1: 

2: \documentstyle[amssymb]{article}

3: 

4: \hoffset -1.5truecm

5: \voffset -2truecm

6: \textheight 24truecm

7: \textwidth 16truecm

8: \pagestyle {plain}

9: \pagenumbering {arabic}

10: \renewcommand{\baselinestretch}{2}

11: \input{tcilatex}

12: 

13: \begin{document}

14: 

15: \title{Equilibrium and nonequilibrium solitons in a lossy split-step system with

16: lumped amplification}

17: \author{Rodislav Driben and Boris A. Malomed \\

18: %EndAName

19: Department of Interdisciplinary Studies,\\

20: Faculty of Engineering, Tel Aviv University, Tel Aviv 69978,\\

21: Israel}

22: \maketitle

23: 

24: \begin{center}

25: {\bf ABSTRACT}

26: \end{center}

27: 

28: We propose a more realistic version of the recently introduced split-step

29: model (SSM), which consists of periodically alternating dispersive and

30: nonlinear segments, by adding uniformly distributed loss and {\it lumped}

31: gain to it. In the case when the loss is exactly balanced by gain, a family

32: of stable equilibrium solitons (ESs) is found. Unless the system's period $L$

33: is very small, saturation is observed in the dependence of the amplitude of

34: the established ES vs. that of the initial pulse. Stable nonequilibrium

35: solitons (NESs) are found in the case when the gain slightly exceeds (by up

36: to $\simeq 3\%$) the value necessary to balance the loss. The existence of

37: NESs is possible as the excessive energy pump is offset by permanent

38: radiation losses, which is confirmed by computation of the corresponding

39: Poynting vector. Unlike ESs that form a continuous family of solutions, NES

40: is an isolated solution, which disappears in the limit $L\rightarrow 0$,

41: i.e., it cannot be found in the overpumped nonlinear Schr\"{o}dinger

42: equation. Interactions between ESs turn out to be essentially the same as in

43: SSM without loss and gain, while interactions between NESs are different:

44: two NESs perturb each other by the radiation jets emanating from them, even

45: if they are separated by a large distance. Moving NESs survive collisions,

46: changing their velocities.

47: 

48: \newpage

49: 

50: \section{Introduction}

51: 

52: It is commonly known that the observation and use of solitons in various

53: systems, such as fiber-optic telecommunication links \cite{HK}, makes it

54: necessary to periodically compensate losses. If the system is long, which is

55: the case for telecommunication links, or if it has the shape of a closed

56: loop which is composed of different elements (examples are fiber lasers

57: generating the so-called stretched pulses \cite{Haus}, and a ferromagnetic

58: film combined with an amplifier, which is able to support periodic trains of

59: propagating magnetic solitons \cite{Slavin}), the loss-compensating gain is

60: usually provided for by short (effectively, point-like) amplifying elements

61: which are periodically placed ({\it lumped}) in the case of the long link.

62: 

63: If the system itself, except for the lumped amplifiers, is uniform, the

64: discreteness of the gain is not an important factor: in that case, it is

65: enough to demand that the gain averaged along the link is in balance with

66: the uniformly distributed loss. In particular, if the undamped uniform

67: nonlinear system supports solitons, it is easy to demonstrate that the

68: balance condition provides for the stable existence of solitons in the

69: presence of loss and amplification \cite{review}. On the other hand, a

70: number of strongly heterogeneous systems, based on periodic alternation of

71: very different elements, have been recently introduced, chiefly in the

72: context of nonlinear optics. These include the well-known dispersion

73: management (i.e., alternation of fiber segments with opposite signs of the

74: group-velocity dispersion \cite{DM}), one- or multidimensional {\it tandem

75: systems} concatenating linear and quadratically nonlinear waveguide sections 

76: \cite{tandem}, media composed of alternating layers with different values of

77: the Kerr coefficient \cite{layered}, {\it alternate waveguides}, in which

78: waveguiding and anti-waveguiding nonlinear segments are juxtaposed \cite

79: {Gisin}, dynamical lattices subject to {\it diffraction management} \cite

80: {Mark}, and some others.

81: 

82: A new system belonging to this class was recently proposed in the form of a 

83: {\it split-step model} (SSM) \cite{SSM}. In SSM, pieces of a dispersive

84: optical fiber with negligible nonlinearity (in fact, these are not

85: necessarily fibers; they may instead be short elements with a very strong

86: group-velocity dispersion, such as a fiber Bragg grating \cite

87: {Braggcompensator} or a photonic crystal \cite{photcrystal}) periodically

88: alternate with nonlinear fiber segments operating close to the

89: zero-dispersion point, in which the dispersion is negligible (these may be

90: replaced by short elements in which strong effective Kerr nonlinearity is

91: created by means of the cascading mechanism in quadratically nonlinear

92: optical crystals \cite{Wise}).

93: 

94: In the limit when lengths of both dispersive and nonlinear segments are much

95: smaller than the soliton's dispersion length, the SSM model is tantamount to

96: the commonly known split-step algorithm simulating the nonlinear

97: Schr\"{o}dinger (NLS) equation \cite{NM}. However, on the contrary to the

98: split-step algorithm, in SSM the separation between dispersive and nonlinear

99: segments is a physical feature, rather than a numerical trick, so that

100: lengths of the segments are not small, being {\em comparable to or larger

101: than} the dispersion length of a pulse (soliton) that may propagate in the

102: system.

103: 

104: In the above-mentioned works, it has been demonstrated, by means of direct

105: numerical simulations, that, in the absence of loss, all those strongly

106: inhomogeneous periodic systems support very robust quasi-soliton pulses,

107: although it may be quite difficult to prove their existence and stability in

108: a rigorous form. The introduction of losses and compensating gain poses a

109: new problem for solitons in these systems, the most fundamental issue being

110: whether the straightforward lumped amplification, with the average gain

111: compensating the loss, may support {\em stable} soliton-like pulses.

112: 

113: The objective of the present work is to look for stable solitons in what

114: seems to be the simplest and, possibly, most fundamental model of the type

115: described above: SSM with uniform losses and periodically placed point-like

116: amplifiers. Very recently, it has been demonstrated that, in a more general

117: case, when the average gain considerably exceeds the loss rate (by several

118: per cent), SSM solitons {\em cannot} be stable \cite{Alex}. It was found

119: that, nevertheless, efficient stabilization of solitons can be achieved by

120: means of synchronous intensity modulators (these time-domain control

121: elements are used in fiber-optic systems \cite{SIM}).

122: 

123: In the present work, we focus on a more special case, that, by itself, may

124: be less relevant to direct applications to telecommunications, but appears

125: to have a fundamental interest for the understanding of pulse dynamics. It

126: is a case when the gain exactly compensates the loss in SSM, or the relative

127: excess gain $\Delta G/G$ is small enough (in fact, it must be $\,\Delta

128: G/G\,\,_{\sim }^{<}\,0.03$), and no control elements (intensity modulators

129: or frequency-domain filters) are added. We find stable solitons of two

130: different types. First, in the case when the average gain exactly

131: compensates the loss ($\Delta G=0$), we find stable {\it equilibrium

132: solitons }(ESs). Usually, stationary solitary pulses in dissipative systems,

133: supported by the balance between loss and gain, are isolated solutions,

134: typical examples being an exact unstable solitary-pulse solution to the

135: cubic complex Ginzburg-Landau (GL) equation \cite{PS} and its stable

136: counterpart (also an exact solution) in a system of linearly coupled cubic

137: and linear GL equations \cite{Javid}. A difference of ESs found in the

138: present model is that they form a {\em continuous family}, which may be

139: parameterized by their amplitude. More accurately, the amplitude $\eta _{

140: {\rm fin}}$ of the soliton in the established state is nearly equal to the

141: initial amplitude $\eta _{{\rm in}}$, provided that $\eta _{{\rm in}}$ is

142: not too large. For larger values of $\eta _{{\rm in}}$, saturation in the

143: dependence $\eta _{{\rm fin}}\left( \eta _{{\rm in}}\right) $ is observed.

144: 

145: If small excess gain is present, being limited to the above-mentioned

146: interval $\Delta G/G\,\,_{\sim }^{<}\,0.03$, stable {\it nonequilibrium

147: solitons} (NESs) are found. Unlike the equilibrium ones, they are indeed

148: isolated solutions with a uniquely defined value of the amplitude. Their

149: existence in the presence of the excess gain is explained by the fact that

150: they permanently emit small-amplitude radiation waves. We check that the

151: size of the corresponding Poynting vector (energy flux) exactly matches the

152: rate of the energy pump into the soliton.

153: 

154: In this paper, we also demonstrate the existence of stable moving ESs and

155: NESs in the same model, and consider collisions between them, which turn out

156: to be inelastic in the latter case, although NESs survive collisions.

157: Interactions between immobile solitons are also considered, with a

158: conclusion that NESs strongly perturb each other, via the radiation jets

159: that they continuously emit.

160: 

161: In fact, ESs are quite simple pulses, which smoothly go over into the usual

162: nonlinear Schr\"{o}dinger (NLS) solitons as the system's period is

163: vanishing. On the other hand, NESs appear to be novel solitary pulses which

164: are supported by the stable balance between gain, dissipative loss, {\em and}

165: the emission of radiation. NESs have no counterparts in the usual uniform

166: NLS equation: we demonstrate that they disappear if the SSM period vanishes

167: (they disappear too if the period becomes too large). They also disappear if

168: the loss parameter is very large, and they only exist within the

169: above-mentioned narrow margin of values of the excess gain.

170: 

171: \section{The model}

172: 

173: The linear lossy dispersive segment of the system is described by the damped

174: linear Schr\"{o}dinger equation, 

175: \begin{equation}

176: iu_{z}+\frac{1}{2}u_{\tau \tau }=-\alpha _{D}u.  \label{dispersion}

177: \end{equation}

178: In an optical fiber, $u$, $z$ and $\tau $ are, respectively, the local

179: amplitude of the electromagnetic waves, propagation distance, and local time

180: ($\tau \equiv t-z/V$, where $t$ and $V$ are the time proper and mean group

181: velocity of the wave packet), the dispersion coefficient is normalized to be 

182: $1$, and $\alpha _{D}>0$ is a loss constant. Obviously, Eq. (1) can be

183: solved by means of the Fourier transform in $\tau $ (which is used as a part

184: of the standard split-step numerical algorithm \cite{NM}).

185: 

186: The nonlinear segment of the system is governed by the dispersionless NLS\

187: equation, 

188: \begin{equation}

189: iu_{z}+|u|^{2}u=-i\alpha _{N}u,  \label{nonlinearity}

190: \end{equation}

191: where the nonlinearity coefficient is normalized to be $1$, and $\alpha

192: _{N}>0$ is the loss constant in the nonlinear segment (we assume that the

193: third-order dispersion may be neglected). A solution to Eq. (\ref

194: {nonlinearity}) is obvious, 

195: \begin{equation}

196: u(z,\tau )=u(0,\tau )\exp \left( -\alpha _{D}\,z\right) \exp \left[ i\frac{

197: |u(0,\tau )|^{2}}{2\alpha _{N}}\left[ 1-\exp \left( -2\alpha _{N}z\right) 

198: \right] \right] .  \label{phase}

199: \end{equation}

200: 

201: We define the system's elementary cell as an interval between midpoints of

202: two neighboring nonlinear segments. Point-like amplifiers are set at

203: junctions between cells. They acts on the wave field so that 

204: \begin{equation}

205: u(\tau )\mapsto u(\tau )\cdot e^{G}\,,  \label{G}

206: \end{equation}

207: where $G$ is the gain (the gain defined in terms of the signal's power and

208: measured in dB is $8.69G$). Thus, the full transformation ({\it map}) for

209: the pulse passing a cell can be represented as a superposition of two

210: nonlinear phase transformations (\ref{phase}) corresponding to the nonlinear

211: half-segments at the cell's edges, and the Fourier transform corresponding

212: to the dispersive segment in the middle of the cell, which are followed by

213: the multiplication as per Eq. (\ref{G}).

214: 

215: In order to prevent radiation from re-entering the integration domain, the

216: linear equation (\ref{dispersion}) was solved directly by means of numerical

217: methods in a broad integration interval $\Delta T$, placing absorbers at its

218: edges. In fact, the absorbers emulate a real physical effect, namely, the

219: fact that the energy emitted with radiation is lost.

220: 

221: If the lengths of the dispersive and nonlinear segments are $L_{D}$ and 

222: $L_{N}$, the cell size is $L=L_{D}+L_{N}$. Note that the left-hand sides of

223: Eqs. (\ref{dispersion}) and (\ref{nonlinearity}) are separately invariant

224: with respect to transformations 

225: \begin{equation}

226: \tau \rightarrow \tau /\Lambda _{D},z\rightarrow z/\Lambda _{D}^{2},\,{\rm

227: and}\,\,u\rightarrow \Lambda _{N}u,z\rightarrow z/\Lambda _{N}^{2}\,,

228: \label{scaling}

229: \end{equation}

230: The rescalings (\ref{scaling}) may be used (with $\Lambda _{D}^{2}/\Lambda

231: _{N}^{2}=L_{D}/L_{N}$) to make $L_{D}=L_{N}\equiv L/2\,$, which we assume

232: below to hold.

233: 

234: With regard to the normalizations adopted, an average ({\it distributed})

235: version of the model amounts to a perturbed NLS equation, 

236: \begin{equation}

237: iu_{z}+\frac{1}{4}u_{\tau \tau }+\frac{1}{2}|u|^{2}u=-i\left[ \frac{1}{2}

238: \left( \alpha _{N}+\alpha _{D}\right) -\frac{G}{L}\right] .  \label{NLS}

239: \end{equation}

240: The balance between the loss and gain is expected in the case when the

241: right-hand side of Eq. (\ref{NLS}) vanishes, i.e., when the gain takes the

242: equilibrium value 

243: \begin{equation}

244: G_{{\rm eq}}=\frac{L}{2}\left( \alpha _{N}+\alpha _{D}\right) .  \label{eq}

245: \end{equation}

246: 

247: \section{Equilibrium solitons}

248: 

249: As it was demonstrated in Ref. \cite{SSM}, SSM without loss and gain gives

250: rise to a family of stable solitary pulses. The first objective of the

251: present work is to generate similar pulses in the model introduced in the

252: previous section, in the case when the loss and gain are balanced as per Eq.

253: (\ref{eq}). To this end, we simulated the evolution of initial pulses that

254: were taken as fundamental solitons of the unperturbed NLS equation 

255: (\ref{NLS}), 

256: \begin{equation}

257: u(z=0,\tau )=\eta \,{\rm sech}\left( \eta \tau \right) ,  \label{initial}

258: \end{equation}

259: with an arbitrary amplitude $\eta $.

260: 

261: In a vast parametric range, the initial pulse (\ref{initial}) readily

262: generates a solitary wave, which remains stable as long as simulations could

263: be run, see a typical example in Fig. 1. As is seen in the figure, the pulse

264: changes its shape against the initial configuration (\ref{initial}).

265: Detailed analysis shows that the established pulse very well fits to the

266: expression (\ref{initial}), but with a different (smaller) value of the

267: amplitude. As well as the fundamental NLS soliton, the established pulse has

268: no {\it chirp} (i.e., no intrinsic phase structure \cite{HK}).

269: 

270: The family of {\it equilibrium solitons}, alias ESs [the ones subject to the

271: equilibrium condition \ref{eq})] found this way is characterized by a

272: dependence of the final (``output'') value $\eta _{{\rm out}}$ of the

273: amplitude on the initial (``input'') amplitude $\eta _{{\rm in}}$, which is

274: presented, for different fixed values of the period $L$, and for different

275: values of the loss constant, in Figs. 2(a) and 2(b). As is seen, for small

276: values of $\eta _{{\rm in}}$ the amplitude $\eta _{{\rm out}} $ is virtually

277: identical to it. However, with the increase of $\eta _{{\rm in}}$ the growth

278: of $\eta _{{\rm out}}$ slows down, and, eventually, $\eta _{{\rm out}}$

279: saturates at a constant value $\eta _{{\rm sat}}$, that strongly depends on 

280: $L$, but very weakly on $\alpha $, see Fig. 2(b). An example of the formation

281: of a stable ES with the amplitude $\eta _{{\rm out}}$ which is much smaller

282: than $\eta _{{\rm in}}$ (in the case of strong saturation) is displayed in

283: Fig. 3.

284: 

285: Figure 2(a) shows that, in the limit of very small $L$, the saturation does

286: not take place, and in this case we simply have $\eta _{{\rm out}}=\eta_{

287: {\rm in}}$\thinspace , which has a simple meaning: for small $L$, the

288: averaging applies, transforming the SSM with the balanced loss and gain into

289: the unperturbed NLS equation [see Eq. (\ref{NLS})]. In the latter equation,

290: the initial pulse (\ref{initial}) generates a fundamental soliton with the

291: amplitude $\eta $.

292: 

293: We stress that different branches of the dependence shown in Fig. 2(a) were

294: not aborted arbitrarily: except for the one corresponding to $L=0.1$, they

295: all terminate at points corresponding to a maximum value of $\eta _{{\rm in}

296: } $ for which the initial pulse (\ref{initial}) can generate a soliton. If 

297: $\eta _{{\rm in}}$ exceeds this maximum value, the initial pulse blows up,

298: generating spatio-temporal ``turbulence''. We did not study the latter

299: regime in detail, focusing on the regular behavior.

300: 

301: \section{Nonequilibrium solitons}

302: 

303: Numerical simulations of the model demonstrate that, if the gain slightly

304: exceeds the equilibrium value (\ref{eq}), stable solitons are still

305: generated in a narrow interval of values of the excess gain: $\left( G-G_{

306: {\rm eq}}\right) /G_{{\rm eq}}\lesssim 0.03$. A typical example of the

307: formation of such a {\it nonequilibrium} soliton (NES) is shown in Fig. 4,

308: which corresponds to $\left( G-G_{{\rm eq}}\right) /G_{{\rm eq}}=0.024$. If

309: the relative excess gain exceeds the maximum value $\simeq 0.03$, the system

310: blows up.

311: 

312: The apparent existence of the stable pulse shown in Fig. 4 suggests a

313: natural question: how is the energy balance provided for in this case, if

314: the gain exceeds the loss? An answer is that the energy pump into NES by the

315: excess gain is compensated by extra loss in the form of small-amplitude

316: radiation waves emitted by the pulse. Indeed, the background on top of which

317: the NES showed in Fig. 4 is found is not just $u\equiv 0$, but a

318: small-amplitude wave field.

319: 

320: To analyze this issue, we define the norm of the solution (it is what is

321: usually called energy in optics), $E=\int_{-\infty }^{+\infty }\left|

322: u(z,\tau )\right| ^{2}d\tau \,$. The energy is an obvious dynamical

323: invariant in the case of the unperturbed NLS equation, as well as in the SSM

324: without loss and gain. In the present model, the evolution of the energy of

325: a localized pulse is governed by a balance equation. After straightforward

326: calculations which use integration by parts, the balance equation can be

327: cast in the form 

328: \begin{equation}

329: \frac{dE_{{\rm sol}}}{dz}=2\left[ G\sum_{n}\delta (z-nL)-\alpha (z)\right]

330: E_{{\rm sol}}-D(z)\cdot {\rm Im}\left( u^{\ast }u_{\tau }\right) |_{\tau

331: =-T}^{\tau =+T}\,.  \label{balance}

332: \end{equation}

333: Here, the soliton's energy is defined as the integral over some finite

334: interval, $-T<\tau <+T$, which is essentially larger than a proper size of

335: the soliton, but smaller than the total size of the domain between the

336: above-mentioned edge absorbers, $E_{{\rm sol}}\equiv \int_{-T}^{+T}\left|

337: u(z,\tau )\right| ^{2}d\tau \,$. The last term in Eq. (\ref{balance}), which

338: is produced by the integration by parts, represents the {\it Poynting vector}

339: (energy flux) of the field $u$ at the points $\tau =\pm T$. Lastly, Eq. (\ref

340: {balance}) assumes that the amplifiers are placed at the points $z=Ln$, the

341: coefficient $\alpha (z)$ takes values $\alpha _{N}$ and $\alpha _{D}$ in the

342: nonlinear and dispersive segments, respectively, while $D(z)\equiv 1$ inside

343: the dispersive segments, and $\equiv 0$ inside the nonlinear ones.

344: 

345: In the case when the coefficients $\alpha $ and $G$ are small, Eq. (\ref

346: {balance}) can be averaged over a large number of cells, reducing to 

347: \begin{equation}

348: \frac{dE_{{\rm sol}}}{dz}=\left[ 2LG-\left( \alpha _{N}+\alpha _{D}\right)

349: \right] E_{{\rm sol}}-\frac{1}{2}\left( J_{+}+J_{-}\right) \,,

350: \label{Poynting}

351: \end{equation}

352: where $J_{\pm }$ are $z$-averaged values of ${\rm Im}\left( u^{\ast }u_{\tau

353: }\right) $ at the points $\tau =\pm T$. Taking, for instance, the NES

354: displayed in Fig. 4, and assuming that its shape is exactly described by Eq.

355: (\ref{initial}), we find $E_{{\rm sol}}=3.844$, hence the first term on the

356: right-hand side of Eq. (\ref{Poynting}), i.e., the net rate of the energy

357: pump into the NES, is $0.0165$. On the other hand, numerical data yield the

358: value $\left( J_{+}+J_{-}\right) /2\approx 0.0155$ of the second term (the

359: net energy flux across the points $\tau =\pm T$). The comparison shows

360: reasonable agreement between the energy pump and emission rates; the

361: remaining difference may be explained by a deviation of the exact shape of

362: the field from the expression (\ref{initial}).

363: 

364: A drastic difference of NES from ES is that NES is an isolated solution,

365: which is found with a single value of its amplitude, while ESs form a

366: continuous family. In other words, NES is a unique {\em attractor} in the

367: system.

368: 

369: It is instructive to summarize data showing the NES amplitude as a function

370: of the period $L$ and dissipative constant $\alpha $, see Fig. 5. An

371: important inference, suggested by Fig. 5(a) and confirmed by many more

372: simulations, is that the amplitude diverges (i.e., the NES blows up) as 

373: $L\rightarrow 0$. This result means that NES cannot exist in the uniform NLS

374: equation with unbalanced gain, which is quite obvious by itself. Thus, we

375: conclude that the existence of NES is a distinctive feature of SSM. Note

376: that NES does not exist (actually, it blows up) past the point where the

377: curve terminates in Fig. 5(b).

378: 

379: \section{Interactions and collisions between solitons}

380: 

381: Interactions between solitons is the next natural issue to consider. To this

382: end, we note that, although Eq. (\ref{nonlinearity}) and, hence, SSM as a

383: whole, are {\em not} Galilean invariant, moving pulses can be readily

384: generated from the zero-velocity ones considered above, ``pushing'' them by

385: means of a multiplier $\exp \left( -i\chi \tau \right) $ with an arbitrary

386: real constant $\chi $, added at the point $z=0$, which gives rise a

387: soliton's velocity proportional to $\chi $ \cite{SSM}. The result is

388: formation of stable moving solitons, both ESs and NESs, see an example of

389: moving NESs below in Fig. 7. Apparently, the amplitude of a moving NES is

390: somewhat smaller than that of its quiescent counterpart, but this issue was

391: not studied in detail.

392: 

393: Simulations demonstrate that both the interaction between quiescent ESs, as

394: well as collision between moving ones, are almost identical to those which

395: were studied in detail in SSM without loss and gain in Ref. \cite{SSM}.

396: Namely, collisions are quasi-elastic, and two zero-velocity ESs do not

397: interact, provided that the separation between them exceeds a certain

398: minimum value.

399: 

400: The situation is different in the case of NESs. First, even if the

401: separation between two zero-velocity NESs is large, they strongly perturb

402: each other -- not via direct interaction, but rather by the radiation

403: ``jets'' emitted by each soliton, see a typical example in Fig. 6. Note that

404: each soliton has a smaller amplitude than in isolation. Second, moving NESs

405: survive collisions, but, as it is seen in the example displayed in Fig. 7,

406: the pulses reappear after the collision with different values of the

407: velocity [in fact, the slope of their trajectories in the $\left( \tau

408: ,z\right) $ plane increases], and with a smaller amplitude. However, the

409: amplitude gradually relaxes to the unique value selected by NES moving at a

410: given velocity.

411: 

412: \section{Conclusion}

413: 

414: We have added loss and gain to a model of a long periodic system consisting

415: of separated dispersive and nonlinear segments. It was shown that, if the

416: loss is exactly balanced by gain, a family of stable equilibrium solitons

417: exists. Unless the system's period is very small, saturation is observed in

418: the dependence of the amplitude of the established soliton vs. that of the

419: initial pulse. Stable nonequilibrium solitons were found in the case when

420: the gain slightly exceeds (by up to $\simeq 3\%$) the value necessary to

421: balance the loss. The existence of NESs is possible as the excessive energy

422: pump is offset by permanent radiation losses, which is confirmed by

423: computation of the corresponding Poynting vector. Unlike ESs that exist as a

424: continuous family of solutions, NES is an isolated one, and it disappears

425: when the system's period vanishes. Interactions between ESs were found to be

426: essentially the same as in the model without loss and gain, while

427: interactions between NESs are different: two NESs perturb each other by

428: their radiation jets, even if they are separated by a large distance. Moving

429: NESs survive collisions, with a change of the velocity.

430: 

431: \section*{Acknowledgements}

432: 

433: We appreciate useful discussions with M.J. Ablowitz and P.L. Chu.

434: 

435: \newpage

436: 

437: \begin{thebibliography}{99}

438: \bibitem{HK}  A. Hasegawa and Y. Kodama. {\it Solitons in optical

439: communications} (Oxford University Press: Oxford, UK, 1995).

440: 

441: \bibitem{Haus}  H.A.~Haus, K.~Tamura, L.E.~Nelson, and E.P.~Ippen, IEEE\ J.

442: Quant. Electr. {\bf 31} (1995) 591.

443: 

444: \bibitem{Slavin}  G.A. Melkov, A.A. Serga, V.S. Tiberkevich, Y.V.

445: Kobljanskij, and A.N. Slavin, Phys. Rev. E {\bf 63} (2001) 066607; G.A.

446: Melkov, Y.V. Kobljanskyj, A.A. Serga AA, V.S. Tiberkevich, and A.N. Slavin,

447: Phys. Stat. Sol. A {\bf 189} (2002) 1007.

448: 

449: \bibitem{review}  Yu.S. Kivshar and B.A. Malomed, Rev. Mod. Phys. {\bf 61}

450: (1989)\thinspace 763.

451: 

452: \bibitem{DM}  J.H.B. Nijhof, N.J. Doran, W. Forysiak and F.M. Knox,

453: Electron.~Lett. {\bf 33} (1997) 1726.

454: 

455: \bibitem{tandem}  L. Torner, IEEE Photon. Techn. Lett. {\bf 11} (1999) 1268;

456: L. Torner, S. Carrasco, J.P. Torres, L.-C. Crasovan, and D. Mihalache, Opt.

457: Commun. {\bf 199} (2001) 217.

458: 

459: \bibitem{layered}  L. Berg\'{e}, V.K. Mezentsev, J. Juul Rasmussen, P.L.

460: Christiansen, and Yu.B. Gaididei, Opt. Lett. {\bf 25} (2000) 1037; I. Towers

461: and B.A. Malomed, J. Opt. Soc. Am. B {\bf 19} (2002) 537.

462: 

463: \bibitem{Gisin}  A. Kaplan, B.V. Gisin, and B.A. Malomed, J. Opt. Soc. Am. B 

464: {\bf 19} (2002) 522.

465: 

466: \bibitem{Mark}  M.J. Ablowitz and Z.H. Musslimani, Phys. Rev. Lett. {\bf 87}

467: (2001) 254102.

468: 

469: \bibitem{SSM}  R. Driben and B.A. Malomed, Opt. Commun. {\bf 185} (2000) 439

470: (2000); {\it ibid}. {\bf 197} (2001) 481.

471: 

472: \bibitem{Braggcompensator}  G. Lenz, B.J. Eggleton, C.K. Madsen, and R.E.

473: Slusher, IEEE\ J. Quant. Electr. {\bf QE37} (2001) 525.

474: 

475: \bibitem{photcrystal}  M. Notomi, K. Yamada, A. Shinya, J. Takahashi, C.

476: Takahashi, and I. Yokohama, Phys. Rev. Lett. {\bf 87} (2001) 253902.

477: 

478: \bibitem{Wise}  L.J. Qian, X. Liu, and F.W. Wise, Opt. Lett. {\bf 24} (1999)

479: 166; X. Liu, L. Qian, and F. Wise, Opt. Lett. {\bf 24} (1999) 1777.

480: 

481: \bibitem{NM}  A.C. Newell and J.V. Moloney. {\it Nonlinear Optics}

482: (Addison-Wesley\TEXTsymbol{\vert}: 1992).

483: 

484: \bibitem{Alex}  R. Driben, A. Orenstein, and B.A. Malomed, Opt. Commun. {\bf

485: 205} (2002) 139.

486: 

487: \bibitem{SIM}  K. Suzuki and M. Nakazawa, Opt. Fiber Techn. {\bf 1} (1995)

488: 289 (1995); T.Yamamoto and M. Nakazawa, Opt. Lett. {\bf 26} (2001) 647

489: (2001); E. Pincemin, O. Audouin, B. Dany, and S. Wabnitz, IEEE J. Light.

490: Techn. {\bf 19} (2001) 624.

491: 

492: \bibitem{PS}  N.R. Pereira and L. Stenflo, Phys. Fluids {\bf 20} (1977) 1733.

493: 

494: \bibitem{Javid}  J. Atai and B.A. Malomed, Phys. Lett. A {\bf 246} (1998)

495: 412.

496: \end{thebibliography}

497: 

498: \newpage

499: 

500: \section{Figure captions}

501: 

502: Fig. 1. An example of the establishment of a stable pulse of the equilibrium

503: type in SSM with the balanced loss and gain [$\alpha _{N}=\alpha _{D}=0.035$,

504: $L=1$, and, in accord with Eq. (\ref{eq}), $G=0.035$]. The initial pulse

505: is taken as the NLS soliton (\ref{initial}) with $\eta =1$. In this and

506: subsequent figures, $|u|^{2}$ (``power'') is shown vs. $\tau $ at the end of

507: each cell.

508: 

509: Fig. 2. The established (``output'') amplitude of the equilibrium soliton

510: vs. the initial (``input'') amplitude. In (a), $\alpha _{N}=\alpha _{D}$ is

511: fixed to be $0.035$; in (b), $L$ is fixed to be $1$, and $\alpha _{N}=\alpha

512: _{D}\equiv \alpha $.

513: 

514: Fig. 3. Formation of a stable equilibrium soliton in the case when Fig. 2

515: shows strong saturation. The initial amplitude in Eq. (\ref{initial}) is 

516: $\eta =4$, $L=1$, $\alpha _{N}=\alpha _{D}=0.035$.

517: 

518: Fig. 4. An example of the formation of a nonequilibrium soliton. The

519: amplitude of the input pulse and the SSM's parameters are the same as in the

520: case shown in Fig. 1, except for $G=0.0\,\allowbreak 358\,5$, which

521: corresponds to the relative excess gain $\left( G-G_{{\rm eq}}\right) /G_{

522: {\rm eq}}\approx 0.024$.

523: 

524: Fig. 5. The amplitude of the stable nonequilibrium soliton vs. the system's

525: period (``stepsize'') $L$ (a) and the loss parameter $\alpha _{N}=\alpha

526: _{D}\equiv \alpha $ (b). In both panels, the relative excess gain is fixed

527: to be $0.024$; in (a), $\alpha =0.035$, and in (b), $L=1$.

528: 

529: Fig. 6. Interaction of two nonequilibrium solitons (the same ones as in Fig.

530: 4). The initial separation between them is $\Delta \tau =18$, the

531: full-width-at-half-maximum \cite{HK} of each one being initially $T_{{\rm

532: FWHM}}=2$. (a) The distribution of $\left| u(z,\tau )\right| ^{2}$; (b) the

533: same shown by means of level contours.

534: 

535: Fig. 7. An example of the collision between two nonequilibrium solitons.

536: Each moving soliton is generated from the quiescent one shown in Fig. 4,

537: multiplying it by $\exp \left( \pm i\tau /2\right) $. The side view of the

538: distribution of $\left| u(z,\tau )\right| ^{2}$ is displayed [in this case,

539: a full three-dimensional picture like those in Figs. 1, 3, 4, and 6(a) would

540: seem messy], the inset showing trajectories of centers of the two solitons

541: on the plane $\left( \tau ,z\right) $.

542: 

543: \end{document}

544: