shown in Fig. 7. The splitting
happens very quickly, and in this case it is, actually, hard to
say if the original soliton splits into two secondary ones, or
simply into two wave packets of radiation.
Another major unexpected fact is that, keeping increasing
E, we once again encounter a stability domain, which is an
extension of the stability area contiguous to the axis E = 0.
Subsequent increase of E up to the value 2.0 did not reveal
any new splitting threshold bounding this domain. Thus,
instead of pulling the instability border closer to the axis
E = 0, which would be a naive expectation, increase of E
pushes it off. An example of the quasi-stable soliton,

Fig. 6. Stable solitons at the left and right edges of the “stability isthmus” Fig. 8. A typical example of the restabilized soliton to the right of the ulti-
from Fig. 5: (a) E = 1.0, E’ = 2.5; (b) E = 1.2, E’ = 2.5. mate splitting border (at large E ) : E = 2.0, E’ = 2.5.

Physica Scripta 53
392 R. Grimshaw, J . H e and B. A . Malomed

obtained just beyond this ultimate stability border, is dis- complicated dynamical structure was found in soliton-
played in Fig. 8. This is a breather-like soliton seemingly soliton collisions in the nonintegrable system of two coupled
quite similar to those found in other stability regions (cf. NLS equations describing a bimodal nonlinear optical fiber.
Figs 2 and 6). As concerns the latter system, similar but less detailed
T o conclude this section, it is relevant to mention that the numerical results were earlier reported in [29]. In the work
set of Figs 1, 2, 3(b), 4, 6(a), 6(b), 7 and 8 furnishes a cut of [29], the observed complicated dynamical pattern was qual-
the parametric space along the constant value of the itatively explained in terms of collision-induced resonant
soliton’s peak power E’ = 2.5, which allows us to directly excitation of the soliton’s internal eigenmodes (see more
trace the change of the soliton’s behavior with the increase details in [30]).
of E.
5. Conclusion
4. Comparison with other models
In this work, we have reported results of systematic numeri-
Existence of “stability windows”, in which a complicated cal simulations of evolution of a soliton governed by the
dynamical system retrieves a relatively simple behaviour NLS eq. (1) with a periodically modulated dispersion coeffi-
after having performed transition to a more complex evolu- cient, which is the simplest model of a periodically inhomo-
tion pattern, is known in models describing soliton-soliton geneous nonlinear waveguide. Motivation for this study was
collisions in nonintegrable systems, such as the double sine- suggested by the work [lo], where the soliton’s evolution in
Gordon or the 9’ model [25]. Later, a similar phenomenon the same model was considered by means of the semi-
was considered in detail in the framework of a simpler analytical approximation based on a variational principle.
problem, viz., collision of a single soliton with a local inho- We have found that broad low-energy solitons, whose dis-
mogeneity (defect) [26]. In these problems, the most essen- persion length is much larger than the modulation period,
tial control parameter was a collision velocity (in the 94 are very stable in this model, even in the case when the local
model, this was the only control parameter, while in the dispersion coefficient periodically changes its sign. These
double sine-Gordon model one has an additional param- broad solitons very slowly decay into radiation, suffering no
eter, which is a coefficient in front of the perturbation term catastrophic changes. Contrary to this, solitons with the dis-
in the underlying equation). The “windows” were found as persion length comparable to the modulation period
intervals of values of the velocity in which the soliton- demonstrate a sudden splitting into a pair of secondary soli-
soliton or soliton-defect collisions recovered their quasi- tons and radiation, provided that the modulation amplitude
elastic character, so that free soliton(s) reappeared after the exceeds a qxtain threshold. Duration of the initial period of
collision, although with smaller velocities and with some steady evolution preceding the sudden splitting is larger just
internal oscillations excited; also emission of radiation took beyond the threshold, and becomes smaller further in the
place. This elasticity recovery happened after the collisions transcritical region. This fundamental result qualitatively
had already been inelastic (resulting in capture of the agrees with the prediction made on the basis of the varia-
solitons) at larger velocities. tional approximation in [lo], although this approximation
All these models were dealing with topological solitons could not predict that formation of a pair of the secondary
(the so-called kinks), which in all cases had an internal solitons would be a generic outcome of the splitting. The
degree of freedom. A qualitative explanation for existence of critical value of the modulation amplitude which renders the
the “windows” has been given on the basis of analysis of a splitting possible proves to be close to the value predicted
resonant energy exchange between the motion of the kink as by the variational approximation, although somewhat
a whole and its internal degree of freedom [25, 261. It was smaller.
demonstrated (both numerically and semi-analytically) that Another remarkable fact revealed by the simulations is
the “windows” had a very complicated fine structure, alter- that, with a further increase of the modulation amplitude,
nating with lacunas in which the collisions were again there is an “isthmus” in which the soliton retrieves its stabil-
inelastic. Of course, only quasi-elastic collisions were pos- ity. At still larger modulation amplitudes, the splitting sets
sible at very large velocities, and all the collisions were in again. However, subsequent increase of the modulation
inelastic at very small velocities, after closing the last amplitude leads to an uttermost stability area, which seems
“window”. to be unlimited. The restoration of the quasi-stability of the
As we mentioned above, the soliton in the NLS equation solitons at large values of the modulation amplitude E sug-
has a quasi-mode of internal vibrations, which may help to gests consideration of the model
understand the complicated structures in the parametric
plane of our model, the existence of which is suggested by iu, + )(sin z)u,, + 1 U /’U = 0, (11)
Fig. 5 . Our hypothesis that the genuine structure of the which formally corresponds to E = CO, [cf. (l)]. Note that
splitting border should be fractal is, as a matter of fact, sug- the only control parameter in the model (11) would be the
gested by analogy with the above-mentioned works [25] soliton’s initial energy. Evolution of the soliton in this
and [26] (however, there was no rigorous proof that the model will be considered elsewhere. Note, however, that the
parametric structures dealt with in these works were fractal). model does not take into account the higher-order disper-
Lastly, complex interaction patterns have been found very sion (a term with the third derivative [l]), so that its physi-
recently for the usual non-topological solitons in numerical cal relevance may be limited.
simulations of different perturbed versions of the NLS equa- Returning to the model considered in this work, it is rele-
tion. In [27], this was found for interaction of the soliton vant to recall two more features found in this work. One of
with a local defect in a discrete NLS equation, and in [28] a them is the existence of the system of stability islands adja-
Physica Scripta 53
 Decay of a Fundamental Soliton in a Periodically Modulated Nonlinear Waveguide 393

cent to the first splitting border. We have put forward a 2. Tajima, K., Opt. Lett. 12, 54 (1987); Kuehl, H. H., J. Opt. Soc. Am. B
5, 709 (1988).
conjecture that the genuine structure of this border, as well
3. Bogatyrev, V. A. et al., IEEE J. Lightwave Technol. 9, 561 (1991).
as the set of stability islands, should be fractal. It is less clear 4. Chernikov, S. V., Dianov, E. M., Richardson, D. J. and Payne, D. N.,
if the higher splitting-unsplitting borders in Fig. 5 should Opt. Lett. 18,476 (1993).
also have a fractal structure. Another noteworthy feature is 5. Forysiak, W., Doran, N. and Knox, F. M., Opt. Lett. 19, 174 (1994);
that, in most cases, the quasi-stable solitons are actually Malomed, B. A., Opt. Lett. 19, 341 (1994).
6. Shipulin, A. V., Fursa, D. G., Golovchenko, E. A. and Dianov, E. M.,
breathers performing long-period shape oscillations. A
Electron. Lett. 29, 1401 (1993).
better understanding of structure of the breathers could be 7. Lighthill, J., “Waves in Fluids” (Cambridge University Press, Cam-
achieved by monitoring their soliton contents, i.e., applying bridge 1978).
to them the inverse scattering transform [14] and following 8. Miropolsky, Yu. Z., “Dynamics of Internal Gravity Waves in the
change of the corresponding discrete scattering data us. the Ocean” (Gidrometeoizdat Publishers, Leningrad 1981) (in Russian).
evolution variable z. This, however, requires very extensive 9. Malomed, B. A. and Shrira, V. I., Physica D 52, 157 (1991).
10. Malomed, B. A., Parker, D . F. and Smyth, N. F., Phys. Rev. E 48,
numerical computations. 1418 (1993).
A challenging problem for application of analytical 11. Aceves, A. B. and Wabnitz, S., Phys. Lett. A 141, 37 (1989);Martijn de
methods is explanation of the main features produced by Sterke, C. and Sipe, J. E., Opt. Lett. 14, 871 (1989).
our numerical simulations. One possibility might be to 12. Satsuma, J. and Yajima, N., Progr. Theor. Phys. Suppl. 55, 284 (1974).
13. Anderson, D., Phys. Rev. A 27, 3135 (1983); Anderson, D., Lisak, M.
make use of the fact that the linearized eq. (1) has an
and Reichel, T., J. Opt. Soc. Am. B 5, 207 (1988).
obvious relation to the Mathieu equation. It is well known 14. Zakharov, V. E., Manakov, S. V., Novikov, S. P. and Pitaevsky, L. P.,
that the latter equation gives rise to alternating stable and “Theory of Solitons” (Nauka Publishers, Moscow 1980) (in Russian)
unstable zones [31], which could be somehow related to [English translation: Consultants Bureau, New York 19841.
alternating stability and splitting regions in Fig. 5. However, 15. Abdullaev, F. Kh., Caputo, J. G. and Flytzanis, N., Phys. Rev. E 50,
using this idea is impeded by a rather wide margin that 1552 (1994).
16. Malomed, B. A. and Smyth, N. F., Phys. Rev. E 50, 1535 (1994).
separates the splitting range from the axis E = 0 corre- 17. Ueda, T. and Kath, W. L., Phys. Rev. A 42, 563 (1990).
sponding to the linearized eq. (1)(Fig. 5). 18. Kaup, D.J., Malomed, B. A. and Tasgal, R. S., Phys. Rev. E 48, 3049
The analysis presented in this work suggests an idea for (1993).
experiments with subpicosecond optical solitons, as the 19. Bauer, R. G. and Melnikov, L. A., Opt. Commun. 115, 190 (1995).
technology allowing to fabricate long silica fibers with con- 20. Hasegawa, A. and Kodama, Y., Phys. Rev. Lett. 66, 161 (1991).
21. Anderson, D., Lisak, M. and Reichel T., Phys. Rev. A 38, 1618 (1988).
trolled variable dispersion is now available, as well as 22. Barthelemy, A., Maneuf, S. and Froehly, S., Optics Commun. 55, 201
powerful and stable laser sources of the picosecond solitons. (1985); Maeuf, S., Desailly, R. and Froehly S., Optics Commun. 65,
Another implication of this work is that it lends additional 193 (1988); Maneuf, S. and Reynaud, P., Optics Commun. 66, 325
credit to the variational approximation, demonstrating that, (1988).
although it fails to grasp many important details, it is able 23. Malomed, B. A., Physica Scripta 47, 797 (1993).
24. Desaix, M., Anderson, D. and Lisak, M., Opt. Lett. 15, 18 (1990).
to qualitatively correctly predict gross features of dynamical 25. Campbell, D. K., Schonfeld, J. F. and Wingate, C. A., Physica D 9, 1
processes in complex systems. (1983); Peyrard, M. and Campbell, D. K., Physica D 9, 33 (1983);
Campbell, D. K., Peyrard, M. and Sodano, P., Physica D 19, 165
(1986).
Acknowledgements 26. Kivshar, Yu. S., Fei Zhang and Vkquez, L., Phys. Rev. Lett. 67, 1177
We are indebted to M. Quiroga-Teixeiro, who has performed first tentative (1991); Phys. Rev. A 45, 6019 (1992); Fei Zhang, Kivshar, Yu. S.,
runs of numerical simulations for (1) and allowed us to make comparison Malomed, B. A. and Vazquez, L., Phys. Lett. A 159, 318 (1991);
with his results (they accorded with ours). We appreciate a useful discussion Malomed, B. A., J. Phys. A, Math. Gen. 25, 755 (1992); Malomed, B.
of the results with A. S. Kovalev, who attracted our attention to the fact A., Campbell, D. K., Knowles, N. and Flesch, R. J., Phys. Lett. A 178,
that (1) may be related to the Mathieu equation. One of the authors 271 (1993).
(B.A.M.) appreciates support from the Department of Mathematics at the 27. Forinash, K., Peyrard, M. and Malomed, B. A., Phys. Rev. E 49, 3400
Monash University (Clayton, Australia). (1994).
28. Cao, X.and Meerhofer, D. D., J. Opt. Soc. Am. B 11, 380 (1994); Opt.
Lett. 19, 1711 (1994).
References 29. Malomed, B. A. and Wabnitz, S., Opt. Lett. 16, 1388 (1991).
30. Malomed, B. A., J. Opt. Soc. Am. B 9,2075 (1992).
1. Agrawal, G. P., “Nonlinear Fiber Optics” (Academic, Boston 1989); 31. “Handbook of Mathematical Functions” (Edited by M. Abramowitz
Hasegawa, A. and Kodama, Y., “Solitons in Optical Communica- and I. A. Stegun) (National Bureau of Standards, 1964).
tions” (Oxford University Press, Oxford 1995).

Physica Scripta 53