**Localized Dispersive States in Nonlinear Coupled M o d e Equations for Light **
**Propagation in Fiber Bragg Gratings* **

C. Martel M. Higuera and J. D. Carrasco

**Abstract.** Dispersión effects induce new instabilities and dynamics in the weakly nonlinear description of light
propagation in fiber Bragg gratings. A new family of dispersive localized pulses that propágate with
the group velocity is numerically found, and its stability is also analyzed. The unavoidable different
asymptotic order of transport and dispersión effects plays a crucial role in the determination of
these localized states. These results are also interesting from the point of view of general pattern
formation since this asymptotic imbalance is a generic situation in any transport-dominated (i.e..
nonzero group velocity) spatially extended system.

**Key words.** fiber Bragg gratings, nonlinear coupled mode equations, envelope equations, múltiple scale
meth-ods, localized solutions

A M S subject classifications. 34E13, 74J30, 74J34, 37K45, 78A60

**1. Introduction.** Fiber Bragg gratings (FBG) are microstructured optical fibers that
present a spatially periodic variation of the refractive index. The combination of the guiding
properties of the periodic media with the Kerr nonlinearity of the fiber results in the very
par-ticular light propagation characteristic of these elements, which make them very promising for
many technological applications that range from optical Communications (wavelength división,
dispersión management, optical buffers and storing devices, etc.) to fiber sensing (structural
stress measure in aircraft components and buildings, temperature change detection, etc.); see.
e.g., the recent review [12].

The amplitude equations that are commonly used in the literature to model one di-mensional light propagation in an FBG are the so-called nonlinear coupled mode equations (NLCME) [24, 7, 9, 8, 1, 10], which, conveniently scaled, can be written as

*(1.1) A+ -A+ = ÍKA~ + iA+(a\A+\2 + \A~\2): *

*(1.2) A¿ +A~ =mA+ + \A-(a\A-\2 + \A+\2): *

*where A^1* are the envelopes of the two counterpropagating wavetrains that resonate with the

The NLCME can be obtained from the full Maxwell-Lorentz equations using múltiple scales techniques in the limit of small grating depth, small light intensity, and slow spatial and temporal dependence of the field envelopes (see [10] for a detailed description of this derivation process).

It has been recently shown [16, 17] that light propagation in FBG can develop dispersive structures that are not accounted for in the NLCME formulation, and that, to correctly describe the weakly nonlinear dynamics of the system, the NLCME have to be completed with material dispersión terms:

*(1.3) A+ -A+ = ÍKA~ + \A+(a\A+\2 + \A~\2) + ieA+x, *

*(1.4) A¿ + A~ =\KA+ + \A~{a\A~\2 + \A+\2) + \eA~x. *

The dispersive nonlinear coupled mode equations (NLCMEd) above are scaled as the NLCME:
the characteristic length scale is the slow scale that results from the balance of the advection
term with the small effect of the grating, the characteristic time is the corresponding transport
time scale (which sets to one the scaled group velocity), and the characteristic size of the
wavetrains is the one resulting from the saturation of the small nonlinear terms. The slow
envelope assumption forces the dispersive terms to always be small as compared with the
advection terms; in other words, second derivatives of the slow amplitudes are much smaller
than their first derivatives. In the scaled equations this effect is contained in the scaled
*dispersión coefficient e (which measures the dispersión to transport ratio), and therefore, in *
order to be consistent with the slow envelope assumption, the NLCMEd must be considered
*in the limit e —> 0. The NLCMEd were already analyzed in [6], but they considered only the *
*case e ~ 1 that, as explained above, does not correspond to a generic situation from the point *
of view of large scale pattern formation in extended systems.

The NLCMEd can be regarded as asymptotically nonuniform, in the sense that the
*NLCMEd is an asymptotic model obtained in the e —> 0 (weakly nonlinear, slow envelope) *
*limit that still contains the small parameter e. This is the unavoidable consequence of *
simulta-neously considering two balances of different asymptotic order: one induced by the dominant
effect of the transport at the group velocity (balance described by NLCME) and a second one
that is associated with the underlying dispersive, nonlinear, Schródinger-like dynamics of the
system. The small dispersive terms in the NLCMEd are essential to describe the dynamics
of the system when it develops small dispersive scales ¿disp ~ A / Ñ - AS ^ w a s shown in [16],
*the NLCMEd in the e —> 0 limit constitute a singular perturbation problem (cf. [14]), and *
the onset of the dispersive scales is not a higher order, longer time effect; it takes place in the
*same time scale of the NLCME, no matter how small the dispersión coefficient e is. *

**I^**

**+**

**I**

** \A~ **

**\A~**

*o X* 1* o x* 1

**Figure 1.*** Space-time representation of a solution of the NLCMEd exhibiting small dispersive scales all *
*over the domaín (a = 1/2, n = 2, e = — 1CP*3*, and períodícíty boundary conditions). See the accompanyíng file *

(69822_01.gif* [22.4MB]) for an animation of the onset of the dispersive scales. *

*scales we have repeated the NLCMEd simulation but with a reduced dispersión e = —10*_3/4.
The result is plotted in Figure 2, where it can be seen that the small scales again flll the entire
*domain, but their typical size, ¿disp ~ A/Ñ> ^S n o w* approximately one half of that in Figure 1

(see also the corresponding animations 69822_01.gif [22.4MB] and 69822_02.gif [20.1MB]). The results in [10] rigorously establish that the solutions of the NLCME remain asymptot-ically cióse to solutions of the original Maxwell equations. This proximity result would appear to be in contradiction with the results just presented in Figures 1 and 2, which show NLCMEd solutions drifting away from the corresponding NLCME solutions because of the destabiliza-tion and development of the dispersive scales. There is no contradicdestabiliza-tion because the result in [10] does not give information about the stability of the solutions involved: when a stable NLCME solution is dispersively unstable, then the result in [10] simply ensures that there is a corresponding cióse solution of the Maxwell equations, but what happens in the Maxwell equation dynamics is that this solution is also unstable, and the physical system likes to move away from this solution cióse to the NLCME and develops dispersive scales, as the NLCMEd correctly predict. In other words, this dispersive destabilization is present in the Maxwell equations and is captured by the NLCMEd, but it is simply not seen in the framework of the zero-dispersion NLCME. This point was also numerically verified in [17], where the solutions of the NLCMEd and the Maxwell equations were computed and compared.

*X* 1

**Figure 2.*** Space-time representation of a solution of the NLCMEd exhibiting small dispersive scales all *
*over the domaín (a = 1/2, n = 2, e = — lCP*3*/4, and períodícíty boundary condítíons). See the accompanyíng *

*file* (69822_02.gif* [20.1MB]) for an animation of the onset of the dispersive scales. *

can be found in [17], and a similar derivation in the context of the oscillatory instability in extended dissipative systems is given in [19].)

The Maxwell-Lorentz equations for the original physical problem of light propagation in an FBG can be written as (using the same nondimensionalization as in [10])

(1.5)

(1.6)

*d2{E + P) d2E *

**di****2 **

*d2p *
*di2 *

*dx2 '' *

*-u2(l - 2Ancos(2¿))P + üú2{n2Q - l)E + u^P3 *

where the only small parameter is the intensity of the grating An ~ e < l . The basic spatial
*scale of the periodic grating is Xg = ir, and the rest of the parameters are of order 1 and *

*independent of e. We look for weakly nonlinear solutions that are a superposition of the *
grating resonant wavetrains

(1.7) *(E, P) - (A+é + ÍX-\-íüjt + A~e -ÍX-\-íüjt\ * + C.C. +

where c e . stands for eomplex conjúgate. The only assumptions required to obtain the envelope equations are

**«**

** \A%\ « \Á±| «**

**\A%\ « \Á±| «**

** 1^1 « 1, **

*í±*

_{< \Af | < 1^1 < 1, and An ~ e < 1. }*quantities An, Á^, Á~, Á~~,... is obtained. After applying solvability conditions to the *
res-onant problems at each order, we obtain the terms of the resulting envelope equations. We
now retain the first order nonlinear terms, the first order effect of the grating, and the first
*two linear effects (transport and dispersión), and, after rescaling x, t, and A^1* as in [10], we

end up with the NLCMEd equations (1.3)-(1.4).

*Note that if one sets a priori the scaling x,t ~ 1/e and l ^ l ~ y/e, then the standard *
NLCME (1.1)—(1.2) are obtained. However, if we try a dispersive nonlinear Schrodinger
*(NLS)-like scaling, x ~ 1/ \/e and l ^ l ~ y/e, then at first order just propagation with the *
*group velocity for times t ~ l/\/e is obtained. Also, if the length of the domain is of order *

*L ~ 1/y/e and we have periodicity boundary conditions, one ends up, in the longer time scale *
*t ~ 1/e, with two averaged NLS equations like the ones analyzed in* [15] and [18]. On the other
hand, the NLCMEd analyzed in this paper allow us to cover a more general problem: the
*onset, in the standard NLCME scenario with typical spatial scale x ~ 1/e, of small dispersive *
*scales x ~ 1/x/e- In this case, all scales in between those two can appear, and in fact they do *
appear, as is shown in Figures 1 and .

The main goal of this paper is to show that, in addition to complex spatio-temporal patterns, the dispersión effects can also give rise to new, purely dispersive localized states, which might be of interest from the optical Communications point of view. It is interesting to note that the results in this paper apply also to Bose-Einstein condensates in optical lattices (a system that has recently received very much attention [22, 25]) and, in general, to any dissipationless propagative system, extended in one spatial direction, reflection- and transla-tion invariant, and with a small superimposed spatial periodic modulatransla-tion of its background, since the NLCMEd are the appropriate envelope equations for the description of the weakly nonlinear resonant dynamics of this kind of system.

In order to show that light propagation in an FBG can happen in the form of dispersive pulses, in section 2 we derive and solve numerically an asymptotic equation for a family of symmetric pulses, and in section 3 we perform some numerical integrations of the complete NLCMEd to show that some of the pulses in this family do propágate as stable localized structures. Finally, some concluding remarks are drawn in section 4.

**2. Dispersive pulses.** The starting point is the continuous wave (CW) family of constant
uniform modulus solutions of the NLCME (1.1)—(1.2).

*(2.1) A+W = p eos 6 eiüjt+imx, *

*(2.2) A¿w = p sin 6 eiwt+imx, *

*where p > 0 is the light intensity flowing through the fiber, #£[—§,•§] measures the relative *
amount of both wavetrains, and the frequeney and wavenumber of the amplitudes are given
by

£ < 1

**A**

** cw **

**cw**

*A~ *

**- x *

**Figure 3.*** Sketch of a dispersive pulse on top of a continuous wave. *

*The CWs with |w| ~ 1 and \m\ ~ 1 are approximate solutions of the NLCMEd (up to order *

*e corrections) and their stability was first analyzed in [8] and then completed in* [16], where
it was found that, for both signs of the dispersión coefficient, there are dispersively unstable
CW which are stable in the dispersionless context of the NLCME.

The NLCMEd (1.3)-(1.4)* have to be considered in the small dispersión limit e —> 0. When *
looking for small-dispersion induced solutions it seems somehow natural to try to proceed in
a similar way as in the semiclassical limit of the NLS, i.e., looking for solutions with a fast
dispersive phase extended all over the domain (see, e.g., [4, 11]). But in this case, due to the
necessary dominance of the transport terms, one obtains at first order just linear
propaga-tion with constant group velocity (±1) for the phases, and therefore the phenomenology of
the semiclassical limit of the NLS equation (shocks, singularities, etc.) that comes from the
resulting nonlinear equation for the phase is not present in the regime explored here.

Instead, we look here for a different type of solution: small-dispersion induced solutions
in which the dispersión effects are not extended but localized. More precisely, we look for
*dispersive pulses of width ~ \/\e\ that propágate on top of a stable uniform CW, as sketched *
in Figure 3. In order to turn the background CW into a constant, it is convenient to first
perform in the NLCMEd the change of variables

*J^+ _ p+ J.üJt+imx *

*A— rp— iu>t-\-imx *

to obtain

*(2.5) F + - F+ + i(w - m)F+ = ÍKF~ + iF+(a\F+\2* + | F " |2*) + isF+: *

*(2.6) Ff + F~ + i(w + m)F- = ÍKF+ + iF-(a\F-\2* + |F+|2*) + \eF~x. *

*A localized dispersive pulse on F+ depends on the fast spatial scale X = xj\f\e\, and, *

according to (2.5)-(2.6),* in the short time scale T = t/^/\e\ ~ 1 it just propagates with the *
group velocity, suggesting that we have to look for solutions of the form

F + = F0+(r?,í) + - - - ,

F - = F0- ( r?, í ) + - - - ,

*with r¡ = X + T. Inserting the above ansatz into* (2.5)-(2.6) yields

*which means that F remains in first approximation equal to the unperturbed CW. Similarly. *
the following equation is obtained for F0+:

*(2.7) F+ + i(w - m)F+ = i/cpsin 9 + iF+(a\F+\2 + p2* sin2* 9) ± iF*0+^;

*where the + (—) sign corresponds to e positive (negative), together with the boundary *
con-ditions

(2.8) Fo*+ - » p c o s 0 for r¡ ->• ±oo. *

which ensure that the background CW is recovered away from the pulse.

*For dispersive pulses propagating over a zero background we have to set to zero p, w, and m *
in (2.7). A standard NLS equation is then obtained, which is known to exhibit localized pulses
*(solitons) in the focusing case of positive dispersión (recall that a > 0). Note that the effect of *
the grating on the dispersive pulses is felt only through the background CW, and therefore it
is completely gone in this case. In the scaling of the original problem of light propagation on
an FBG ((1.5)—(1.6)), the only nonzero Fourier spectrum components of these NLS solitons
*correspond, in first approximation, to dispersive wavenumbers (k = fc*resonant + Afc, with

*Ak ~ \f\s\) that are so off-resonance that they simply do not feel the grating (recall that *

fcresonant = ± 1 ; See ( 1 . 7 ) ) .

Equation (2.7) is an NLS equation with a direct forcing term coming from the effect of the grating. The steady solutions of this equation and their stability properties were analyzed in [2, 3], where explicit analytic expressions for the steady pulses were found. In this paper we consider the more general family of traveling localized solutions. More precisely, we look for traveling pulses of the form

F0*+ = pcos9(l + a(r¡ + vt)). *

*where v represents, in the original variables, a small correction of the group velocity. The *
resulting boundary valué problem for a, after making use of relations (2.3) and (2.4) and the
*rescaled variable £ = (?y + vt){^peos 9), can be written as *

*(2.9) a^ = —iva^ + aa — (\a\2 + a + a)(l + a), *

*(2.10) a ->• 0 as £ ->• ±oo. *

*where v = v/{^fapcos9) and a = ktan(9)/(ap2eos2 9). We are restricting our search to the *

focusing case of positive sign in (2.7) and, in order to have a dispersively stable background
*CW, we have to consider only the range 0 < 9 < | (see [16]), which implies that we have *
to look for pulses in (2.9)-(2.10)* only for a > 0. On the other hand,* (2.9)-(2.10) remain
*invariant under the transformation i) —> —i) and a —> a, and therefore we can also set v > 0. *

The solutions of (2.9)-(2.10) correspond to traveling dispersive pulses and can be regarded
*as homoclinic orbits (in the variable £) connecting the trivial state (Le., a = 0) back to itself. *

To analyze the existence of such connections we first consider the linearized system around the zero state:

*a *

**(c) **

**(d) **

**Figure 4.*** Eigenvalues of the trivial state as a function of the parameters a and v2. Sketches show the *

*distribution of the four eigenvalues in each of the distinct regions. *

where uT = (ar, a¿, (ar*)g, (a¿)g), ar* and a¿ are the real and imaginary part of a, (ar*)g = dar/d( *
and (a¿)g = da¿/d{, and the matrix A ^ is given by

**(2.12) ** **^ o o — **

**/ **

**V **

0 0 1 o \ 0 0 0 1

*a-2 0 0 v *

*0 a -v 0 j *

*The eigenvalues of this system are given by \± = ^/r]± and A|. = — y/rj±, with *

2a - (í>2 + 2) ± ^/(í)2 + 2)2 - 4í)2a

(2.13) **»7± = **

/2-Figure 4* shows the behavior of the four eigenvalues Aj_ and A|. in the (a, v2) plañe. There *

*are four distinct regions, separated by the boundaries a = 2 and a = (v2* + 2)2/4í*2: (a) four
real eigenvalues, (b) two pairs of complex conjúgate eigenvalues, (c) four purely imaginary
eigenvalues, and (d) two real eigenvalues along with two purely imaginary eigenvalues. In
regions (a) and (b) the unstable and stable manifolds of the origin are two dimensional,
while the equilibrium is nonhyperbolic in regions (c) and (d) where the center manifold is,
respectively, four- and two dimensional. Homoclinic orbits belong to both the stable and
unstable manifold of the origin. We investígate below the presence of homoclinic solutions in
cases (a) and (b), where these correspond to the intersections of two dimensional stable and
unstable manifolds in a four dimensional space [23].

The problem (2.9)-(2.10) is invariant under the symmetry

(2.14)

_{£^-£}

_{£^-£}

_{; }that comes from the time reversing (Hamiltonian) and spatial reflection symmetries of the NLCMEd. We further restrict our search for dispersive pulses to the case of reflection-symmetric pulses, Le., to pulses that satisfy

a(-0-If we now set the symmetry axis to £ = 0, we can reduce the problem to a semi-infinite interval £ € [0, +00[ with the boundary conditions

(2.15) (ar,a¿) = (a0,0) and ((ar)g, (a¿)?) = (0,60) at £ = 0,

(2.16) (or.Oi) ->• (0,0) as £ ->• 00.

Finally, to numerically compute the profiles of the symmetric pulses, we replace the infinite
interval by a finite one, [0,L]. Following [13],* the resulting boundary conditions at £ = L *
are obtained by requiring that the solution be in the subspace spanned by the eigenvectors
associated with the decaying eigenvalues of the matrix A ^ (see (2.11)). In summary, the
boundary valué problem that we intégrate numerically is given by (2.9), which we rewrite as
*a real first order system of four equations in ]0, L[ together with the four boundary conditions *

(2.17) C0u = 0 at £ = 0 and

(2.18) CooU = 0 at £ = L,

where uT* = (ar, a¿, (a*r)?, (a¿)?),

*and CQO is a matrix whose rows are the left eigenvectors of Aoo associated with the *
exponen-tially growing directions

(r?+* — a — v2) av/^/rj^ (r¡+ — a)/^/rj^ v *
*(r?_ — a — v2) av/' y/rfl (r?_ — a)/' yfrjl v *

*For each valué of a this problem is solved using a shooting method. We start from the known *
*solutions for v = 0 obtained in [3] and apply numerical continuation techniques to lócate the *
*propagating pulses with v > 0. This procedure for setting the boundary conditions at £ = L, *
*rather than simply imposing a{L) = 0, allows the shooting method to converge faster, and *
*the results obtained are found to be essentially independent of L for L > 10. *

The left panel of Figure 5 shows several families of homoclinic orbits represented in the

*(ao,v) plañe for different valúes of a and corresponding to the case ao > 0. The dashed *

line separates the regions where the homoclinic orbit connects to a saddle point and where
it connects to a saddle-focus, while the open circles correspond to the solutions shown in the
right panels. In Figure 5(II)(a)-(c) the solutions can be seen to develop oscillations as we move
towards ao = 0. This corresponds to moving along a horizontal line in Figure 4* (constant a) *
*and to the right, approaching región (c) (precisely at ao = 0, v = ^fa + V*a — 2) where the
eigenvalues become purely imaginary.

This oscillatory behavior, however, is not observed for the families of pulse solutions found
for ao < 0, as seen in Figure 6. Instead, these curves display turning points and, to better
*appreciate these limit points, the results have been plotted in the plañe (E,v), where E is the *
positive quantity

(2-2*° ) Coo = 1 , „ ,0, *

**E**

** = \¡j**

**= \¡j**

**Q**** (M**

**2**

** + M**

**2**

**)d£. **

(I)

**ftn **

U0.8 0.7

0.6

0.5

0.4

0.3

0.2

0.1

1

1

1

1 —•

1

*** **

**3.5 **

**3 **
**4 **

**\ ( a ) **

**\( b )" **

**V**

**c)**

**V **

*a*

*a *

0.5 1 1.5

**a****r**** o **

**Figure 5.*** (I) The solid Unes correspondí to homoclinic cycles to the origin with ao > 0 for the indicated *

*valúes of a. The dashed Une bounds the regions where the origin is a saddle node (on the left) and saddle-focus *
*(on the right). (II) Spatial profiles of three pulses for the valúes marked in (I) with open circles. *

**Figure 6.*** (I) Curves of homoclinic cycles to the origin withao < 0 for the indicated valúes of a. (II) Spatial *

*profiles of two pulses corresponding to the valúes marked in (I) with open circles. *

**3. NLCMEd simulations.*** After having found a two-parameter (a,v) family of symmetric *
*dispersive pulses (DP) that can be numerically continued from the solutions for v = 0 obtained *
in [3], we now proceed to study their stability.

**I^**

**+**

**I**

** \A~ **

**\A~**

**Figure 7.*** Space-tíme representatíon of the solution of the NLGMEd (a = 1/2, n = 1, e = 1CP*B*, and *

*periodicity boundary conditions) for an unstable pulse propagating on top of a CW (p = 1 and O = TT/4). *
*The pulse parameters are a = 4 and v = 1, and it corresponds to the point labeled (a) in Figure* 6.* See the *
*accompanying file* (69822_03.gif* [15.8MB]) for an animation of this pulse destabilization. *

associated with the small dispersión coefficients are integrated implicitly, and the nonlinear
terms are computed in physical space with the usual 2/3 rule to avoid aliasing effects [5].
*(The máximum required resolution for the simulations in this paper was Np = 4096 and *
Ai = .0005.)

**Figure 8.*** Space-time representatíon of the solution of the NLCMEd, (a = 1/2, K = 1, e = 1CP*B*, and *

*periodicity boundary conditions) for a stable pulse propagating on top of a CW (p = 1 and O = TT/4:). The pulse *
*parameters are a = 4 and v = 2.6, and it corresponds to the point labeled (a) in Figure* 5.* See the accompanying *

*file* (69822_04.gif* [3.6MB]) for an animation of this propagating pulse. *

Figure 6 (see also the corresponding animation 69822_03.gif [15.8MB]). Note that the size of
*the pulse first grows (at t = 5.5 it is larger than at t = 3), and then it decays again at t = 7.0. *
A few time units later the oscillatory tails spread, and the pulse structure is eventually lost
*(not shown in the figure). All DP explored for ÜQ < 0 propagated over the same simple CW *
*with parameters p = 1 and 0 = 7r/4 (cf.* (2.1)-(2.2)),* which corresponds to a = 4 in* (2.14).

*and all were found to be unstable regardless of the propagation speed v. *

*On the other hand, for ÜQ > 0 and for the same background CW (Le., a = 4), the *
*DP are found to be unstable approximately for 0 < -0 < 2.2 and stable for v > 2.2. The *
evolution of two stable pulses is shown in Figures 8 and 9,* which correspond to v = 2.6 and *

*v = 3.2, respectively, where the structure of the slightly perturbed pulses is seen to remain *

now virtually unaltered after more than 40 time units (see also the corresponding animations

69822_04.gif [3.6MB] and 69822_05.gif [3.5MB]).

**Figure 9.*** Space-time representatíon of the solution of the NLCMEd, (a = 1/2, K = 1, e = 1CP*B*, and *

*periodicity boundary conditions) for a stable pulse propagating on top of a CW (p = 1 and O = TT/4:). The pulse *
*parameters are a = 4 and v = 3.2, and it corresponds to the point labeled (b) in Figure* 5.* See the accompanying *

*file* (69822_05.gif* [3.5MB]) for an animation of this propagating pulse. *

is the relevant one in the NLCME dynamics.

Another very interesting feature of the DP that is worth mentioning is the fact that they are somehow transparent to each other: two DP propagating in opposite directions just pass through each other without distortion. This is illustrated in Figure 10 (see also the animation

69822_06.gif [3.9MB]), where two stable DP (corresponding to those in Figures 8 and 9) are
sent towards each other and after 40 time units (approximately 80 collisions) they still remain
practically undistorted. The reason behind this behavior is the dominant character of the
transport effect induced by the group velocity. If we rewrite the NLCMEd for a DP with
*short (dispersive) spatial and temporal scales X = x/' \/e ~ 1 and T = t/y/e ~ 1, they take. *
at first order, the form of two uncoupled wave equations:

*AT = Ax* + • • • ;

It is clear then that the DP traveling in opposite directions simply propágate through different channels, and are (in first approximation) completely independent.

**Figure 10.*** Space-tíme representation of the solution of the NLCMEd, (a = 1/2, K = 1, e = 10~*B*, and *

*periodicity boundary conditions) showing the simultaneous propagation in opposite directions of the two stable *
*pulses from Figures* 8* and* 9.* See the accompanying file* (69822_06.gif* [3.9MB]) for an animation of this pulse *

*interaction. *

states (dispersive pulses, DP) that propágate on top of a CW. These DP are approximately advected by the group velocity, but this transport effect does not play any role in the determi-nation of their infernal structure, which results basically from a balance between nonlinearity and dispersión. It is also important to emphasize that this type of dynamics is not contained in the standard dispersionless NLCME (f.f)-(f.2) formulation usually applied to model light propagation in FBG.

In the context of light propagation on FBG the localized structures that have been most
extensively analyzed are the so-called gap solitons (GS). The GS are well-known pulse-like
solutions of the NLCME (1.1)—(1.2) that have the striking property that they can propágate
at any velocity between zero and the speed of light of the bare fiber (see, e.g., [1, 10] and
references therein). The new family of localized structures presented in this paper, the DP,
appear in the same regime of weakly nonlinear light propagation in FBG where the GS exist
(the most frequently analyzed regime). There are robust stable DP, and their dynamics evolve
in the same time scale as the GS, but they exhibit several essentially different characteristics
that are worth mentioning: (i) the GS propágate over the zero intensity state, while the
DP propágate on top of a saturated uniform continuous wave; (ii) both GS and DP have
amplitudes of the same size, but the DP are much more narrow pulses (in the scalings used in
*this paper, ¿DP ~ ¿disp ^ \/e <^1 and 5QS ~ 1); (üi) the DP propágate with the speed of light *
*of the bare fiber (v = ±1) and only on one of the amplitudes, A+ or A~, while the GS have *

a priori, looks like it could be interesting from the point of view of information transmission. The ideas used in the analysis performed in this paper go beyond the context of light propagation on fiber gratings and apply to the (systematically not considered in the litera-ture) generic case of pattern formation in propagative spatially extended systems. The main ingredients are (i) the unavoidable different asymptotic order of transport (first order spatial derivatives) and dispersion/dissipation (second derivatives), and (ii) the need to consider these two effects simultaneously in order to correctly account for the weakly nonlinear dynamics of the system. The envelope equations are necessarily asymptotically nonuniform; that is, the small parameter is not gone from the envelope equation formulation, reflecting the impossi-bility of balancing the two effects mentioned above. The resulting dynamics is much richer, and new patterns (as the pulses obtained in this paper) and instabilities can be found that have shorter spatial scales but that are still slow as compared with the basic linear wavetrains and therefore well accounted for by the envelope equation formalism. All these new dynam-ical states are not detected if the usual formulation that only takes into account transport effects is applied. Finally, it is also interesting to mention that similar effects have been pre-viously described in the context of the oscillatory instability in dissipative systems [20] and in parametrically forced surface waves [21].

**REFERENCES **

*[1] A. AcEVES, Optical gap solitons: Past, presera and future; theory and experiment, Chaos, 10 (2000), pp. *
584-589.

*[2] I. BARASHENKOV AND Y. S. SMIRNOV, Existence and stability chart for the AC-driven, damped nonlinear *

*Schroedínger solitons, Phys. Rev. E, 54 (1996), pp. 5707-5725. *

*[3] I. BARASHENKOV, T. ZHANLAV, AND M. BOGDAN, Instabilities and solíton structures in the dríven *

*nonlinear Schroedínger equation, in Nonlinear World Vol. 1, Proceedings of the IV International *

Workshop on Nonlinear and Turbulent Processes in Physics, V. Bar'yakhtar, V. Chernousenko, N.
Erokhin, A. Sitenko, and V. Zakharov, eds., World Scientific, River Edge, NJ, 1989, pp. 3-9.
*[4] J. C. BRONSKY AND D. W. McLAUGHLIN, Semiclassical behavior in the NLS equation: Optical *

*shocks-focusing instabilities, in Singular Limits of Dispersive Waves, NATO Adv. Sci. Inst. Ser. B Phys. 320, *

N. Ercolani, I. Gabitov, C. Levermore, and D. Serré, eds., Plenum, New York, 1994, pp. 21-38.
*[5] C. CANUTO, H. HUSSANI, A. QUARTERONI, AND T. ZANG, Spectral Methods in Fluid Mechanics, Springer *

Ser. Comput. Phys., Springer-Verlag, New York, Berlin, 1988.

*[6] A. CHAMPNEYS, B. MALOMED, AND M. FRIEDMAN, Thirring solitons in the presence of dispersión, Phys. *
Rev. Lett., 80 (1998), pp. 4169-4172.

*. CHRISTODOULIDES AND R. JOSEPH, Slow Bragg solitons in nonlinear periodic structures, Phys. Rev. *
Lett., 62 (1989), pp. 1746-1749.

*. DE STERKE, Theory of modulational instability in fiber Bragg gratings, J. Opt. Soc. Amer. B Opt. *
Phys., 15 (1998), pp. 2660-2667.

*. DE STERKE AND J. SlPE, Gap solitons, Progr. Optics, 33 (1994), pp. 203-260. *

*. GOODMAN, M. WEINSTEIN, AND P. HOLMES, Nonlinear propagation of light in one-dimensional *

*periodic structures, J. Nonlinear Sci., 11 (2001), pp. 123-168. *

*JlN, C. LEVERMORE, AND D. W. MCLAUGHLIN, The behavior of solutions of the NLS equation in the *

*semiclassical limit, in Singular Limits of Dispersive Waves, NATO Adv. Sci. Inst. Ser. B Phys. 320, *

N. Ercolani, I. Gabitov, C. Levermore, and D. Serré, eds., Plenum, New York, 1994, pp. 235-255.
*. KASHYAP, Fiber Bragg Gratings, Optics and Photonics, Academic Press, New York, 1999. *

*. KELLER, Numerical Solution of Two Point Boundary Valué Problems, CBMS-NSF Regional Conf. *
Ser. in Appl. Math. 24, SIAM, Philadelphia, 1976.

*[14] J. KEVORKIAN AND J. COLÉ, Múltiple Scale and Singular Perturbation Methods, Appl. Math. Sci. 114, *
Springer-Verlag, New York, Berlin, 1996.

*E. KNOBOCH AND J. GlBBON, Coupled NLS equations for counterpropagating waves in systems with *

*reflection symmetry, Phys. Lett. A, 154 (1991), pp. 353-356. *

*[16] C. MARTEL, Dispersive destabilization of nonlinear light propagation in fiber Bragg gratings, Chaos, 15 *
(2005), paper 013701.

*[17] C. MARTEL AND C. CASAS, Dispersive destabilization of nonlinear light propagation in fiber Bragg *

*grat-ings: A numerical verification, Chaos, 17 (2007), paper 013114. *

*[18] C. MARTEL, E. KNOBOCH, AND J. VEGA, Dynamics of counterpropagating waves in parametrically driven *

*systems, Phys. D, 137 (2000), pp. 94-123. *

*[19] C. MARTEL AND J. V E G A , Finite size effects near the onset of the oscülatory instability, Nonlinearity, 9 *
(1996), pp. 1129-1171.

*[20] C. MARTEL AND J. V E G A , Dynamics of a hyperbolic system that applies at the onset of the oscülatory *

*instability, Nonlinearity, 11 (1998), pp. 105-142. *

*[21] C. MARTEL, J. VEGA, AND E. KNOBOCH, Dynamics of counterpropagating waves in parametrically driven *

*systems: Dispersión vs. advection, Phys. D, 174 (2003), pp. 198-217. *

*[22] H. SAKAGUCHI AND B. MALOMED, Dynamics of positive- and negative-mass solitons in optical lattices *

*and inverted traps, J. Phys. B, 37 (2004), pp. 1443-1459. *

*[23] S. WlGGINS, Global Bifurcation and Chaos, Appl. Math. Ser. 73, Springer-Verlag, New York, Berlin, 1988. *
*[24] H. WlNFUL AND G. COOPERMAN, Self-pulsing and chaos in distributed feedback bistable optical devices, *

Appl. Phys. Lett., 40 (1982), pp. 298-300.