392 J. Opt. Soc. Am. B/Vol. 5, No. 2/February 1988 Curtis R.

Menyuk

Stability of solitons in birefringent optical fibers.

II. Arbitrary amplitudes

Curtis R. Menyuk*
Science Applications International Corporation, 1710 Goodridge Drive, McLean, Virginia 22102

Received March 31, 1987; accepted October 6, 1987

Propagation of short pulses in birefringent single-mode fibers is considered. Initial pulses are assumed to be
linearly polarized at an arbitrary angle with respect to the polarization axes. The Kerr nonlinearity leads to a
substantial interaction between the partial pulses in each of the two polarizations. When the amplitudes of the
partial pulses are equal, it is found that above a certain amplitude threshold, whose size increases with birefringence,
the two partial pulses lock together and travel as one unit. This unit can be a single soliton or, at higher amplitudes,
a breather. At the same time, the central frequencies of both polarizations shift just far enough so that, if a rapid os-
cillation is ignored, their group velocities become identical. When the initial amplitudes are unequal, it is found as
before that above a certain threshold one or more solitons emerge from the initial pulse. However, the breathers
that appeared when the amplitudes were equal are unstable; they break up into two distinct solitons moving at
different velocities when the amplitudes become slightly unequal. It is further shown that realistic fiber attenua-
tion has little effect on these results. The numerical method used to obtain these results is described in detail.

1. INTRODUCTION The second direction in which we extend our previous

results is to consider somewhat higher birefringences than
It is well known that so-called "single-mode" fibers are real- previously and amplitudes corresponding to that of two soli-
ly bimodal because of the presence of linear birefringence.' tons. In the case of larger birefringences, it is found, if equal
Hence, in addition to spreading as a result of dispersion, the initial amplitudes are assumed, that below a certain ampli-
partial pulses in each of the two polarizations will tend to tude threshold the initial pulse splits into two solitons, each
split apart because of the birefringence. It is also well of which consists primarily of one polarization but that can
known that it is possible to compensate for the effect of have a substantial fraction of the other polarization mixed
linear dispersion by use of the Kerr nonlinearity in the in. Above threshold, a multiple-soliton structure that un-
anomalous-dispersion regime. 2 3 Solitons are created in dergoes complicated oscillations, a breather, is formed. The
which spreading does not occur. We recently showed that a evolution of this breather is reminiscent of that of the
similar result holds true for the birefringent walk-off.4 breathers (higher-order solitons) that are observed in the
Above a certain amplitude threshold, whose size increases solution of the nonlinear Schr6dinger equation; however,the
with the strength of the linear birefringence, the two polar- evolution of this new type of breather appears to be chaotic,
izations strongly interact, and solitons consisting of both not regular. The breather ceases to exist when the initial
polarizations are formed. It is interesting to note that other amplitudes of the two pulses differ only slightly. Instead,
authors 5 6 showed previously that solitons can exist in multi- the initial pulse breaks up into two separate solitons. Of
mode fibers. A problem with these theories is that in actual these two solitons, one is larger and consists primarily of the
fibers, the intermodal dispersion is too large to lead to soli- higher-intensity polarization. Above some threshold, which
tons at realistic pulse intensities. However, a birefringent is close to that found for breathers to form at equal initial
fiber can be viewed as a multimode fiber with sufficiently amplitudes, the amplitude of the portion of the lower-inten-
weak intermodal dispersion to make the generation of soli- sity polarization traveling with the larger soliton exceeds the
tons practical. amplitude of the portion traveling with the smaller soliton.
In previous work, we derived a pair of coupled partial When the initial amplitudes differ strongly, then the higher-
differential equations that describe propagation in birefrin- intensity polarization will once again exhibit breatherlike
gent optical fibers with the effect of Kerr nonlinearity in- behavior, while the lower-intensity polarization will exhibit
cluded. 7 We then solved these equations numerically, de- continuum behavior.
termining the threshold amplitudes at which solitons propa- Finally, we consider the effect of fiber attenuation in more
gate stably. In this work, we extend our previous results in detail than before4 and describe in more detail the numerical
two important directions. First, we study what happens method used to calculate our results.
when the initial amplitudes of the two partial pulses in each There are two different motivations for studying the effect
polarization are unequal and the total amplitude corre- of birefringence on soliton propagation in fibers. The first is
sponds to that of a single soliton. It is found that the practical. It has been proposed to use solitons as bits in a
amplitude threshold for creating a soliton decreases some- communication system. 8 9 Clearly, it is of interest to know
what as the initial amplitudes of the two partial pulses be- the limits that birefringence imposes on this scheme. The
come increasingly unequal. It is also found that the pulse present model does not account for inhomogeneities, which
evolves in a markedly different manner. are expected to play an important role in real fibers, but it

0740-3224/88/020392-11$02.00 c 1988 Optical Society of America

Curtis R. Menyuk Vol. 5, No. 2/February 1988/J. Opt. Soc. Am. B 393

does represent the first step toward a complete investigation wo. The derivatives k' = ak/aw, k" = a2k/aC 2 , 1' = aiiac,and
of this issue. 1"= a2liiW2 are evaluated at w = wo. The quantity x is the
The second motivation is more fundamental. It might Kerr coefficientwith a geometric factor due to the finite core
seem surprising at first that arbitrary initial conditions, size taken into account. The quantity F is the attentuation
above a given integrated intensity threshold, will in general coefficient. Finally, the quantities U and V represent the
produce solitons. That is to say, the central frequencies of amplitudes of the partial pulse envelopes in each polariza-
each partial pulse must shift in just such a way that the two tion. The linear terms in Eqs. (1) are derived from a slowly
partial pulses will move together. Yet it is so. This result is varying approximation with contributions from third- and
one manifestation of what appears to be a general physical higher-order derivatives neglected. The nonlinear terms
principle: Moderate (or, more precisely, sufficiently small) are due to the cubic (Kerr) interaction. The factor 2/3,
Hamiltonian deformations of nonlinear, integrable systems which appears before IV12in Eq. (la) and UL12 in Eq. (lb),
do not destroy solitons if these deformations are autono- leads to nonlinear birefringence. 2 0 The exponentially vary-
mous in the sense that they have no explicit time or space ing terms in Eqs. (la) and (lb) are rapidly varying whenever
dependence.' 0 "'1 The equations that we will solve4 7 are an the birefringent beat length is much less than one soliton
autonomous Hamiltonian deformation of the coupled non- period and can be neglected. That will always be the case
linear Schr6dinger equation whose integrability was demon- when the pulse width is larger than approximately 1 psec.
strated by Manakov.12 The existence of solitons could thus Retaining these exponentially varying terms in the limit of
have been anticipated a priori-and, in fact, was. Another zero birefringence, we find that Eqs. (1) are invariant under
manifestation of this principle in fibers is that solitons can the transformation
be launched at the zero-dispersion point.1 3 We note, howev-
U= UcosO + Vsin0,
er, that this principle does not apply only to fibers but is
quite general and manifests itself in a wide array of different -V=Vcos0- U sin O. (2)
physical contexts.10
In order to observe the phenomena described in this pa- This invariance is a fundamental symmetry requirement.
4
per, the most natural approach is to use the soliton laser1 or We now normalize Eqs. (1), using variables closely related
one of a number of other recently invented schemes for to those used by Mollenauer, Stolen, and co-workers.4 2' We
producing short pulses in the appropriate wavelength re- first assume that k" = 1"and write
gime.1 5-' 8 In this way, one avoids dealing with inhomoge-
neities and other difficulties associated with long propaga- k = 1" = ° D(A), (3)
tion lengths. On the other hand, as we will show, fibers with
higher than normal birefringence must be used. To obtain
the required birefringence, specially manufactured or where D(X) is evaluated at X = Xo,the carrier wavelength
mounted fiber is needed.1 9 Our principal motivation for 27rc/wo. We also assume k' - 1' = (ko - lo)/wo. We define
investigating the relatively complicated two-soliton solu- 2 2 2
7rZ 7r c to
tions is that these lie in the parameter regime that would be to = 0.568r,
most easily reached in soliton laser experiments.' 9
The remainder of this paper is organized as follows: In _ 2
S =- t-- ==( D(X)
1/U ,
Section 2 we introduce the basic equations, and, in Section 3, gk' + '
to \ vg
the numerical approach used to solve them. In Section 4 we
discuss cases in which the partial pulses in each polarization X 1/2V
=
have the same amplitude and describe the effect of attenua- =2 V
tion. In Section 5 we discuss cases in which the partial
pulses no longer have the same initial amplitude. Section 6
k'-I' T rcAn 8TrC 2zor
,
21hk"Ito D(A)X to, R
= to, I= 7r
contains the conclusions. X0

where r is the FWHM of the pulse intensity and An = (k -

2. BASIC EQUATIONS l)c/wo is the difference between the indices of refraction;
Eqs. (1) become
We have shown that pulse propagation in a birefringent
optical fiber is described by the equations 7
iau au) 1 a'u X2 2 2
I
G~ as) 2as' (IuI' + -3-IvI'
' 3/
i -d +k'---a + 2 IV2)U
- 2 k"-at2 + X(IU12
az 2\ 3 / + 1 v2u* exp(-iR) = -iyu, (4a)
3
+ XV2U*exp[-2i(ko- 0)z] = -iru, (la) av au 2av ~
6 i (d _ 5 v + 1 v +-(2 JU12 + ,12)

/ a'v +X2J1 a+ as) 2 as2 I3 /

(aV±1 av) V2
2
(az at) 2 at 2 3 + ± I u2 v* exp(iRlb)= -i 7 v. (4b)
3
+ XU2V* 0 - 1)z1 =-irv, (lb)
exp[2i(k We now determine the values of 5 and R and show that for
6 the cases of interest to us in this paper, the exponentially
where ko = k(co) and 10= l(coo)are the wave numbers of each varying terms can be neglected. We use the experimentally
of the two polarizations evaluated at the carrier frequency determined value21 of D(X) = 6.5 X 10-3 at Xo = 1.55 ,um.
 394 J. Opt. Soc. Am. B/Vol. 5, No. 2/February 1988 Curtis R. Menyuk

According to Ref. i, the values of An vary between 5 X 10-9

2
and 8 X 10-4. Typical values tend to be concentrated in the i (u + da + u = '
range 10-6-10-5. In soliton communication-system designs
that have been proposed to date, 6 7 it is assumed that T - 5
psec. If we let r = 5 psec, we find that the total range of 3 is 2
(9)
ad as as2
1.3 X 10- 3 -2 X 102,while typical values are concentrated in
the range 6 = 0.3-3.0. We also find that R = 1.4 X 104, which Our purpose is to determine the condition for birefringent
implies that R6 >>1 over the entire range. Hence the final splitting to dominate over dispersive broadening. We solve
terms in Eqs. (4) are negligible. If we let = 250 fsec, the Eqs. (9) by using Fourier methods combined with the meth-
size produced by the soliton laser,1 4 we find that the total od of steepest descent, 24 taking Eqs. (6) as our initial condi-
range of values is 7 X 10-5-10, typical values are concen- tion with a = 450. We then obtain, in the limit >>1,
trated in the range = 0.013-0.13, and R = 700. It follows
that R6 < 1 when 6 < 1.4 X10- 3 , which is below the typical u(, s) = (e)/ (1-i) exp[i(s - )2]sech[ 7 (s -I)
range but well within the range of allowed values. Thus,
when the birefringence is small, the final terms in Eqs. (4)
can no longer be ignored. These terms can lead to an insta- v(t, s) = (r)12 (1-i) 2 exp[i(s + 34)2]sech[t (s + 6#)
bility in which the slower-moving partial pulse transfers (10)
energy to the faster.2 2
When the final terms in Eqs. (4) can be ignored, Eqs. (4) If we take as our criterion for birefringence to dominate over
become dispersion that the intensity develop a double-humped
structure, we find that > 0.42. The locking together of the
. - - u (J + 2I12 u = -iyu, partial pulses in the two polarizations, which we will show
a (5a)
as 2 a2 + +3 ) occurs nonlinearly above some amplitude threshold, will be
most easily observed in experiments when this condition is
met.
(O as)+ 2 s
2 3
(5b)

3. NUMERICAL APPROACH
We will be concentrating on this case in the remainder of this
paper. For our initial conditions, we choose We now describe the numerical approach used to solve Eqs.
(5). We first define the Fourier-transform variables:
u ( = ) = A cos a sech s,
v ( = 0) = A sin a sech s. (6)
fl(c, t) = 1 dsei'u(s, t),

The angle a thus determines the relative strengths of the

partial pulses in each of the two polarizations. Equations (, 0) = 21 | dse1' (s, ). (11)
(5) are not consistent with the symmetry requirement, Eqs.
(2), which is not surprising since this requirement applies The linear portion of Eqs. (5) can be solved exactly over any
only in the limit of zero birefringence where Eqs. (5) are no finite distance interval by writing
longer valid. As a consequence, when a = 0 and = 0, Eqs.
d(w, + A4)= uf(, )exp[iw( - 0/2)A - yAfl,
(5) have the following single-soliton solutions:
+ A)
A, D( , ()exj[-iw( + /2)A - yAf]. (12)
u(, s) = exp[2 (1 + 62)t - is]sech s, Similarly, the nonlinear portion of Eqs. (5) can be solved
exactly over any finite distance interval by writing
V(Q,s) = 0. (7)
u(s, + A) = u(s, )exp[i(uI2 + 2/3IdI2)A~I,
When a = 450, we find that v(s, t + Ai) = v(s, t)exp[i(2 /3 uI2 + Iv12 )AZ]. (13)

= (3 )1/2 exp[ (1 + 62)t - is]sech The quantities 1u12 and 12 are constant functions of t over
the interval (Q, t + At). We can use Eqs. (12) and (13) to
exp[ 2 (1 + devise a semispectral, leapfrog scheme, which is second-
uVQ,s= (3)1 32)t + ibs]sech s. (8)
order accurate in A. We initially use the operators in Eqs.
(13) to advance the equations one-half spatial step from =
Hence the intensity is a factor of 6/5 larger for a given pulse 0 to = /2. We then use the operators in Eqs. (12) to
width when a = 450 than when a = 0. The effective advance the equations from t = 0 to = A
At. We next use the
nonlinearity is weaker at 450, so that a higher amplitude is operators in Eqs. (13) to advance the equations from Ai/2 to
needed to balance the effect of dispersion. A similar effect 3At/2, and so on. Hence the name leapfrog. On each time
has already been noted for continuous waves by Botineau step, we transform once from the temporal to the spectral
and Stolen.2 3 domain using the fast Fourier transform and once back using
We now briefly consider the linear regime with zero atten- its inverse; hence the name semispectral. To save computer
uation, where time, we replace the operators in qs. (13),
Curtis R. Menyuk Vol. 5, No. 2/February 1988/J. Opt. Soc. Am. B 395

N = exp[iflul
2
+ 2 /3 IVI2 )Lt] This special symmetry no longer holds when af Fd 45°. We
now define snax() to be the s value at which u(s, 0)1reaches
2 2 2
N = exp~i( /3 Iu1 + lvl )Ag], (14) its maximum value; we define Snax() analogously. From
Eq. (19), it then follows that
with the operators
[1+i 2 2/
3 I2)L/4l2 Slax() = Smax(). (20)

I- i(j1u2 + 2/Ivl2)/4J When the two partial pulses split apart because of birefrin-
gence, snax() increases without bound. When, however, the
2 2 2
1 + i( /3 1u1+ JV1 )Ag/4 12 two partial pulses stay together, Smax is bounded. We also
Nv = I . (15)
2 2 2
[1 - i( /31u1 + Jv1 )zA/4j define the frequency centroids,

These operators are consistent with Eqs. (14) through sec- J d@@&(., )12
ond order in At and are unitary, i.e., they preserve 1u12and
Iv12.We chose square operators and, in fact, advanced the
equations twice per spatial step, using NU112 and N, 112 in
Xcent(0 =
Jc
EA dwli(w,
dwi2cc t)12

order to have the full solutions to Eqs. (5) available at each

time step. The added computational overhead (two multi- X dwwlD(w, )2
plications per node per time step) is slight. Variants of the ocent(0) X
(21)
semispectral method have been used by many authors in EJ dwID(, t)12
problems closely related to ours. In problems like ours in
which the boundary conditions are simple and the functions
From Eq. (19), it followsthat
to be calculated are smooth, numerical tests indicate that
this approach is generally the most computationally effi- WCent(0)= W4ent(0) (22)
cient.2 5
Our numerical computations can be carried out only over a and
When the two partial pulses stay together, wcuent(Q)
finite region of s space. Something must be done to damp a wvent() shift just enough on average so that the group veloci-
signal that has left the computational region so that it does ties of the two partial pulses are the same.
not return as a result of the periodic boundary conditions In Fig. 1 we show S' ax and Wcentwith = 0.15 at various
assumed by the fast Fourier transform. The computations values of A. Below a certain threshold, A = 0.7, smaxin-
described in this paper are displayed in the range s = (-40, appears to be reaching an asymp-
creases steadily, while wceent
40). To deal with edge effects we enlarge the computational totic value. Beyond this threshold, snax is seen to reach a
region, usually by a factor of 2 to s = (-80, 80), and introduce maximum value and then decrease, indicating that the par-
edge damping in the form tial pulses are mutually bound. Over a longer length scale
u(new) = ru(old), V(new)= rV(old) than shown, both sMaxand c'nt oscillate. In Fig. 2 we show
(16)
similar results with a = 0.5. Here the threshold value is A =
1.0. The transition is sharper in this case, and oscillations
once per time step, where when A > 1.0 are clearly visible. In Figs. 3 and 4 we set 6 =
F= - sech[O(s - Sedge)] (17)
(a) 3.0
I11 A = 5
and edge indicates the location of the nearest edge. For all
the computations described in this paper e = 1/2. We also
set the number of node points N = 512 at amplitudes where 06
at most one soliton appears and N = 1024at higher ampli-
tudes. A = 07
In order to verify the accuracy of our results, we performed
standard numerical tests. In selected instances, we halved A
the spatial step, doubled the number of node points, doubled 0.0
the factor / in Eq. (17), doubled the computational region,
and verified that no significant changes occurred in our (b)
results.

4. EQUAL AMPLITUDES: a = 45°

We now solve Eqs. (5), assuming that ar = 450 so that the
amplitudes of the two partial pulses are initially equal.
From Eqs. (5) and (6), it then follows that
v(s,0) = u(-s, 0) (18) 20

and Fig. 1. Variation of (a) the maximum location s'ax and (b) the
frequency centroid Wcentwith distance along the fiber measured in
D(co, = t(-, . (19) soliton periods (6 = 0.15, y = 0.0).
396 J. Opt. Soc. Am. B/Vol. 5, No. 2/February 1988 Curtis R. Menyuk

A = 0.9 inferred from Eq. (18). One of the effects of increasing the
(a) 5.0
attenuation is slightly to increase the amplitude threshold
E ._
= , beyond which the two partial pulses are mutually bound.
xr When y = 0.0, the two partial pulses initially move apart as
o they spread, as a result of dispersion, but, beyond some
point, they begin to move together again as if they were tied
0.0 by a spring. The nonlinearity acts to sharpen the total pulse
-1.0 as well. When that pulse has traveled 50 soliton periods,
one finds that the maxima of the partial pulses are almost
coincident and that there is a central portion, almost as
sharp as the original component, sitting on a broad back-
(b) 7.0
ground of dispersive waves. By contrast, the attenuated
or
FOURIER
AMPLITUDE AMPLITUDE
0.08 T II I I I I I T I ' -_
() I

0.0.
0 Soliton Periods 20
Fig. 2. Variation of (a) the maximum location Smaxand (b) the
frequency centroid co', with distance along the fiber (6 = 0.5, Y
0.0).
0.00 I

0.5 and compare details of the pulse evolution when A = 0.1

nR
ll.uo

and when A = 1.0. When A = 0.1, the pulse evolution is . I I I

.(C)
essentially linear. Both the u and v partial pulses spread
because of dispersion as they separate because of birefrin-
gence. By contrast, when A = 1.0, the effect of nonlinearity
constrains both the spreading that is due to dispersion and
the splitting that is due to birefringence. The accompany-
- I
ing frequency shift is also visible. nn
n Bw
us
When we consider higher values of 6, qualitatively similar -40 0 40 -5 0 5
behavior is seen in snax and wu'nt Below threshold, suax S a)
shows a steady increase, while wuet monotonically tends Fig. 3. Details of the pulse evolution (A = 0.1, = 0.5, y = 0.0).
towards an asymptote; above threshold, both quantities os- Solid lines indicate u and ; dashed lines indicate v and . (A) u(s)
and v(s), = 0; (B) (w) and v(w), = 0; (C) u(s) and u(s), = 57r(10
cillate. When = 0.75, A = 1.5 is the threshold. When 6 =
soliton periods); (D) (w) and v(w), = 57r.
1.0, A = 2.0 is the threshold. The behavior of Smaxand cent
near threshold when 6 = 1.0 is shown in Fig. 5. The thresh-
old becomes increasingly sharp as increases. Note that the FOURIER
maximum values of the u and v partial pulses can change AMPLITUDE AMPLITUDE
discontinuously as the breather oscillates, as shown in Fig. 5.
0.8 . . . .
(A) I (B)
When 6 = 0.75 or 6 = 1.0, one finds that below threshold
two solitons are created and move in opposite directions
relative to the point s = 0. Each consists primarily of one 0.81 -
polarization but contains a substantial contribution from 0.81 1 1 1 I : I I I I 1
the other. Above threshold a breather appears. This
breather is a two-soliton structure, which oscillates in a com-
plicated way as it propagates in . The u polarization is
0.0 . .L1.L A,

shown in Fig. 6; the v polarization is its mirror image about 0.8 , .

the point s = 0. This behavior contrasts with that observed
-(C)
when 6 = 0.15 or 6 = 0.5. In these cases, spreading, not
soliton generation, is observed below threshold, while above i .l
threshold only a single soliton appears.
We now determine the effect of attenuation by comparing /I
/ I\

the cases y = 0.0 and y = 0.0105; the latter case corresponds I ''

to 2 dB/km for 5-psec pulses. In general, we find that the 0.0 I- 0.0.,,,,,I,,
evolution is essentially unchanged by attenuation during the -40 0 40 -5 0 5
S Co
first 10 soliton periods but differs ubstantially at distances
greater than 20 soliton periods Fig. 4. Details of the pulse evolution (A = 1.0, = 0.5, y = 0.0).
In Fig. 7, we display this
Solid linies indicate u and al;dashed lines indicate v and . (A) u(s)
comparison with A = 1.0 and b = 0.5. Only the quantity u(s, and (s), = 0; (B) 5(w)and (U), = 0; (C) u(s) and v(s), = r (10
t) is exhibited, but the shape of v(s, ) can be immediately soliton periods); (D) (w) and (w), t = 5r.
Curtis R. Menyuk Vol. 5, No. 2/February 1988/J. Opt. Soc. Am. B 397

(a) 3.0 0.8

0
0
0

E
E
0.0

-1.0
nu.D | -

(b) (B)
1.5
0)

lul

a
0.0
LL

0.0
0 10 - (C)
SolitonPeriods
Fig. 5. Variation of (a) the maximum location s.. and (b) the
frequency centroid c', with distance along the fiber ( = 1.0, y = lul
0.0).

u-AMPLITUDE 0.0
2.0 -40 0 40
S
Fig. 7. Pulse evolution (A = 1.0, = 0.5). Solid lines indicate y =
0.0; dashed lines indicate y = 0.0105. (A) = 0, (B) = 10r (20
soliton periods), (C) = 257r.

10

0.0 a0 y = 0.0105 .

on
0
- (b) Ex
E
Y = 0.0

nn L
0 £
-40 0 40 -
_s

S o Soliton Periods 50
Fig. 8. Variation of the maximum location S' ax (A = 1.0, 6 = 0.5).
Fig. 6. Details of the pulse structure at 10 soliton periods, = 57r (6
= 1.0). Only the u polarization is shown. (a) A = 2.0, (b) A = 2.05.
Table 1. Threshold Values of A at Which the Kerr
Nonlinearity Is Sufficient to Compensate for Linear
pulse decreases too rapidly for the nonlinearity to have a Birefringence at a = 450
chance to affect the partial pulses, and they simply separate
6 -y A
and spread. This difference is quite visible in Fig. 8, where
S aX,iS plotted in the two cases. 0.15 0.0 0.7
As A is increased, we have already seen that the oscillation 0.15 0.0105 0.8
periods of S'maxand °cent rapidly decrease. As a consequence, 0.5 0.0 1.0
when A = 1.1, the nonlinearity is already large enough for 0.5 0.0105 1.1
the two partial pulses to be bound together. When 6 = 0.75 0.75 0.0 1.5
0.75 0.0105 1.5
or 6 = 1.0, the threshold difference introduced by attenua-
1.0 0.0 2.0
tion is almost undetectable. The threshold results in all the 0.0105 2.0
1.0
cases that we considered are summarized in Table 1.
398 J. Opt. Soc. Am. B/Vol. 5, No. 2/February 1988 Curtis R. Menyuk

AMPLITUDE
0.8 . ., I l .,I . ft| a, <94| X (, t) 0 (26)
-(A)
and

d J dcow(W(c, )J2 + I(, 4)i2 ) =° (27)

which can in turn be derived from Fourier transforms of the

evolution equations, Eqs. (5), setting y = 0.
0.0
The first case that we consider in detail is 6 = 0.15, A = 0.8,
0.8 . . . I I . I a = 30°. The spatial evolution is shown in Fig. 9. Two
. (B) points should be apparent. First, the peak of the v pulse
travels to the right along with the peak of the u pulse; in
effect, the smaller partial pulse has been captured by the
larger. Second, while the width of the u pulse has increased,
the width of the v pulse has increased even more. Indeed,
there was a strong tendency throughout all our computa-
tions for the largest-amplitude structure, whether a soliton
0.0 - or a breather, to end up primarily in the polarization that is
initially more intense, while the continuum ends up primari-
ly in the polarization that is initially less intense. We stress
that intensity-dependent birefringence, without dispersion,
is not sufficient to explain this result. The point is that the
more intense, less wide partial pulse remains consistently in
one polarization as a function of distance along the fiber,

(a)
-40 0 40
S E
= 0
Fig. 9. Pulse evolution (A = 0.8, = 0.15, y = 0.0, a = 30°). Solid .E
lines indicate the u polarization; dashed lines indicate the v polar- x .;'M
ization. (A) = 0, (B) = 5r (10 soliton periods), (C) = 107r.
co
c O

5. UNEQUAL AMPLITUDES: a 450

In order to study what happens in the general case when a #d
450, it is useful to define the partial pulse widths Wu and WV.
Writing (b) o.

dsslu(s, )I2 J| dss2lu(s, t)12

SaUvg
= J d avg
| dslu(s, t)1 j dslu(s, ) 12

J dsslv(s, t)12 J dss 2 lv(s, t)12

su g=I 2
Sav (S )avg X (23)
j dslv(s, t)12 E| dslv(s, t)12
(C) 1.

we let
0
WV= [(S2) g - (Su )2]1/2

WV = [(S2 )vg- (Svg)2 1/2. (24) a.O

No special symmetry relations were found to relate sax to

I- ................................
,u ,, or u to WVat arbitrary values of a, but there is a special U-9 _ , .

relation between wv ,t and uCentwhen attenuation can be ° Soliton Periods 20

neglected: Fig. 10. Parameter variation with distance along the fiber (A = 0.8,
* = 0.15, y = 0.0, a = 30°). (a) The maximum locations Suaxand
cent() = -(cot r )e cent( (25) ,vnax,(b) the pulse widths wuand wv, (c) the frequency centroids Swuent
and Wcent. Solid lines indicate the u polarization; dotted lines indi-
To derive Eq. (25), we use the relations cate the v polarization.
Curtis R. M\;enyuk Vol. 5, No. 2/February 1988/J. Opt. Soc. Am. B 399

AMPLITUDE while the less intense, wider pulse remains consistently in

1.2 , , I i I the other polarization. It is likely that this effect was occur-
-(A)
ring in the experiments of Mollenauer et al. 2 6 on picosecond
pulse narrowing, but a definite conclusion is not possible, as
all their data was taken at a fixed fiber length. In Fig. 10,
the trends that we have just pointed out are clearly visible.
In particular, s' axinitially decreases but, after two soliton
periods, reverses direction and moves with s ax Also, wfcent
and Wcent are found to be related according to Eq. (25).
0.0. 1 ~ -
The next case that we examine is 6 = 0.5, A = 1.1, a = 300.
This case is also slightly above threshold, but the birefrin-
gence is larger, and, as a consequence, the evolution is quali-

AMPLITUDE
0.8

0.8
(B)

-4u 0 4U
S
Fig. 11. Pulse evolution (A = 1.1, 6 = 0.5, y = 0.0, a = 30°). Solid - . , . I . I I
lines indicate the u polarization; dashed lines indicate the v polar-
0.0
ization. (A) t = 0, (B) t = 57r (10 soliton periods), (C) = 107r. -40 40
S
Fig. 13. Pulse evolution (A = 0.8, a = 0.15, y = 0.0, a = 15°). Solid
lines indicate the u polarization; dashed lines indicate the v polar-
ization. (A) t = 0, (B) t = 107r(20 soliton periods).

(a) 15

AMPLITUDE

4.

(0 I

(b) 3

U -o

U-
-7
0 Soliton Periods 20
Fig. 12. Parameter variation with distance along the fiber (A = 1.1, S
S= 0.5, y = 0.0, a = 300). (a) The maximum locations s'ax and s'nax,
(b) the pulse widths wuand wV,and (c) the frequency centroids wcent Fig. 14. Pulse evolution (A = 1.0, 6 = 0.5, -y= 0.0, a = 150). Solid
and wcent, Solid lines indicate the u polarization; dotted lines indi- lines indicate the u polarization; dashed lines indicate the Dpolar-
cate the v polarization. ization. (A) 0 = 0, (B) t = 107r (20 soliton periods).
 400 J. Opt. Soc. Am. B/Vol. 5, No. 2/February 1988 Curtis R. Menyuk

AMPLITUDE pulse continues to increase as A increases. If we choose as

our threshold the amplitude required to generate a soliton,
we find that A = 0.9. We can fix the amplitude and vary a
instead to determine the limit beyond which solitons can no
longer form. Setting A = 0.9, we find that when a 400,

u AMPLITUDE v AMPLITUDE
2.5 I I I I I I

(a) (b)

1.0
(b)

-F--
-

-
0.0 a I XII I I

2.5

- (d)
0.0 I I
1.0 -1- rj- I I I
()
0.0 1 I I

2.5 I I I I I I II II I

- (e) (1)
,:_~ I -I
nn
v.v

-40 0 40
S
Fig. 15. Details of the pulse structure at 20 soliton periods, = l0ir
1 0.0 I I1a
e
(A = 0.9, 6 = 0.5). Solid lines indicate the u polarization; dashed -40 40 -40 0 40
lines indicate the v polarization. (a) a = 30°, (b) a = 350, (c) a
400. S S
Fig. 16. Details of the pulse structure at 10 soliton periods, t = 57r
(A = 2.1, = 1.0). (a) u polarization, a = 44.5°; (b) v polarization, a
tatively different, as is shown in Fig. 11. The original peak = 44.5°; (c) u polarization, a 44.8°; (d) v polarization, a = 44.8°;
of the v pulse is not captured by the u pulse; it moves steadily (e) u polarization, a = 450; (f) v polarization, a = 45°.
to the left. At the same time, a new peak is created from the
background of the v pulse by the interaction with the peak of AMPLITUDE
the u pulse. This new peak moves to the left with the u 3.0
pulse. The v pulse thus consists of two parts-a dispersive
wave component that moves to the left, spreading and di-
minishing in amplitude as it moves, and a portion that con-
tributes to the soliton. The effect on the widths is clearly
visible in Fig. 12. The width of the v pulse grows steadily,
while the width of the u pulse is almost constant. Once
again, we find that cnt and w'entare related by Eqs. (23).
0.0
Similar results are found when a = 15°, as shown in Figs.
13 and 14. Here, once again, we show the pulse evolution 3.0
when (5= 0.15 and 6 = 0.5. The amplitude in both cases is
just above threshold. In the first case, the central peak of - (b)
the v pulse is captured by the u pulse; in the second case, a
new peak is created from the background of the v pulse, I I r I
which then moves with the u pulse.
The designation of an exact threshold is somewhat arbi-
trary in these cases, as no sharp transition in the pulse
II
behavior is observed. As the amplitude is increased at 300 .0
with = 0.5, one finds that when A = 0.8, a double-peaked 40
structure forms on the v pulse, but no soliton is created.
S
When A = 0.9, the fraction of the v pulse that ends up
Fig. 17. Details of the pulse structure at 10 soliton periods, 5r
traveling with the u pulse is larger than before, and a soliton (A = 2.0, 6 = 1.0). Solid lines indicate the u polarization; dashed
is created. The fraction of the v pulse that travels with the u 0
lines indicate the v polarization. (a) a = 25 ; (b) a = 30°.
 Vol. 5, No. 2/February 1988/J. Opt. Soc. Am. B 401
Curtis R. Menyuk

u AMPLITUDE v AMPLITUDE 6. CONCLUSIONS

3.01 1 1 1 1 1 1 1
We have examined the behavior of optical pulses in birefrin-
- (a) - (b)
gent single-mode optical fibers. We have shown that the
nonlinear Kerr effect leads to soliton formation even in the
presence of substantial birefringence. These solitons gener-
ally consist of both polarizations mixed together. In any
given soliton, the two partial pulses in each of the polariza-
0.0
3.0
.

n (c)I===J
..

l
-INJ
.

l
.. .
I,,
. _

-(d)_--?>M - tions shift their central frequencies in just such a way that
the two partial pulses move together.
The number of solitons produced and their subsequent
evolution depend on the birefringence 6,the total amplitude
A, and the polarization angle a. Depending on the regime,
we have found that the initial pulse can produce continuum
n
(C) waves, a single soliton, two solitons that propagate at differ-
ent velocities, or two solitons that propagate as a single
unit-a breather. This behavior is very similar to that of
-40 0 40 -40 0 40 equations such as the nonlinear Schr6dinger equation,
which can be solved using spectral transform methods.
S S That is not surprising since our equations are an autono-
Fig. 18. Details of the pulse structure at 10 soliton periods, = 557r
mous Hamiltonian deformation of an integrable system.
(6 = 1.0, a = 40°). (a) u polarization, A = 2.2; (b) v polarization, A = The problem that we have considered here is of both
2.2; (c) u polarization, A = 2.3; (d) v polarization, A = 2.3. fundamental and practical interest. Pulse evolution in bire-
fringent fibers is intrinsically interesting and, in the regimes
that we have considered, demonstrate what appears to be a
only a continuum is produced, as is shown in Fig. 15. In general physical principle: Moderate, autonomous Hamil-
general, as we decrease a, the amplitude at which solitons tonian deformations do not destroy solitons. The behavior
first form decreases, reducing to A = 0.5 at a = 00, where the that we have predicted can be observed most straightfor-
nonlinear Schrddinger equation becomes valid. wardly by using a device that produces short pulses in the
When we turn to larger values of 3,the behavior becomes anomalous-dispersion regime, such as the soliton laser, and
more complex. When a = 450, we previously found that by using fibers with larger than normal birefringence.
above some amplitude threshold a breather appears. This From the practical standpoint, the effects that we have
breather no longer appears when a differs only slightly from described can be important in normal fibers for 5-psec
450. We consider first A = 1.6 and 3= 0.75. When a = 440, pulses. This pulse size has been proposed for communica-
oscillatory, breatherlike behavior persists for at least 10 soli- tion systems. Thus it is of interest to know that solitons can
ton periods, but when a = 430, two solitons separate by 10 still be produced in the presence of significant birefringence.
soliton periods. When A = 2.1 and 3 = 1.0, this effect is even Of course, much work remains to be done to explore the
sharper. Even at a = 44.80, two solitons are clearly seen effect of fiber inhomogeneities on pulse propagation before
separating after 10 soliton periods, as shown in Fig. 16. any definitive conclusions can be drawn.
Two sorts of transition can be found in the solutions.
First, below some value of a, instead of breaking up into two
separate solitons, the pulse breaks up into continuum and a ACKNOWLEDGMENTS
breather. The breather consists primarily of the u polariza-
The author gratefully acknowledges the careful reading by
tion, and the continuum consists primarily of the v polariza-
Roger Stolen of his manuscript and Dr. Stolen's many useful
tion. When A = 1.4 and 6 = 0.75, this transition occurs
criticisms and comments. This research has been support-
between a = 350 and a = 400. When A = 2.0 and 6 = 1.0, this
ed in part by Science Applications International Corpora-
transition occurs between a = 250 and a = 300. The latter
tion.
transition is shown in Fig. 17. Note the strong tendency of
the continuum to end up primarily in the weaker-intensity * Permanent address, Department of Electrical Engineer-
polarization. As was mentioned previously, it seems likely ing, University of Maryland, Baltimore, Maryland 21228.
26
that Mollenauer et al. have observed this phenomenon.
The second transition occurs when we fix the angle a and
vary the total amplitude A. If a < 450 but above the thresh- REFERENCES
old for creating two solitons, then the v polarization consists 1. I. P. Kaminow, "Polarization in optical fibers," IEEE J. Quan-
of two portions. The first moves toward increasing s with tum Electron. QE-17, 15-22 (1981).
the larger soliton, and the second moves toward decreasing s 2. A. Hasegawa and F. Tappert, "Transmission of stationary non-
with the smaller soliton. As A increases, the amplitude of linear pulses in dispersive dielectric fibers. I. Anomalous dis-
the portion moving toward increasing s increases and ulti- persion," Appl. Phys. Lett. 23, 142-144 (1973).
3. L. F. Mollenauer, R. H. Stolen, and J. P. Gordon, "Experimental
mately becomes greater than the amplitude of the portion observation of picosecond pulse narrowing and solitons in opti-
moving toward decreasings. When 3= 0.75 and a = 400, the cal fibers," Phys. Rev. Lett. 45, 1095-1098 (1980).
transition occurs at A = 2.2, somewhat higher than the value 4. C. R. Menyuk, "Stability of solitons in birefringent optical fi-
at which the a = 450 transition occurs. The 3 = 1.0, a = 400 bers. I: Equal propagation amplitudes," Opt. Lett. 12, 614-
transition is shown in Fig. 18. 616 (1987).
 402 J. Opt. Soc. Am. B/Vol. 5, No. 2/February 1988
Curtis R. Menyuk

5. A. Hasegawa, "Self-confinement of multimode optical pulse in a 16. B. Zysset, P. Beaud, W. Hodel, and H. P. Weber, "80-fs soliton-
glass fiber," Opt. Lett. 5, 416-417 (1980). like pulses from an optical nonlinear fiber resonator," in Ultra-
6. B. Crosignani and P. Di Porto, "Soliton propagation in multi- fast Phenomena V, G. R. Fleming and A. E. Siegman, eds.
mode fibers," Opt. Lett. 6, 329-330 (1981). (Springer-Verlag, Berlin, 1986), pp. 54-57.
7. C. R. Menyuk, "Nonlinear pulse propagation in birefringent 17. J. D. Kafka and T. Baer, "Fiber soliton Raman laser pumped by
optical fibers," IEEE J. Quantum Electron. QE-23, 174-176 a Nd:YAG laser," Opt. Lett. 12, 181-183 (1987).
(1987). 18. A. S. Gouveia-Neto, A. S. L. Gomes, and J. R. Taylor, "Higher-
8. A. Hasegawa and Y. Kodama, "Signal transmission by optical order soliton pulse compression and splitting at 1.32 um in a
solitons in monomode fibers," Proc. IEEE 69,1145-1150 (1981). single-mode optical fiber," IEEE J. Quantum Electron. QE-23,
9. L. F. Mollenauer, J. P. Gordon, and M. N. Islam, "Soliton 1193-1198 (1987).
propagation i long fibers with periodically compensated loss," 19. R. Stolen, AT&T Bell Laboratories, Holmdel, New Jersey 07733
IEEE J. Quantum Electron. QE-22, 157-173 (1986). (personal communication).
10. C. R. Menyuk, "Origin of solitons in the 'real' world," Phys. Rev. 20. R. H. Stolen, J. Botineau, and A.Ashkin, "Intensity discrimina-
A 33, 4367-4374 (1986). tion of optical pulses with birefringent fibers," Opt. Lett. 7,512-
11. C. R. Menyuk, P. K. A. Wai, H. H. Chen, and Y. C. Lee, "Hamil- 514 (1982).
tonian deformations of integrable, nonlinear field equations 21. R. H. Stolen, L. F. Mollenauer, and W. J. Tomlinson, "Observa-
(with applications to optical fibers)," in Published Proceedings tion of pulse restoration at the soliton period in optical fibers,"
of the 4th Conference on Applied Mathematics and Comput- Opt. Lett. 8, 186-188 (1983).
ing, F. Dressel, ed., U.S. Army Research Office Rep. 87-1 (U.S. 22. K. J. Blow,N. J. Doran, and D. Wood, "Polarization instabilities
Government Printing Office, Washington, D.C., 1987), pp. 373- for solitons in birefringent fibers," Opt.' Lett. 12, 202-204
386. (1987).
12. S. V. Manakov, "On the theory of two-dimensional self-focusing 23. J. Botineau and R. H. Stolen, "Effect of polarization on spectral
of electromagnetic waves," Zh. Eksp. Teor. Fiz. 65, 505-516 broadening in optical fibers," J. Opt. Soc. Am. 72, 1592-1596
(1973) [Sov. Phys. JETP 38, 248-253 (1974)]. (1982).
13. P. K. A. Wai, C. R. Menyuk, H. H. Chen, and Y. C. Lee, "Soliton 24. See, e.g., P. M. Morse and H. Feshbach, Methods of Theoretical
propagation at the zero dispersion wavelength of single mode Physics, Part I (McGraw-Hill, New York, 1953), pp. 437-440.
optical fibers," Opt. Lett. 12, 628-630 (1987). 25. See, e.g., T. R. Taha and M. J. Ablowitz, "Analytical and nu-
14. L. F. Mollenauer and . H. Stolen, "The soliton laser," Opt. merical aspects of certain nonlinear evolution equations. II.
Lett. 9, 13-15 (1984). Numerical, nonlinear Schrddinger equation," J. Comp. Phys.
15. M. N. Islam, L. F. Mollenauer, and R. H. Stolen, "Fiber Raman 55, 203-230 (1984).
amplification soliton-laser (FRASL)," in Ultrafast Phenomena 26. L. F. Mollenauer, R. H. Stolen, J. P. Gordon, and W. J. Tomlin-
V, G. R. Fleming and A. E. Siegman, eds. (Springer-Verlag, son, "Extreme picosecond pulse narrowing by means of soliton
Berlin, 1986), pp. 46-50. effect in single-mode fibers," Opt. Lett. 8, 289-291 (1983).

Curtis R. Menyuk
Curtis R. Menyuk was born in Newton,
Massachusetts, on March 26, 1954. He
received the B.S. and M.S. degrees from
the Massachusetts Institute of Technol-
- ogy in 1976 and the Ph.D. degree from
&* the University of California, Los Ange-
les, in 1981, where he was a Fannie and
John Hertz Foundation Fellow. Since
graduating, his work has been concen-
trated primarily in the areas of nonlinear
dynamics and plasma physics. Howev-
er, in the past two to three years his in-
terests have shifted toward nonlinear
optics, and his publications are increasingly in this field. He was
previously at Science Applications International Corporation in
McLean, Virginia, and is now an associate professor in the electrical
engineering department at the University of Maryland, College
Park and Baltimore. He has been charged with coordinating a pro-
gram in photonics and communication at the Baltimore campus. He
is a member of the Society for Industrial and Applied Mathematics,
the American Physical Society, and the Optical Society of America.